0
0
Fork 0

Replace constant buffer access on shader with new Load instruction (#4646)

This commit is contained in:
gdkchan 2023-05-20 16:19:26 -03:00 committed by GitHub
parent fb27042e01
commit 402f05b8ef
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
42 changed files with 788 additions and 625 deletions

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 = 5027;
private const uint CodeGenVersion = 4646;
private const string SharedTocFileName = "shared.toc";
private const string SharedDataFileName = "shared.data";

View file

@ -4,6 +4,7 @@ using Ryujinx.Graphics.Gpu.Engine.Threed;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Shader;
using Ryujinx.Graphics.Shader.Translation;
using System;
namespace Ryujinx.Graphics.Gpu.Shader
{
@ -16,6 +17,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
private readonly ResourceCounts _resourceCounts;
private readonly int _stageIndex;
private readonly int[] _constantBufferBindings;
/// <summary>
/// Creates a new GPU accessor.
/// </summary>
@ -25,6 +28,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
_context = context;
_resourceCounts = resourceCounts;
_stageIndex = stageIndex;
if (context.Capabilities.Api != TargetApi.Vulkan)
{
_constantBufferBindings = new int[Constants.TotalGpUniformBuffers];
_constantBufferBindings.AsSpan().Fill(-1);
}
}
public int QueryBindingConstantBuffer(int index)
@ -36,7 +45,15 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
else
{
return _resourceCounts.UniformBuffersCount++;
int binding = _constantBufferBindings[index];
if (binding < 0)
{
binding = _resourceCounts.UniformBuffersCount++;
_constantBufferBindings[index] = binding;
}
return binding;
}
}

View file

@ -3,6 +3,7 @@ using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
using System.Globalization;
using System.Linq;
using System.Numerics;
@ -102,13 +103,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.AppendLine();
}
var cBufferDescriptors = context.Config.GetConstantBufferDescriptors();
if (cBufferDescriptors.Length != 0)
{
DeclareUniforms(context, cBufferDescriptors);
context.AppendLine();
}
DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
var sBufferDescriptors = context.Config.GetStorageBufferDescriptors();
if (sBufferDescriptors.Length != 0)
@ -265,18 +260,12 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
scaleElements++; // Also includes render target scale, for gl_FragCoord.
}
DeclareSupportUniformBlock(context, context.Config.Stage, scaleElements);
if (context.Config.UsedFeatures.HasFlag(FeatureFlags.IntegerSampling) && scaleElements != 0)
{
AppendHelperFunction(context, $"Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/TexelFetchScale_{stage}.glsl");
context.AppendLine();
}
}
else if (isFragment || context.Config.Stage == ShaderStage.Vertex)
{
DeclareSupportUniformBlock(context, context.Config.Stage, 0);
}
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.AtomicMinMaxS32Shared) != 0)
@ -389,36 +378,38 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
};
}
private static void DeclareUniforms(CodeGenContext context, BufferDescriptor[] descriptors)
private static void DeclareConstantBuffers(CodeGenContext context, IEnumerable<BufferDefinition> buffers)
{
string ubSize = "[" + NumberFormatter.FormatInt(Constants.ConstantBufferSize / 16) + "]";
if (context.Config.UsedFeatures.HasFlag(FeatureFlags.CbIndexing))
foreach (BufferDefinition buffer in buffers)
{
string ubName = OperandManager.GetShaderStagePrefix(context.Config.Stage);
ubName += "_" + DefaultNames.UniformNamePrefix;
string blockName = $"{ubName}_{DefaultNames.BlockSuffix}";
context.AppendLine($"layout (binding = {context.Config.FirstConstantBufferBinding}, std140) uniform {blockName}");
context.EnterScope();
context.AppendLine("vec4 " + DefaultNames.DataName + ubSize + ";");
context.LeaveScope($" {ubName}[{NumberFormatter.FormatInt(descriptors.Max(x => x.Slot) + 1)}];");
}
else
{
foreach (var descriptor in descriptors)
string layout = buffer.Layout switch
{
string ubName = OperandManager.GetShaderStagePrefix(context.Config.Stage);
BufferLayout.Std140 => "std140",
_ => "std430"
};
ubName += "_" + DefaultNames.UniformNamePrefix + descriptor.Slot;
context.AppendLine($"layout (binding = {buffer.Binding}, {layout}) uniform _{buffer.Name}");
context.EnterScope();
context.AppendLine($"layout (binding = {descriptor.Binding}, std140) uniform {ubName}");
context.EnterScope();
context.AppendLine("vec4 " + OperandManager.GetUbName(context.Config.Stage, descriptor.Slot, false) + ubSize + ";");
context.LeaveScope(";");
foreach (StructureField field in buffer.Type.Fields)
{
if (field.Type.HasFlag(AggregateType.Array))
{
string typeName = GetVarTypeName(context, field.Type & ~AggregateType.Array);
string arraySize = field.ArrayLength.ToString(CultureInfo.InvariantCulture);
context.AppendLine($"{typeName} {field.Name}[{arraySize}];");
}
else
{
string typeName = GetVarTypeName(context, field.Type);
context.AppendLine($"{typeName} {field.Name};");
}
}
context.LeaveScope($" {buffer.Name};");
context.AppendLine();
}
}
@ -759,39 +750,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.AppendLine($"layout (location = {location}) patch out vec4 {name};");
}
private static void DeclareSupportUniformBlock(CodeGenContext context, ShaderStage stage, int scaleElements)
{
bool needsSupportBlock = stage == ShaderStage.Fragment ||
(context.Config.LastInVertexPipeline && context.Config.GpuAccessor.QueryViewportTransformDisable());
if (!needsSupportBlock && scaleElements == 0)
{
return;
}
context.AppendLine($"layout (binding = 0, std140) uniform {DefaultNames.SupportBlockName}");
context.EnterScope();
switch (stage)
{
case ShaderStage.Fragment:
case ShaderStage.Vertex:
context.AppendLine($"uint {DefaultNames.SupportBlockAlphaTestName};");
context.AppendLine($"bool {DefaultNames.SupportBlockIsBgraName}[{SupportBuffer.FragmentIsBgraCount}];");
context.AppendLine($"vec4 {DefaultNames.SupportBlockViewportInverse};");
context.AppendLine($"int {DefaultNames.SupportBlockFragmentScaleCount};");
break;
case ShaderStage.Compute:
context.AppendLine($"uint s_reserved[{SupportBuffer.ComputeRenderScaleOffset / SupportBuffer.FieldSize}];");
break;
}
context.AppendLine($"float {DefaultNames.SupportBlockRenderScaleName}[{SupportBuffer.RenderScaleMaxCount}];");
context.LeaveScope(";");
context.AppendLine();
}
private static void AppendHelperFunction(CodeGenContext context, string filename)
{
string code = EmbeddedResources.ReadAllText(filename);

View file

@ -15,18 +15,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public const string DataName = "data";
public const string SupportBlockName = "support_block";
public const string SupportBlockAlphaTestName = "s_alpha_test";
public const string SupportBlockIsBgraName = "s_is_bgra";
public const string SupportBlockViewportInverse = "s_viewport_inverse";
public const string SupportBlockFragmentScaleCount = "s_frag_scale_count";
public const string SupportBlockRenderScaleName = "s_render_scale";
public const string BlockSuffix = "block";
public const string UniformNamePrefix = "c";
public const string UniformNameSuffix = "data";
public const string LocalMemoryName = "local_mem";
public const string SharedMemoryName = "shared_mem";

