emu-mirrors/ryujinx-final
Archived
0
0
Fork 0
Code Activity

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

Browse source 
* 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
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
View file

@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
private const ushort FileFormatVersionMajor = 1;
private const ushort FileFormatVersionMinor = 2;
private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
private const uint CodeGenVersion = 5080;
private const uint CodeGenVersion = 5241;
private const string SharedTocFileName = "shared.toc";
private const string SharedDataFileName = "shared.data";

75
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
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())
{

3
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
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";

21
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl
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);
}

8
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs
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";
}
}

23
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl
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);
}

37
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
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);

8
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
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);

86
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
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:

15
src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs
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:

5
src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
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[]>();

60
src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs
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)

176
src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs
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,

109
src/Ryujinx.Graphics.Shader/Instructions/InstEmitMemory.cs
View file

@ -10,12 +10,6 @@ namespace Ryujinx.Graphics.Shader.Instructions
{
static partial class InstEmit
{
private enum MemoryRegion
{
Local,
Shared
}
public static void Atom(EmitterContext context)
{
InstAtom op = context.GetOp<InstAtom>();
@ -51,7 +45,8 @@ namespace Ryujinx.Graphics.Shader.Instructions
_ => AtomSize.U32
};
Operand res = EmitAtomicOp(context, StorageKind.SharedMemory, op.AtomOp, size, offset, Const(0), value);
Operand id = Const(context.Config.ResourceManager.SharedMemoryId);
Operand res = EmitAtomicOp(context, StorageKind.SharedMemory, op.AtomOp, size, id, offset, value);
context.Copy(GetDest(op.Dest), res);
}
@ -114,14 +109,14 @@ namespace Ryujinx.Graphics.Shader.Instructions
{
InstLdl op = context.GetOp<InstLdl>();
EmitLoad(context, MemoryRegion.Local, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
EmitLoad(context, StorageKind.LocalMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
}
public static void Lds(EmitterContext context)
{
InstLds op = context.GetOp<InstLds>();
EmitLoad(context, MemoryRegion.Shared, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
EmitLoad(context, StorageKind.SharedMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
}
public static void Red(EmitterContext context)
@ -144,14 +139,14 @@ namespace Ryujinx.Graphics.Shader.Instructions
{
InstStl op = context.GetOp<InstStl>();
EmitStore(context, MemoryRegion.Local, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
EmitStore(context, StorageKind.LocalMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
}
public static void Sts(EmitterContext context)
{
InstSts op = context.GetOp<InstSts>();
EmitStore(context, MemoryRegion.Shared, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
EmitStore(context, StorageKind.SharedMemory, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
}
private static Operand EmitLoadConstant(EmitterContext context, Operand slot, Operand offset)
@ -192,8 +187,8 @@ namespace Ryujinx.Graphics.Shader.Instructions
StorageKind storageKind,
AtomOp op,
AtomSize type,
Operand addrLow,
Operand addrHigh,
Operand e0,
Operand e1,
Operand value)
{
Operand res = Const(0);
@ -203,7 +198,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.Add:
if (type == AtomSize.S32 || type == AtomSize.U32)
{
res = context.AtomicAdd(storageKind, addrLow, addrHigh, value);
res = context.AtomicAdd(storageKind, e0, e1, value);
}
else
{
@ -213,7 +208,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.And:
if (type == AtomSize.S32 || type == AtomSize.U32)
{
res = context.AtomicAnd(storageKind, addrLow, addrHigh, value);
res = context.AtomicAnd(storageKind, e0, e1, value);
}
else
{
@ -223,7 +218,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.Xor:
if (type == AtomSize.S32 || type == AtomSize.U32)
{
res = context.AtomicXor(storageKind, addrLow, addrHigh, value);
