Implement Load/Store Local/Shared and Atomic shared using new instructions (#5241)

* Implement Load/Store Local/Shared and Atomic shared using new instructions

* Remove now unused code

* Fix base offset register overwrite

* Fix missing storage buffer set index when generating GLSL for Vulkan

* Shader cache version bump

* Remove more unused code

* Some PR feedback
This commit is contained in:
gdkchan 2023-06-15 17:31:53 -03:00 committed by GitHub
parent 32d21ddf17
commit f92921a6d1
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
30 changed files with 475 additions and 567 deletions

View file

@ -71,40 +71,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.AppendLine($"const int {DefaultNames.UndefinedName} = 0;");
context.AppendLine();
if (context.Config.Stage == ShaderStage.Compute)
{
int localMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeLocalMemorySize(), 4);
if (localMemorySize != 0)
{
string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
context.AppendLine();
}
int sharedMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeSharedMemorySize(), 4);
if (sharedMemorySize != 0)
{
string sharedMemorySizeStr = NumberFormatter.FormatInt(sharedMemorySize);
context.AppendLine($"shared uint {DefaultNames.SharedMemoryName}[{sharedMemorySizeStr}];");
context.AppendLine();
}
}
else if (context.Config.LocalMemorySize != 0)
{
int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
context.AppendLine();
}
DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
DeclareStorageBuffers(context, context.Config.Properties.StorageBuffers.Values);
DeclareMemories(context, context.Config.Properties.LocalMemories.Values, isShared: false);
DeclareMemories(context, context.Config.Properties.SharedMemories.Values, isShared: true);
var textureDescriptors = context.Config.GetTextureDescriptors();
if (textureDescriptors.Length != 0)
@ -238,11 +208,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.AppendLine();
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.AtomicMinMaxS32Shared) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl");
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighS32) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighS32.glsl");
@ -273,11 +238,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleXor.glsl");
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.StoreSharedSmallInt) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl");
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.SwizzleAdd) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/SwizzleAdd.glsl");
@ -358,7 +318,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
_ => "std430"
};
context.AppendLine($"layout (binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
string set = string.Empty;
if (context.Config.Options.TargetApi == TargetApi.Vulkan)
{
set = $"set = {buffer.Set}, ";
}
context.AppendLine($"layout ({set}binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
context.EnterScope();
foreach (StructureField field in buffer.Type.Fields)
@ -391,6 +358,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
}
}
private static void DeclareMemories(CodeGenContext context, IEnumerable<MemoryDefinition> memories, bool isShared)
{
string prefix = isShared ? "shared " : string.Empty;
foreach (MemoryDefinition memory in memories)
{
string typeName = GetVarTypeName(context, memory.Type & ~AggregateType.Array);
if (memory.ArrayLength > 0)
{
string arraySize = memory.ArrayLength.ToString(CultureInfo.InvariantCulture);
context.AppendLine($"{prefix}{typeName} {memory.Name}[{arraySize}];");
}
else
{
context.AppendLine($"{prefix}{typeName} {memory.Name}[];");
}
}
}
private static void DeclareSamplers(CodeGenContext context, TextureDescriptor[] descriptors)
{
int arraySize = 0;
@ -717,7 +705,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
string code = EmbeddedResources.ReadAllText(filename);
code = code.Replace("\t", CodeGenContext.Tab);
code = code.Replace("$SHARED_MEM$", DefaultNames.SharedMemoryName);
if (context.Config.GpuAccessor.QueryHostSupportsShaderBallot())
{

View file

@ -11,9 +11,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public const string IAttributePrefix = "in_attr";
public const string OAttributePrefix = "out_attr";
public const string LocalMemoryName = "local_mem";
public const string SharedMemoryName = "shared_mem";
public const string ArgumentNamePrefix = "a";
public const string UndefinedName = "undef";

View file

@ -1,21 +0,0 @@
int Helper_AtomicMaxS32(int offset, int value)
{
uint oldValue, newValue;
do
{
oldValue = $SHARED_MEM$[offset];
newValue = uint(max(int(oldValue), value));
} while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
return int(oldValue);
}
int Helper_AtomicMinS32(int offset, int value)
{
uint oldValue, newValue;
do
{
oldValue = $SHARED_MEM$[offset];
newValue = uint(min(int(oldValue), value));
} while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
return int(oldValue);
}

