mirrors/ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

New shader cache implementation (#3194)

Browse source 
* New shader cache implementation

* Remove some debug code

* Take transform feedback varying count into account

* Create shader cache directory if it does not exist + fragment output map related fixes

* Remove debug code

* Only check texture descriptors if the constant buffer is bound

* Also check CPU VA on GetSpanMapped

* Remove more unused code and move cache related code

* XML docs + remove more unused methods

* Better codegen for TransformFeedbackDescriptor.AsSpan

* Support migration from old cache format, remove more unused code

Shader cache rebuild now also rewrites the shared toc and data files

* Fix migration error with BRX shaders

* Add a limit to the async translation queue

 Avoid async translation threads not being able to keep up and the queue growing very large

* Re-create specialization state on recompile

This might be required if a new version of the shader translator requires more or less state, or if there is a bug related to the GPU state access

* Make shader cache more error resilient

* Add some missing XML docs and move GpuAccessor docs to the interface/use inheritdoc

* Address early PR feedback

* Fix rebase

* Remove IRenderer.CompileShader and IShader interface, replace with new ShaderSource struct passed to CreateProgram directly

* Handle some missing exceptions

* Make shader cache purge delete both old and new shader caches

* Register textures on new specialization state

* Translate and compile shaders in forward order (eliminates diffs due to different binding numbers)

* Limit in-flight shader compilation to the maximum number of compilation threads

* Replace ParallelDiskCacheLoader state changed event with a callback function

* Better handling for invalid constant buffer 1 data length

* Do not create the old cache directory structure if the old cache does not exist

* Constant buffer use should be per-stage. This change will invalidate existing new caches (file format version was incremented)

* Replace rectangle texture with just coordinate normalization

* Skip incompatible shaders that are missing texture information, instead of crashing

This is required if we, for example, support new texture instruction to the shader translator, and then they allow access to textures that were not accessed before. In this scenario, the old cache entry is no longer usable

* Fix coordinates normalization on cubemap textures

* Check if title ID is null before combining shader cache path

* More robust constant buffer address validation on spec state

* More robust constant buffer address validation on spec state (2)

* Regenerate shader cache with one stream, rather than one per shader.

* Only create shader cache directory during initialization

* Logging improvements

* Proper shader program disposal

* PR feedback, and add a comment on serialized structs

* XML docs for RegisterTexture

Co-authored-by: riperiperi <rhy3756547@hotmail.com>
This commit is contained in:
gdkchan 2022-04-10 10:49:44 -03:00 • committed by GitHub
parent 26a881176e
commit 43ebd7a9bb
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
81 changed files with 6421 additions and 2406 deletions
Download patch file Download diff file Expand all files Collapse all files