res = context.AtomicXor(storageKind, e0, e1, value);
}
else
{
@ -233,7 +228,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.Or:
if (type == AtomSize.S32 || type == AtomSize.U32)
{
res = context.AtomicOr(storageKind, addrLow, addrHigh, value);
res = context.AtomicOr(storageKind, e0, e1, value);
}
else
{
@ -243,11 +238,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.Max:
if (type == AtomSize.S32)
{
res = context.AtomicMaxS32(storageKind, addrLow, addrHigh, value);
res = context.AtomicMaxS32(storageKind, e0, e1, value);
}
else if (type == AtomSize.U32)
{
res = context.AtomicMaxU32(storageKind, addrLow, addrHigh, value);
res = context.AtomicMaxU32(storageKind, e0, e1, value);
}
else
{
@ -257,11 +252,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
case AtomOp.Min:
if (type == AtomSize.S32)
{
res = context.AtomicMinS32(storageKind, addrLow, addrHigh, value);
res = context.AtomicMinS32(storageKind, e0, e1, value);
}
else if (type == AtomSize.U32)
{
res = context.AtomicMinU32(storageKind, addrLow, addrHigh, value);
res = context.AtomicMinU32(storageKind, e0, e1, value);
}
else
{
@ -275,7 +270,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
private static void EmitLoad(
EmitterContext context,
MemoryRegion region,
StorageKind storageKind,
LsSize2 size,
Operand srcA,
int rd,
@ -287,19 +282,19 @@ namespace Ryujinx.Graphics.Shader.Instructions
return;
}
int id = storageKind == StorageKind.LocalMemory
? context.Config.ResourceManager.LocalMemoryId
: context.Config.ResourceManager.SharedMemoryId;
bool isSmallInt = size < LsSize2.B32;
int count = 1;
switch (size)
int count = size switch
{
case LsSize2.B64: count = 2; break;
case LsSize2.B128: count = 4; break;
}
LsSize2.B64 => 2,
LsSize2.B128 => 4,
_ => 1
};
Operand baseOffset = context.IAdd(srcA, Const(offset));
Operand wordOffset = context.ShiftRightU32(baseOffset, Const(2)); // Word offset = byte offset / 4 (one word = 4 bytes).
Operand bitOffset = GetBitOffset(context, baseOffset);
Operand baseOffset = context.Copy(srcA);
for (int index = 0; index < count; index++)
{
@ -310,14 +305,10 @@ namespace Ryujinx.Graphics.Shader.Instructions
break;
}
Operand elemOffset = context.IAdd(wordOffset, Const(index));
Operand value = null;
switch (region)
{
case MemoryRegion.Local: value = context.LoadLocal(elemOffset); break;
case MemoryRegion.Shared: value = context.LoadShared(elemOffset); break;
}
Operand byteOffset = context.IAdd(baseOffset, Const(offset + index * 4));
Operand wordOffset = context.ShiftRightU32(byteOffset, Const(2)); // Word offset = byte offset / 4 (one word = 4 bytes).
Operand bitOffset = GetBitOffset(context, byteOffset);
Operand value = context.Load(storageKind, id, wordOffset);
if (isSmallInt)
{
@ -360,7 +351,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
private static void EmitStore(
EmitterContext context,
MemoryRegion region,
StorageKind storageKind,
LsSize2 size,
Operand srcA,
int rd,
@ -372,52 +363,54 @@ namespace Ryujinx.Graphics.Shader.Instructions
return;
}
int id = storageKind == StorageKind.LocalMemory
? context.Config.ResourceManager.LocalMemoryId
: context.Config.ResourceManager.SharedMemoryId;
bool isSmallInt = size < LsSize2.B32;
int count = 1;
switch (size)
int count = size switch
{
case LsSize2.B64: count = 2; break;
case LsSize2.B128: count = 4; break;
}
LsSize2.B64 => 2,
LsSize2.B128 => 4,
_ => 1
};
Operand baseOffset = context.IAdd(srcA, Const(offset));
Operand wordOffset = context.ShiftRightU32(baseOffset, Const(2));
Operand bitOffset = GetBitOffset(context, baseOffset);
Operand baseOffset = context.Copy(srcA);
for (int index = 0; index < count; index++)
{
bool isRz = rd + index >= RegisterConsts.RegisterZeroIndex;
Operand value = Register(isRz ? rd : rd + index, RegisterType.Gpr);
Operand elemOffset = context.IAdd(wordOffset, Const(index));
Operand byteOffset = context.IAdd(baseOffset, Const(offset + index * 4));
Operand wordOffset = context.ShiftRightU32(byteOffset, Const(2));
Operand bitOffset = GetBitOffset(context, byteOffset);
if (isSmallInt && region == MemoryRegion.Local)
if (isSmallInt && storageKind == StorageKind.LocalMemory)
{
Operand word = context.LoadLocal(elemOffset);
Operand word = context.Load(storageKind, id, wordOffset);
value = InsertSmallInt(context, (LsSize)size, bitOffset, word, value);
}
if (region == MemoryRegion.Local)
if (storageKind == StorageKind.LocalMemory)
{
context.StoreLocal(elemOffset, value);
context.Store(storageKind, id, wordOffset, value);
}
else if (region == MemoryRegion.Shared)
else if (storageKind == StorageKind.SharedMemory)
{
switch (size)
{
case LsSize2.U8:
case LsSize2.S8:
context.StoreShared8(baseOffset, value);
context.Store(StorageKind.SharedMemory8, id, byteOffset, value);
break;
case LsSize2.U16:
case LsSize2.S16:
context.StoreShared16(baseOffset, value);
context.Store(StorageKind.SharedMemory16, id, byteOffset, value);
break;
default:
context.StoreShared(elemOffset, value);
context.Store(storageKind, id, wordOffset, value);
break;
}
}