View file

@ -2,9 +2,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{
static class HelperFunctionNames
{
public static string AtomicMaxS32 = "Helper_AtomicMaxS32";
public static string AtomicMinS32 = "Helper_AtomicMinS32";
public static string MultiplyHighS32 = "Helper_MultiplyHighS32";
public static string MultiplyHighU32 = "Helper_MultiplyHighU32";
@ -13,10 +10,5 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public static string ShuffleUp = "Helper_ShuffleUp";
public static string ShuffleXor = "Helper_ShuffleXor";
public static string SwizzleAdd = "Helper_SwizzleAdd";
public static string StoreShared16 = "Helper_StoreShared16";
public static string StoreShared8 = "Helper_StoreShared8";
public static string StoreStorage16 = "Helper_StoreStorage16";
public static string StoreStorage8 = "Helper_StoreStorage8";
}
}

View file

@ -1,23 +0,0 @@
void Helper_StoreShared16(int offset, uint value)
{
int wordOffset = offset >> 2;
int bitOffset = (offset & 3) * 8;
uint oldValue, newValue;
do
{
oldValue = $SHARED_MEM$[wordOffset];
newValue = bitfieldInsert(oldValue, value, bitOffset, 16);
} while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
}
void Helper_StoreShared8(int offset, uint value)
{
int wordOffset = offset >> 2;
int bitOffset = (offset & 3) * 8;
uint oldValue, newValue;
do
{
oldValue = $SHARED_MEM$[wordOffset];
newValue = bitfieldInsert(oldValue, value, bitOffset, 8);
} while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
}

View file

@ -68,7 +68,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
string args = string.Empty;
if (atomic && operation.StorageKind == StorageKind.StorageBuffer)
if (atomic && (operation.StorageKind == StorageKind.StorageBuffer || operation.StorageKind == StorageKind.SharedMemory))
{
args = GenerateLoadOrStore(context, operation, isStore: false);
@ -81,23 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
args += ", " + GetSoureExpr(context, operation.GetSource(argIndex), dstType);
}
}
else if (atomic && operation.StorageKind == StorageKind.SharedMemory)
{
args = LoadShared(context, operation);
// For shared memory access, the second argument is unused and should be ignored.
// It is there to make both storage and shared access have the same number of arguments.
// For storage, both inputs are consumed when the argument index is 0, so we should skip it here.
for (int argIndex = 2; argIndex < arity; argIndex++)
{
args += ", ";
AggregateType dstType = GetSrcVarType(inst, argIndex);
args += GetSoureExpr(context, operation.GetSource(argIndex), dstType);
}
}
else
{
for (int argIndex = 0; argIndex < arity; argIndex++)
@ -179,12 +162,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.Load:
return Load(context, operation);
case Instruction.LoadLocal:
return LoadLocal(context, operation);
case Instruction.LoadShared:
return LoadShared(context, operation);
case Instruction.Lod:
return Lod(context, operation);
@ -200,18 +177,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.Store:
return Store(context, operation);
case Instruction.StoreLocal:
return StoreLocal(context, operation);
case Instruction.StoreShared:
return StoreShared(context, operation);
case Instruction.StoreShared16:
return StoreShared16(context, operation);
case Instruction.StoreShared8:
return StoreShared8(context, operation);
case Instruction.TextureSample:
return TextureSample(context, operation);