331
Ryujinx.Graphics.Gpu/Shader/Cache/CacheHelper.cs
View file

@ -2,11 +2,8 @@
using Ryujinx.Common;
using Ryujinx.Common.Configuration;
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Memory;
using Ryujinx.Graphics.Gpu.Shader.Cache.Definition;
using Ryujinx.Graphics.Shader;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
using System.IO;
@ -20,70 +17,6 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache
/// </summary>
static class CacheHelper
{
/// <summary>
/// Try to read the manifest header from a given file path.
/// </summary>
/// <param name="manifestPath">The path to the manifest file</param>
/// <param name="header">The manifest header read</param>
/// <returns>Return true if the manifest header was read</returns>
public static bool TryReadManifestHeader(string manifestPath, out CacheManifestHeader header)
{
header = default;
if (File.Exists(manifestPath))
{
Memory<byte> rawManifest = File.ReadAllBytes(manifestPath);
if (MemoryMarshal.TryRead(rawManifest.Span, out header))
{
return true;
}
}
return false;
}
/// <summary>
/// Try to read the manifest from a given file path.
/// </summary>
/// <param name="manifestPath">The path to the manifest file</param>
/// <param name="graphicsApi">The graphics api used by the cache</param>
/// <param name="hashType">The hash type of the cache</param>
/// <param name="header">The manifest header read</param>
/// <param name="entries">The entries read from the cache manifest</param>
/// <returns>Return true if the manifest was read</returns>
public static bool TryReadManifestFile(string manifestPath, CacheGraphicsApi graphicsApi, CacheHashType hashType, out CacheManifestHeader header, out HashSet<Hash128> entries)
{
header = default;
entries = new HashSet<Hash128>();
if (File.Exists(manifestPath))
{
Memory<byte> rawManifest = File.ReadAllBytes(manifestPath);
if (MemoryMarshal.TryRead(rawManifest.Span, out header))
{
Memory<byte> hashTableRaw = rawManifest.Slice(Unsafe.SizeOf<CacheManifestHeader>());
bool isValid = header.IsValid(graphicsApi, hashType, hashTableRaw.Span);
if (isValid)
{
ReadOnlySpan<Hash128> hashTable = MemoryMarshal.Cast<byte, Hash128>(hashTableRaw.Span);
foreach (Hash128 hash in hashTable)
{
entries.Add(hash);
}
}
return isValid;
}
}
return false;
}
/// <summary>
/// Compute a cache manifest from runtime data.
/// </summary>
@ -246,82 +179,23 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache
return null;
}
/// <summary>
/// Compute the guest program code for usage while dumping to disk or hash.
/// </summary>
/// <param name="cachedShaderEntries">The guest shader entries to use</param>
/// <param name="tfd">The transform feedback descriptors</param>
/// <param name="forHashCompute">Used to determine if the guest program code is generated for hashing</param>
/// <returns>The guest program code for usage while dumping to disk or hash</returns>
private static byte[] ComputeGuestProgramCode(ReadOnlySpan<GuestShaderCacheEntry> cachedShaderEntries, TransformFeedbackDescriptor[] tfd, bool forHashCompute = false)
{
using (MemoryStream stream = new MemoryStream())
{
BinaryWriter writer = new BinaryWriter(stream);
foreach (GuestShaderCacheEntry cachedShaderEntry in cachedShaderEntries)
{
if (cachedShaderEntry != null)
{
// Code (and Code A if present)
stream.Write(cachedShaderEntry.Code);
if (forHashCompute)
{
// Guest GPU accessor header (only write this for hashes, already present in the header for dumps)
writer.WriteStruct(cachedShaderEntry.Header.GpuAccessorHeader);
}
// Texture descriptors
foreach (GuestTextureDescriptor textureDescriptor in cachedShaderEntry.TextureDescriptors.Values)
{
writer.WriteStruct(textureDescriptor);
}
}
}
// Transform feedback
if (tfd != null)
{
foreach (TransformFeedbackDescriptor transform in tfd)
{
writer.WriteStruct(new GuestShaderCacheTransformFeedbackHeader(transform.BufferIndex, transform.Stride, transform.VaryingLocations.Length));
writer.Write(transform.VaryingLocations);