6
src/Ryujinx.Graphics.Shader/IntermediateRepresentation/Instruction.cs
View file

@ -79,8 +79,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
ImageAtomic,
IsNan,
Load,
LoadLocal,
LoadShared,
Lod,
LogarithmB2,
LogicalAnd,
@ -115,10 +113,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
Sine,
SquareRoot,
Store,
StoreLocal,
StoreShared,
StoreShared16,
StoreShared8,
Subtract,
SwizzleAdd,
TextureSample,

11
src/Ryujinx.Graphics.Shader/IntermediateRepresentation/StorageKind.cs
View file

@ -11,12 +11,13 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
StorageBuffer,
LocalMemory,
SharedMemory,
SharedMemory8, // TODO: Remove this and store type as a field on the Operation class itself.
SharedMemory16, // TODO: Remove this and store type as a field on the Operation class itself.
GlobalMemory,
// TODO: Remove those and store type as a field on the Operation class itself.
GlobalMemoryS8,
GlobalMemoryS16,
GlobalMemoryU8,
GlobalMemoryU16
GlobalMemoryS8, // TODO: Remove this and store type as a field on the Operation class itself.
GlobalMemoryS16, // TODO: Remove this and store type as a field on the Operation class itself.
GlobalMemoryU8, // TODO: Remove this and store type as a field on the Operation class itself.
GlobalMemoryU16 // TODO: Remove this and store type as a field on the Operation class itself.
}
static class StorageKindExtensions

2
src/Ryujinx.Graphics.Shader/Ryujinx.Graphics.Shader.csproj
View file

@ -10,14 +10,12 @@
</ItemGroup>
<ItemGroup>
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\AtomicMinMaxS32Shared.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\MultiplyHighS32.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\MultiplyHighU32.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\Shuffle.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleDown.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleUp.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\ShuffleXor.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\StoreSharedSmallInt.glsl" />
<EmbeddedResource Include="CodeGen\Glsl\HelperFunctions\SwizzleAdd.glsl" />
</ItemGroup>

18
src/Ryujinx.Graphics.Shader/StructuredIr/HelperFunctionsMask.cs
View file

@ -5,15 +5,13 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
[Flags]
enum HelperFunctionsMask
{
AtomicMinMaxS32Shared = 1 << 0,
MultiplyHighS32 = 1 << 2,
MultiplyHighU32 = 1 << 3,
Shuffle = 1 << 4,
ShuffleDown = 1 << 5,
ShuffleUp = 1 << 6,
ShuffleXor = 1 << 7,
StoreSharedSmallInt = 1 << 8,
SwizzleAdd = 1 << 10,
FSI = 1 << 11
MultiplyHighS32 = 1 << 2,
MultiplyHighU32 = 1 << 3,
Shuffle = 1 << 4,
ShuffleDown = 1 << 5,
ShuffleUp = 1 << 6,
ShuffleXor = 1 << 7,
SwizzleAdd = 1 << 10,
FSI = 1 << 11
}
}

