ryujinx/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs

2324 lines
100 KiB
C#
Raw Normal View History

Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Numerics;
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
using static Spv.Specification;
namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
using SpvInstruction = Spv.Generator.Instruction;
using SpvLiteralInteger = Spv.Generator.LiteralInteger;
static class Instructions
{
private const MemorySemanticsMask DefaultMemorySemantics =
MemorySemanticsMask.ImageMemory |
MemorySemanticsMask.AtomicCounterMemory |
MemorySemanticsMask.WorkgroupMemory |
MemorySemanticsMask.UniformMemory |
MemorySemanticsMask.AcquireRelease;
private static readonly Func<CodeGenContext, AstOperation, OperationResult>[] InstTable;
static Instructions()
{
InstTable = new Func<CodeGenContext, AstOperation, OperationResult>[(int)Instruction.Count];
Add(Instruction.Absolute, GenerateAbsolute);
Add(Instruction.Add, GenerateAdd);
Add(Instruction.AtomicAdd, GenerateAtomicAdd);
Add(Instruction.AtomicAnd, GenerateAtomicAnd);
Add(Instruction.AtomicCompareAndSwap, GenerateAtomicCompareAndSwap);
Add(Instruction.AtomicMinS32, GenerateAtomicMinS32);
Add(Instruction.AtomicMinU32, GenerateAtomicMinU32);
Add(Instruction.AtomicMaxS32, GenerateAtomicMaxS32);
Add(Instruction.AtomicMaxU32, GenerateAtomicMaxU32);
Add(Instruction.AtomicOr, GenerateAtomicOr);
Add(Instruction.AtomicSwap, GenerateAtomicSwap);
Add(Instruction.AtomicXor, GenerateAtomicXor);
Add(Instruction.Ballot, GenerateBallot);
Add(Instruction.Barrier, GenerateBarrier);
Add(Instruction.BitCount, GenerateBitCount);
Add(Instruction.BitfieldExtractS32, GenerateBitfieldExtractS32);
Add(Instruction.BitfieldExtractU32, GenerateBitfieldExtractU32);
Add(Instruction.BitfieldInsert, GenerateBitfieldInsert);
Add(Instruction.BitfieldReverse, GenerateBitfieldReverse);
Add(Instruction.BitwiseAnd, GenerateBitwiseAnd);
Add(Instruction.BitwiseExclusiveOr, GenerateBitwiseExclusiveOr);
Add(Instruction.BitwiseNot, GenerateBitwiseNot);
Add(Instruction.BitwiseOr, GenerateBitwiseOr);
Add(Instruction.Call, GenerateCall);
Add(Instruction.Ceiling, GenerateCeiling);
Add(Instruction.Clamp, GenerateClamp);
Add(Instruction.ClampU32, GenerateClampU32);
Add(Instruction.Comment, GenerateComment);
Add(Instruction.CompareEqual, GenerateCompareEqual);
Add(Instruction.CompareGreater, GenerateCompareGreater);
Add(Instruction.CompareGreaterOrEqual, GenerateCompareGreaterOrEqual);
Add(Instruction.CompareGreaterOrEqualU32, GenerateCompareGreaterOrEqualU32);
Add(Instruction.CompareGreaterU32, GenerateCompareGreaterU32);
Add(Instruction.CompareLess, GenerateCompareLess);
Add(Instruction.CompareLessOrEqual, GenerateCompareLessOrEqual);
Add(Instruction.CompareLessOrEqualU32, GenerateCompareLessOrEqualU32);
Add(Instruction.CompareLessU32, GenerateCompareLessU32);
Add(Instruction.CompareNotEqual, GenerateCompareNotEqual);
Add(Instruction.ConditionalSelect, GenerateConditionalSelect);
Add(Instruction.ConvertFP32ToFP64, GenerateConvertFP32ToFP64);
Add(Instruction.ConvertFP32ToS32, GenerateConvertFP32ToS32);
Add(Instruction.ConvertFP32ToU32, GenerateConvertFP32ToU32);
Add(Instruction.ConvertFP64ToFP32, GenerateConvertFP64ToFP32);
Add(Instruction.ConvertFP64ToS32, GenerateConvertFP64ToS32);
Add(Instruction.ConvertFP64ToU32, GenerateConvertFP64ToU32);
Add(Instruction.ConvertS32ToFP32, GenerateConvertS32ToFP32);
Add(Instruction.ConvertS32ToFP64, GenerateConvertS32ToFP64);
Add(Instruction.ConvertU32ToFP32, GenerateConvertU32ToFP32);
Add(Instruction.ConvertU32ToFP64, GenerateConvertU32ToFP64);
Add(Instruction.Cosine, GenerateCosine);
Add(Instruction.Ddx, GenerateDdx);
Add(Instruction.Ddy, GenerateDdy);
Add(Instruction.Discard, GenerateDiscard);
Add(Instruction.Divide, GenerateDivide);
Add(Instruction.EmitVertex, GenerateEmitVertex);
Add(Instruction.EndPrimitive, GenerateEndPrimitive);
Add(Instruction.ExponentB2, GenerateExponentB2);
Add(Instruction.FSIBegin, GenerateFSIBegin);
Add(Instruction.FSIEnd, GenerateFSIEnd);
Add(Instruction.FindLSB, GenerateFindLSB);
Add(Instruction.FindMSBS32, GenerateFindMSBS32);
Add(Instruction.FindMSBU32, GenerateFindMSBU32);
Add(Instruction.Floor, GenerateFloor);
Add(Instruction.FusedMultiplyAdd, GenerateFusedMultiplyAdd);
Add(Instruction.GroupMemoryBarrier, GenerateGroupMemoryBarrier);
Add(Instruction.ImageAtomic, GenerateImageAtomic);
Add(Instruction.ImageLoad, GenerateImageLoad);
Add(Instruction.ImageStore, GenerateImageStore);
Add(Instruction.IsNan, GenerateIsNan);
Add(Instruction.LoadAttribute, GenerateLoadAttribute);
Add(Instruction.LoadConstant, GenerateLoadConstant);
Add(Instruction.LoadLocal, GenerateLoadLocal);
Add(Instruction.LoadShared, GenerateLoadShared);
Add(Instruction.LoadStorage, GenerateLoadStorage);
Add(Instruction.Lod, GenerateLod);
Add(Instruction.LogarithmB2, GenerateLogarithmB2);
Add(Instruction.LogicalAnd, GenerateLogicalAnd);
Add(Instruction.LogicalExclusiveOr, GenerateLogicalExclusiveOr);
Add(Instruction.LogicalNot, GenerateLogicalNot);
Add(Instruction.LogicalOr, GenerateLogicalOr);
Add(Instruction.LoopBreak, GenerateLoopBreak);
Add(Instruction.LoopContinue, GenerateLoopContinue);
Add(Instruction.Maximum, GenerateMaximum);
Add(Instruction.MaximumU32, GenerateMaximumU32);
Add(Instruction.MemoryBarrier, GenerateMemoryBarrier);
Add(Instruction.Minimum, GenerateMinimum);
Add(Instruction.MinimumU32, GenerateMinimumU32);
Add(Instruction.Multiply, GenerateMultiply);
Add(Instruction.MultiplyHighS32, GenerateMultiplyHighS32);
Add(Instruction.MultiplyHighU32, GenerateMultiplyHighU32);
Add(Instruction.Negate, GenerateNegate);
Add(Instruction.PackDouble2x32, GeneratePackDouble2x32);
Add(Instruction.PackHalf2x16, GeneratePackHalf2x16);
Add(Instruction.ReciprocalSquareRoot, GenerateReciprocalSquareRoot);
Add(Instruction.Return, GenerateReturn);
Add(Instruction.Round, GenerateRound);
Add(Instruction.ShiftLeft, GenerateShiftLeft);
Add(Instruction.ShiftRightS32, GenerateShiftRightS32);
Add(Instruction.ShiftRightU32, GenerateShiftRightU32);
Add(Instruction.Shuffle, GenerateShuffle);
Add(Instruction.ShuffleDown, GenerateShuffleDown);
Add(Instruction.ShuffleUp, GenerateShuffleUp);
Add(Instruction.ShuffleXor, GenerateShuffleXor);
Add(Instruction.Sine, GenerateSine);
Add(Instruction.SquareRoot, GenerateSquareRoot);
Add(Instruction.StoreAttribute, GenerateStoreAttribute);
Add(Instruction.StoreLocal, GenerateStoreLocal);
Add(Instruction.StoreShared, GenerateStoreShared);
Add(Instruction.StoreShared16, GenerateStoreShared16);
Add(Instruction.StoreShared8, GenerateStoreShared8);
Add(Instruction.StoreStorage, GenerateStoreStorage);
Add(Instruction.StoreStorage16, GenerateStoreStorage16);
Add(Instruction.StoreStorage8, GenerateStoreStorage8);
Add(Instruction.Subtract, GenerateSubtract);
Add(Instruction.SwizzleAdd, GenerateSwizzleAdd);
Add(Instruction.TextureSample, GenerateTextureSample);
Add(Instruction.TextureSize, GenerateTextureSize);
Add(Instruction.Truncate, GenerateTruncate);
Add(Instruction.UnpackDouble2x32, GenerateUnpackDouble2x32);
Add(Instruction.UnpackHalf2x16, GenerateUnpackHalf2x16);
Add(Instruction.VectorExtract, GenerateVectorExtract);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
Add(Instruction.VoteAll, GenerateVoteAll);
Add(Instruction.VoteAllEqual, GenerateVoteAllEqual);
Add(Instruction.VoteAny, GenerateVoteAny);
}
private static void Add(Instruction inst, Func<CodeGenContext, AstOperation, OperationResult> handler)
{
InstTable[(int)(inst & Instruction.Mask)] = handler;
}
public static OperationResult Generate(CodeGenContext context, AstOperation operation)
{
var handler = InstTable[(int)(operation.Inst & Instruction.Mask)];
if (handler != null)
{
return handler(context, operation);
}
else
{
throw new NotImplementedException(operation.Inst.ToString());
}
}
private static OperationResult GenerateAbsolute(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslFAbs, context.Delegates.GlslSAbs);
}
private static OperationResult GenerateAdd(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.FAdd, context.Delegates.IAdd);
}
private static OperationResult GenerateAtomicAdd(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicIAdd);
}
private static OperationResult GenerateAtomicAnd(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicAnd);
}
private static OperationResult GenerateAtomicCompareAndSwap(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryCas(context, operation);
}
private static OperationResult GenerateAtomicMinS32(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicSMin);
}
private static OperationResult GenerateAtomicMinU32(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicUMin);
}
private static OperationResult GenerateAtomicMaxS32(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicSMax);
}
private static OperationResult GenerateAtomicMaxU32(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicUMax);
}
private static OperationResult GenerateAtomicOr(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicOr);
}
private static OperationResult GenerateAtomicSwap(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicExchange);
}
private static OperationResult GenerateAtomicXor(CodeGenContext context, AstOperation operation)
{
return GenerateAtomicMemoryBinary(context, operation, context.Delegates.AtomicXor);
}
private static OperationResult GenerateBallot(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
var uvec4Type = context.TypeVector(context.TypeU32(), 4);
var execution = context.Constant(context.TypeU32(), Scope.Subgroup);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var maskVector = context.GroupNonUniformBallot(uvec4Type, execution, context.Get(AggregateType.Bool, source));
var mask = context.CompositeExtract(context.TypeU32(), maskVector, (SpvLiteralInteger)0);
return new OperationResult(AggregateType.U32, mask);
}
private static OperationResult GenerateBarrier(CodeGenContext context, AstOperation operation)
{
context.ControlBarrier(
context.Constant(context.TypeU32(), Scope.Workgroup),
context.Constant(context.TypeU32(), Scope.Workgroup),
context.Constant(context.TypeU32(), MemorySemanticsMask.WorkgroupMemory | MemorySemanticsMask.AcquireRelease));
return OperationResult.Invalid;
}
private static OperationResult GenerateBitCount(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryS32(context, operation, context.Delegates.BitCount);