}
}
return stream.ToArray();
}
}
/// <summary>
/// Compute a guest hash from shader entries.
/// </summary>
/// <param name="cachedShaderEntries">The guest shader entries to use</param>
/// <param name="tfd">The optional transform feedback descriptors</param>
/// <returns>A guest hash from shader entries</returns>
public static Hash128 ComputeGuestHashFromCache(ReadOnlySpan<GuestShaderCacheEntry> cachedShaderEntries, TransformFeedbackDescriptor[] tfd = null)
{
return XXHash128.ComputeHash(ComputeGuestProgramCode(cachedShaderEntries, tfd, true));
}
/// <summary>
/// Read transform feedback descriptors from guest.
/// </summary>
/// <param name="data">The raw guest transform feedback descriptors</param>
/// <param name="header">The guest shader program header</param>
/// <returns>The transform feedback descriptors read from guest</returns>
public static TransformFeedbackDescriptor[] ReadTransformFeedbackInformation(ref ReadOnlySpan<byte> data, GuestShaderCacheHeader header)
public static TransformFeedbackDescriptorOld[] ReadTransformFeedbackInformation(ref ReadOnlySpan<byte> data, GuestShaderCacheHeader header)
{
if (header.TransformFeedbackCount != 0)
{
TransformFeedbackDescriptor[] result = new TransformFeedbackDescriptor[header.TransformFeedbackCount];
TransformFeedbackDescriptorOld[] result = new TransformFeedbackDescriptorOld[header.TransformFeedbackCount];
for (int i = 0; i < result.Length; i++)
{
GuestShaderCacheTransformFeedbackHeader feedbackHeader = MemoryMarshal.Read<GuestShaderCacheTransformFeedbackHeader>(data);
result[i] = new TransformFeedbackDescriptor(feedbackHeader.BufferIndex, feedbackHeader.Stride, data.Slice(Unsafe.SizeOf<GuestShaderCacheTransformFeedbackHeader>(), feedbackHeader.VaryingLocationsLength).ToArray());
result[i] = new TransformFeedbackDescriptorOld(feedbackHeader.BufferIndex, feedbackHeader.Stride, data.Slice(Unsafe.SizeOf<GuestShaderCacheTransformFeedbackHeader>(), feedbackHeader.VaryingLocationsLength).ToArray());
data = data.Slice(Unsafe.SizeOf<GuestShaderCacheTransformFeedbackHeader>() + feedbackHeader.VaryingLocationsLength);
}
@ -332,205 +206,6 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache
return null;
}
/// <summary>
/// Builds gpu state flags using information from the given gpu accessor.
/// </summary>
/// <param name="gpuAccessor">The gpu accessor</param>
/// <returns>The gpu state flags</returns>
private static GuestGpuStateFlags GetGpuStateFlags(IGpuAccessor gpuAccessor)
{
GuestGpuStateFlags flags = 0;
if (gpuAccessor.QueryEarlyZForce())
{
flags |= GuestGpuStateFlags.EarlyZForce;
}
return flags;
}
/// <summary>
/// Packs the tessellation parameters from the gpu accessor.
/// </summary>
/// <param name="gpuAccessor">The gpu accessor</param>
/// <returns>The packed tessellation parameters</returns>
private static byte GetTessellationModePacked(IGpuAccessor gpuAccessor)
{
byte value;
value = (byte)((int)gpuAccessor.QueryTessPatchType() & 3);
value |= (byte)(((int)gpuAccessor.QueryTessSpacing() & 3) << 2);
if (gpuAccessor.QueryTessCw())
{
value |= 0x10;
}
return value;
}
/// <summary>
/// Create a new instance of <see cref="GuestGpuAccessorHeader"/> from an gpu accessor.
/// </summary>
/// <param name="gpuAccessor">The gpu accessor</param>
/// <returns>A new instance of <see cref="GuestGpuAccessorHeader"/></returns>
public static GuestGpuAccessorHeader CreateGuestGpuAccessorCache(IGpuAccessor gpuAccessor)
{
return new GuestGpuAccessorHeader
{
ComputeLocalSizeX = gpuAccessor.QueryComputeLocalSizeX(),
ComputeLocalSizeY = gpuAccessor.QueryComputeLocalSizeY(),
ComputeLocalSizeZ = gpuAccessor.QueryComputeLocalSizeZ(),
ComputeLocalMemorySize = gpuAccessor.QueryComputeLocalMemorySize(),
ComputeSharedMemorySize = gpuAccessor.QueryComputeSharedMemorySize(),
PrimitiveTopology = gpuAccessor.QueryPrimitiveTopology(),
TessellationModePacked = GetTessellationModePacked(gpuAccessor),
StateFlags = GetGpuStateFlags(gpuAccessor)