View file

@ -17,9 +17,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.AtomicAdd, InstType.AtomicBinary, "atomicAdd");
Add(Instruction.AtomicAnd, InstType.AtomicBinary, "atomicAnd");
Add(Instruction.AtomicCompareAndSwap, InstType.AtomicTernary, "atomicCompSwap");
Add(Instruction.AtomicMaxS32, InstType.CallTernary, HelperFunctionNames.AtomicMaxS32);
Add(Instruction.AtomicMaxU32, InstType.AtomicBinary, "atomicMax");
Add(Instruction.AtomicMinS32, InstType.CallTernary, HelperFunctionNames.AtomicMinS32);
Add(Instruction.AtomicMinU32, InstType.AtomicBinary, "atomicMin");
Add(Instruction.AtomicOr, InstType.AtomicBinary, "atomicOr");
Add(Instruction.AtomicSwap, InstType.AtomicBinary, "atomicExchange");
@ -83,8 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.ImageAtomic, InstType.Special);
Add(Instruction.IsNan, InstType.CallUnary, "isnan");
Add(Instruction.Load, InstType.Special);
Add(Instruction.LoadLocal, InstType.Special);
Add(Instruction.LoadShared, InstType.Special);
Add(Instruction.Lod, InstType.Special);
Add(Instruction.LogarithmB2, InstType.CallUnary, "log2");
Add(Instruction.LogicalAnd, InstType.OpBinaryCom, "&&", 9);
@ -118,10 +114,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.Sine, InstType.CallUnary, "sin");
Add(Instruction.SquareRoot, InstType.CallUnary, "sqrt");
Add(Instruction.Store, InstType.Special);
Add(Instruction.StoreLocal, InstType.Special);
Add(Instruction.StoreShared, InstType.Special);
Add(Instruction.StoreShared16, InstType.Special);
Add(Instruction.StoreShared8, InstType.Special);
Add(Instruction.Subtract, InstType.OpBinary, "-", 2);
Add(Instruction.SwizzleAdd, InstType.CallTernary, HelperFunctionNames.SwizzleAdd);
Add(Instruction.TextureSample, InstType.Special);

View file

@ -191,25 +191,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
return GenerateLoadOrStore(context, operation, isStore: false);
}
public static string LoadLocal(CodeGenContext context, AstOperation operation)
{
return LoadLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
}
public static string LoadShared(CodeGenContext context, AstOperation operation)
{
return LoadLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
}
private static string LoadLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
{
IAstNode src1 = operation.GetSource(0);
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
return $"{arrayName}[{offsetExpr}]";
}
public static string Lod(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
@ -263,58 +244,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
return GenerateLoadOrStore(context, operation, isStore: true);
}
public static string StoreLocal(CodeGenContext context, AstOperation operation)
{
return StoreLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
}
public static string StoreShared(CodeGenContext context, AstOperation operation)
{
return StoreLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
}
private static string StoreLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
return $"{arrayName}[{offsetExpr}] = {src}";
}
public static string StoreShared16(CodeGenContext context, AstOperation operation)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
return $"{HelperFunctionNames.StoreShared16}({offsetExpr}, {src})";
}
public static string StoreShared8(CodeGenContext context, AstOperation operation)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
return $"{HelperFunctionNames.StoreShared8}({offsetExpr}, {src})";
}
public static string TextureSample(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
@ -675,6 +604,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
varType = field.Type;
break;
case StorageKind.LocalMemory:
case StorageKind.SharedMemory:
if (!(operation.GetSource(srcIndex++) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
MemoryDefinition memory = storageKind == StorageKind.LocalMemory
? context.Config.Properties.LocalMemories[bindingId.Value]
: context.Config.Properties.SharedMemories[bindingId.Value];
varName = memory.Name;
varType = memory.Type;
break;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:

View file

@ -113,7 +113,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
if (node is AstOperation operation)
{
if (operation.Inst == Instruction.Load)
if (operation.Inst == Instruction.Load || operation.Inst.IsAtomic())
{
switch (operation.StorageKind)
{
@ -136,6 +136,19 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
return field.Type & AggregateType.ElementTypeMask;
case StorageKind.LocalMemory:
case StorageKind.SharedMemory:
if (!(operation.GetSource(0) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {operation.StorageKind} storage must be a constant operand.");
}
MemoryDefinition memory = operation.StorageKind == StorageKind.LocalMemory
? context.Config.Properties.LocalMemories[bindingId.Value]
: context.Config.Properties.SharedMemories[bindingId.Value];
return memory.Type & AggregateType.ElementTypeMask;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:

View file

@ -25,8 +25,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
public Dictionary<int, Instruction> ConstantBuffers { get; } = new Dictionary<int, Instruction>();
public Dictionary<int, Instruction> StorageBuffers { get; } = new Dictionary<int, Instruction>();
public Instruction LocalMemory { get; set; }
public Instruction SharedMemory { get; set; }
public Dictionary<int, Instruction> LocalMemories { get; } = new Dictionary<int, Instruction>();
public Dictionary<int, Instruction> SharedMemories { get; } = new Dictionary<int, Instruction>();
public Dictionary<TextureMeta, SamplerType> SamplersTypes { get; } = new Dictionary<TextureMeta, SamplerType>();
public Dictionary<TextureMeta, (Instruction, Instruction, Instruction)> Samplers { get; } = new Dictionary<TextureMeta, (Instruction, Instruction, Instruction)>();
public Dictionary<TextureMeta, (Instruction, Instruction)> Images { get; } = new Dictionary<TextureMeta, (Instruction, Instruction)>();
@ -35,7 +35,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
public Dictionary<IoDefinition, Instruction> InputsPerPatch { get; } = new Dictionary<IoDefinition, Instruction>();
public Dictionary<IoDefinition, Instruction> OutputsPerPatch { get; } = new Dictionary<IoDefinition, Instruction>();
public Instruction CoordTemp { get; set; }
public StructuredFunction CurrentFunction { get; set; }
private readonly Dictionary<AstOperand, Instruction> _locals = new Dictionary<AstOperand, Instruction>();
private readonly Dictionary<int, Instruction[]> _localForArgs = new Dictionary<int, Instruction[]>();