}
private static OperationResult GenerateBitfieldExtractS32(CodeGenContext context, AstOperation operation)
{
return GenerateTernaryS32(context, operation, context.Delegates.BitFieldSExtract);
}
private static OperationResult GenerateBitfieldExtractU32(CodeGenContext context, AstOperation operation)
{
return GenerateTernaryS32(context, operation, context.Delegates.BitFieldUExtract);
}
private static OperationResult GenerateBitfieldInsert(CodeGenContext context, AstOperation operation)
{
return GenerateQuaternaryS32(context, operation, context.Delegates.BitFieldInsert);
}
private static OperationResult GenerateBitfieldReverse(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryS32(context, operation, context.Delegates.BitReverse);
}
private static OperationResult GenerateBitwiseAnd(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.BitwiseAnd);
}
private static OperationResult GenerateBitwiseExclusiveOr(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.BitwiseXor);
}
private static OperationResult GenerateBitwiseNot(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryS32(context, operation, context.Delegates.Not);
}
private static OperationResult GenerateBitwiseOr(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.BitwiseOr);
}
private static OperationResult GenerateCall(CodeGenContext context, AstOperation operation)
{
AstOperand funcId = (AstOperand)operation.GetSource(0);
Debug.Assert(funcId.Type == OperandType.Constant);
(var function, var spvFunc) = context.GetFunction(funcId.Value);
var args = new SpvInstruction[operation.SourcesCount - 1];
var spvLocals = context.GetLocalForArgsPointers(funcId.Value);
for (int i = 0; i < args.Length; i++)
{
var operand = (AstOperand)operation.GetSource(i + 1);
if (i >= function.InArguments.Length)
{
args[i] = context.GetLocalPointer(operand);
}
else
{
var type = function.GetArgumentType(i);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var value = context.Get(type, operand);
var spvLocal = spvLocals[i];
context.Store(spvLocal, value);
args[i] = spvLocal;
}
}
var retType = function.ReturnType;
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var result = context.FunctionCall(context.GetType(retType), spvFunc, args);
return new OperationResult(retType, result);
}
private static OperationResult GenerateCeiling(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslCeil, null);
}
private static OperationResult GenerateClamp(CodeGenContext context, AstOperation operation)
{
return GenerateTernary(context, operation, context.Delegates.GlslFClamp, context.Delegates.GlslSClamp);
}
private static OperationResult GenerateClampU32(CodeGenContext context, AstOperation operation)
{
return GenerateTernaryU32(context, operation, context.Delegates.GlslUClamp);
}
private static OperationResult GenerateComment(CodeGenContext context, AstOperation operation)
{
return OperationResult.Invalid;
}
private static OperationResult GenerateCompareEqual(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdEqual, context.Delegates.IEqual);
}
private static OperationResult GenerateCompareGreater(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdGreaterThan, context.Delegates.SGreaterThan);
}
private static OperationResult GenerateCompareGreaterOrEqual(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdGreaterThanEqual, context.Delegates.SGreaterThanEqual);
}
private static OperationResult GenerateCompareGreaterOrEqualU32(CodeGenContext context, AstOperation operation)
{
return GenerateCompareU32(context, operation, context.Delegates.UGreaterThanEqual);
}
private static OperationResult GenerateCompareGreaterU32(CodeGenContext context, AstOperation operation)
{
return GenerateCompareU32(context, operation, context.Delegates.UGreaterThan);
}
private static OperationResult GenerateCompareLess(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdLessThan, context.Delegates.SLessThan);
}
private static OperationResult GenerateCompareLessOrEqual(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdLessThanEqual, context.Delegates.SLessThanEqual);
}
private static OperationResult GenerateCompareLessOrEqualU32(CodeGenContext context, AstOperation operation)
{
return GenerateCompareU32(context, operation, context.Delegates.ULessThanEqual);
}
private static OperationResult GenerateCompareLessU32(CodeGenContext context, AstOperation operation)
{
return GenerateCompareU32(context, operation, context.Delegates.ULessThan);
}
private static OperationResult GenerateCompareNotEqual(CodeGenContext context, AstOperation operation)
{
return GenerateCompare(context, operation, context.Delegates.FOrdNotEqual, context.Delegates.INotEqual);
}
private static OperationResult GenerateConditionalSelect(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
var cond = context.Get(AggregateType.Bool, src1);
if (operation.Inst.HasFlag(Instruction.FP64))
{
return new OperationResult(AggregateType.FP64, context.Select(context.TypeFP64(), cond, context.GetFP64(src2), context.GetFP64(src3)));
}
else if (operation.Inst.HasFlag(Instruction.FP32))
{
return new OperationResult(AggregateType.FP32, context.Select(context.TypeFP32(), cond, context.GetFP32(src2), context.GetFP32(src3)));
}
else
{
return new OperationResult(AggregateType.S32, context.Select(context.TypeS32(), cond, context.GetS32(src2), context.GetS32(src3)));
}
}
private static OperationResult GenerateConvertFP32ToFP64(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP64, context.FConvert(context.TypeFP64(), context.GetFP32(source)));
}
private static OperationResult GenerateConvertFP32ToS32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.S32, context.ConvertFToS(context.TypeS32(), context.GetFP32(source)));
}
private static OperationResult GenerateConvertFP32ToU32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.U32, context.ConvertFToU(context.TypeU32(), context.GetFP32(source)));
}
private static OperationResult GenerateConvertFP64ToFP32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP32, context.FConvert(context.TypeFP32(), context.GetFP64(source)));
}
private static OperationResult GenerateConvertFP64ToS32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.S32, context.ConvertFToS(context.TypeS32(), context.GetFP64(source)));
}
private static OperationResult GenerateConvertFP64ToU32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.U32, context.ConvertFToU(context.TypeU32(), context.GetFP64(source)));
}
private static OperationResult GenerateConvertS32ToFP32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP32, context.ConvertSToF(context.TypeFP32(), context.GetS32(source)));
}
private static OperationResult GenerateConvertS32ToFP64(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP64, context.ConvertSToF(context.TypeFP64(), context.GetS32(source)));
}
private static OperationResult GenerateConvertU32ToFP32(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP32, context.ConvertUToF(context.TypeFP32(), context.GetU32(source)));
}
private static OperationResult GenerateConvertU32ToFP64(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP64, context.ConvertUToF(context.TypeFP64(), context.GetU32(source)));
}
private static OperationResult GenerateCosine(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslCos, null);
}
private static OperationResult GenerateDdx(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryFP32(context, operation, context.Delegates.DPdx);
}
private static OperationResult GenerateDdy(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryFP32(context, operation, context.Delegates.DPdy);
}
private static OperationResult GenerateDiscard(CodeGenContext context, AstOperation operation)
{
context.Kill();
return OperationResult.Invalid;
}
private static OperationResult GenerateDivide(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.FDiv, context.Delegates.SDiv);
}
private static OperationResult GenerateEmitVertex(CodeGenContext context, AstOperation operation)
{
context.EmitVertex();
return OperationResult.Invalid;
}
private static OperationResult GenerateEndPrimitive(CodeGenContext context, AstOperation operation)
{
context.EndPrimitive();
return OperationResult.Invalid;
}
private static OperationResult GenerateExponentB2(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslExp2, null);
}
private static OperationResult GenerateFSIBegin(CodeGenContext context, AstOperation operation)
{
if (context.Config.GpuAccessor.QueryHostSupportsFragmentShaderInterlock())
{
context.BeginInvocationInterlockEXT();
}
return OperationResult.Invalid;
}
private static OperationResult GenerateFSIEnd(CodeGenContext context, AstOperation operation)
{
if (context.Config.GpuAccessor.QueryHostSupportsFragmentShaderInterlock())
{
context.EndInvocationInterlockEXT();
}
return OperationResult.Invalid;
}
private static OperationResult GenerateFindLSB(CodeGenContext context, AstOperation operation)
{
var source = context.GetU32(operation.GetSource(0));
return new OperationResult(AggregateType.U32, context.GlslFindILsb(context.TypeU32(), source));
}
private static OperationResult GenerateFindMSBS32(CodeGenContext context, AstOperation operation)
{
var source = context.GetS32(operation.GetSource(0));
return new OperationResult(AggregateType.U32, context.GlslFindSMsb(context.TypeU32(), source));
}
private static OperationResult GenerateFindMSBU32(CodeGenContext context, AstOperation operation)
{
var source = context.GetU32(operation.GetSource(0));
return new OperationResult(AggregateType.U32, context.GlslFindUMsb(context.TypeU32(), source));
}
private static OperationResult GenerateFloor(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslFloor, null);
}
private static OperationResult GenerateFusedMultiplyAdd(CodeGenContext context, AstOperation operation)
{
return GenerateTernary(context, operation, context.Delegates.GlslFma, null);
}
private static OperationResult GenerateGroupMemoryBarrier(CodeGenContext context, AstOperation operation)
{
context.MemoryBarrier(context.Constant(context.TypeU32(), Scope.Workgroup), context.Constant(context.TypeU32(), DefaultMemorySemantics));
return OperationResult.Invalid;
}
private static OperationResult GenerateImageAtomic(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
var componentType = texOp.Format.GetComponentType();
// TODO: Bindless texture support. For now we just return 0/do nothing.