6
src/Ryujinx.Graphics.Shader/StructuredIr/InstructionInfo.cs
View file

@ -90,8 +90,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Add(Instruction.ImageAtomic, AggregateType.S32);
Add(Instruction.IsNan, AggregateType.Bool, AggregateType.Scalar);
Add(Instruction.Load, AggregateType.FP32);
Add(Instruction.LoadLocal, AggregateType.U32, AggregateType.S32);
Add(Instruction.LoadShared, AggregateType.U32, AggregateType.S32);
Add(Instruction.Lod, AggregateType.FP32);
Add(Instruction.LogarithmB2, AggregateType.Scalar, AggregateType.Scalar);
Add(Instruction.LogicalAnd, AggregateType.Bool, AggregateType.Bool, AggregateType.Bool);
@ -121,10 +119,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Add(Instruction.Sine, AggregateType.Scalar, AggregateType.Scalar);
Add(Instruction.SquareRoot, AggregateType.Scalar, AggregateType.Scalar);
Add(Instruction.Store, AggregateType.Void);
Add(Instruction.StoreLocal, AggregateType.Void, AggregateType.S32, AggregateType.U32);
Add(Instruction.StoreShared, AggregateType.Void, AggregateType.S32, AggregateType.U32);
Add(Instruction.StoreShared16, AggregateType.Void, AggregateType.S32, AggregateType.U32);
Add(Instruction.StoreShared8, AggregateType.Void, AggregateType.S32, AggregateType.U32);
Add(Instruction.Subtract, AggregateType.Scalar, AggregateType.Scalar, AggregateType.Scalar);
Add(Instruction.SwizzleAdd, AggregateType.FP32, AggregateType.FP32, AggregateType.FP32, AggregateType.S32);
Add(Instruction.TextureSample, AggregateType.FP32);

18
src/Ryujinx.Graphics.Shader/StructuredIr/MemoryDefinition.cs Normal file
View file

@ -0,0 +1,18 @@
using Ryujinx.Graphics.Shader.Translation;
namespace Ryujinx.Graphics.Shader.StructuredIr
{
readonly struct MemoryDefinition
{
public string Name { get; }
public AggregateType Type { get; }
public int ArrayLength { get; }
public MemoryDefinition(string name, AggregateType type, int arrayLength = 1)
{
Name = name;
Type = type;
ArrayLength = arrayLength;
}
}
}

22
src/Ryujinx.Graphics.Shader/StructuredIr/ShaderProperties.cs
View file

@ -6,14 +6,20 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{
private readonly Dictionary<int, BufferDefinition> _constantBuffers;
private readonly Dictionary<int, BufferDefinition> _storageBuffers;
private readonly Dictionary<int, MemoryDefinition> _localMemories;
private readonly Dictionary<int, MemoryDefinition> _sharedMemories;
public IReadOnlyDictionary<int, BufferDefinition> ConstantBuffers => _constantBuffers;
public IReadOnlyDictionary<int, BufferDefinition> StorageBuffers => _storageBuffers;
public IReadOnlyDictionary<int, MemoryDefinition> LocalMemories => _localMemories;
public IReadOnlyDictionary<int, MemoryDefinition> SharedMemories => _sharedMemories;
public ShaderProperties()
{
_constantBuffers = new Dictionary<int, BufferDefinition>();
_storageBuffers = new Dictionary<int, BufferDefinition>();
_localMemories = new Dictionary<int, MemoryDefinition>();
_sharedMemories = new Dictionary<int, MemoryDefinition>();
}
public void AddConstantBuffer(int binding, BufferDefinition definition)
@ -25,5 +31,21 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{
_storageBuffers[binding] = definition;
}
public int AddLocalMemory(MemoryDefinition definition)
{
int id = _localMemories.Count;
_localMemories.Add(id, definition);
return id;
}
public int AddSharedMemory(MemoryDefinition definition)
{
int id = _sharedMemories.Count;
_sharedMemories.Add(id, definition);
return id;
}
}
}

11
src/Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs
View file

@ -274,13 +274,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
// decide which helper functions are needed on the final generated code.
switch (operation.Inst)
{
case Instruction.AtomicMaxS32:
case Instruction.AtomicMinS32:
if (operation.StorageKind == StorageKind.SharedMemory)
{
context.Info.HelperFunctionsMask |= HelperFunctionsMask.AtomicMinMaxS32Shared;
}
break;
case Instruction.MultiplyHighS32:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighS32;
break;
@ -299,10 +292,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
case Instruction.ShuffleXor:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleXor;
break;
case Instruction.StoreShared16:
case Instruction.StoreShared8:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.StoreSharedSmallInt;
break;
case Instruction.SwizzleAdd:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.SwizzleAdd;
break;