View file

@ -1,6 +1,6 @@
ivec2 Helper_TexelFetchScale(ivec2 inputVec, int samplerIndex)
{
float scale = s_render_scale[samplerIndex];
float scale = support_buffer.s_render_scale[1 + samplerIndex];
if (scale == 1.0)
{
return inputVec;
@ -10,7 +10,7 @@
int Helper_TextureSizeUnscale(int size, int samplerIndex)
{
float scale = s_render_scale[samplerIndex];
float scale = support_buffer.s_render_scale[1 + samplerIndex];
if (scale == 1.0)
{
return size;

View file

@ -1,6 +1,6 @@
ivec2 Helper_TexelFetchScale(ivec2 inputVec, int samplerIndex)
{
float scale = s_render_scale[1 + samplerIndex];
float scale = support_buffer.s_render_scale[1 + samplerIndex];
if (scale == 1.0)
{
return inputVec;
@ -17,7 +17,7 @@
int Helper_TextureSizeUnscale(int size, int samplerIndex)
{
float scale = abs(s_render_scale[1 + samplerIndex]);
float scale = abs(support_buffer.s_render_scale[1 + samplerIndex]);
if (scale == 1.0)
{
return size;

View file

@ -1,6 +1,6 @@
ivec2 Helper_TexelFetchScale(ivec2 inputVec, int samplerIndex)
{
float scale = abs(s_render_scale[1 + samplerIndex + s_frag_scale_count]);
float scale = abs(support_buffer.s_render_scale[1 + samplerIndex + support_buffer.s_frag_scale_count]);
if (scale == 1.0)
{
return inputVec;
@ -11,7 +11,7 @@
int Helper_TextureSizeUnscale(int size, int samplerIndex)
{
float scale = abs(s_render_scale[1 + samplerIndex + s_frag_scale_count]);
float scale = abs(support_buffer.s_render_scale[1 + samplerIndex + support_buffer.s_frag_scale_count]);
if (scale == 1.0)
{
return size;

View file

@ -167,9 +167,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.Load:
return Load(context, operation);
case Instruction.LoadConstant:
return LoadConstant(context, operation);
case Instruction.LoadLocal:
return LoadLocal(context, operation);

View file

@ -83,7 +83,6 @@ 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.LoadConstant, InstType.Special);
Add(Instruction.LoadLocal, InstType.Special);
Add(Instruction.LoadShared, InstType.Special);
Add(Instruction.LoadStorage, InstType.Special);

View file

@ -215,29 +215,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
return GenerateLoadOrStore(context, operation, isStore: false);
}
public static string LoadConstant(CodeGenContext context, AstOperation operation)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
offsetExpr = Enclose(offsetExpr, src2, Instruction.ShiftRightS32, isLhs: true);
var config = context.Config;
bool indexElement = !config.GpuAccessor.QueryHostHasVectorIndexingBug();
if (src1 is AstOperand operand && operand.Type == OperandType.Constant)
{
bool cbIndexable = config.UsedFeatures.HasFlag(Translation.FeatureFlags.CbIndexing);
return OperandManager.GetConstantBufferName(operand.Value, offsetExpr, config.Stage, cbIndexable, indexElement);
}
else
{
string slotExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
return OperandManager.GetConstantBufferName(slotExpr, offsetExpr, config.Stage, indexElement);
}
}
public static string LoadLocal(CodeGenContext context, AstOperation operation)
{
return LoadLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
@ -809,9 +786,29 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
string varName;
AggregateType varType;
int srcIndex = 0;
int inputsCount = isStore ? operation.SourcesCount - 1 : operation.SourcesCount;
switch (storageKind)
{
case StorageKind.ConstantBuffer:
if (!(operation.GetSource(srcIndex++) is AstOperand bindingIndex) || bindingIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
int binding = bindingIndex.Value;
BufferDefinition buffer = context.Config.Properties.ConstantBuffers[binding];
if (!(operation.GetSource(srcIndex++) is AstOperand fieldIndex) || fieldIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"Second input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
StructureField field = buffer.Type.Fields[fieldIndex.Value];
varName = $"{buffer.Name}.{field.Name}";
varType = field.Type;
break;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:
@ -864,40 +861,39 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
varName = $"gl_out[{expr}].{varName}";
}
}
int firstSrcIndex = srcIndex;
int inputsCount = isStore ? operation.SourcesCount - 1 : operation.SourcesCount;
for (; srcIndex < inputsCount; srcIndex++)
{
IAstNode src = operation.GetSource(srcIndex);
if ((varType & AggregateType.ElementCountMask) != 0 &&
srcIndex == inputsCount - 1 &&
src is AstOperand elementIndex &&
elementIndex.Type == OperandType.Constant)
{
varName += "." + "xyzw"[elementIndex.Value & 3];
}
else if (srcIndex == firstSrcIndex && context.Config.Stage == ShaderStage.TessellationControl && storageKind == StorageKind.Output)
{
// GLSL requires that for tessellation control shader outputs,
// that the index expression must be *exactly* "gl_InvocationID",
// otherwise the compilation fails.
// TODO: Get rid of this and use expression propagation to make sure we generate the correct code from IR.
varName += "[gl_InvocationID]";
}
else
{
varName += $"[{GetSoureExpr(context, src, AggregateType.S32)}]";
}
}
break;
default:
throw new InvalidOperationException($"Invalid storage kind {storageKind}.");
}
int firstSrcIndex = srcIndex;
for (; srcIndex < inputsCount; srcIndex++)
{
IAstNode src = operation.GetSource(srcIndex);
if ((varType & AggregateType.ElementCountMask) != 0 &&
srcIndex == inputsCount - 1 &&
src is AstOperand elementIndex &&
elementIndex.Type == OperandType.Constant)
{
varName += "." + "xyzw"[elementIndex.Value & 3];
}
else if (srcIndex == firstSrcIndex && context.Config.Stage == ShaderStage.TessellationControl && storageKind == StorageKind.Output)
{
// GLSL requires that for tessellation control shader outputs,
// that the index expression must be *exactly* "gl_InvocationID",
// otherwise the compilation fails.
// TODO: Get rid of this and use expression propagation to make sure we generate the correct code from IR.
varName += "[gl_InvocationID]";
}
else
{
varName += $"[{GetSoureExpr(context, src, AggregateType.S32)}]";
}
}
if (isStore)
{
varType &= AggregateType.ElementTypeMask;

View file

@ -27,7 +27,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
IoVariable.FragmentCoord => ("gl_FragCoord", AggregateType.Vector4 | AggregateType.FP32),
IoVariable.FragmentOutputColor => GetFragmentOutputColorVariableName(config, location),
IoVariable.FragmentOutputDepth => ("gl_FragDepth", AggregateType.FP32),
IoVariable.FragmentOutputIsBgra => (DefaultNames.SupportBlockIsBgraName, AggregateType.Array | AggregateType.Bool),
IoVariable.FrontColorDiffuse => ("gl_FrontColor", AggregateType.Vector4 | AggregateType.FP32), // Deprecated.
IoVariable.FrontColorSpecular => ("gl_FrontSecondaryColor", AggregateType.Vector4 | AggregateType.FP32), // Deprecated.
IoVariable.FrontFacing => ("gl_FrontFacing", AggregateType.Bool),
@ -46,8 +45,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
IoVariable.SubgroupLaneId => GetSubgroupInvocationIdVariableName(config),
IoVariable.SubgroupLeMask => GetSubgroupMaskVariableName(config, "Le"),
IoVariable.SubgroupLtMask => GetSubgroupMaskVariableName(config, "Lt"),
IoVariable.SupportBlockRenderScale => (DefaultNames.SupportBlockRenderScaleName, AggregateType.Array | AggregateType.FP32),
IoVariable.SupportBlockViewInverse => (DefaultNames.SupportBlockViewportInverse, AggregateType.Vector2 | AggregateType.FP32),
IoVariable.TessellationCoord => ("gl_TessCoord", AggregateType.Vector3 | AggregateType.FP32),
IoVariable.TessellationLevelInner => ("gl_TessLevelInner", AggregateType.Array | AggregateType.FP32),
IoVariable.TessellationLevelOuter => ("gl_TessLevelOuter", AggregateType.Array | AggregateType.FP32),