if (isBindless)
{
return new OperationResult(componentType, componentType switch
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
{
AggregateType.S32 => context.Constant(context.TypeS32(), 0),
AggregateType.U32 => context.Constant(context.TypeU32(), 0u),
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
_ => context.Constant(context.TypeFP32(), 0f),
});
}
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
int srcIndex = isBindless ? 1 : 0;
SpvInstruction Src(AggregateType type)
{
return context.Get(type, texOp.GetSource(srcIndex++));
}
SpvInstruction index = null;
if (isIndexed)
{
index = Src(AggregateType.S32);
}
int coordsCount = texOp.Type.GetDimensions();
int pCount = coordsCount + (isArray ? 1 : 0);
SpvInstruction pCoords;
if (pCount > 1)
{
SpvInstruction[] elems = new SpvInstruction[pCount];
for (int i = 0; i < pCount; i++)
{
elems[i] = Src(AggregateType.S32);
}
var vectorType = context.TypeVector(context.TypeS32(), pCount);
pCoords = context.CompositeConstruct(vectorType, elems);
}
else
{
pCoords = Src(AggregateType.S32);
}
SpvInstruction value = Src(componentType);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
(var imageType, var imageVariable) = context.Images[new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format)];
var image = context.Load(imageType, imageVariable);
SpvInstruction resultType = context.GetType(componentType);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
SpvInstruction imagePointerType = context.TypePointer(StorageClass.Image, resultType);
var pointer = context.ImageTexelPointer(imagePointerType, imageVariable, pCoords, context.Constant(context.TypeU32(), 0));
var one = context.Constant(context.TypeU32(), 1);
var zero = context.Constant(context.TypeU32(), 0);
var result = (texOp.Flags & TextureFlags.AtomicMask) switch
{
TextureFlags.Add => context.AtomicIAdd(resultType, pointer, one, zero, value),
TextureFlags.Minimum => componentType == AggregateType.S32
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
? context.AtomicSMin(resultType, pointer, one, zero, value)
: context.AtomicUMin(resultType, pointer, one, zero, value),
TextureFlags.Maximum => componentType == AggregateType.S32
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
? context.AtomicSMax(resultType, pointer, one, zero, value)
: context.AtomicUMax(resultType, pointer, one, zero, value),
TextureFlags.Increment => context.AtomicIIncrement(resultType, pointer, one, zero),
TextureFlags.Decrement => context.AtomicIDecrement(resultType, pointer, one, zero),
TextureFlags.BitwiseAnd => context.AtomicAnd(resultType, pointer, one, zero, value),
TextureFlags.BitwiseOr => context.AtomicOr(resultType, pointer, one, zero, value),
TextureFlags.BitwiseXor => context.AtomicXor(resultType, pointer, one, zero, value),
TextureFlags.Swap => context.AtomicExchange(resultType, pointer, one, zero, value),
TextureFlags.CAS => context.AtomicCompareExchange(resultType, pointer, one, zero, zero, Src(componentType), value),
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
_ => context.AtomicIAdd(resultType, pointer, one, zero, value),
};
return new OperationResult(componentType, result);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
private static OperationResult GenerateImageLoad(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
var componentType = texOp.Format.GetComponentType();
// TODO: Bindless texture support. For now we just return 0/do nothing.
if (isBindless)
{
return GetZeroOperationResult(context, texOp, componentType, isVector: true);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
int srcIndex = isBindless ? 1 : 0;
SpvInstruction Src(AggregateType type)
{
return context.Get(type, texOp.GetSource(srcIndex++));
}
SpvInstruction index = null;
if (isIndexed)
{
index = Src(AggregateType.S32);
}
int coordsCount = texOp.Type.GetDimensions();
int pCount = coordsCount + (isArray ? 1 : 0);
SpvInstruction pCoords;
if (pCount > 1)
{
SpvInstruction[] elems = new SpvInstruction[pCount];
for (int i = 0; i < pCount; i++)
{
elems[i] = Src(AggregateType.S32);
}
var vectorType = context.TypeVector(context.TypeS32(), pCount);
pCoords = context.CompositeConstruct(vectorType, elems);
}
else
{
pCoords = Src(AggregateType.S32);
}
pCoords = ScalingHelpers.ApplyScaling(context, texOp, pCoords, intCoords: true, isBindless, isIndexed, isArray, pCount);
(var imageType, var imageVariable) = context.Images[new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format)];
var image = context.Load(imageType, imageVariable);
var imageComponentType = context.GetType(componentType);
var swizzledResultType = texOp.GetVectorType(componentType);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var texel = context.ImageRead(context.TypeVector(imageComponentType, 4), image, pCoords, ImageOperandsMask.MaskNone);
var result = GetSwizzledResult(context, texel, swizzledResultType, texOp.Index);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(componentType, result);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
private static OperationResult GenerateImageStore(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
// TODO: Bindless texture support. For now we just return 0/do nothing.
if (isBindless)
{
return OperationResult.Invalid;
}
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
int srcIndex = isBindless ? 1 : 0;
SpvInstruction Src(AggregateType type)
{
return context.Get(type, texOp.GetSource(srcIndex++));
}
SpvInstruction index = null;
if (isIndexed)
{
index = Src(AggregateType.S32);
}
int coordsCount = texOp.Type.GetDimensions();
int pCount = coordsCount + (isArray ? 1 : 0);
SpvInstruction pCoords;
if (pCount > 1)
{
SpvInstruction[] elems = new SpvInstruction[pCount];
for (int i = 0; i < pCount; i++)
{
elems[i] = Src(AggregateType.S32);
}
var vectorType = context.TypeVector(context.TypeS32(), pCount);
pCoords = context.CompositeConstruct(vectorType, elems);
}
else
{
pCoords = Src(AggregateType.S32);
}
var componentType = texOp.Format.GetComponentType();
const int ComponentsCount = 4;
SpvInstruction[] cElems = new SpvInstruction[ComponentsCount];
for (int i = 0; i < ComponentsCount; i++)
{
if (srcIndex < texOp.SourcesCount)
{
cElems[i] = Src(componentType);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
else
{
cElems[i] = componentType switch
{
AggregateType.S32 => context.Constant(context.TypeS32(), 0),
AggregateType.U32 => context.Constant(context.TypeU32(), 0u),
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
_ => context.Constant(context.TypeFP32(), 0f),
};
}
}
var texel = context.CompositeConstruct(context.TypeVector(context.GetType(componentType), ComponentsCount), cElems);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
(var imageType, var imageVariable) = context.Images[new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format)];
var image = context.Load(imageType, imageVariable);
context.ImageWrite(image, pCoords, texel, ImageOperandsMask.MaskNone);
return OperationResult.Invalid;
}
private static OperationResult GenerateIsNan(CodeGenContext context, AstOperation operation)
{
var source = operation.GetSource(0);
SpvInstruction result;
if (operation.Inst.HasFlag(Instruction.FP64))
{
result = context.IsNan(context.TypeBool(), context.GetFP64(source));
}
else
{
result = context.IsNan(context.TypeBool(), context.GetFP32(source));
}
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateLoadAttribute(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
if (!(src1 is AstOperand baseAttr) || baseAttr.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {nameof(Instruction.LoadAttribute)} must be a constant operand.");
}
var index = context.Get(AggregateType.S32, src3);
var resultType = AggregateType.FP32;
if (src2 is AstOperand operand && operand.Type == OperandType.Constant)
{
int attrOffset = (baseAttr.Value & AttributeConsts.Mask) + (operand.Value << 2);
bool isOutAttr = (baseAttr.Value & AttributeConsts.LoadOutputMask) != 0;
return new OperationResult(resultType, context.GetAttribute(resultType, attrOffset, isOutAttr, index));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
else
{
var attr = context.Get(AggregateType.S32, src2);
return new OperationResult(resultType, context.GetAttribute(resultType, attr, isOutAttr: false, index));
}
}
private static OperationResult GenerateLoadConstant(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = context.Get(AggregateType.S32, operation.GetSource(1));
var i1 = context.Constant(context.TypeS32(), 0);
var i2 = context.ShiftRightArithmetic(context.TypeS32(), src2, context.Constant(context.TypeS32(), 2));
var i3 = context.BitwiseAnd(context.TypeS32(), src2, context.Constant(context.TypeS32(), 3));
SpvInstruction value = null;
if (context.Config.GpuAccessor.QueryHostHasVectorIndexingBug())
{
// Test for each component individually.
for (int i = 0; i < 4; i++)
{
var component = context.Constant(context.TypeS32(), i);
SpvInstruction elemPointer;
if (context.UniformBuffersArray != null)
{
var ubVariable = context.UniformBuffersArray;
var i0 = context.Get(AggregateType.S32, src1);
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i0, i1, i2, component);
}
else
{
var ubVariable = context.UniformBuffers[((AstOperand)src1).Value];
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i1, i2, component);
}
SpvInstruction newValue = context.Load(context.TypeFP32(), elemPointer);
value = value != null ? context.Select(context.TypeFP32(), context.IEqual(context.TypeBool(), i3, component), newValue, value) : newValue;
}
}
else
{
SpvInstruction elemPointer;
if (context.UniformBuffersArray != null)
{
var ubVariable = context.UniformBuffersArray;
var i0 = context.Get(AggregateType.S32, src1);
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i0, i1, i2, i3);
}
else
{
var ubVariable = context.UniformBuffers[((AstOperand)src1).Value];
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i1, i2, i3);
}
value = context.Load(context.TypeFP32(), elemPointer);
}
return new OperationResult(AggregateType.FP32, value);
}
private static OperationResult GenerateLoadLocal(CodeGenContext context, AstOperation operation)
{
return GenerateLoadLocalOrShared(context, operation, StorageClass.Private, context.LocalMemory);
}
private static OperationResult GenerateLoadShared(CodeGenContext context, AstOperation operation)
{
return GenerateLoadLocalOrShared(context, operation, StorageClass.Workgroup, context.SharedMemory);
}
private static OperationResult GenerateLoadLocalOrShared(
CodeGenContext context,
AstOperation operation,
StorageClass storageClass,
SpvInstruction memory)
{
var offset = context.Get(AggregateType.S32, operation.GetSource(0));
var elemPointer = context.AccessChain(context.TypePointer(storageClass, context.TypeU32()), memory, offset);
var value = context.Load(context.TypeU32(), elemPointer);
return new OperationResult(AggregateType.U32, value);
}
private static OperationResult GenerateLoadStorage(CodeGenContext context, AstOperation operation)
{
var elemPointer = GetStorageElemPointer(context, operation);
var value = context.Load(context.TypeU32(), elemPointer);
return new OperationResult(AggregateType.U32, value);
}
private static OperationResult GenerateLod(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return new OperationResult(AggregateType.S32, context.Constant(context.TypeS32(), 0));
}
int srcIndex = 0;
SpvInstruction Src(AggregateType type)
{
return context.Get(type, texOp.GetSource(srcIndex++));
}
SpvInstruction index = null;
if (isIndexed)
{
index = Src(AggregateType.S32);
}
int pCount = texOp.Type.GetDimensions();
SpvInstruction pCoords;
if (pCount > 1)
{
SpvInstruction[] elems = new SpvInstruction[pCount];
for (int i = 0; i < pCount; i++)
{
elems[i] = Src(AggregateType.FP32);
}
var vectorType = context.TypeVector(context.TypeFP32(), pCount);
pCoords = context.CompositeConstruct(vectorType, elems);
}
else
{
pCoords = Src(AggregateType.FP32);
}
var meta = new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format);
(_, var sampledImageType, var sampledImageVariable) = context.Samplers[meta];
var image = context.Load(sampledImageType, sampledImageVariable);
var resultType = context.TypeVector(context.TypeFP32(), 2);
var packed = context.ImageQueryLod(resultType, image, pCoords);
var result = context.CompositeExtract(context.TypeFP32(), packed, (SpvLiteralInteger)texOp.Index);
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateLogarithmB2(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslLog2, null);
}
private static OperationResult GenerateLogicalAnd(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryBool(context, operation, context.Delegates.LogicalAnd);
}
private static OperationResult GenerateLogicalExclusiveOr(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryBool(context, operation, context.Delegates.LogicalNotEqual);
}
private static OperationResult GenerateLogicalNot(CodeGenContext context, AstOperation operation)
{
return GenerateUnaryBool(context, operation, context.Delegates.LogicalNot);
}
private static OperationResult GenerateLogicalOr(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryBool(context, operation, context.Delegates.LogicalOr);
}
private static OperationResult GenerateLoopBreak(CodeGenContext context, AstOperation operation)
{
AstBlock loopBlock = context.CurrentBlock;
while (loopBlock.Type != AstBlockType.DoWhile)
{
loopBlock = loopBlock.Parent;
}
context.Branch(context.GetNextLabel(loopBlock.Parent));
return OperationResult.Invalid;
}
private static OperationResult GenerateLoopContinue(CodeGenContext context, AstOperation operation)
{
AstBlock loopBlock = context.CurrentBlock;
while (loopBlock.Type != AstBlockType.DoWhile)
{
loopBlock = loopBlock.Parent;
}
(var loopTarget, var continueTarget) = context.LoopTargets[loopBlock];
context.Branch(continueTarget);
return OperationResult.Invalid;