40
src/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
View file

@ -67,6 +67,11 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.AtomicAnd, storageKind, Local(), Const(binding), e0, e1, value);
}
public static Operand AtomicCompareAndSwap(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand compare, Operand value)
{
return context.Add(Instruction.AtomicCompareAndSwap, storageKind, Local(), Const(binding), e0, compare, value);
}
public static Operand AtomicCompareAndSwap(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1, Operand compare, Operand value)
{
return context.Add(Instruction.AtomicCompareAndSwap, storageKind, Local(), Const(binding), e0, e1, compare, value);
@ -661,16 +666,6 @@ namespace Ryujinx.Graphics.Shader.Translation
: context.Load(storageKind, (int)ioVariable, arrayIndex, elemIndex);
}
public static Operand LoadLocal(this EmitterContext context, Operand a)
{
return context.Add(Instruction.LoadLocal, Local(), a);
}
public static Operand LoadShared(this EmitterContext context, Operand a)
{
return context.Add(Instruction.LoadShared, Local(), a);
}
public static Operand MemoryBarrier(this EmitterContext context)
{
return context.Add(Instruction.MemoryBarrier);
@ -753,6 +748,11 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.Store, storageKind, null, e0, e1, value);
}
public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand value)
{
return context.Add(Instruction.Store, storageKind, null, Const(binding), e0, value);
}
public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1, Operand value)
{
return context.Add(Instruction.Store, storageKind, null, Const(binding), e0, e1, value);
@ -797,26 +797,6 @@ namespace Ryujinx.Graphics.Shader.Translation
: context.Add(Instruction.Store, storageKind, null, Const((int)ioVariable), arrayIndex, elemIndex, value);
}
public static Operand StoreLocal(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.StoreLocal, null, a, b);
}
public static Operand StoreShared(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.StoreShared, null, a, b);
}
public static Operand StoreShared16(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.StoreShared16, null, a, b);
}
public static Operand StoreShared8(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.StoreShared8, null, a, b);
}
public static Operand UnpackDouble2x32High(this EmitterContext context, Operand a)
{
return UnpackDouble2x32(context, a, 1);