View file

@ -36,63 +36,12 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{
OperandType.Argument => GetArgumentName(operand.Value),
OperandType.Constant => NumberFormatter.FormatInt(operand.Value),
OperandType.ConstantBuffer => GetConstantBufferName(operand, context.Config),
OperandType.LocalVariable => _locals[operand],
OperandType.Undefined => DefaultNames.UndefinedName,
_ => throw new ArgumentException($"Invalid operand type \"{operand.Type}\".")
};
}
private static string GetConstantBufferName(AstOperand operand, ShaderConfig config)
{
return GetConstantBufferName(operand.CbufSlot, operand.CbufOffset, config.Stage, config.UsedFeatures.HasFlag(FeatureFlags.CbIndexing));
}
public static string GetConstantBufferName(int slot, int offset, ShaderStage stage, bool cbIndexable)
{
return $"{GetUbName(stage, slot, cbIndexable)}[{offset >> 2}].{GetSwizzleMask(offset & 3)}";
}
private static string GetVec4Indexed(string vectorName, string indexExpr, bool indexElement)
{
if (indexElement)
{
return $"{vectorName}[{indexExpr}]";
}
string result = $"{vectorName}.x";
for (int i = 1; i < 4; i++)
{
result = $"(({indexExpr}) == {i}) ? ({vectorName}.{GetSwizzleMask(i)}) : ({result})";
}
return $"({result})";
}
public static string GetConstantBufferName(int slot, string offsetExpr, ShaderStage stage, bool cbIndexable, bool indexElement)
{
return GetVec4Indexed(GetUbName(stage, slot, cbIndexable) + $"[{offsetExpr} >> 2]", offsetExpr + " & 3", indexElement);
}
public static string GetConstantBufferName(string slotExpr, string offsetExpr, ShaderStage stage, bool indexElement)
{
return GetVec4Indexed(GetUbName(stage, slotExpr) + $"[{offsetExpr} >> 2]", offsetExpr + " & 3", indexElement);
}
public static string GetUbName(ShaderStage stage, int slot, bool cbIndexable)
{
if (cbIndexable)
{
return GetUbName(stage, NumberFormatter.FormatInt(slot, AggregateType.S32));
}
return $"{GetShaderStagePrefix(stage)}_{DefaultNames.UniformNamePrefix}{slot}_{DefaultNames.UniformNameSuffix}";
}
private static string GetUbName(ShaderStage stage, string slotExpr)
{
return $"{GetShaderStagePrefix(stage)}_{DefaultNames.UniformNamePrefix}[{slotExpr}].{DefaultNames.DataName}";
}
public static string GetSamplerName(ShaderStage stage, AstTextureOperation texOp, string indexExpr)
{
return GetSamplerName(stage, texOp.CbufSlot, texOp.Handle, texOp.Type.HasFlag(SamplerType.Indexed), indexExpr);
@ -168,6 +117,22 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{
switch (operation.StorageKind)
{
case StorageKind.ConstantBuffer:
if (!(operation.GetSource(0) is AstOperand bindingIndex) || bindingIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {operation.StorageKind} storage must be a constant operand.");
}
if (!(operation.GetSource(1) is AstOperand fieldIndex) || fieldIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"Second input of {operation.Inst} with {operation.StorageKind} storage must be a constant operand.");
}
BufferDefinition buffer = context.Config.Properties.ConstantBuffers[bindingIndex.Value];
StructureField field = buffer.Type.Fields[fieldIndex.Value];
return field.Type & AggregateType.ElementTypeMask;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:

View file

@ -23,9 +23,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
public int InputVertices { get; }
public Dictionary<int, Instruction> UniformBuffers { get; } = new Dictionary<int, Instruction>();
public Instruction SupportBuffer { get; set; }
public Instruction UniformBuffersArray { get; set; }
public Dictionary<int, Instruction> ConstantBuffers { get; } = new Dictionary<int, Instruction>();
public Instruction StorageBuffersArray { get; set; }
public Instruction LocalMemory { get; set; }
public Instruction SharedMemory { get; set; }
@ -217,7 +215,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
IrOperandType.Argument => GetArgument(type, operand),
IrOperandType.Constant => GetConstant(type, operand),
IrOperandType.ConstantBuffer => GetConstantBuffer(type, operand),
IrOperandType.LocalVariable => GetLocal(type, operand),
IrOperandType.Undefined => GetUndefined(type),
_ => throw new ArgumentException($"Invalid operand type \"{operand.Type}\".")
@ -274,31 +271,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
};
}
public Instruction GetConstantBuffer(AggregateType type, AstOperand operand)
{
var i1 = Constant(TypeS32(), 0);
var i2 = Constant(TypeS32(), operand.CbufOffset >> 2);
var i3 = Constant(TypeU32(), operand.CbufOffset & 3);
Instruction elemPointer;
if (UniformBuffersArray != null)
{
var ubVariable = UniformBuffersArray;
var i0 = Constant(TypeS32(), operand.CbufSlot);
elemPointer = AccessChain(TypePointer(StorageClass.Uniform, TypeFP32()), ubVariable, i0, i1, i2, i3);
}
else
{
var ubVariable = UniformBuffers[operand.CbufSlot];
elemPointer = AccessChain(TypePointer(StorageClass.Uniform, TypeFP32()), ubVariable, i1, i2, i3);
}
return BitcastIfNeeded(type, AggregateType.FP32, Load(TypeFP32(), elemPointer));
}
public Instruction GetLocalPointer(AstOperand local)
{
return _locals[local];