}
private static OperationResult GenerateMaximum(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.GlslFMax, context.Delegates.GlslSMax);
}
private static OperationResult GenerateMaximumU32(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryU32(context, operation, context.Delegates.GlslUMax);
}
private static OperationResult GenerateMemoryBarrier(CodeGenContext context, AstOperation operation)
{
context.MemoryBarrier(context.Constant(context.TypeU32(), Scope.Device), context.Constant(context.TypeU32(), DefaultMemorySemantics));
return OperationResult.Invalid;
}
private static OperationResult GenerateMinimum(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.GlslFMin, context.Delegates.GlslSMin);
}
private static OperationResult GenerateMinimumU32(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryU32(context, operation, context.Delegates.GlslUMin);
}
private static OperationResult GenerateMultiply(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.FMul, context.Delegates.IMul);
}
private static OperationResult GenerateMultiplyHighS32(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var resultType = context.TypeStruct(false, context.TypeS32(), context.TypeS32());
var result = context.SMulExtended(resultType, context.GetS32(src1), context.GetS32(src2));
result = context.CompositeExtract(context.TypeS32(), result, 1);
return new OperationResult(AggregateType.S32, result);
}
private static OperationResult GenerateMultiplyHighU32(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var resultType = context.TypeStruct(false, context.TypeU32(), context.TypeU32());
var result = context.UMulExtended(resultType, context.GetU32(src1), context.GetU32(src2));
result = context.CompositeExtract(context.TypeU32(), result, 1);
return new OperationResult(AggregateType.U32, result);
}
private static OperationResult GenerateNegate(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.FNegate, context.Delegates.SNegate);
}
private static OperationResult GeneratePackDouble2x32(CodeGenContext context, AstOperation operation)
{
var value0 = context.GetU32(operation.GetSource(0));
var value1 = context.GetU32(operation.GetSource(1));
var vector = context.CompositeConstruct(context.TypeVector(context.TypeU32(), 2), value0, value1);
var result = context.GlslPackDouble2x32(context.TypeFP64(), vector);
return new OperationResult(AggregateType.FP64, result);
}
private static OperationResult GeneratePackHalf2x16(CodeGenContext context, AstOperation operation)
{
var value0 = context.GetFP32(operation.GetSource(0));
var value1 = context.GetFP32(operation.GetSource(1));
var vector = context.CompositeConstruct(context.TypeVector(context.TypeFP32(), 2), value0, value1);
var result = context.GlslPackHalf2x16(context.TypeU32(), vector);
return new OperationResult(AggregateType.U32, result);
}
private static OperationResult GenerateReciprocalSquareRoot(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslInverseSqrt, null);
}
private static OperationResult GenerateReturn(CodeGenContext context, AstOperation operation)
{
context.Return();
return OperationResult.Invalid;
}
private static OperationResult GenerateRound(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslRoundEven, null);
}
private static OperationResult GenerateShiftLeft(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.ShiftLeftLogical);
}
private static OperationResult GenerateShiftRightS32(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.ShiftRightArithmetic);
}
private static OperationResult GenerateShiftRightU32(CodeGenContext context, AstOperation operation)
{
return GenerateBinaryS32(context, operation, context.Delegates.ShiftRightLogical);
}
private static OperationResult GenerateShuffle(CodeGenContext context, AstOperation operation)
{
var x = context.GetFP32(operation.GetSource(0));
var index = context.GetU32(operation.GetSource(1));
var mask = context.GetU32(operation.GetSource(2));
var const31 = context.Constant(context.TypeU32(), 31);
var const8 = context.Constant(context.TypeU32(), 8);
var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
var notSegMask = context.Not(context.TypeU32(), segMask);
var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
var indexNotSegMask = context.BitwiseAnd(context.TypeU32(), index, notSegMask);
var threadId = context.GetAttribute(AggregateType.U32, AttributeConsts.LaneId, false);
var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
var srcThreadId = context.BitwiseOr(context.TypeU32(), indexNotSegMask, minThreadId);
var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var result = context.Select(context.TypeFP32(), valid, value, x);
var validLocal = (AstOperand)operation.GetSource(3);
context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateShuffleDown(CodeGenContext context, AstOperation operation)
{
var x = context.GetFP32(operation.GetSource(0));
var index = context.GetU32(operation.GetSource(1));
var mask = context.GetU32(operation.GetSource(2));
var const31 = context.Constant(context.TypeU32(), 31);
var const8 = context.Constant(context.TypeU32(), 8);
var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
var notSegMask = context.Not(context.TypeU32(), segMask);
var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
var threadId = context.GetAttribute(AggregateType.U32, AttributeConsts.LaneId, false);
var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
var srcThreadId = context.IAdd(context.TypeU32(), threadId, index);
var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var result = context.Select(context.TypeFP32(), valid, value, x);
var validLocal = (AstOperand)operation.GetSource(3);
context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateShuffleUp(CodeGenContext context, AstOperation operation)
{
var x = context.GetFP32(operation.GetSource(0));
var index = context.GetU32(operation.GetSource(1));
var mask = context.GetU32(operation.GetSource(2));
var const31 = context.Constant(context.TypeU32(), 31);
var const8 = context.Constant(context.TypeU32(), 8);
var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
var threadId = context.GetAttribute(AggregateType.U32, AttributeConsts.LaneId, false);
var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
var srcThreadId = context.ISub(context.TypeU32(), threadId, index);
var valid = context.SGreaterThanEqual(context.TypeBool(), srcThreadId, minThreadId);
var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var result = context.Select(context.TypeFP32(), valid, value, x);
var validLocal = (AstOperand)operation.GetSource(3);
context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateShuffleXor(CodeGenContext context, AstOperation operation)
{
var x = context.GetFP32(operation.GetSource(0));
var index = context.GetU32(operation.GetSource(1));
var mask = context.GetU32(operation.GetSource(2));
var const31 = context.Constant(context.TypeU32(), 31);
var const8 = context.Constant(context.TypeU32(), 8);
var clamp = context.BitwiseAnd(context.TypeU32(), mask, const31);
var segMask = context.BitwiseAnd(context.TypeU32(), context.ShiftRightLogical(context.TypeU32(), mask, const8), const31);
var notSegMask = context.Not(context.TypeU32(), segMask);
var clampNotSegMask = context.BitwiseAnd(context.TypeU32(), clamp, notSegMask);
var threadId = context.GetAttribute(AggregateType.U32, AttributeConsts.LaneId, false);
var minThreadId = context.BitwiseAnd(context.TypeU32(), threadId, segMask);
var maxThreadId = context.BitwiseOr(context.TypeU32(), minThreadId, clampNotSegMask);
var srcThreadId = context.BitwiseXor(context.TypeU32(), threadId, index);
var valid = context.ULessThanEqual(context.TypeBool(), srcThreadId, maxThreadId);
var value = context.GroupNonUniformShuffle(context.TypeFP32(), context.Constant(context.TypeU32(), (int)Scope.Subgroup), x, srcThreadId);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var result = context.Select(context.TypeFP32(), valid, value, x);
var validLocal = (AstOperand)operation.GetSource(3);
context.Store(context.GetLocalPointer(validLocal), context.BitcastIfNeeded(validLocal.VarType, AggregateType.Bool, valid));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateSine(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslSin, null);
}
private static OperationResult GenerateSquareRoot(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslSqrt, null);
}
private static OperationResult GenerateStoreAttribute(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
if (!(src1 is AstOperand baseAttr) || baseAttr.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {nameof(Instruction.StoreAttribute)} must be a constant operand.");
}
SpvInstruction elemPointer;
AggregateType elemType;
if (src2 is AstOperand operand && operand.Type == OperandType.Constant)
{
int attrOffset = (baseAttr.Value & AttributeConsts.Mask) + (operand.Value << 2);
elemPointer = context.GetAttributeElemPointer(attrOffset, isOutAttr: true, index: null, out elemType);
}
else
{
var attr = context.Get(AggregateType.S32, src2);
elemPointer = context.GetAttributeElemPointer(attr, isOutAttr: true, index: null, out elemType);
}
var value = context.Get(elemType, src3);
context.Store(elemPointer, value);
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreLocal(CodeGenContext context, AstOperation operation)
{
return GenerateStoreLocalOrShared(context, operation, StorageClass.Private, context.LocalMemory);
}
private static OperationResult GenerateStoreShared(CodeGenContext context, AstOperation operation)
{
return GenerateStoreLocalOrShared(context, operation, StorageClass.Workgroup, context.SharedMemory);
}
private static OperationResult GenerateStoreLocalOrShared(
CodeGenContext context,
AstOperation operation,
StorageClass storageClass,
SpvInstruction memory)
{
var offset = context.Get(AggregateType.S32, operation.GetSource(0));
var value = context.Get(AggregateType.U32, operation.GetSource(1));
var elemPointer = context.AccessChain(context.TypePointer(storageClass, context.TypeU32()), memory, offset);
context.Store(elemPointer, value);
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreShared16(CodeGenContext context, AstOperation operation)
{
GenerateStoreSharedSmallInt(context, operation, 16);
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreShared8(CodeGenContext context, AstOperation operation)
{
GenerateStoreSharedSmallInt(context, operation, 8);
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreStorage(CodeGenContext context, AstOperation operation)
{
var elemPointer = GetStorageElemPointer(context, operation);
context.Store(elemPointer, context.Get(AggregateType.U32, operation.GetSource(2)));
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreStorage16(CodeGenContext context, AstOperation operation)
{
GenerateStoreStorageSmallInt(context, operation, 16);
return OperationResult.Invalid;
}
private static OperationResult GenerateStoreStorage8(CodeGenContext context, AstOperation operation)
{
GenerateStoreStorageSmallInt(context, operation, 8);
return OperationResult.Invalid;
}
private static OperationResult GenerateSubtract(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.FSub, context.Delegates.ISub);
}
private static OperationResult GenerateSwizzleAdd(CodeGenContext context, AstOperation operation)
{
var x = context.Get(AggregateType.FP32, operation.GetSource(0));
var y = context.Get(AggregateType.FP32, operation.GetSource(1));
var mask = context.Get(AggregateType.U32, operation.GetSource(2));
var v4float = context.TypeVector(context.TypeFP32(), 4);
var one = context.Constant(context.TypeFP32(), 1.0f);
var minusOne = context.Constant(context.TypeFP32(), -1.0f);
var zero = context.Constant(context.TypeFP32(), 0.0f);
var xLut = context.ConstantComposite(v4float, one, minusOne, one, zero);
var yLut = context.ConstantComposite(v4float, one, one, minusOne, one);
var three = context.Constant(context.TypeU32(), 3);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var threadId = context.GetAttribute(AggregateType.U32, AttributeConsts.LaneId, false);
var shift = context.BitwiseAnd(context.TypeU32(), threadId, three);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
shift = context.ShiftLeftLogical(context.TypeU32(), shift, context.Constant(context.TypeU32(), 1));
var lutIdx = context.ShiftRightLogical(context.TypeU32(), mask, shift);
lutIdx = context.BitwiseAnd(context.TypeU32(), lutIdx, three);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
var xLutValue = context.VectorExtractDynamic(context.TypeFP32(), xLut, lutIdx);
var yLutValue = context.VectorExtractDynamic(context.TypeFP32(), yLut, lutIdx);
var xResult = context.FMul(context.TypeFP32(), x, xLutValue);
var yResult = context.FMul(context.TypeFP32(), y, yLutValue);
var result = context.FAdd(context.TypeFP32(), xResult, yResult);
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateTextureSample(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
bool isGather = (texOp.Flags & TextureFlags.Gather) != 0;
bool hasDerivatives = (texOp.Flags & TextureFlags.Derivatives) != 0;
bool intCoords = (texOp.Flags & TextureFlags.IntCoords) != 0;
bool hasLodBias = (texOp.Flags & TextureFlags.LodBias) != 0;
bool hasLodLevel = (texOp.Flags & TextureFlags.LodLevel) != 0;
bool hasOffset = (texOp.Flags & TextureFlags.Offset) != 0;
bool hasOffsets = (texOp.Flags & TextureFlags.Offsets) != 0;
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
bool isMultisample = (texOp.Type & SamplerType.Multisample) != 0;
bool isShadow = (texOp.Type & SamplerType.Shadow) != 0;
bool colorIsVector = isGather || !isShadow;
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return GetZeroOperationResult(context, texOp, AggregateType.FP32, colorIsVector);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
// This combination is valid, but not available on GLSL.