};
}
/// <summary>
/// Create guest shader cache entries from the runtime contexts.
/// </summary>
/// <param name="channel">The GPU channel in use</param>
/// <param name="shaderContexts">The runtime contexts</param>
/// <returns>Guest shader cahe entries from the runtime contexts</returns>
public static GuestShaderCacheEntry[] CreateShaderCacheEntries(GpuChannel channel, ReadOnlySpan<TranslatorContext> shaderContexts)
{
MemoryManager memoryManager = channel.MemoryManager;
int startIndex = shaderContexts.Length > 1 ? 1 : 0;
GuestShaderCacheEntry[] entries = new GuestShaderCacheEntry[shaderContexts.Length - startIndex];
for (int i = startIndex; i < shaderContexts.Length; i++)
{
TranslatorContext context = shaderContexts[i];
if (context == null)
{
continue;
}
GpuAccessor gpuAccessor = context.GpuAccessor as GpuAccessor;
ulong cb1DataAddress;
int cb1DataSize = gpuAccessor?.Cb1DataSize ?? 0;
if (context.Stage == ShaderStage.Compute)
{
cb1DataAddress = channel.BufferManager.GetComputeUniformBufferAddress(1);
}
else
{
int stageIndex = context.Stage switch
{
ShaderStage.TessellationControl => 1,
ShaderStage.TessellationEvaluation => 2,
ShaderStage.Geometry => 3,
ShaderStage.Fragment => 4,
_ => 0
};
cb1DataAddress = channel.BufferManager.GetGraphicsUniformBufferAddress(stageIndex, 1);
}
int size = context.Size;
TranslatorContext translatorContext2 = i == 1 ? shaderContexts[0] : null;
int sizeA = translatorContext2 != null ? translatorContext2.Size : 0;
byte[] code = new byte[size + cb1DataSize + sizeA];
memoryManager.GetSpan(context.Address, size).CopyTo(code);
if (cb1DataAddress != 0 && cb1DataSize != 0)
{
memoryManager.Physical.GetSpan(cb1DataAddress, cb1DataSize).CopyTo(code.AsSpan(size, cb1DataSize));
}
if (translatorContext2 != null)
{
memoryManager.GetSpan(translatorContext2.Address, sizeA).CopyTo(code.AsSpan(size + cb1DataSize, sizeA));
}
GuestGpuAccessorHeader gpuAccessorHeader = CreateGuestGpuAccessorCache(context.GpuAccessor);
if (gpuAccessor != null)
{
gpuAccessorHeader.TextureDescriptorCount = context.TextureHandlesForCache.Count;
}
GuestShaderCacheEntryHeader header = new GuestShaderCacheEntryHeader(
context.Stage,
size + cb1DataSize,
sizeA,
cb1DataSize,
gpuAccessorHeader);
GuestShaderCacheEntry entry = new GuestShaderCacheEntry(header, code);
if (gpuAccessor != null)
{
foreach (int textureHandle in context.TextureHandlesForCache)
{
GuestTextureDescriptor textureDescriptor = ((Image.TextureDescriptor)gpuAccessor.GetTextureDescriptor(textureHandle, -1)).ToCache();
textureDescriptor.Handle = (uint)textureHandle;
entry.TextureDescriptors.Add(textureHandle, textureDescriptor);
}
}
entries[i - startIndex] = entry;
}
return entries;
}
/// <summary>
/// Create a guest shader program.
/// </summary>
/// <param name="shaderCacheEntries">The entries composing the guest program dump</param>
/// <param name="tfd">The transform feedback descriptors in use</param>
/// <returns>The resulting guest shader program</returns>
public static byte[] CreateGuestProgramDump(GuestShaderCacheEntry[] shaderCacheEntries, TransformFeedbackDescriptor[] tfd = null)
{
using (MemoryStream resultStream = new MemoryStream())
{
BinaryWriter resultStreamWriter = new BinaryWriter(resultStream);
byte transformFeedbackCount = 0;
if (tfd != null)
{
transformFeedbackCount = (byte)tfd.Length;
}
// Header
resultStreamWriter.WriteStruct(new GuestShaderCacheHeader((byte)shaderCacheEntries.Length, transformFeedbackCount));
// Write all entries header
foreach (GuestShaderCacheEntry entry in shaderCacheEntries)
{
if (entry == null)
{
resultStreamWriter.WriteStruct(new GuestShaderCacheEntryHeader());
}
else
{
resultStreamWriter.WriteStruct(entry.Header);
}
}
// Finally, write all program code and all transform feedback information.
resultStreamWriter.Write(ComputeGuestProgramCode(shaderCacheEntries, tfd));
return resultStream.ToArray();
}
}
/// <summary>
/// Save temporary files not in archive.
/// </summary>