View file

@ -98,8 +98,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
DeclareLocalMemory(context, localMemorySize);
}
DeclareSupportBuffer(context);
DeclareUniformBuffers(context, context.Config.GetConstantBufferDescriptors());
DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
DeclareStorageBuffers(context, context.Config.GetStorageBufferDescriptors());
DeclareSamplers(context, context.Config.GetTextureDescriptors());
DeclareImages(context, context.Config.GetImageDescriptors());
@ -127,84 +126,63 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
return variable;
}
private static void DeclareSupportBuffer(CodeGenContext context)
private static void DeclareConstantBuffers(CodeGenContext context, IEnumerable<BufferDefinition> buffers)
{
if (!context.Config.Stage.SupportsRenderScale() && !(context.Config.LastInVertexPipeline && context.Config.GpuAccessor.QueryViewportTransformDisable()))
HashSet<SpvInstruction> decoratedTypes = new HashSet<SpvInstruction>();
foreach (BufferDefinition buffer in buffers)
{
return;
}
int alignment = buffer.Layout == BufferLayout.Std140 ? 16 : 4;
int alignmentMask = alignment - 1;
int offset = 0;
var isBgraArrayType = context.TypeArray(context.TypeU32(), context.Constant(context.TypeU32(), SupportBuffer.FragmentIsBgraCount));
var viewportInverseVectorType = context.TypeVector(context.TypeFP32(), 4);
var renderScaleArrayType = context.TypeArray(context.TypeFP32(), context.Constant(context.TypeU32(), SupportBuffer.RenderScaleMaxCount));
SpvInstruction[] structFieldTypes = new SpvInstruction[buffer.Type.Fields.Length];
int[] structFieldOffsets = new int[buffer.Type.Fields.Length];
context.Decorate(isBgraArrayType, Decoration.ArrayStride, (LiteralInteger)SupportBuffer.FieldSize);
context.Decorate(renderScaleArrayType, Decoration.ArrayStride, (LiteralInteger)SupportBuffer.FieldSize);
for (int fieldIndex = 0; fieldIndex < buffer.Type.Fields.Length; fieldIndex++)
{
StructureField field = buffer.Type.Fields[fieldIndex];
int fieldSize = (field.Type.GetSizeInBytes() + alignmentMask) & ~alignmentMask;
var supportBufferStructType = context.TypeStruct(false, context.TypeU32(), isBgraArrayType, viewportInverseVectorType, context.TypeS32(), renderScaleArrayType);
structFieldTypes[fieldIndex] = context.GetType(field.Type, field.ArrayLength);
structFieldOffsets[fieldIndex] = offset;
context.MemberDecorate(supportBufferStructType, 0, Decoration.Offset, (LiteralInteger)SupportBuffer.FragmentAlphaTestOffset);
context.MemberDecorate(supportBufferStructType, 1, Decoration.Offset, (LiteralInteger)SupportBuffer.FragmentIsBgraOffset);
context.MemberDecorate(supportBufferStructType, 2, Decoration.Offset, (LiteralInteger)SupportBuffer.ViewportInverseOffset);
context.MemberDecorate(supportBufferStructType, 3, Decoration.Offset, (LiteralInteger)SupportBuffer.FragmentRenderScaleCountOffset);
context.MemberDecorate(supportBufferStructType, 4, Decoration.Offset, (LiteralInteger)SupportBuffer.GraphicsRenderScaleOffset);
context.Decorate(supportBufferStructType, Decoration.Block);
if (field.Type.HasFlag(AggregateType.Array))
{
// We can't decorate the type more than once.
if (decoratedTypes.Add(structFieldTypes[fieldIndex]))
{
context.Decorate(structFieldTypes[fieldIndex], Decoration.ArrayStride, (LiteralInteger)fieldSize);
}
var supportBufferPointerType = context.TypePointer(StorageClass.Uniform, supportBufferStructType);
var supportBufferVariable = context.Variable(supportBufferPointerType, StorageClass.Uniform);
offset += fieldSize * field.ArrayLength;
}
else
{
offset += fieldSize;
}
}
context.Decorate(supportBufferVariable, Decoration.DescriptorSet, (LiteralInteger)0);
context.Decorate(supportBufferVariable, Decoration.Binding, (LiteralInteger)0);
var ubStructType = context.TypeStruct(false, structFieldTypes);
context.AddGlobalVariable(supportBufferVariable);
if (decoratedTypes.Add(ubStructType))
{
context.Decorate(ubStructType, Decoration.Block);
context.SupportBuffer = supportBufferVariable;
}
for (int fieldIndex = 0; fieldIndex < structFieldOffsets.Length; fieldIndex++)
{
context.MemberDecorate(ubStructType, fieldIndex, Decoration.Offset, (LiteralInteger)structFieldOffsets[fieldIndex]);
}
}
private static void DeclareUniformBuffers(CodeGenContext context, BufferDescriptor[] descriptors)
{
if (descriptors.Length == 0)
{
return;
}
uint ubSize = Constants.ConstantBufferSize / 16;
var ubArrayType = context.TypeArray(context.TypeVector(context.TypeFP32(), 4), context.Constant(context.TypeU32(), ubSize), true);
context.Decorate(ubArrayType, Decoration.ArrayStride, (LiteralInteger)16);
var ubStructType = context.TypeStruct(true, ubArrayType);
context.Decorate(ubStructType, Decoration.Block);
context.MemberDecorate(ubStructType, 0, Decoration.Offset, (LiteralInteger)0);
if (context.Config.UsedFeatures.HasFlag(FeatureFlags.CbIndexing))
{
int count = descriptors.Max(x => x.Slot) + 1;
var ubStructArrayType = context.TypeArray(ubStructType, context.Constant(context.TypeU32(), count));
var ubPointerType = context.TypePointer(StorageClass.Uniform, ubStructArrayType);
var ubPointerType = context.TypePointer(StorageClass.Uniform, ubStructType);
var ubVariable = context.Variable(ubPointerType, StorageClass.Uniform);
context.Name(ubVariable, $"{GetStagePrefix(context.Config.Stage)}_u");
context.Decorate(ubVariable, Decoration.DescriptorSet, (LiteralInteger)0);
context.Decorate(ubVariable, Decoration.Binding, (LiteralInteger)context.Config.FirstConstantBufferBinding);
context.Name(ubVariable, buffer.Name);
context.Decorate(ubVariable, Decoration.DescriptorSet, (LiteralInteger)buffer.Set);
context.Decorate(ubVariable, Decoration.Binding, (LiteralInteger)buffer.Binding);
context.AddGlobalVariable(ubVariable);
context.UniformBuffersArray = ubVariable;
}
else
{
var ubPointerType = context.TypePointer(StorageClass.Uniform, ubStructType);
foreach (var descriptor in descriptors)
{
var ubVariable = context.Variable(ubPointerType, StorageClass.Uniform);
context.Name(ubVariable, $"{GetStagePrefix(context.Config.Stage)}_c{descriptor.Slot}");
context.Decorate(ubVariable, Decoration.DescriptorSet, (LiteralInteger)0);
context.Decorate(ubVariable, Decoration.Binding, (LiteralInteger)descriptor.Binding);
context.AddGlobalVariable(ubVariable);
context.UniformBuffers.Add(descriptor.Slot, ubVariable);
}
context.ConstantBuffers.Add(buffer.Binding, ubVariable);
}
}
@ -394,25 +372,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
foreach (var ioDefinition in info.IoDefinitions)
{
var ioVariable = ioDefinition.IoVariable;
// Those are actually from constant buffer, rather than being actual inputs or outputs,
// so we must ignore them here as they are declared as part of the support buffer.
// TODO: Delete this after we represent this properly on the IR (as a constant buffer rather than "input").
if (ioVariable == IoVariable.FragmentOutputIsBgra ||
ioVariable == IoVariable.SupportBlockRenderScale ||
ioVariable == IoVariable.SupportBlockViewInverse)
{
continue;
}
bool isOutput = ioDefinition.StorageKind.IsOutput();
bool isPerPatch = ioDefinition.StorageKind.IsPerPatch();
PixelImap iq = PixelImap.Unused;
if (context.Config.Stage == ShaderStage.Fragment)
{
var ioVariable = ioDefinition.IoVariable;
if (ioVariable == IoVariable.UserDefined)
{
iq = context.Config.ImapTypes[ioDefinition.Location].GetFirstUsedType();
@ -429,6 +393,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
}
}
bool isOutput = ioDefinition.StorageKind.IsOutput();
bool isPerPatch = ioDefinition.StorageKind.IsPerPatch();
DeclareInputOrOutput(context, ioDefinition, isOutput, isPerPatch, iq);
}
}