109
src/Ryujinx.Graphics.Shader/Translation/HelperFunctionManager.cs
View file

@ -9,13 +9,13 @@ namespace Ryujinx.Graphics.Shader.Translation
class HelperFunctionManager
{
private readonly List<Function> _functionList;
private readonly Dictionary<HelperFunctionName, int> _functionIds;
private readonly Dictionary<int, int> _functionIds;
private readonly ShaderStage _stage;
public HelperFunctionManager(List<Function> functionList, ShaderStage stage)
{
_functionList = functionList;
_functionIds = new Dictionary<HelperFunctionName, int>();
_functionIds = new Dictionary<int, int>();
_stage = stage;
}
@ -29,14 +29,30 @@ namespace Ryujinx.Graphics.Shader.Translation
public int GetOrCreateFunctionId(HelperFunctionName functionName)
{
if (_functionIds.TryGetValue(functionName, out int functionId))
if (_functionIds.TryGetValue((int)functionName, out int functionId))
{
return functionId;
}
Function function = GenerateFunction(functionName);
functionId = AddFunction(function);
_functionIds.Add(functionName, functionId);
_functionIds.Add((int)functionName, functionId);
return functionId;
}
public int GetOrCreateFunctionId(HelperFunctionName functionName, int id)
{
int key = (int)functionName | (id << 16);
if (_functionIds.TryGetValue(key, out int functionId))
{
return functionId;
}
Function function = GenerateFunction(functionName, id);
functionId = AddFunction(function);
_functionIds.Add(key, functionId);
return functionId;
}
@ -140,6 +156,67 @@ namespace Ryujinx.Graphics.Shader.Translation
return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, "ConvertFloatToDouble", false, 1, 2);
}
private static Function GenerateFunction(HelperFunctionName functionName, int id)
{
return functionName switch
{
HelperFunctionName.SharedAtomicMaxS32 => GenerateSharedAtomicSigned(id, isMin: false),
HelperFunctionName.SharedAtomicMinS32 => GenerateSharedAtomicSigned(id, isMin: true),
HelperFunctionName.SharedStore8 => GenerateSharedStore8(id),
HelperFunctionName.SharedStore16 => GenerateSharedStore16(id),
_ => throw new ArgumentException($"Invalid function name {functionName}")
};
}
private static Function GenerateSharedAtomicSigned(int id, bool isMin)
{
EmitterContext context = new EmitterContext();
Operand wordOffset = Argument(0);
Operand value = Argument(1);
Operand result = GenerateSharedAtomicCasLoop(context, wordOffset, id, (memValue) =>
{
return isMin
? context.IMinimumS32(memValue, value)
: context.IMaximumS32(memValue, value);
});
context.Return(result);
return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, $"SharedAtomic{(isMin ? "Min" : "Max")}_{id}", true, 2, 0);
}
private static Function GenerateSharedStore8(int id)
{
return GenerateSharedStore(id, 8);
}
private static Function GenerateSharedStore16(int id)
{
return GenerateSharedStore(id, 16);
}
private static Function GenerateSharedStore(int id, int bitSize)
{
EmitterContext context = new EmitterContext();
Operand offset = Argument(0);
Operand value = Argument(1);
Operand wordOffset = context.ShiftRightU32(offset, Const(2));
Operand bitOffset = GetBitOffset(context, offset);
GenerateSharedAtomicCasLoop(context, wordOffset, id, (memValue) =>
{
return context.BitfieldInsert(memValue, value, bitOffset, Const(bitSize));
});
context.Return();
return new Function(ControlFlowGraph.Create(context.GetOperations()).Blocks, $"SharedStore{bitSize}_{id}", false, 2, 0);
}
private Function GenerateTexelFetchScaleFunction()
{
EmitterContext context = new EmitterContext();
@ -226,5 +303,29 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.IAdd(Const(1), index);
}
}
public static Operand GetBitOffset(EmitterContext context, Operand offset)
{
return context.ShiftLeft(context.BitwiseAnd(offset, Const(3)), Const(3));
}
private static Operand GenerateSharedAtomicCasLoop(EmitterContext context, Operand wordOffset, int id, Func<Operand, Operand> opCallback)
{
Operand lblLoopHead = Label();
context.MarkLabel(lblLoopHead);
Operand oldValue = context.Load(StorageKind.SharedMemory, id, wordOffset);
Operand newValue = opCallback(oldValue);
Operand casResult = context.AtomicCompareAndSwap(StorageKind.SharedMemory, id, wordOffset, oldValue, newValue);
Operand casFail = context.ICompareNotEqual(casResult, oldValue);
context.BranchIfTrue(lblLoopHead, casFail);
return oldValue;
}
}
}

4
src/Ryujinx.Graphics.Shader/Translation/HelperFunctionName.cs
View file

@ -4,6 +4,10 @@ namespace Ryujinx.Graphics.Shader.Translation
{
ConvertDoubleToFloat,
ConvertFloatToDouble,
SharedAtomicMaxS32,
SharedAtomicMinS32,
SharedStore8,
SharedStore16,
TexelFetchScale,
TextureSizeUnscale
}

32
src/Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs
View file

@ -244,7 +244,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
node = nextNode;
}
}
else if (operation.Inst == Instruction.StoreShared || operation.Inst == Instruction.StoreLocal)
else if (operation.Inst == Instruction.Store &&
(operation.StorageKind == StorageKind.SharedMemory ||
operation.StorageKind == StorageKind.LocalMemory))
{
// The NVIDIA compiler can sometimes use shared or local memory as temporary
// storage to place the base address and size on, so we need
@ -874,7 +876,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
if (bitSize < 32)
{
Operand bitOffset = GetBitOffset(context, offset);
Operand bitOffset = HelperFunctionManager.GetBitOffset(context, offset);
GenerateAtomicCasLoop(context, wordOffset, binding, (memValue) =>
{
@ -892,7 +894,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
if (IsSmallInt(storageKind))
{
Operand bitOffset = GetBitOffset(context, offset);
Operand bitOffset = HelperFunctionManager.GetBitOffset(context, offset);
switch (storageKind)
{
@ -921,11 +923,6 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
return true;
}
private static Operand GetBitOffset(EmitterContext context, Operand offset)
{
return context.ShiftLeft(context.BitwiseAnd(offset, Const(3)), Const(3));
}
private static Operand GenerateAtomicCasLoop(EmitterContext context, Operand wordOffset, int binding, Func<Operand, Operand> opCallback)
{
Operand lblLoopHead = Label();
@ -1070,15 +1067,18 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
{
baseOffset = null;
if (operation.Inst == Instruction.LoadShared || operation.Inst == Instruction.StoreShared)
if (operation.Inst == Instruction.Load || operation.Inst == Instruction.Store)
{
type = LsMemoryType.Shared;
return TryGetSharedMemoryOffsets(operation, out baseOffset, out constOffset);
}
else if (operation.Inst == Instruction.LoadLocal || operation.Inst == Instruction.StoreLocal)
{
type = LsMemoryType.Local;
return TryGetLocalMemoryOffset(operation, out constOffset);
if (operation.StorageKind == StorageKind.SharedMemory)
{
type = LsMemoryType.Shared;
return TryGetSharedMemoryOffsets(operation, out baseOffset, out constOffset);
}
else if (operation.StorageKind == StorageKind.LocalMemory)
{
type = LsMemoryType.Local;
return TryGetLocalMemoryOffset(operation, out constOffset);
}
}
type = default;