2
Ryujinx.Graphics.Gpu/Shader/Cache/CacheManager.cs
View file

@ -47,8 +47,6 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache
string baseCacheDirectory = CacheHelper.GetBaseCacheDirectory(titleId);
CacheMigration.Run(baseCacheDirectory, graphicsApi, hashType, shaderProvider);
_guestProgramCache = new CacheCollection(baseCacheDirectory, _hashType, CacheGraphicsApi.Guest, "", "program", GuestCacheVersion);
_hostProgramCache = new CacheCollection(baseCacheDirectory, _hashType, _graphicsApi, _shaderProvider, "host", shaderCodeGenVersion);
}

175
Ryujinx.Graphics.Gpu/Shader/Cache/CacheMigration.cs
View file

@ -1,175 +0,0 @@
using ICSharpCode.SharpZipLib.Zip;
using Ryujinx.Common;
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Shader.Cache.Definition;
using System;
using System.Collections.Generic;
using System.IO;
namespace Ryujinx.Graphics.Gpu.Shader.Cache
{
/// <summary>
/// Class handling shader cache migrations.
/// </summary>
static class CacheMigration
{
/// <summary>
/// Check if the given cache version need to recompute its hash.
/// </summary>
/// <param name="version">The version in use</param>
/// <param name="newVersion">The new version after migration</param>
/// <returns>True if a hash recompute is needed</returns>
public static bool NeedHashRecompute(ulong version, out ulong newVersion)
{
const ulong TargetBrokenVersion = 1717;
const ulong TargetFixedVersion = 1759;
newVersion = TargetFixedVersion;
if (version == TargetBrokenVersion)
{
return true;
}
return false;
}
private class StreamZipEntryDataSource : IStaticDataSource
{
private readonly ZipFile Archive;
private readonly ZipEntry Entry;
public StreamZipEntryDataSource(ZipFile archive, ZipEntry entry)
{
Archive = archive;
Entry = entry;
}
public Stream GetSource()
{
return Archive.GetInputStream(Entry);
}
}
/// <summary>
/// Move a file with the name of a given hash to another in the cache archive.
/// </summary>
/// <param name="archive">The archive in use</param>
/// <param name="oldKey">The old key</param>
/// <param name="newKey">The new key</param>
private static void MoveEntry(ZipFile archive, Hash128 oldKey, Hash128 newKey)
{
ZipEntry oldGuestEntry = archive.GetEntry($"{oldKey}");
if (oldGuestEntry != null)
{
archive.Add(new StreamZipEntryDataSource(archive, oldGuestEntry), $"{newKey}", CompressionMethod.Deflated);
archive.Delete(oldGuestEntry);
}
}
/// <summary>
/// Recompute all the hashes of a given cache.
/// </summary>
/// <param name="guestBaseCacheDirectory">The guest cache directory path</param>
/// <param name="hostBaseCacheDirectory">The host cache directory path</param>
/// <param name="graphicsApi">The graphics api in use</param>
/// <param name="hashType">The hash type in use</param>
/// <param name="newVersion">The version to write in the host and guest manifest after migration</param>
private static void RecomputeHashes(string guestBaseCacheDirectory, string hostBaseCacheDirectory, CacheGraphicsApi graphicsApi, CacheHashType hashType, ulong newVersion)
{
string guestManifestPath = CacheHelper.GetManifestPath(guestBaseCacheDirectory);
string hostManifestPath = CacheHelper.GetManifestPath(hostBaseCacheDirectory);
if (CacheHelper.TryReadManifestFile(guestManifestPath, CacheGraphicsApi.Guest, hashType, out _, out HashSet<Hash128> guestEntries))
{
CacheHelper.TryReadManifestFile(hostManifestPath, graphicsApi, hashType, out _, out HashSet<Hash128> hostEntries);