View file

@ -98,7 +98,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
Add(Instruction.ImageStore, GenerateImageStore);
Add(Instruction.IsNan, GenerateIsNan);
Add(Instruction.Load, GenerateLoad);
Add(Instruction.LoadConstant, GenerateLoadConstant);
Add(Instruction.LoadLocal, GenerateLoadLocal);
Add(Instruction.LoadShared, GenerateLoadShared);
Add(Instruction.LoadStorage, GenerateLoadStorage);
@ -313,10 +312,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
for (int i = 0; i < args.Length; i++)
{
var operand = (AstOperand)operation.GetSource(i + 1);
var operand = operation.GetSource(i + 1);
if (i >= function.InArguments.Length)
{
args[i] = context.GetLocalPointer(operand);
args[i] = context.GetLocalPointer((AstOperand)operand);
}
else
{
@ -867,68 +867,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
return GenerateLoadOrStore(context, operation, isStore: false);
}
private static OperationResult GenerateLoadConstant(CodeGenContext context, AstOperation operation)
{
var src1 = operation.GetSource(0);
var src2 = context.Get(AggregateType.S32, operation.GetSource(1));
var i1 = context.Constant(context.TypeS32(), 0);
var i2 = context.ShiftRightArithmetic(context.TypeS32(), src2, context.Constant(context.TypeS32(), 2));
var i3 = context.BitwiseAnd(context.TypeS32(), src2, context.Constant(context.TypeS32(), 3));
SpvInstruction value = null;
if (context.Config.GpuAccessor.QueryHostHasVectorIndexingBug())
{
// Test for each component individually.
for (int i = 0; i < 4; i++)
{
var component = context.Constant(context.TypeS32(), i);
SpvInstruction elemPointer;
if (context.UniformBuffersArray != null)
{
var ubVariable = context.UniformBuffersArray;
var i0 = context.Get(AggregateType.S32, src1);
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i0, i1, i2, component);
}
else
{
var ubVariable = context.UniformBuffers[((AstOperand)src1).Value];
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i1, i2, component);
}
SpvInstruction newValue = context.Load(context.TypeFP32(), elemPointer);
value = value != null ? context.Select(context.TypeFP32(), context.IEqual(context.TypeBool(), i3, component), newValue, value) : newValue;
}
}
else
{
SpvInstruction elemPointer;
if (context.UniformBuffersArray != null)
{
var ubVariable = context.UniformBuffersArray;
var i0 = context.Get(AggregateType.S32, src1);
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i0, i1, i2, i3);
}
else
{
var ubVariable = context.UniformBuffers[((AstOperand)src1).Value];
elemPointer = context.AccessChain(context.TypePointer(StorageClass.Uniform, context.TypeFP32()), ubVariable, i1, i2, i3);
}
value = context.Load(context.TypeFP32(), elemPointer);
}
return new OperationResult(AggregateType.FP32, value);
}
private static OperationResult GenerateLoadLocal(CodeGenContext context, AstOperation operation)
{
return GenerateLoadLocalOrShared(context, operation, StorageClass.Private, context.LocalMemory);
@ -1990,12 +1928,32 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
StorageKind storageKind = operation.StorageKind;
SpvInstruction pointer;
StorageClass storageClass;
SpvInstruction baseObj;
AggregateType varType;
int srcIndex = 0;
switch (storageKind)
{
case StorageKind.ConstantBuffer:
if (!(operation.GetSource(srcIndex++) is AstOperand bindingIndex) || bindingIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
if (!(operation.GetSource(srcIndex) is AstOperand fieldIndex) || fieldIndex.Type != OperandType.Constant)
{
throw new InvalidOperationException($"Second input of {operation.Inst} with {storageKind} storage must be a constant operand.");
}
BufferDefinition buffer = context.Config.Properties.ConstantBuffers[bindingIndex.Value];
StructureField field = buffer.Type.Fields[fieldIndex.Value];
storageClass = StorageClass.Uniform;
varType = field.Type & AggregateType.ElementTypeMask;
baseObj = context.ConstantBuffers[bindingIndex.Value];
break;
case StorageKind.Input:
case StorageKind.InputPerPatch:
case StorageKind.Output:
@ -2038,33 +1996,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
varType = context.Config.GetFragmentOutputColorType(location);
}
else if (ioVariable == IoVariable.FragmentOutputIsBgra)
{
var pointerType = context.TypePointer(StorageClass.Uniform, context.TypeU32());
var elemIndex = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(pointerType, context.SupportBuffer, context.Constant(context.TypeU32(), 1), elemIndex);
varType = AggregateType.U32;
break;
}
else if (ioVariable == IoVariable.SupportBlockRenderScale)
{
var pointerType = context.TypePointer(StorageClass.Uniform, context.TypeFP32());
var elemIndex = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(pointerType, context.SupportBuffer, context.Constant(context.TypeU32(), 4), elemIndex);
varType = AggregateType.FP32;
break;
}
else if (ioVariable == IoVariable.SupportBlockViewInverse)
{
var pointerType = context.TypePointer(StorageClass.Uniform, context.TypeFP32());
var elemIndex = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(pointerType, context.SupportBuffer, context.Constant(context.TypeU32(), 2), elemIndex);
varType = AggregateType.FP32;
break;
}
else
{
(_, varType) = IoMap.GetSpirvBuiltIn(ioVariable);
@ -2072,55 +2003,57 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
varType &= AggregateType.ElementTypeMask;
int inputsCount = (isStore ? operation.SourcesCount - 1 : operation.SourcesCount) - srcIndex;
var storageClass = isOutput ? StorageClass.Output : StorageClass.Input;
storageClass = isOutput ? StorageClass.Output : StorageClass.Input;
var ioDefinition = new IoDefinition(storageKind, ioVariable, location, component);
var dict = isPerPatch
? (isOutput ? context.OutputsPerPatch : context.InputsPerPatch)
: (isOutput ? context.Outputs : context.Inputs);
SpvInstruction baseObj = dict[ioDefinition];
SpvInstruction e0, e1, e2;
switch (inputsCount)
{
case 0:
pointer = baseObj;
break;
case 1:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0);
break;
case 2:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e1 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0, e1);
break;
case 3:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e1 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e2 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0, e1, e2);
break;
default:
var indexes = new SpvInstruction[inputsCount];
int index = 0;
for (; index < inputsCount; srcIndex++, index++)
{
indexes[index] = context.Get(AggregateType.S32, operation.GetSource(srcIndex));
}
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, indexes);
break;
}
baseObj = dict[ioDefinition];
break;
default:
throw new InvalidOperationException($"Invalid storage kind {storageKind}.");
}
int inputsCount = (isStore ? operation.SourcesCount - 1 : operation.SourcesCount) - srcIndex;
SpvInstruction e0, e1, e2;
SpvInstruction pointer;
switch (inputsCount)
{
case 0:
pointer = baseObj;
break;
case 1:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0);
break;
case 2:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e1 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0, e1);
break;
case 3:
e0 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e1 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
e2 = context.Get(AggregateType.S32, operation.GetSource(srcIndex++));
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, e0, e1, e2);
break;
default:
var indexes = new SpvInstruction[inputsCount];
int index = 0;
for (; index < inputsCount; srcIndex++, index++)
{
indexes[index] = context.Get(AggregateType.S32, operation.GetSource(srcIndex));
}
pointer = context.AccessChain(context.TypePointer(storageClass, context.GetType(varType)), baseObj, indexes);
break;
}
if (isStore)
{
context.Store(pointer, context.Get(varType, operation.GetSource(srcIndex)));