View file

@ -6,7 +6,6 @@ using Spv.Generator;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Numerics;
using static Spv.Specification;
using SpvInstruction = Spv.Generator.Instruction;
@ -44,13 +43,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
context.AddLocalVariable(spvLocal);
context.DeclareLocal(local, spvLocal);
}
var ivector2Type = context.TypeVector(context.TypeS32(), 2);
var coordTempPointerType = context.TypePointer(StorageClass.Function, ivector2Type);
var coordTemp = context.Variable(coordTempPointerType, StorageClass.Function);
context.AddLocalVariable(coordTemp);
context.CoordTemp = coordTemp;
}
public static void DeclareLocalForArgs(CodeGenContext context, List<StructuredFunction> functions)
@ -77,54 +69,30 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
public static void DeclareAll(CodeGenContext context, StructuredProgramInfo info)
{
if (context.Config.Stage == ShaderStage.Compute)
{
int localMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeLocalMemorySize(), 4);
if (localMemorySize != 0)
{
DeclareLocalMemory(context, localMemorySize);
}
int sharedMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeSharedMemorySize(), 4);
if (sharedMemorySize != 0)
{
DeclareSharedMemory(context, sharedMemorySize);
}
}
else if (context.Config.LocalMemorySize != 0)
{
int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
DeclareLocalMemory(context, localMemorySize);
}
DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
DeclareStorageBuffers(context, context.Config.Properties.StorageBuffers.Values);
DeclareMemories(context, context.Config.Properties.LocalMemories, context.LocalMemories, StorageClass.Private);
DeclareMemories(context, context.Config.Properties.SharedMemories, context.SharedMemories, StorageClass.Workgroup);
DeclareSamplers(context, context.Config.GetTextureDescriptors());
DeclareImages(context, context.Config.GetImageDescriptors());
DeclareInputsAndOutputs(context, info);
}
private static void DeclareLocalMemory(CodeGenContext context, int size)
private static void DeclareMemories(
CodeGenContext context,
IReadOnlyDictionary<int, MemoryDefinition> memories,
Dictionary<int, SpvInstruction> dict,
StorageClass storage)
{
context.LocalMemory = DeclareMemory(context, StorageClass.Private, size);
}
foreach ((int id, MemoryDefinition memory) in memories)
{
var pointerType = context.TypePointer(storage, context.GetType(memory.Type, memory.ArrayLength));
var variable = context.Variable(pointerType, storage);
private static void DeclareSharedMemory(CodeGenContext context, int size)
{
context.SharedMemory = DeclareMemory(context, StorageClass.Workgroup, size);
}
context.AddGlobalVariable(variable);
private static SpvInstruction DeclareMemory(CodeGenContext context, StorageClass storage, int size)
{
var arrayType = context.TypeArray(context.TypeU32(), context.Constant(context.TypeU32(), size));
var pointerType = context.TypePointer(storage, arrayType);
var variable = context.Variable(pointerType, storage);
context.AddGlobalVariable(variable);
return variable;
dict.Add(id, variable);
}
}
private static void DeclareConstantBuffers(CodeGenContext context, IEnumerable<BufferDefinition> buffers)