Logger.Info?.Print(LogClass.Gpu, "Shader cache hashes need to be recomputed, performing migration...");
string guestArchivePath = CacheHelper.GetArchivePath(guestBaseCacheDirectory);
string hostArchivePath = CacheHelper.GetArchivePath(hostBaseCacheDirectory);
ZipFile guestArchive = new ZipFile(File.Open(guestArchivePath, FileMode.OpenOrCreate, FileAccess.ReadWrite, FileShare.None));
ZipFile hostArchive = new ZipFile(File.Open(hostArchivePath, FileMode.OpenOrCreate, FileAccess.ReadWrite, FileShare.None));
CacheHelper.EnsureArchiveUpToDate(guestBaseCacheDirectory, guestArchive, guestEntries);
CacheHelper.EnsureArchiveUpToDate(hostBaseCacheDirectory, hostArchive, hostEntries);
int programIndex = 0;
HashSet<Hash128> newEntries = new HashSet<Hash128>();
foreach (Hash128 oldHash in guestEntries)
{
byte[] guestProgram = CacheHelper.ReadFromArchive(guestArchive, oldHash);
Logger.Info?.Print(LogClass.Gpu, $"Migrating shader {oldHash} ({programIndex + 1} / {guestEntries.Count})");
if (guestProgram != null)
{
ReadOnlySpan<byte> guestProgramReadOnlySpan = guestProgram;
ReadOnlySpan<GuestShaderCacheEntry> cachedShaderEntries = GuestShaderCacheEntry.Parse(ref guestProgramReadOnlySpan, out GuestShaderCacheHeader fileHeader);
TransformFeedbackDescriptor[] tfd = CacheHelper.ReadTransformFeedbackInformation(ref guestProgramReadOnlySpan, fileHeader);
Hash128 newHash = CacheHelper.ComputeGuestHashFromCache(cachedShaderEntries, tfd);
if (newHash != oldHash)
{
MoveEntry(guestArchive, oldHash, newHash);
MoveEntry(hostArchive, oldHash, newHash);
}
else
{
Logger.Warning?.Print(LogClass.Gpu, $"Same hashes for shader {oldHash}");
}
newEntries.Add(newHash);
}
programIndex++;
}
byte[] newGuestManifestContent = CacheHelper.ComputeManifest(newVersion, CacheGraphicsApi.Guest, hashType, newEntries);
byte[] newHostManifestContent = CacheHelper.ComputeManifest(newVersion, graphicsApi, hashType, newEntries);
File.WriteAllBytes(guestManifestPath, newGuestManifestContent);
File.WriteAllBytes(hostManifestPath, newHostManifestContent);
guestArchive.CommitUpdate();
hostArchive.CommitUpdate();
guestArchive.Close();
hostArchive.Close();
}
}
/// <summary>
/// Check and run cache migration if needed.
/// </summary>
/// <param name="baseCacheDirectory">The base path of the cache</param>
/// <param name="graphicsApi">The graphics api in use</param>
/// <param name="hashType">The hash type in use</param>
/// <param name="shaderProvider">The shader provider name of the cache</param>
public static void Run(string baseCacheDirectory, CacheGraphicsApi graphicsApi, CacheHashType hashType, string shaderProvider)
{
string guestBaseCacheDirectory = CacheHelper.GenerateCachePath(baseCacheDirectory, CacheGraphicsApi.Guest, "", "program");
string hostBaseCacheDirectory = CacheHelper.GenerateCachePath(baseCacheDirectory, graphicsApi, shaderProvider, "host");
string guestArchivePath = CacheHelper.GetArchivePath(guestBaseCacheDirectory);
string hostArchivePath = CacheHelper.GetArchivePath(hostBaseCacheDirectory);
bool isReadOnly = CacheHelper.IsArchiveReadOnly(guestArchivePath) || CacheHelper.IsArchiveReadOnly(hostArchivePath);
if (!isReadOnly && CacheHelper.TryReadManifestHeader(CacheHelper.GetManifestPath(guestBaseCacheDirectory), out CacheManifestHeader header))
{
if (NeedHashRecompute(header.Version, out ulong newVersion))
{
RecomputeHashes(guestBaseCacheDirectory, hostBaseCacheDirectory, graphicsApi, hashType, newVersion);
}
}
}
}
}