View file

@ -63,7 +63,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
if (context.Config.Stage == ShaderStage.Vertex)
{
var scaleCountPointerType = context.TypePointer(StorageClass.Uniform, context.TypeS32());
var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.SupportBuffer, context.Constant(context.TypeU32(), 3));
var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.ConstantBuffers[0], context.Constant(context.TypeU32(), 3));
var scaleCount = context.Load(context.TypeS32(), scaleCountElemPointer);
scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, scaleCount);
@ -71,7 +71,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, context.Constant(context.TypeU32(), 1));
var scaleElemPointer = context.AccessChain(pointerType, context.SupportBuffer, fieldIndex, scaleIndex);
var scaleElemPointer = context.AccessChain(pointerType, context.ConstantBuffers[0], fieldIndex, scaleIndex);
var scale = context.Load(context.TypeFP32(), scaleElemPointer);
var ivector2Type = context.TypeVector(context.TypeS32(), 2);
@ -201,7 +201,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
if (context.Config.Stage == ShaderStage.Vertex)
{
var scaleCountPointerType = context.TypePointer(StorageClass.Uniform, context.TypeS32());
var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.SupportBuffer, context.Constant(context.TypeU32(), 3));
var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.ConstantBuffers[0], context.Constant(context.TypeU32(), 3));
var scaleCount = context.Load(context.TypeS32(), scaleCountElemPointer);
scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, scaleCount);
@ -209,7 +209,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, context.Constant(context.TypeU32(), 1));
var scaleElemPointer = context.AccessChain(pointerType, context.SupportBuffer, fieldIndex, scaleIndex);
var scaleElemPointer = context.AccessChain(pointerType, context.ConstantBuffers[0], fieldIndex, scaleIndex);
var scale = context.GlslFAbs(context.TypeFP32(), context.Load(context.TypeFP32(), scaleElemPointer));
var passthrough = context.FOrdEqual(context.TypeBool(), scale, context.Constant(context.TypeFP32(), 1f));

View file

@ -160,7 +160,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
{
// FragCoord X/Y must be divided by the render target scale, if resolution scaling is active,
// because the shader code is not expecting scaled values.
res = context.FPDivide(res, context.Load(StorageKind.Input, IoVariable.SupportBlockRenderScale, null, Const(0)));
res = context.FPDivide(res, context.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.RenderScale), Const(0)));
}
else if (op.Imm10 == AttributeConsts.FrontFacing && context.Config.GpuAccessor.QueryHostHasFrontFacingBug())
{

View file

@ -1,6 +1,7 @@
using Ryujinx.Graphics.Shader.Decoders;
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.Translation;
using System.Numerics;
using static Ryujinx.Graphics.Shader.Instructions.InstEmitHelper;
using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
@ -80,7 +81,6 @@ namespace Ryujinx.Graphics.Shader.Instructions
Operand addr = context.IAdd(srcA, Const(Imm16ToSInt(op.CbufOffset)));
Operand wordOffset = context.ShiftRightU32(addr, Const(2));
Operand bitOffset = GetBitOffset(context, addr);
for (int index = 0; index < count; index++)
{
@ -92,11 +92,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
}
Operand offset = context.IAdd(wordOffset, Const(index));
Operand value = context.LoadConstant(slot, offset);
Operand value = EmitLoadConstant(context, slot, offset);
if (isSmallInt)
{
value = ExtractSmallInt(context, (LsSize)op.LsSize, bitOffset, value);
value = ExtractSmallInt(context, (LsSize)op.LsSize, GetBitOffset(context, addr), value);
}
context.Copy(Register(dest), value);
@ -154,6 +154,39 @@ namespace Ryujinx.Graphics.Shader.Instructions
EmitStore(context, MemoryRegion.Shared, op.LsSize, GetSrcReg(context, op.SrcA), op.Dest, Imm24ToSInt(op.Imm24));
}
private static Operand EmitLoadConstant(EmitterContext context, Operand slot, Operand offset)
{
Operand vecIndex = context.ShiftRightU32(offset, Const(2));
Operand elemIndex = context.BitwiseAnd(offset, Const(3));
if (slot.Type == OperandType.Constant)
{
int binding = context.Config.ResourceManager.GetConstantBufferBinding(slot.Value);
return context.Load(StorageKind.ConstantBuffer, binding, Const(0), vecIndex, elemIndex);
}
else
{
Operand value = Const(0);
uint cbUseMask = context.Config.GpuAccessor.QueryConstantBufferUse();
while (cbUseMask != 0)
{
int cbIndex = BitOperations.TrailingZeroCount(cbUseMask);
int binding = context.Config.ResourceManager.GetConstantBufferBinding(cbIndex);
Operand isCurrent = context.ICompareEqual(slot, Const(cbIndex));
Operand currentValue = context.Load(StorageKind.ConstantBuffer, binding, Const(0), vecIndex, elemIndex);
value = context.ConditionalSelect(isCurrent, currentValue, value);
cbUseMask &= ~(1u << cbIndex);
}
return value;
}
}
private static Operand EmitAtomicOp(
EmitterContext context,
StorageKind storageKind,

View file

@ -79,7 +79,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
ImageAtomic,
IsNan,
Load,
LoadConstant,
LoadGlobal,
LoadLocal,
LoadShared,

View file

@ -15,7 +15,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
FragmentCoord,
FragmentOutputColor,
FragmentOutputDepth,
FragmentOutputIsBgra, // TODO: Remove and use constant buffer access.
FrontColorDiffuse,
FrontColorSpecular,
FrontFacing,
@ -34,8 +33,6 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
SubgroupLaneId,
SubgroupLeMask,
SubgroupLtMask,
SupportBlockViewInverse, // TODO: Remove and use constant buffer access.
SupportBlockRenderScale, // TODO: Remove and use constant buffer access.
TessellationCoord,
TessellationLevelInner,
TessellationLevelOuter,

View file

@ -15,9 +15,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
public int Value { get; }
public int CbufSlot { get; }
public int CbufOffset { get; }
private AstOperand()
{
Defs = new HashSet<IAstNode>();
@ -29,16 +26,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
public AstOperand(Operand operand) : this()
{
Type = operand.Type;
if (Type == OperandType.ConstantBuffer)
{
CbufSlot = operand.GetCbufSlot();
CbufOffset = operand.GetCbufOffset();
}
else
{
Value = operand.Value;
}
Value = operand.Value;
}
public AstOperand(OperandType type, int value = 0) : this()

View file

@ -0,0 +1,20 @@
namespace Ryujinx.Graphics.Shader.StructuredIr
{
readonly struct BufferDefinition
{
public BufferLayout Layout { get; }
public int Set { get; }
public int Binding { get; }
public string Name { get; }
public StructureType Type { get; }
public BufferDefinition(BufferLayout layout, int set, int binding, string name, StructureType type)
{
Layout = layout;
Set = set;
Binding = binding;
Name = name;
Type = type;
}
}
}

View file

@ -0,0 +1,8 @@
namespace Ryujinx.Graphics.Shader.StructuredIr
{
enum BufferLayout
{
Std140,
Std430
}
}

View file

@ -90,7 +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.LoadConstant, AggregateType.FP32, AggregateType.S32, AggregateType.S32);
Add(Instruction.LoadGlobal, AggregateType.U32, AggregateType.S32, AggregateType.S32);
Add(Instruction.LoadLocal, AggregateType.U32, AggregateType.S32);
Add(Instruction.LoadShared, AggregateType.U32, AggregateType.S32);

View file

@ -24,7 +24,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{
OperandType.Argument => AggregateType.S32,
OperandType.Constant => AggregateType.S32,
OperandType.ConstantBuffer => AggregateType.FP32,
OperandType.Undefined => AggregateType.S32,
_ => throw new ArgumentException($"Invalid operand type \"{type}\".")
};

View file

@ -0,0 +1,21 @@
using System.Collections.Generic;
namespace Ryujinx.Graphics.Shader.StructuredIr
{
class ShaderProperties
{
private readonly Dictionary<int, BufferDefinition> _constantBuffers;
public IReadOnlyDictionary<int, BufferDefinition> ConstantBuffers => _constantBuffers;
public ShaderProperties()
{
_constantBuffers = new Dictionary<int, BufferDefinition>();
}
public void AddConstantBuffer(int binding, BufferDefinition definition)
{
_constantBuffers[binding] = definition;
}
}
}