// For now, ignore the LOD level and do a normal sample.
// TODO: How to implement it properly?
if (hasLodLevel && isArray && isShadow)
{
hasLodLevel = false;
}
int srcIndex = isBindless ? 1 : 0;
SpvInstruction Src(AggregateType type)
{
return context.Get(type, texOp.GetSource(srcIndex++));
}
SpvInstruction index = null;
if (isIndexed)
{
index = Src(AggregateType.S32);
}
int coordsCount = texOp.Type.GetDimensions();
int pCount = coordsCount;
int arrayIndexElem = -1;
if (isArray)
{
arrayIndexElem = pCount++;
}
AggregateType coordType = intCoords ? AggregateType.S32 : AggregateType.FP32;
SpvInstruction AssemblePVector(int count)
{
if (count > 1)
{
SpvInstruction[] elems = new SpvInstruction[count];
for (int index = 0; index < count; index++)
{
if (arrayIndexElem == index)
{
elems[index] = Src(AggregateType.S32);
if (!intCoords)
{
elems[index] = context.ConvertSToF(context.TypeFP32(), elems[index]);
}
}
else
{
elems[index] = Src(coordType);
}
}
var vectorType = context.TypeVector(intCoords ? context.TypeS32() : context.TypeFP32(), count);
return context.CompositeConstruct(vectorType, elems);
}
else
{
return Src(coordType);
}
}
SpvInstruction pCoords = AssemblePVector(pCount);
pCoords = ScalingHelpers.ApplyScaling(context, texOp, pCoords, intCoords, isBindless, isIndexed, isArray, pCount);
SpvInstruction AssembleDerivativesVector(int count)
{
if (count > 1)
{
SpvInstruction[] elems = new SpvInstruction[count];
for (int index = 0; index < count; index++)
{
elems[index] = Src(AggregateType.FP32);
}
var vectorType = context.TypeVector(context.TypeFP32(), count);
return context.CompositeConstruct(vectorType, elems);
}
else
{
return Src(AggregateType.FP32);
}
}
SpvInstruction dRef = null;
if (isShadow)
{
dRef = Src(AggregateType.FP32);
}
SpvInstruction[] derivatives = null;
if (hasDerivatives)
{
derivatives = new[]
{
AssembleDerivativesVector(coordsCount), // dPdx
AssembleDerivativesVector(coordsCount) // dPdy
};
}
SpvInstruction sample = null;
SpvInstruction lod = null;
if (isMultisample)
{
sample = Src(AggregateType.S32);
}
else if (hasLodLevel)
{
lod = Src(coordType);
}
SpvInstruction AssembleOffsetVector(int count)
{
if (count > 1)
{
SpvInstruction[] elems = new SpvInstruction[count];
for (int index = 0; index < count; index++)
{
elems[index] = Src(AggregateType.S32);
}
var vectorType = context.TypeVector(context.TypeS32(), count);
return context.ConstantComposite(vectorType, elems);
}
else
{
return Src(AggregateType.S32);
}
}
SpvInstruction[] offsets = null;
if (hasOffset)
{
offsets = new[] { AssembleOffsetVector(coordsCount) };
}
else if (hasOffsets)
{
offsets = new[]
{
AssembleOffsetVector(coordsCount),
AssembleOffsetVector(coordsCount),
AssembleOffsetVector(coordsCount),
AssembleOffsetVector(coordsCount)
};
}
SpvInstruction lodBias = null;
if (hasLodBias)
{
lodBias = Src(AggregateType.FP32);
}
SpvInstruction compIdx = null;
// textureGather* optional extra component index,
// not needed for shadow samplers.
if (isGather && !isShadow)
{
compIdx = Src(AggregateType.S32);
}
var operandsList = new List<SpvInstruction>();
var operandsMask = ImageOperandsMask.MaskNone;
if (hasLodBias)
{
operandsMask |= ImageOperandsMask.Bias;
operandsList.Add(lodBias);
}
if (!isMultisample && hasLodLevel)
{
operandsMask |= ImageOperandsMask.Lod;
operandsList.Add(lod);
}
if (hasDerivatives)
{
operandsMask |= ImageOperandsMask.Grad;
operandsList.Add(derivatives[0]);
operandsList.Add(derivatives[1]);
}
if (hasOffset)
{
operandsMask |= ImageOperandsMask.ConstOffset;
operandsList.Add(offsets[0]);
}
else if (hasOffsets)
{
operandsMask |= ImageOperandsMask.ConstOffsets;
SpvInstruction arrayv2 = context.TypeArray(context.TypeVector(context.TypeS32(), 2), context.Constant(context.TypeU32(), 4));
operandsList.Add(context.ConstantComposite(arrayv2, offsets[0], offsets[1], offsets[2], offsets[3]));
}
if (isMultisample)
{
operandsMask |= ImageOperandsMask.Sample;
operandsList.Add(sample);
}
var resultType = colorIsVector ? context.TypeVector(context.TypeFP32(), 4) : context.TypeFP32();
var meta = new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format);
(var imageType, var sampledImageType, var sampledImageVariable) = context.Samplers[meta];
var image = context.Load(sampledImageType, sampledImageVariable);
if (intCoords)
{
image = context.Image(imageType, image);
}
var operands = operandsList.ToArray();
SpvInstruction result;
if (intCoords)
{
result = context.ImageFetch(resultType, image, pCoords, operandsMask, operands);
}
else if (isGather)
{
if (isShadow)
{
result = context.ImageDrefGather(resultType, image, pCoords, dRef, operandsMask, operands);
}
else
{
result = context.ImageGather(resultType, image, pCoords, compIdx, operandsMask, operands);
}
}
else if (isShadow)
{
if (hasLodLevel)
{
result = context.ImageSampleDrefExplicitLod(resultType, image, pCoords, dRef, operandsMask, operands);
}
else
{
result = context.ImageSampleDrefImplicitLod(resultType, image, pCoords, dRef, operandsMask, operands);
}
}
else if (hasDerivatives || hasLodLevel)
{
result = context.ImageSampleExplicitLod(resultType, image, pCoords, operandsMask, operands);
}
else
{
result = context.ImageSampleImplicitLod(resultType, image, pCoords, operandsMask, operands);
}
var swizzledResultType = AggregateType.FP32;
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
if (colorIsVector)
{
swizzledResultType = texOp.GetVectorType(swizzledResultType);
result = GetSwizzledResult(context, result, swizzledResultType, texOp.Index);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
return new OperationResult(swizzledResultType, result);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
private static OperationResult GenerateTextureSize(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return new OperationResult(AggregateType.S32, context.Constant(context.TypeS32(), 0));
}
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
SpvInstruction index = null;
if (isIndexed)
{
index = context.GetS32(texOp.GetSource(0));
}
var meta = new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format);
(var imageType, var sampledImageType, var sampledImageVariable) = context.Samplers[meta];
var image = context.Load(sampledImageType, sampledImageVariable);
image = context.Image(imageType, image);
if (texOp.Index == 3)
{
return new OperationResult(AggregateType.S32, context.ImageQueryLevels(context.TypeS32(), image));
}
else
{
var type = context.SamplersTypes[meta];
bool hasLod = !type.HasFlag(SamplerType.Multisample) && type != SamplerType.TextureBuffer;
int dimensions = (type & SamplerType.Mask) == SamplerType.TextureCube ? 2 : type.GetDimensions();
if (type.HasFlag(SamplerType.Array))
{
dimensions++;
}
var resultType = dimensions == 1 ? context.TypeS32() : context.TypeVector(context.TypeS32(), dimensions);
SpvInstruction result;
if (hasLod)
{
int lodSrcIndex = isBindless || isIndexed ? 1 : 0;
var lod = context.GetS32(operation.GetSource(lodSrcIndex));
result = context.ImageQuerySizeLod(resultType, image, lod);
}
else
{
result = context.ImageQuerySize(resultType, image);
}
if (dimensions != 1)
{
result = context.CompositeExtract(context.TypeS32(), result, (SpvLiteralInteger)texOp.Index);
}
if (texOp.Index < 2 || (type & SamplerType.Mask) == SamplerType.Texture3D)
{
result = ScalingHelpers.ApplyUnscaling(context, texOp.WithType(type), result, isBindless, isIndexed);
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
}
return new OperationResult(AggregateType.S32, result);
}
}
private static OperationResult GenerateTruncate(CodeGenContext context, AstOperation operation)
{
return GenerateUnary(context, operation, context.Delegates.GlslTrunc, null);
}
private static OperationResult GenerateUnpackDouble2x32(CodeGenContext context, AstOperation operation)
{
var value = context.GetFP64(operation.GetSource(0));
var vector = context.GlslUnpackDouble2x32(context.TypeVector(context.TypeU32(), 2), value);
var result = context.CompositeExtract(context.TypeU32(), vector, operation.Index);
return new OperationResult(AggregateType.U32, result);
}
private static OperationResult GenerateUnpackHalf2x16(CodeGenContext context, AstOperation operation)
{
var value = context.GetU32(operation.GetSource(0));
var vector = context.GlslUnpackHalf2x16(context.TypeVector(context.TypeFP32(), 2), value);
var result = context.CompositeExtract(context.TypeFP32(), vector, operation.Index);
return new OperationResult(AggregateType.FP32, result);
}
private static OperationResult GenerateVectorExtract(CodeGenContext context, AstOperation operation)
{
var vector = context.GetWithType(operation.GetSource(0), out AggregateType vectorType);
var scalarType = vectorType & ~AggregateType.ElementCountMask;
var resultType = context.GetType(scalarType);
SpvInstruction result;
if (operation.GetSource(1) is AstOperand indexOperand && indexOperand.Type == OperandType.Constant)
{
result = context.CompositeExtract(resultType, vector, (SpvLiteralInteger)indexOperand.Value);
}
else
{
var index = context.Get(AggregateType.S32, operation.GetSource(1));
result = context.VectorExtractDynamic(resultType, vector, index);
}
return new OperationResult(scalarType, result);
}
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
private static OperationResult GenerateVoteAll(CodeGenContext context, AstOperation operation)
{
var execution = context.Constant(context.TypeU32(), Scope.Subgroup);
var result = context.GroupNonUniformAll(context.TypeBool(), execution, context.Get(AggregateType.Bool, operation.GetSource(0)));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateVoteAllEqual(CodeGenContext context, AstOperation operation)
{
var execution = context.Constant(context.TypeU32(), Scope.Subgroup);
var result = context.GroupNonUniformAllEqual(context.TypeBool(), execution, context.Get(AggregateType.Bool, operation.GetSource(0)));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateVoteAny(CodeGenContext context, AstOperation operation)
{
var execution = context.Constant(context.TypeU32(), Scope.Subgroup);
var result = context.GroupNonUniformAny(context.TypeBool(), execution, context.Get(AggregateType.Bool, operation.GetSource(0)));
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateCompare(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitF,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitI)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
SpvInstruction result;
if (operation.Inst.HasFlag(Instruction.FP64))
{
result = emitF(context.TypeBool(), context.GetFP64(src1), context.GetFP64(src2));
}
else if (operation.Inst.HasFlag(Instruction.FP32))
{
result = emitF(context.TypeBool(), context.GetFP32(src1), context.GetFP32(src2));
}
else
{
result = emitI(context.TypeBool(), context.GetS32(src1), context.GetS32(src2));
}
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateCompareU32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var result = emitU(context.TypeBool(), context.GetU32(src1), context.GetU32(src2));