3
Ryujinx.Graphics.Gpu/Shader/Cache/Definition/HostShaderCacheEntry.cs
View file

@ -96,6 +96,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache.Definition
SBuffers,
Textures,
Images,
default,
Header.UseFlags.HasFlag(UseFlags.InstanceId),
Header.UseFlags.HasFlag(UseFlags.RtLayer),
Header.ClipDistancesWritten,
@ -160,7 +161,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.Cache.Definition
/// <param name="programCode">The host shader program</param>
/// <param name="codeHolders">The shaders code holder</param>
/// <returns>Raw data of a new host shader cache file</returns>
internal static byte[] Create(ReadOnlySpan<byte> programCode, ShaderCodeHolder[] codeHolders)
internal static byte[] Create(ReadOnlySpan<byte> programCode, CachedShaderStage[] codeHolders)
{
HostShaderCacheHeader header = new HostShaderCacheHeader((byte)codeHolders.Length, programCode.Length);

255
Ryujinx.Graphics.Gpu/Shader/Cache/Migration.cs Normal file
View file

@ -0,0 +1,255 @@
using Ryujinx.Common;
using Ryujinx.Common.Logging;
using Ryujinx.Common.Memory;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Engine.Threed;
using Ryujinx.Graphics.Gpu.Shader.Cache.Definition;
using Ryujinx.Graphics.Gpu.Shader.DiskCache;
using Ryujinx.Graphics.Shader;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
using System.Runtime.InteropServices;
namespace Ryujinx.Graphics.Gpu.Shader.Cache
{
/// <summary>
/// Class handling shader cache migrations.
/// </summary>
static class Migration
{
// Last codegen version before the migration to the new cache.
private const ulong ShaderCodeGenVersion = 3054;
/// <summary>
/// Migrates from the old cache format to the new one.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="hostStorage">Disk cache host storage (used to create the new shader files)</param>
/// <returns>Number of migrated shaders</returns>
public static int MigrateFromLegacyCache(GpuContext context, DiskCacheHostStorage hostStorage)
{
string baseCacheDirectory = CacheHelper.GetBaseCacheDirectory(GraphicsConfig.TitleId);
string cacheDirectory = CacheHelper.GenerateCachePath(baseCacheDirectory, CacheGraphicsApi.Guest, "", "program");
// If the directory does not exist, we have no old cache.
// Exist early as the CacheManager constructor will create the directories.
if (!Directory.Exists(cacheDirectory))
{
return 0;
}
if (GraphicsConfig.EnableShaderCache && GraphicsConfig.TitleId != null)
{
CacheManager cacheManager = new CacheManager(CacheGraphicsApi.OpenGL, CacheHashType.XxHash128, "glsl", GraphicsConfig.TitleId, ShaderCodeGenVersion);
bool isReadOnly = cacheManager.IsReadOnly;
HashSet<Hash128> invalidEntries = null;
if (isReadOnly)
{
Logger.Warning?.Print(LogClass.Gpu, "Loading shader cache in read-only mode (cache in use by another program!)");
}
else
{
invalidEntries = new HashSet<Hash128>();
}
ReadOnlySpan<Hash128> guestProgramList = cacheManager.GetGuestProgramList();
for (int programIndex = 0; programIndex < guestProgramList.Length; programIndex++)
{
Hash128 key = guestProgramList[programIndex];
byte[] guestProgram = cacheManager.GetGuestProgramByHash(ref key);
if (guestProgram == null)
{
Logger.Error?.Print(LogClass.Gpu, $"Ignoring orphan shader hash {key} in cache (is the cache incomplete?)");
continue;
}
ReadOnlySpan<byte> guestProgramReadOnlySpan = guestProgram;
ReadOnlySpan<GuestShaderCacheEntry> cachedShaderEntries = GuestShaderCacheEntry.Parse(ref guestProgramReadOnlySpan, out GuestShaderCacheHeader fileHeader);
if (cachedShaderEntries[0].Header.Stage == ShaderStage.Compute)
{
Debug.Assert(cachedShaderEntries.Length == 1);
GuestShaderCacheEntry entry = cachedShaderEntries[0];
byte[] code = entry.Code.AsSpan(0, entry.Header.Size - entry.Header.Cb1DataSize).ToArray();
Span<byte> codeSpan = entry.Code;
byte[] cb1Data = codeSpan.Slice(codeSpan.Length - entry.Header.Cb1DataSize).ToArray();
ShaderProgramInfo info = new ShaderProgramInfo(
Array.Empty<BufferDescriptor>(),
Array.Empty<BufferDescriptor>(),
Array.Empty<TextureDescriptor>(),
Array.Empty<TextureDescriptor>(),
ShaderStage.Compute,
false,
false,
0,
0);
GpuChannelComputeState computeState = new GpuChannelComputeState(
entry.Header.GpuAccessorHeader.ComputeLocalSizeX,
entry.Header.GpuAccessorHeader.ComputeLocalSizeY,
entry.Header.GpuAccessorHeader.ComputeLocalSizeZ,
entry.Header.GpuAccessorHeader.ComputeLocalMemorySize,
entry.Header.GpuAccessorHeader.ComputeSharedMemorySize);