View file

@ -0,0 +1,28 @@
using Ryujinx.Graphics.Shader.Translation;
namespace Ryujinx.Graphics.Shader.StructuredIr
{
struct StructureField
{
public AggregateType Type { get; }
public string Name { get; }
public int ArrayLength { get; }
public StructureField(AggregateType type, string name, int arrayLength = 1)
{
Type = type;
Name = name;
ArrayLength = arrayLength;
}
}
class StructureType
{
public StructureField[] Fields { get; }
public StructureType(StructureField[] fields)
{
Fields = fields;
}
}
}

View file

@ -73,27 +73,34 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Instruction inst = operation.Inst;
StorageKind storageKind = operation.StorageKind;
if ((inst == Instruction.Load || inst == Instruction.Store) && storageKind.IsInputOrOutput())
if (inst == Instruction.Load || inst == Instruction.Store)
{
IoVariable ioVariable = (IoVariable)operation.GetSource(0).Value;
bool isOutput = storageKind.IsOutput();
bool perPatch = storageKind.IsPerPatch();
int location = 0;
int component = 0;
if (context.Config.HasPerLocationInputOrOutput(ioVariable, isOutput))
if (storageKind.IsInputOrOutput())
{
location = operation.GetSource(1).Value;
IoVariable ioVariable = (IoVariable)operation.GetSource(0).Value;
bool isOutput = storageKind.IsOutput();
bool perPatch = storageKind.IsPerPatch();
int location = 0;
int component = 0;
if (operation.SourcesCount > 2 &&
operation.GetSource(2).Type == OperandType.Constant &&
context.Config.HasPerLocationInputOrOutputComponent(ioVariable, location, operation.GetSource(2).Value, isOutput))
if (context.Config.HasPerLocationInputOrOutput(ioVariable, isOutput))
{
component = operation.GetSource(2).Value;
}
}
location = operation.GetSource(1).Value;
context.Info.IoDefinitions.Add(new IoDefinition(storageKind, ioVariable, location, component));
if (operation.SourcesCount > 2 &&
operation.GetSource(2).Type == OperandType.Constant &&
context.Config.HasPerLocationInputOrOutputComponent(ioVariable, location, operation.GetSource(2).Value, isOutput))
{
component = operation.GetSource(2).Value;
}
}
context.Info.IoDefinitions.Add(new IoDefinition(storageKind, ioVariable, location, component));
}
else if (storageKind == StorageKind.ConstantBuffer && operation.GetSource(0).Type == OperandType.Constant)
{
context.Config.ResourceManager.SetUsedConstantBufferBinding(operation.GetSource(0).Value);
}
}
bool vectorDest = IsVectorDestInst(inst);
@ -105,7 +112,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
for (int index = 0; index < operation.SourcesCount; index++)
{
sources[index] = context.GetOperand(operation.GetSource(index));
sources[index] = context.GetOperandOrCbLoad(operation.GetSource(index));
}
for (int index = 0; index < outDestsCount; index++)

View file

@ -298,6 +298,33 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
return newTemp;
}
public IAstNode GetOperandOrCbLoad(Operand operand)
{
if (operand.Type == OperandType.ConstantBuffer)
{
int cbufSlot = operand.GetCbufSlot();
int cbufOffset = operand.GetCbufOffset();
int binding = Config.ResourceManager.GetConstantBufferBinding(cbufSlot);
int vecIndex = cbufOffset >> 2;
int elemIndex = cbufOffset & 3;
Config.ResourceManager.SetUsedConstantBufferBinding(binding);
IAstNode[] sources = new IAstNode[]
{
new AstOperand(OperandType.Constant, binding),
new AstOperand(OperandType.Constant, 0),
new AstOperand(OperandType.Constant, vecIndex),
new AstOperand(OperandType.Constant, elemIndex)
};
return new AstOperation(Instruction.Load, StorageKind.ConstantBuffer, sources, sources.Length);
}
return GetOperand(operand);
}
public AstOperand GetOperand(Operand operand)
{
if (operand == null)
@ -307,11 +334,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
if (operand.Type != OperandType.LocalVariable)
{
if (operand.Type == OperandType.ConstantBuffer)
{
Config.SetUsedConstantBuffer(operand.GetCbufSlot());
}
return new AstOperand(operand);
}

View file

@ -1,4 +1,6 @@
using Ryujinx.Common.Memory;
using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation;
using System.Runtime.CompilerServices;
namespace Ryujinx.Graphics.Shader
@ -11,8 +13,20 @@ namespace Ryujinx.Graphics.Shader
public T W;
}
enum SupportBufferField
{
// Must match the order of the fields on the struct.
FragmentAlphaTest,
FragmentIsBgra,
ViewportInverse,
FragmentRenderScaleCount,
RenderScale
}
public struct SupportBuffer
{
internal const int Binding = 0;
public static int FieldSize;
public static int RequiredSize;
@ -47,6 +61,18 @@ namespace Ryujinx.Graphics.Shader
ComputeRenderScaleOffset = GraphicsRenderScaleOffset + FieldSize;
}
internal static StructureType GetStructureType()
{
return new StructureType(new[]
{
new StructureField(AggregateType.U32, "s_alpha_test"),
new StructureField(AggregateType.Array | AggregateType.U32, "s_is_bgra", FragmentIsBgraCount),
new StructureField(AggregateType.Vector4 | AggregateType.FP32, "s_viewport_inverse"),
new StructureField(AggregateType.S32, "s_frag_scale_count"),
new StructureField(AggregateType.Array | AggregateType.FP32, "s_render_scale", RenderScaleMaxCount)
});
}
public Vector4<int> FragmentAlphaTest;
public Array8<Vector4<int>> FragmentIsBgra;
public Vector4<float> ViewportInverse;

View file

@ -22,4 +22,35 @@
Array = 1 << 10
}
static class AggregateTypeExtensions
{
public static int GetSizeInBytes(this AggregateType type)
{
int elementSize = (type & AggregateType.ElementTypeMask) switch
{
AggregateType.Bool or
AggregateType.FP32 or
AggregateType.S32 or
AggregateType.U32 => 4,
AggregateType.FP64 => 8,
_ => 0
};
switch (type & AggregateType.ElementCountMask)
{
case AggregateType.Vector2:
elementSize *= 2;
break;
case AggregateType.Vector3:
elementSize *= 3;
break;
case AggregateType.Vector4:
elementSize *= 4;
break;
}
return elementSize;
}
}
}

View file

@ -234,8 +234,8 @@ namespace Ryujinx.Graphics.Shader.Translation
{
Operand x = this.Load(StorageKind.Output, IoVariable.Position, null, Const(0));
Operand y = this.Load(StorageKind.Output, IoVariable.Position, null, Const(1));
Operand xScale = this.Load(StorageKind.Input, IoVariable.SupportBlockViewInverse, null, Const(0));
Operand yScale = this.Load(StorageKind.Input, IoVariable.SupportBlockViewInverse, null, Const(1));
Operand xScale = this.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.ViewportInverse), Const(0));
Operand yScale = this.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.ViewportInverse), Const(1));
Operand negativeOne = ConstF(-1.0f);
this.Store(StorageKind.Output, IoVariable.Position, null, Const(0), this.FPFusedMultiplyAdd(x, xScale, negativeOne));
@ -420,7 +420,7 @@ namespace Ryujinx.Graphics.Shader.Translation
// Perform B <-> R swap if needed, for BGRA formats (not supported on OpenGL).
if (!supportsBgra && (component == 0 || component == 2))
{
Operand isBgra = this.Load(StorageKind.Input, IoVariable.FragmentOutputIsBgra, null, Const(rtIndex));
Operand isBgra = this.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.FragmentIsBgra), Const(rtIndex));
Operand lblIsBgra = Label();
Operand lblEnd = Label();