return new OperationResult(AggregateType.Bool, result);
}
private static OperationResult GenerateAtomicMemoryBinary(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
{
var value = context.GetU32(operation.GetSource(2));
SpvInstruction elemPointer;
Instruction mr = operation.Inst & Instruction.MrMask;
if (mr == Instruction.MrStorage)
{
elemPointer = GetStorageElemPointer(context, operation);
}
else if (mr == Instruction.MrShared)
{
var offset = context.GetU32(operation.GetSource(0));
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
}
else
{
throw new InvalidOperationException($"Invalid storage class \"{mr}\".");
}
var one = context.Constant(context.TypeU32(), 1);
var zero = context.Constant(context.TypeU32(), 0);
return new OperationResult(AggregateType.U32, emitU(context.TypeU32(), elemPointer, one, zero, value));
}
private static OperationResult GenerateAtomicMemoryCas(CodeGenContext context, AstOperation operation)
{
var value0 = context.GetU32(operation.GetSource(2));
var value1 = context.GetU32(operation.GetSource(3));
SpvInstruction elemPointer;
Instruction mr = operation.Inst & Instruction.MrMask;
if (mr == Instruction.MrStorage)
{
elemPointer = GetStorageElemPointer(context, operation);
}
else if (mr == Instruction.MrShared)
{
var offset = context.GetU32(operation.GetSource(0));
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
}
else
{
throw new InvalidOperationException($"Invalid storage class \"{mr}\".");
}
var one = context.Constant(context.TypeU32(), 1);
var zero = context.Constant(context.TypeU32(), 0);
return new OperationResult(AggregateType.U32, context.AtomicCompareExchange(context.TypeU32(), elemPointer, one, zero, zero, value1, value0));
}
private static void GenerateStoreSharedSmallInt(CodeGenContext context, AstOperation operation, int bitSize)
{
var offset = context.Get(AggregateType.U32, operation.GetSource(0));
var value = context.Get(AggregateType.U32, operation.GetSource(1));
var wordOffset = context.ShiftRightLogical(context.TypeU32(), offset, context.Constant(context.TypeU32(), 2));
var bitOffset = context.BitwiseAnd(context.TypeU32(), offset, context.Constant(context.TypeU32(), 3));
bitOffset = context.ShiftLeftLogical(context.TypeU32(), bitOffset, context.Constant(context.TypeU32(), 3));
var memory = context.SharedMemory;
var elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), memory, wordOffset);
GenerateStoreSmallInt(context, elemPointer, bitOffset, value, bitSize);
}
private static void GenerateStoreStorageSmallInt(CodeGenContext context, AstOperation operation, int bitSize)
{
var i0 = context.Get(AggregateType.S32, operation.GetSource(0));
var offset = context.Get(AggregateType.U32, operation.GetSource(1));
var value = context.Get(AggregateType.U32, operation.GetSource(2));
var wordOffset = context.ShiftRightLogical(context.TypeU32(), offset, context.Constant(context.TypeU32(), 2));
var bitOffset = context.BitwiseAnd(context.TypeU32(), offset, context.Constant(context.TypeU32(), 3));
bitOffset = context.ShiftLeftLogical(context.TypeU32(), bitOffset, context.Constant(context.TypeU32(), 3));
var sbVariable = context.StorageBuffersArray;
var i1 = context.Constant(context.TypeS32(), 0);
var elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeU32()), sbVariable, i0, i1, wordOffset);
GenerateStoreSmallInt(context, elemPointer, bitOffset, value, bitSize);
}
private static void GenerateStoreSmallInt(
CodeGenContext context,
SpvInstruction elemPointer,
SpvInstruction bitOffset,
SpvInstruction value,
int bitSize)
{
var loopStart = context.Label();
var loopEnd = context.Label();
context.Branch(loopStart);
context.AddLabel(loopStart);
var oldValue = context.Load(context.TypeU32(), elemPointer);
var newValue = context.BitFieldInsert(context.TypeU32(), oldValue, value, bitOffset, context.Constant(context.TypeU32(), bitSize));
var one = context.Constant(context.TypeU32(), 1);
var zero = context.Constant(context.TypeU32(), 0);
var result = context.AtomicCompareExchange(context.TypeU32(), elemPointer, one, zero, zero, newValue, oldValue);
var failed = context.INotEqual(context.TypeBool(), result, oldValue);
context.LoopMerge(loopEnd, loopStart, LoopControlMask.MaskNone);
context.BranchConditional(failed, loopStart, loopEnd);
context.AddLabel(loopEnd);
}
private static OperationResult GetZeroOperationResult(
CodeGenContext context,
AstTextureOperation texOp,
AggregateType scalarType,
bool isVector)
{
var zero = scalarType switch
{
AggregateType.S32 => context.Constant(context.TypeS32(), 0),
AggregateType.U32 => context.Constant(context.TypeU32(), 0u),
_ => context.Constant(context.TypeFP32(), 0f),
};
if (isVector)
{
AggregateType outputType = texOp.GetVectorType(scalarType);
if ((outputType & AggregateType.ElementCountMask) != 0)
{
int componentsCount = BitOperations.PopCount((uint)texOp.Index);
SpvInstruction[] values = new SpvInstruction[componentsCount];
values.AsSpan().Fill(zero);
return new OperationResult(outputType, context.ConstantComposite(context.GetType(outputType), values));
}
}
return new OperationResult(scalarType, zero);
}
private static SpvInstruction GetSwizzledResult(CodeGenContext context, SpvInstruction vector, AggregateType swizzledResultType, int mask)
{
if ((swizzledResultType & AggregateType.ElementCountMask) != 0)
{
SpvLiteralInteger[] components = new SpvLiteralInteger[BitOperations.PopCount((uint)mask)];
int componentIndex = 0;
for (int i = 0; i < 4; i++)
{
if ((mask & (1 << i)) != 0)
{
components[componentIndex++] = i;
}
}
return context.VectorShuffle(context.GetType(swizzledResultType), vector, vector, components);
}
else
{
int componentIndex = (int)BitOperations.TrailingZeroCount(mask);
return context.CompositeExtract(context.GetType(swizzledResultType), vector, (SpvLiteralInteger)componentIndex);
}
}
Vulkan backend (#2518) * WIP Vulkan implementation * No need to initialize attributes on the SPIR-V backend anymore * Allow multithreading shaderc and vkCreateShaderModule You'll only really see the benefit here with threaded-gal or parallel shader cache compile. Fix shaderc multithreaded changes Thread safety for shaderc Options constructor Dunno how they managed to make a constructor not thread safe, but you do you. May avoid some freezes. * Support multiple levels/layers for blit. Fixes MK8D when scaled, maybe a few other games. AMD software "safe" blit not supported right now. * TextureStorage should hold a ref of the foreign storage, otherwise it might be freed while in use * New depth-stencil blit method for AMD * Workaround for AMD driver bug * Fix some tessellation related issues (still doesn't work?) * Submit command buffer before Texture GetData. (UE4 fix) * DrawTexture support * Fix BGRA on OpenGL backend * Fix rebase build break * Support format aliasing on SetImage * Fix uniform buffers being lost when bindings are out of order * Fix storage buffers being lost when bindings are out of order (also avoid allocations when changing bindings) * Use current command buffer for unscaled copy (perf) Avoids flushing commands and renting a command buffer when fulfilling copy dependencies and when games do unscaled copies. * Update to .net6 * Update Silk.NET to version 2.10.1 Somehow, massive performance boost. Seems like their vtable for looking up vulkan methods was really slow before. * Fix PrimitivesGenerated query, disable Transform Feedback queries for now Lets Splatoon 2 work on nvidia. (mostly) * Update counter queue to be similar to the OGL one Fixes softlocks when games had to flush counters. * Don't throw when ending conditional rendering for now This should be re-enabled when conditional rendering is enabled on nvidia etc. * Update findMSB/findLSB to match master's instruction enum * Fix triangle overlay on SMO, Captain Toad, maybe others? * Don't make Intel Mesa pay for Intel Windows bugs * Fix samplers with MinFilter Linear or Nearest (fixes New Super Mario Bros U Deluxe black borders) * Update Spv.Generator * Add alpha test emulation on shader (but no shader specialisation yet...) * Fix R4G4B4A4Unorm texture format permutation * Validation layers should be enabled for any log level other than None * Add barriers around vkCmdCopyImage Write->Read barrier for src image (we want to wait for a write to read it) Write->Read barrier for dst image (we want to wait for the copy to complete before use) * Be a bit more careful with texture access flags, since it can be used for anything * Device local mapping for all buffers May avoid issues with drivers with NVIDIA on linux/older gpus on windows when using large buffers (?) Also some performance things and fixes issues with opengl games loading textures weird. * Cleanup, disable device local buffers for now. * Add single queue support Multiqueue seems to be a bit more responsive on NVIDIA. Should fix texture flush on intel. AMD has been forced to single queue for an experiment. * Fix some validation errors around extended dynamic state * Remove Intel bug workaround, it was fixed on the latest driver * Use circular queue for checking consumption on command buffers Speeds up games that spam command buffers a little. Avoids checking multiple command buffers if multiple are active at once. * Use SupportBufferUpdater, add single layer flush * Fix counter queue leak when game decides to use host conditional rendering * Force device local storage for textures (fixes linux performance) * Port #3019 * Insert barriers around vkCmdBlitImage (may fix some amd flicker) * Fix transform feedback on Intel, gl_Position feedback and clears to inexistent depth buffers * Don't pause transform feedback for multi draw * Fix draw outside of render pass and missing capability * Workaround for wrong last attribute on AMD (affects FFVII, STRIKERS1945, probably more) * Better workaround for AMD vertex buffer size alignment issue * More instructions + fixes on SPIR-V backend * Allow custom aspect ratio on Vulkan * Correct GTK UI status bar positions * SPIR-V: Functions must always end with a return * SPIR-V: Fix ImageQuerySizeLod * SPIR-V: Set DepthReplacing execution mode when FragDepth is modified * SPIR-V: Implement LoopContinue IR instruction * SPIR-V: Geometry shader support * SPIR-V: Use correct binding number on storage buffers array * Reduce allocations for Spir-v serialization Passes BinaryWriter instead of the stream to Write and WriteOperand - Removes creation of BinaryWriter for each instruction - Removes allocations for literal string * Some optimizations to Spv.Generator - Dictionary for lookups of type declarations, constants, extinst - LiteralInteger internal data format -> ushort - Deterministic HashCode implementation to avoid spirv result not being the same between runs - Inline operand list instead of List<T>, falls back to array if many operands. (large performance boost) TODO: improve instruction allocation, structured program