25
src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs
View file

@ -1,3 +1,4 @@
using Ryujinx.Common;
using Ryujinx.Graphics.Shader.StructuredIr;
using System;
using System.Collections.Generic;
@ -22,9 +23,12 @@ namespace Ryujinx.Graphics.Shader.Translation
private readonly HashSet<int> _usedConstantBufferBindings;
public int LocalMemoryId { get; }
public int SharedMemoryId { get; }
public ShaderProperties Properties => _properties;
public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor, ShaderProperties properties)
public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor, ShaderProperties properties, int localMemorySize)
{
_gpuAccessor = gpuAccessor;
_properties = properties;
@ -41,6 +45,25 @@ namespace Ryujinx.Graphics.Shader.Translation
_usedConstantBufferBindings = new HashSet<int>();
properties.AddConstantBuffer(0, new BufferDefinition(BufferLayout.Std140, 0, 0, "support_buffer", SupportBuffer.GetStructureType()));
LocalMemoryId = -1;
SharedMemoryId = -1;
if (localMemorySize != 0)
{
var lmem = new MemoryDefinition("local_memory", AggregateType.Array | AggregateType.U32, BitUtils.DivRoundUp(localMemorySize, sizeof(uint)));
LocalMemoryId = properties.AddLocalMemory(lmem);
}
int sharedMemorySize = stage == ShaderStage.Compute ? gpuAccessor.QueryComputeSharedMemorySize() : 0;
if (sharedMemorySize != 0)
{
var smem = new MemoryDefinition("shared_memory", AggregateType.Array | AggregateType.U32, BitUtils.DivRoundUp(sharedMemorySize, sizeof(uint)));
SharedMemoryId = properties.AddSharedMemory(smem);
}
}
public int GetConstantBufferBinding(int slot)