ShaderSpecializationState specState = new ShaderSpecializationState(computeState);
foreach (var td in entry.TextureDescriptors)
{
var handle = td.Key;
var data = td.Value;
specState.RegisterTexture(
0,
handle,
-1,
data.UnpackFormat(),
data.UnpackSrgb(),
data.UnpackTextureTarget(),
data.UnpackTextureCoordNormalized());
}
CachedShaderStage shader = new CachedShaderStage(info, code, cb1Data);
CachedShaderProgram program = new CachedShaderProgram(null, specState, shader);
hostStorage.AddShader(context, program, ReadOnlySpan<byte>.Empty);
}
else
{
Debug.Assert(cachedShaderEntries.Length == Constants.ShaderStages);
CachedShaderStage[] shaders = new CachedShaderStage[Constants.ShaderStages + 1];
List<ShaderProgram> shaderPrograms = new List<ShaderProgram>();
TransformFeedbackDescriptorOld[] tfd = CacheHelper.ReadTransformFeedbackInformation(ref guestProgramReadOnlySpan, fileHeader);
GuestShaderCacheEntry[] entries = cachedShaderEntries.ToArray();
GuestGpuAccessorHeader accessorHeader = entries[0].Header.GpuAccessorHeader;
TessMode tessMode = new TessMode();
int tessPatchType = accessorHeader.TessellationModePacked & 3;
int tessSpacing = (accessorHeader.TessellationModePacked >> 2) & 3;
bool tessCw = (accessorHeader.TessellationModePacked & 0x10) != 0;
tessMode.Packed = (uint)tessPatchType;
tessMode.Packed |= (uint)(tessSpacing << 4);
if (tessCw)
{
tessMode.Packed |= 0x100;
}
PrimitiveTopology topology = accessorHeader.PrimitiveTopology switch
{
InputTopology.Lines => PrimitiveTopology.Lines,
InputTopology.LinesAdjacency => PrimitiveTopology.LinesAdjacency,
InputTopology.Triangles => PrimitiveTopology.Triangles,
InputTopology.TrianglesAdjacency => PrimitiveTopology.TrianglesAdjacency,
_ => PrimitiveTopology.Points
};
GpuChannelGraphicsState graphicsState = new GpuChannelGraphicsState(
accessorHeader.StateFlags.HasFlag(GuestGpuStateFlags.EarlyZForce),
topology,
tessMode);
TransformFeedbackDescriptor[] tfdNew = null;
if (tfd != null)
{
tfdNew = new TransformFeedbackDescriptor[tfd.Length];
for (int tfIndex = 0; tfIndex < tfd.Length; tfIndex++)
{
Array32<uint> varyingLocations = new Array32<uint>();
Span<byte> varyingLocationsSpan = MemoryMarshal.Cast<uint, byte>(varyingLocations.ToSpan());
tfd[tfIndex].VaryingLocations.CopyTo(varyingLocationsSpan.Slice(0, tfd[tfIndex].VaryingLocations.Length));
tfdNew[tfIndex] = new TransformFeedbackDescriptor(
tfd[tfIndex].BufferIndex,
tfd[tfIndex].Stride,
tfd[tfIndex].VaryingLocations.Length,
ref varyingLocations);
}
}
ShaderSpecializationState specState = new ShaderSpecializationState(graphicsState, tfdNew);
for (int i = 0; i < entries.Length; i++)
{
GuestShaderCacheEntry entry = entries[i];
if (entry == null)
{
continue;
}
ShaderProgramInfo info = new ShaderProgramInfo(
Array.Empty<BufferDescriptor>(),
Array.Empty<BufferDescriptor>(),
Array.Empty<TextureDescriptor>(),
Array.Empty<TextureDescriptor>(),
(ShaderStage)(i + 1),
false,
false,
0,
0);
// NOTE: Vertex B comes first in the shader cache.
byte[] code = entry.Code.AsSpan(0, entry.Header.Size - entry.Header.Cb1DataSize).ToArray();
byte[] code2 = entry.Header.SizeA != 0 ? entry.Code.AsSpan(entry.Header.Size, entry.Header.SizeA).ToArray() : null;
Span<byte> codeSpan = entry.Code;
byte[] cb1Data = codeSpan.Slice(codeSpan.Length - entry.Header.Cb1DataSize).ToArray();
shaders[i + 1] = new CachedShaderStage(info, code, cb1Data);
if (code2 != null)
{
shaders[0] = new CachedShaderStage(null, code2, cb1Data);
}
foreach (var td in entry.TextureDescriptors)
{
var handle = td.Key;
var data = td.Value;
specState.RegisterTexture(
i,
handle,
-1,
data.UnpackFormat(),
data.UnpackSrgb(),
data.UnpackTextureTarget(),
data.UnpackTextureCoordNormalized());
}
}
CachedShaderProgram program = new CachedShaderProgram(null, specState, shaders);
hostStorage.AddShader(context, program, ReadOnlySpan<byte>.Empty);
}
}
return guestProgramList.Length;
}
return 0;
}
}
}

19
Ryujinx.Graphics.Gpu/Shader/Cache/TransformFeedbackDescriptorOld.cs Normal file
View file

@ -0,0 +1,19 @@
using System;
namespace Ryujinx.Graphics.Gpu.Shader.Cache
{
struct TransformFeedbackDescriptorOld
{
public int BufferIndex { get; }
public int Stride { get; }
public byte[] VaryingLocations { get; }
public TransformFeedbackDescriptorOld(int bufferIndex, int stride, byte[] varyingLocations)
{
BufferIndex = bufferIndex;
Stride = stride;
VaryingLocations = varyingLocations ?? throw new ArgumentNullException(nameof(varyingLocations));
}
}
}