creator, ssa? * Pool Spv.Generator resources, cache delegates, spv opts - Pools for Instructions and LiteralIntegers. Can be passed in when creating the generator module. - NewInstruction is called instead of new Instruction() - Ryujinx SpirvGenerator passes in some pools that are static. The idea is for these to be shared between threads eventually. - Estimate code size when creating the output MemoryStream - LiteralInteger pools using ThreadStatic pools that are initialized before and after creation... not sure of a better way since the way these are created is via implicit cast. Also, cache delegates for Spv.Generator for functions that are passed around to GenerateBinary etc, since passing the function raw creates a delegate on each call. TODO: update python spv cs generator to make the coregrammar with NewInstruction and the `params` overloads. * LocalDefMap for Ssa Rewriter Rather than allocating a large array of all registers for each block in the shader, allocate one array of all registers and clear it between blocks. Reduces allocations in the shader translator. * SPIR-V: Transform feedback support * SPIR-V: Fragment shader interlock support (and image coherency) * SPIR-V: Add early fragment tests support * SPIR-V: Implement SwizzleAdd, add missing Triangles ExecutionMode for geometry shaders, remove SamplerType field from TextureMeta * Don't pass depth clip state right now (fix decals) Explicitly disabling it is incorrect. OpenGL currently automatically disables based on depth clamp, which is the behaviour if this state is omitted. * Multisampling support * Multisampling: Use resolve if src samples count > dst samples count * Multisampling: We can only resolve for unscaled copies * SPIR-V: Only add FSI exec mode if used. * SPIR-V: Use ConstantComposite for Texture Offset Vector Fixes a bunch of freezes with SPIR-V on AMD hardware, and validation errors. Note: Obviously assumes input offsets are constant, which they currently are. * SPIR-V: Don't OpReturn if we already OpExit'ed Fixes spir-v parse failure and stack smashing in RADV (obviously you still need bolist) * SPIR-V: Only use input attribute type for input attributes Output vertex attributes should always be of type float. * Multithreaded Pipeline Compilation * Address some feedback * Make this 32 * Update topology with GpuAccessorState * Cleanup for merge (note: disables spir-v) * Make more robust to shader compilation failure - Don't freeze when GLSL compilation fails - Background SPIR-V pipeline compile failure results in skipped draws, similar to GLSL compilation failure. * Fix Multisampling * Only update fragment scale count if a vertex texture needs a scale. Fixes a performance regression introduced by texture scaling in the vertex stage where support buffer updates would be very frequent, even at 1x, if any textures were used on the vertex stage. This check doesn't exactly look cheap (a flag in the shader stage would probably be preferred), but it is much cheaper than uploading scales in both vulkan and opengl, so it will do for now. * Use a bitmap to do granular tracking for buffer uploads. This path is only taken if the much faster check of "is the buffer rented at all" is triggered, so it doesn't actually end up costing too much, and the time saved by not ending render passes (and on gpu for not waiting on barriers) is probably helpful. Avoids ending render passes to update buffer data (not all the time) - 140-180 to 35-45 in SMO metro kingdom (these updates are in the UI) - Very variable 60-150(!) to 16-25 in mario kart 8 (these updates are in the UI) As well as allowing more data to be preloaded persistently, this will also allow more data to be loaded in the preload buffer, which should be faster as it doesn't need to insert barriers between draws. (and on tbdr, does not need to flush and reload tile memory) Improves performance in GPU limited scenarios. Should notably improve performance on TBDR gpus. Still a lot more to do here. * Copy query results after RP ends, rather than ending to copy We need to end the render pass to get the data (submit command buffer) anyways... Reduces render passes created in games that use queries. * Rework Query stuff a bit to avoid render pass end Tries to reset returned queries in background when possible, rather than ending the render pass. Still ends render pass when resetting a counter after draws, but maybe that can be solved too. (by just pulling an empty object off the pool?) * Remove unnecessary lines Was for testing * Fix validation error for query reset Need to think of a better way to do this. * SPIR-V: Fix SwizzleAdd and some validation errors * SPIR-V: Implement attribute indexing and StoreAttribute * SPIR-V: Fix TextureSize for MS and Buffer sampler types * Fix relaunch issues * SPIR-V: Implement LogicalExclusiveOr * SPIR-V: Constant buffer indexing support * Ignore unsupported attributes rather than throwing (matches current GLSL behaviour) * SPIR-V: Implement tessellation support * SPIR-V: Geometry shader passthrough support * SPIR-V: Implement StoreShader8/16 and StoreStorage8/16 * SPIR-V: Resolution scale support and fix TextureSample multisample with LOD bug * SPIR-V: Fix field index for scale count * SPIR-V: Fix another case of wrong field index * SPIRV/GLSL: More scaling related fixes * SPIR-V: Fix ImageLoad CompositeExtract component type * SPIR-V: Workaround for Intel FrontFacing bug * Enable SPIR-V backend by default * Allow null samplers (samplers are not required when only using texelFetch to access the texture) * Fix some validation errors related to texel block view usage flag and invalid image barrier base level * Use explicit subgroup size if we can (might fix some block flickering on AMD) * Take componentMask and scissor into account when clearing framebuffer attachments * Add missing barriers around CmdFillBuffer (fixes Monster Hunter Rise flickering on NVIDIA) * Use ClampToEdge for Clamp sampler address mode on Vulkan (fixes Hollow Knight) Clamp is unsupported on Vulkan, but ClampToEdge behaves almost the same. ClampToBorder on the other hand (which was being used before) is pretty different * Shader specialization for new Vulkan required state (fixes remaining alpha test issues, vertex stretching on AMD on Crash Bandicoot, etc) * Check if the subgroup size is supported before passing a explicit size * Only enable ShaderFloat64 if the GPU supports it * We don't need to recompile shaders if alpha test state changed but alpha test is disabled * Enable shader cache on Vulkan and implement MultiplyHighS32/U32 on SPIR-V (missed those before) * Fix pipeline state saving before it is updated. This should fix a few warnings and potential stutters due to bad pipeline states being saved in the cache. You may need to clear your guest cache. * Allow null samplers on OpenGL backend * _unit0Sampler should be set only for binding 0 * Remove unused PipelineConverter format variable (was causing IOR) * Raise textures limit to 64 on Vulkan * No need to pack the shader binaries if shader cache is disabled * Fix backbuffer not being cleared and scissor not being re-enabled on OpenGL * Do not clear unbound framebuffer color attachments * Geometry shader passthrough emulation * Consolidate UpdateDepthMode and GetDepthMode implementation * Fix A1B5G5R5 texture format and support R4G4 on Vulkan * Add barrier before use of some modified images * Report 32 bit query result on AMD windows (smo issue) * Add texture recompression support (disabled for now) It recompresses ASTC textures into BC7, which might reduce VRAM usage significantly on games that uses ASTC textures * Do not report R4G4 format as supported on Vulkan It was causing mario head to become white on Super Mario 64 (???) * Improvements to -1 to 1 depth mode. - Transformation is only applied on the last stage in the vertex pipeline. - Should fix some issues with geometry and tessellation (hopefully) - Reading back FragCoord Z on fragment will transform back to -1 to 1. * Geometry Shader index count from ThreadsPerInputPrimitive Generally fixes SPIR-V emitting too many triangles, may change games in OpenGL * Remove gl_FragDepth scaling This is always 0-1; the other two issues were causing the problems. Fixes regression with Xenoblade. * Add Gl StencilOp enum values to Vulkan * Update guest cache to v1.1 (due to specialization state changes) This will explode your shader cache from earlier vulkan build, but it must be done. :pensive: * Vulkan/SPIR-V support for viewport inverse * Fix typo * Don't create query pools for unsupported query types * Return of the Vector Indexing Bug One day, everyone will get this right. * Check for transform feedback query support Sometimes transform feedback is supported without the query type. * Fix gl_FragCoord.z transformation FragCoord.z is always in 0-1, even when the real depth range is -1 to 1. Turns out the only bug was geo and tess stage outputs. Fixes Pokemon Sword/Shield, possibly others. * Fix Avalonia Rebase Vulkan is currently not available on Avalonia, but the build does work and you can use opengl. * Fix headless build * Add support for BC6 and BC7 decompression, decompress all BC formats if they are not supported by the host * Fix BCn 4/5 conversion, GetTextureTarget BCn 4/5 could generate invalid data when a line's size in bytes was not divisible by 4, which both backends expect. GetTextureTarget was not creating a view with the replacement format. * Fix dependency * Fix inverse viewport transform vector type on SPIR-V * Do not require null descriptors support * If MultiViewport is not supported, do not try to set more than one viewport/scissor * Bounds check on bitmap add. * Flush queries on attachment change rather than program change Occlusion queries are usually used in a depth only pass so the attachments changing is a better indication of the query block ending. Write mask changes are also considered since some games do depth only pass by setting 0 write mask on all the colour targets. * Add support for avalonia (#6) * add avalonia support * only lock around skia flush * addressed review * cleanup * add fallback size if avalonia attempts to render but the window size is 0. read desktop scale after enabling dpi check * fix getting window handle on linux. skip render is size is 0 * Combine non-buffer with buffer image descriptor sets * Support multisample texture copy with automatic resolve on Vulkan * Remove old CompileShader methods from the Vulkan backend * Add minimal pipeline layouts that only contains used bindings They are used by helper shaders, the intention is avoiding needing to recompile the shaders (from GLSL to SPIR-V) if the bindings changes on the translated guest shaders * Pre-compile helper shader as SPIR-V, and some fixes * Remove pre-compiled shaderc binary for Windows as its no longer needed by default * Workaround RADV crash