View file

@ -97,8 +97,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
Add(Instruction.ImageStore, GenerateImageStore);
Add(Instruction.IsNan, GenerateIsNan);
Add(Instruction.Load, GenerateLoad);
Add(Instruction.LoadLocal, GenerateLoadLocal);
Add(Instruction.LoadShared, GenerateLoadShared);
Add(Instruction.Lod, GenerateLod);
Add(Instruction.LogarithmB2, GenerateLogarithmB2);
Add(Instruction.LogicalAnd, GenerateLogicalAnd);
@ -132,10 +130,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
Add(Instruction.Sine, GenerateSine);
Add(Instruction.SquareRoot, GenerateSquareRoot);
Add(Instruction.Store, GenerateStore);
Add(Instruction.StoreLocal, GenerateStoreLocal);
Add(Instruction.StoreShared, GenerateStoreShared);
Add(Instruction.StoreShared16, GenerateStoreShared16);
Add(Instruction.StoreShared8, GenerateStoreShared8);
Add(Instruction.Subtract, GenerateSubtract);
Add(Instruction.SwizzleAdd, GenerateSwizzleAdd);
Add(Instruction.TextureSample, GenerateTextureSample);
@ -871,30 +865,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
return GenerateLoadOrStore(context, operation, isStore: false);
}
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 GenerateLod(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
@ -1268,45 +1238,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
return GenerateLoadOrStore(context, operation, isStore: true);
}
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 GenerateSubtract(CodeGenContext context, AstOperation operation)
{
return GenerateBinary(context, operation, context.Delegates.FSub, context.Delegates.ISub);
@ -1827,55 +1758,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
AstOperation operation,
Func<SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction, SpvInstruction> emitU)
{
var value = context.GetU32(operation.GetSource(operation.SourcesCount - 1));
SpvInstruction elemPointer = GetStoragePointer(context, operation, out AggregateType varType);
SpvInstruction elemPointer;
if (operation.StorageKind == StorageKind.StorageBuffer)
{
elemPointer = GetStoragePointer(context, operation, out _);
}
else if (operation.StorageKind == StorageKind.SharedMemory)
{
var offset = context.GetU32(operation.GetSource(0));
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
}
else
{
throw new InvalidOperationException($"Invalid storage kind \"{operation.StorageKind}\".");
}
var value = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
var one = context.Constant(context.TypeU32(), 1);
var zero = context.Constant(context.TypeU32(), 0);
return new OperationResult(AggregateType.U32, emitU(context.TypeU32(), elemPointer, one, zero, value));
return new OperationResult(varType, emitU(context.GetType(varType), elemPointer, one, zero, value));
}
private static OperationResult GenerateAtomicMemoryCas(CodeGenContext context, AstOperation operation)
{
var value0 = context.GetU32(operation.GetSource(operation.SourcesCount - 2));
var value1 = context.GetU32(operation.GetSource(operation.SourcesCount - 1));
SpvInstruction elemPointer = GetStoragePointer(context, operation, out AggregateType varType);
SpvInstruction elemPointer;
if (operation.StorageKind == StorageKind.StorageBuffer)
{
elemPointer = GetStoragePointer(context, operation, out _);
}
else if (operation.StorageKind == StorageKind.SharedMemory)
{
var offset = context.GetU32(operation.GetSource(0));
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Workgroup, context.TypeU32()), context.SharedMemory, offset);
}
else
{
throw new InvalidOperationException($"Invalid storage kind \"{operation.StorageKind}\".");
}
var value0 = context.Get(varType, operation.GetSource(operation.SourcesCount - 2));
var value1 = context.Get(varType, operation.GetSource(operation.SourcesCount - 1));
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));
return new OperationResult(varType, context.AtomicCompareExchange(context.GetType(varType), elemPointer, one, zero, zero, value1, value0));
}
private static OperationResult GenerateLoadOrStore(CodeGenContext context, AstOperation operation, bool isStore)
@ -1928,6 +1831,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
: context.StorageBuffers[bindingIndex.Value];
break;
case StorageKind.LocalMemory:
case StorageKind.SharedMemory:
if (!(operation.GetSource(srcIndex++) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
if (storageKind == StorageKind.LocalMemory)
{
storageClass = StorageClass.Private;
varType = context.Config.Properties.LocalMemories[bindingId.Value].Type & AggregateType.ElementTypeMask;
baseObj = context.LocalMemories[bindingId.Value];
}
else
{
storageClass = StorageClass.Workgroup;
varType = context.Config.Properties.SharedMemories[bindingId.Value].Type & AggregateType.ElementTypeMask;
baseObj = context.SharedMemories[bindingId.Value];
}
break;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:
@ -2048,50 +1972,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
return context.Load(context.GetType(varType), context.Inputs[ioDefinition]);
}
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 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,