92
src/Ryujinx.Graphics.Shader/Translation/Rewriter.cs
View file

@ -1,6 +1,8 @@
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation.Optimizations;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
@ -70,6 +72,15 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
}
else
{
node = InsertSharedStoreSmallInt(hfm, node);
if (config.Options.TargetLanguage != TargetLanguage.Spirv)
{
node = InsertSharedAtomicSigned(hfm, node);
}
}
}
}
}
@ -171,6 +182,87 @@ namespace Ryujinx.Graphics.Shader.Translation
operation.TurnIntoCopy(result);
}
private static LinkedListNode<INode> InsertSharedStoreSmallInt(HelperFunctionManager hfm, LinkedListNode<INode> node)
{
Operation operation = (Operation)node.Value;
HelperFunctionName name;
if (operation.StorageKind == StorageKind.SharedMemory8)
{
name = HelperFunctionName.SharedStore8;
}
else if (operation.StorageKind == StorageKind.SharedMemory16)
{
name = HelperFunctionName.SharedStore16;
}
else
{
return node;
}
if (operation.Inst != Instruction.Store)
{
return node;
}
Operand memoryId = operation.GetSource(0);
Operand byteOffset = operation.GetSource(1);
Operand value = operation.GetSource(2);
Debug.Assert(memoryId.Type == OperandType.Constant);
int functionId = hfm.GetOrCreateFunctionId(name, memoryId.Value);
Operand[] callArgs = new Operand[] { Const(functionId), byteOffset, value };
LinkedListNode<INode> newNode = node.List.AddBefore(node, new Operation(Instruction.Call, 0, (Operand)null, callArgs));
Utils.DeleteNode(node, operation);
return newNode;
}
private static LinkedListNode<INode> InsertSharedAtomicSigned(HelperFunctionManager hfm, LinkedListNode<INode> node)
{
Operation operation = (Operation)node.Value;
HelperFunctionName name;
if (operation.Inst == Instruction.AtomicMaxS32)
{
name = HelperFunctionName.SharedAtomicMaxS32;
}
else if (operation.Inst == Instruction.AtomicMinS32)
{
name = HelperFunctionName.SharedAtomicMinS32;
}
else
{
return node;
}
if (operation.StorageKind != StorageKind.SharedMemory)
{
return node;
}
Operand result = operation.Dest;
Operand memoryId = operation.GetSource(0);
Operand byteOffset = operation.GetSource(1);
Operand value = operation.GetSource(2);
Debug.Assert(memoryId.Type == OperandType.Constant);
int functionId = hfm.GetOrCreateFunctionId(name, memoryId.Value);
Operand[] callArgs = new Operand[] { Const(functionId), byteOffset, value };
LinkedListNode<INode> newNode = node.List.AddBefore(node, new Operation(Instruction.Call, 0, result, callArgs));
Utils.DeleteNode(node, operation);
return newNode;
}
private static LinkedListNode<INode> InsertTexelFetchScale(HelperFunctionManager hfm, LinkedListNode<INode> node, ShaderConfig config)
{
TextureOperation texOp = (TextureOperation)node.Value;

16
src/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs
View file

@ -124,7 +124,7 @@ namespace Ryujinx.Graphics.Shader.Translation
private TextureDescriptor[] _cachedTextureDescriptors;
private TextureDescriptor[] _cachedImageDescriptors;
public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options)
public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options, int localMemorySize)
{
Stage = stage;
GpuAccessor = gpuAccessor;
@ -143,7 +143,7 @@ namespace Ryujinx.Graphics.Shader.Translation
_usedTextures = new Dictionary<TextureInfo, TextureMeta>();
_usedImages = new Dictionary<TextureInfo, TextureMeta>();
ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties());
ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties(), localMemorySize);
if (!gpuAccessor.QueryHostSupportsTransformFeedback() && gpuAccessor.QueryTransformFeedbackEnabled())
{
@ -176,14 +176,17 @@ namespace Ryujinx.Graphics.Shader.Translation
OutputTopology outputTopology,
int maxOutputVertices,
IGpuAccessor gpuAccessor,
TranslationOptions options) : this(stage, gpuAccessor, options)
TranslationOptions options) : this(stage, gpuAccessor, options, 0)
{
ThreadsPerInputPrimitive = 1;
OutputTopology = outputTopology;
MaxOutputVertices = maxOutputVertices;
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(header.Stage, gpuAccessor, options)
public ShaderConfig(
ShaderHeader header,
IGpuAccessor gpuAccessor,
TranslationOptions options) : this(header.Stage, gpuAccessor, options, GetLocalMemorySize(header))
{
GpPassthrough = header.Stage == ShaderStage.Geometry && header.GpPassthrough;
ThreadsPerInputPrimitive = header.ThreadsPerInputPrimitive;
@ -197,6 +200,11 @@ namespace Ryujinx.Graphics.Shader.Translation
LastInVertexPipeline = header.Stage < ShaderStage.Fragment;
}
private static int GetLocalMemorySize(ShaderHeader header)
{
return header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize + (header.ShaderLocalMemoryCrsSize / ThreadsPerWarp);
}
private void EnsureTransformFeedbackInitialized()
{
if (HasTransformFeedbackOutputs() && _transformFeedbackOutputs == null)

2
src/Ryujinx.Graphics.Shader/Translation/Translator.cs
View file

@ -107,7 +107,7 @@ namespace Ryujinx.Graphics.Shader.Translation
if (options.Flags.HasFlag(TranslationFlags.Compute))
{
config = new ShaderConfig(ShaderStage.Compute, gpuAccessor, options);
config = new ShaderConfig(ShaderStage.Compute, gpuAccessor, options, gpuAccessor.QueryComputeLocalMemorySize());
program = Decoder.Decode(config, address);
}