Enabling the descriptor indexing extension, even if it is not used, forces the radv driver to use "bolist". * Use RobustBufferAccess on NVIDIA gpus Avoids the SMO waterfall triangle on older NVIDIA gpus. * Implement GPU selector and expose texture recompression on the UI and config * Fix and enable background compute shader compilation Also disables warnings from shader cache pipeline misses. * Fix error due to missing subpass dependency when Attachment Write -> Shader Read barriers are added * If S8D24 is not supported, use D32FS8 * Ensure all fences are destroyed on dispose * Pre-allocate arrays up front on DescriptorSetUpdater, allows the removal of some checks * Add missing clear layer parameter after rebase * Use selected gpu from config for avalonia (#7) * use configured device * address review * Fix D32S8 copy workaround (AMD) Fixes water in Pokemon Legends Arceus on AMD GPUs. Possibly fixes other things. * Use push descriptors for uniform buffer updates (disabled for now) * Push descriptor support check, buffer redundancy checks Should make push descriptors faster, needs more testing though. * Increase light command buffer pool to 2 command buffers, throw rather than returning invalid cbs * Adjust bindings array sizes * Force submit command buffers if memory in use by its resources is high * Add workaround for AMD GCN cubemap view sins `ImageCreateCubeCompatibleBit` seems to generally break 2D array textures with mipmaps... even if they are eventually aliased as a cubemap with mipmaps. Forcing a copy here works around the issue. This could be used in future if enabling this bit reduces performance on certain GPUs. (mobile class is generally a worry) Currently also enabled on Linux as I don't know if they managed to dodge this bug (someone please tell me). Not enabled on Vega at the moment, but easy to add if the issue is there. * Add mobile, non-RX variants to the GCN regex. Also make sure that the 3 digit ones only include numbers starting with 7 or 8. * Increase image limit per stage from 8 to 16 Xenoblade Chronicles 2 was hiting the limit of 8 * Minor code cleanup * Fix NRE caused by SupportBufferUpdater calling pipeline ClearBuffer * Add gpu selector to Avalonia (#8) * Add gpu selector to avalonia settings * show backend label on window * some fixes * address review * Minor changes to the Avalonia UI * Update graphics window UI and locales. (#9) * Update xaml and update locales * locale updates Did my best here but likely needs to be checked by native speakers, especially the use of ampersands in greek, russian and turkish? * Fix locales with more (?) correct translations. * add separator to render widget * fix spanish and portuguese * Add new IdList, replaces buffer list that could not remove elements and had unbounded growth * Don't crash the settings window if Vulkan is not supported * Fix Actions menu not being clickable on GTK UI after relaunch * Rename VulkanGraphicsDevice to VulkanRenderer and Renderer to OpenGLRenderer * Fix IdList and make it not thread safe * Revert useless OpenGL format table changes * Fix headless project build * List throws ArgumentOutOfRangeException * SPIR-V: Fix tessellation * Increase shader cache version due to tessellation fix * Reduce number of Sync objects created (improves perf in some specific titles) * Fix vulkan validation errors for NPOT compressed upload and GCN workaround. * Add timestamp to the shader cache and force rebuild if host cache is outdated * Prefer Mail box present mode for popups (#11) * Prefer Mail box present mode * fix debug * switch present mode when vsync is toggled * only disable vsync on the main window * SPIR-V: Fix geometry shader input load with transform feedback * BC7 Encoder: Prefer more precision on alpha rather than RGB when alpha is 0 * Fix Avalonia build * Address initial PR feedback * Only set transform feedback outputs on last vertex stage * Address riperiperi PR feedback * Remove outdated comment * Remove unused constructor * Only throw for negative results * Throw for QueueSubmit and other errors No point in delaying the inevitable * Transform feedback decorations inside gl_PerVertex struct breaks the NVIDIA compiler * Fix some resolution scale issues * No need for two UpdateScale calls * Fix comments on SPIR-V generator project * Try to fix shader local memory size On DOOM, a shader is using local memory, but both Low and High size are 0, CRS size is 1536, it seems to store on that region? * Remove RectangleF that is now unused * Fix ImageGather with multiple offsets Needs ImageGatherExtended capability, and must use `ConstantComposite` instead of `CompositeConstruct` * Address PR feedback from jD in all projects except Avalonia * Address most of jD PR feedback on Avalonia * Remove unsafe * Fix VulkanSkiaGpu * move present mode request out of Create Swapchain method * split more parts of create swapchain * addressed reviews * addressed review * Address second batch of jD PR feedback * Fix buffer <-> image copy row length and height alignment AlignUp helper does not support NPOT alignment, and ASTC textures can have NPOT block sizes * Better fix for NPOT alignment issue * Use switch expressions on Vulkan EnumConversion Thanks jD * Fix Avalonia build * Add Vulkan selection prompt on startup * Grammar fixes on Vulkan prompt message * Add missing Vulkan migration flag Co-authored-by: riperiperi <rhy3756547@hotmail.com> Co-authored-by: Emmanuel Hansen <emmausssss@gmail.com> Co-authored-by: MutantAura <44103205+MutantAura@users.noreply.github.com>
2022-07-31 17:26:06 -04:00
private static SpvInstruction GetStorageElemPointer(CodeGenContext context, AstOperation operation)
{
var sbVariable = context.StorageBuffersArray;
var i0 = context.Get(AggregateType.S32, operation.GetSource(0));
var i1 = context.Constant(context.TypeS32(), 0);
var i2 = context.Get(AggregateType.S32, operation.GetSource(1));
return context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeU32()), sbVariable, i0, i1, i2);
}
private static OperationResult GenerateUnary(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction> emitF,
Func<SpvInstruction, SpvInstruction, SpvInstruction> emitI)
{
var source = operation.GetSource(0);
if (operation.Inst.HasFlag(Instruction.FP64))
{
return new OperationResult(AggregateType.FP64, emitF(context.TypeFP64(), context.GetFP64(source)));
}
else if (operation.Inst.HasFlag(Instruction.FP32))
{
return new OperationResult(AggregateType.FP32, emitF(context.TypeFP32(), context.GetFP32(source)));
}
else
{
return new OperationResult(AggregateType.S32, emitI(context.TypeS32(), context.GetS32(source)));
}
}
private static OperationResult GenerateUnaryBool(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction> emitB)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.Bool, emitB(context.TypeBool(), context.Get(AggregateType.Bool, source)));
}
private static OperationResult GenerateUnaryFP32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction> emit)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.FP32, emit(context.TypeFP32(), context.GetFP32(source)));
}
private static OperationResult GenerateUnaryS32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction> emitS)
{
var source = operation.GetSource(0);
return new OperationResult(AggregateType.S32, emitS(context.TypeS32(), context.GetS32(source)));
}
private static OperationResult GenerateBinary(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitF,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitI)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
if (operation.Inst.HasFlag(Instruction.FP64))
{
var result = emitF(context.TypeFP64(), context.GetFP64(src1), context.GetFP64(src2));
context.Decorate(result, Decoration.NoContraction);
return new OperationResult(AggregateType.FP64, result);
}
else if (operation.Inst.HasFlag(Instruction.FP32))
{
var result = emitF(context.TypeFP32(), context.GetFP32(src1), context.GetFP32(src2));
context.Decorate(result, Decoration.NoContraction);
return new OperationResult(AggregateType.FP32, result);
}
else
{
return new OperationResult(AggregateType.S32, emitI(context.TypeS32(), context.GetS32(src1), context.GetS32(src2)));
}
}
private static OperationResult GenerateBinaryBool(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitB)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
return new OperationResult(AggregateType.Bool, emitB(context.TypeBool(), context.Get(AggregateType.Bool, src1), context.Get(AggregateType.Bool, src2)));
}
private static OperationResult GenerateBinaryS32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitS)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
return new OperationResult(AggregateType.S32, emitS(context.TypeS32(), context.GetS32(src1), context.GetS32(src2)));
}
private static OperationResult GenerateBinaryU32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
return new OperationResult(AggregateType.U32, emitU(context.TypeU32(), context.GetU32(src1), context.GetU32(src2)));
}
private static OperationResult GenerateTernary(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitF,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitI)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
if (operation.Inst.HasFlag(Instruction.FP64))
{
var result = emitF(context.TypeFP64(), context.GetFP64(src1), context.GetFP64(src2), context.GetFP64(src3));
context.Decorate(result, Decoration.NoContraction);
return new OperationResult(AggregateType.FP64, result);
}
else if (operation.Inst.HasFlag(Instruction.FP32))
{
var result = emitF(context.TypeFP32(), context.GetFP32(src1), context.GetFP32(src2), context.GetFP32(src3));
context.Decorate(result, Decoration.NoContraction);
return new OperationResult(AggregateType.FP32, result);
}
else
{
return new OperationResult(AggregateType.S32, emitI(context.TypeS32(), context.GetS32(src1), context.GetS32(src2), context.GetS32(src3)));
}
}
private static OperationResult GenerateTernaryS32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitS)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
return new OperationResult(AggregateType.S32, emitS(
context.TypeS32(),
context.GetS32(src1),
context.GetS32(src2),
context.GetS32(src3)));
}
private static OperationResult GenerateTernaryU32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
return new OperationResult(AggregateType.U32, emitU(
context.TypeU32(),
context.GetU32(src1),
context.GetU32(src2),
context.GetU32(src3)));
}
private static OperationResult GenerateQuaternaryS32(
CodeGenContext context,
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitS)
{
var src1 = operation.GetSource(0);
var src2 = operation.GetSource(1);
var src3 = operation.GetSource(2);
var src4 = operation.GetSource(3);
return new OperationResult(AggregateType.S32, emitS(
context.TypeS32(),
context.GetS32(src1),
context.GetS32(src2),
context.GetS32(src3),
context.GetS32(src4)));
}
}
}