View file

@ -549,11 +549,31 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(fpType | Instruction.IsNan, Local(), a);
}
public static Operand Load(this EmitterContext context, StorageKind storageKind, int binding)
{
return context.Add(Instruction.Load, storageKind, Local(), Const(binding));
}
public static Operand Load(this EmitterContext context, StorageKind storageKind, int binding, Operand e0)
{
return context.Add(Instruction.Load, storageKind, Local(), Const(binding), e0);
}
public static Operand Load(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1)
{
return context.Add(Instruction.Load, storageKind, Local(), Const(binding), e0, e1);
}
public static Operand Load(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand e1, Operand e2)
{
return context.Add(Instruction.Load, storageKind, Local(), Const(binding), e0, e1, e2);
}
public static Operand Load(this EmitterContext context, StorageKind storageKind, IoVariable ioVariable, Operand primVertex = null)
{
return primVertex != null
? context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable), primVertex)
: context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable));
? context.Load(storageKind, (int)ioVariable, primVertex)
: context.Load(storageKind, (int)ioVariable);
}
public static Operand Load(
@ -564,8 +584,8 @@ namespace Ryujinx.Graphics.Shader.Translation
Operand elemIndex)
{
return primVertex != null
? context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable), primVertex, elemIndex)
: context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable), elemIndex);
? context.Load(storageKind, (int)ioVariable, primVertex, elemIndex)
: context.Load(storageKind, (int)ioVariable, elemIndex);
}
public static Operand Load(
@ -577,22 +597,8 @@ namespace Ryujinx.Graphics.Shader.Translation
Operand elemIndex)
{
return primVertex != null
? context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable), primVertex, arrayIndex, elemIndex)
: context.Add(Instruction.Load, storageKind, Local(), Const((int)ioVariable), arrayIndex, elemIndex);
}
public static Operand LoadConstant(this EmitterContext context, Operand a, Operand b)
{
if (a.Type == OperandType.Constant)
{
context.Config.SetUsedConstantBuffer(a.Value);
}
else
{
context.Config.SetUsedFeature(FeatureFlags.CbIndexing);
}
return context.Add(Instruction.LoadConstant, Local(), a, b);
? context.Load(storageKind, (int)ioVariable, primVertex, arrayIndex, elemIndex)
: context.Load(storageKind, (int)ioVariable, arrayIndex, elemIndex);
}
public static Operand LoadGlobal(this EmitterContext context, Operand a, Operand b)

View file

@ -19,7 +19,6 @@ namespace Ryujinx.Graphics.Shader.Translation
InstanceId = 1 << 3,
DrawParameters = 1 << 4,
RtLayer = 1 << 5,
CbIndexing = 1 << 6,
IaIndexing = 1 << 7,
OaIndexing = 1 << 8,
FixedFuncAttr = 1 << 9

View file

@ -32,25 +32,49 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
continue;
}
if (handleAsgOp.Inst != Instruction.LoadConstant)
if (handleAsgOp.Inst != Instruction.Load ||
handleAsgOp.StorageKind != StorageKind.ConstantBuffer ||
handleAsgOp.SourcesCount != 4)
{
continue;
}
Operand ldcSrc0 = handleAsgOp.GetSource(0);
if (ldcSrc0.Type != OperandType.Constant ||
!config.ResourceManager.TryGetConstantBufferSlot(ldcSrc0.Value, out int src0CbufSlot) ||
src0CbufSlot != 2)
{
continue;
}
Operand ldcSrc1 = handleAsgOp.GetSource(1);
if (ldcSrc0.Type != OperandType.Constant || ldcSrc0.Value != 2)
// We expect field index 0 to be accessed.
if (ldcSrc1.Type != OperandType.Constant || ldcSrc1.Value != 0)
{
continue;
}
if (!(ldcSrc1.AsgOp is Operation shrOp) || shrOp.Inst != Instruction.ShiftRightU32)
Operand ldcSrc2 = handleAsgOp.GetSource(2);
// FIXME: This is missing some checks, for example, a check to ensure that the shift value is 2.
// Might be not worth fixing since if that doesn't kick in, the result will be no texture
// to access anyway which is also wrong.
// Plus this whole transform is fundamentally flawed as-is since we have no way to know the array size.
// Eventually, this should be entirely removed in favor of a implementation that supports true bindless
// texture access.
if (!(ldcSrc2.AsgOp is Operation shrOp) || shrOp.Inst != Instruction.ShiftRightU32)
{
continue;
}
if (!(shrOp.GetSource(0).AsgOp is Operation addOp) || addOp.Inst != Instruction.Add)
if (!(shrOp.GetSource(0).AsgOp is Operation shrOp2) || shrOp2.Inst != Instruction.ShiftRightU32)
{
continue;
}
if (!(shrOp2.GetSource(0).AsgOp is Operation addOp) || addOp.Inst != Instruction.Add)
{
continue;
}

View file

@ -8,7 +8,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
{
static class ConstantFolding
{
public static void RunPass(Operation operation)
public static void RunPass(ShaderConfig config, Operation operation)
{
if (!AreAllSourcesConstant(operation))
{
@ -153,8 +153,21 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
EvaluateFPUnary(operation, (x) => float.IsNaN(x));
break;
case Instruction.LoadConstant:
operation.TurnIntoCopy(Cbuf(operation.GetSource(0).Value, operation.GetSource(1).Value));
case Instruction.Load:
if (operation.StorageKind == StorageKind.ConstantBuffer && operation.SourcesCount == 4)
{
int binding = operation.GetSource(0).Value;
int fieldIndex = operation.GetSource(1).Value;
if (config.ResourceManager.TryGetConstantBufferSlot(binding, out int cbufSlot) && fieldIndex == 0)
{
int vecIndex = operation.GetSource(2).Value;
int elemIndex = operation.GetSource(3).Value;
int cbufOffset = vecIndex * 4 + elemIndex;
operation.TurnIntoCopy(Cbuf(cbufSlot, cbufOffset));
}
}
break;
case Instruction.Maximum:

View file

@ -347,21 +347,23 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
return wordOffset;
}
Operand[] sources = new Operand[operation.SourcesCount];
Operand cbufOffset = GetCbufOffset();
Operand vecIndex = Local();
Operand elemIndex = Local();
node.List.AddBefore(node, new Operation(Instruction.ShiftRightU32, 0, vecIndex, cbufOffset, Const(2)));
node.List.AddBefore(node, new Operation(Instruction.BitwiseAnd, 0, elemIndex, cbufOffset, Const(3)));
Operand[] sources = new Operand[4];
int cbSlot = UbeFirstCbuf + storageIndex;
sources[0] = Const(cbSlot);
sources[1] = GetCbufOffset();
sources[0] = Const(config.ResourceManager.GetConstantBufferBinding(cbSlot));
sources[1] = Const(0);
sources[2] = vecIndex;
sources[3] = elemIndex;
config.SetUsedConstantBuffer(cbSlot);
for (int index = 2; index < operation.SourcesCount; index++)
{
sources[index] = operation.GetSource(index);
}
Operation ldcOp = new Operation(Instruction.LoadConstant, operation.Dest, sources);
Operation ldcOp = new Operation(Instruction.Load, StorageKind.ConstantBuffer, operation.Dest, sources);
for (int index = 0; index < operation.SourcesCount; index++)
{

View file

@ -9,7 +9,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
{
public static void RunPass(BasicBlock[] blocks, ShaderConfig config)
{
RunOptimizationPasses(blocks);
RunOptimizationPasses(blocks, config);
int sbUseMask = 0;
int ubeUseMask = 0;
@ -31,10 +31,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
config.SetAccessibleBufferMasks(sbUseMask, ubeUseMask);
// Run optimizations one last time to remove any code that is now optimizable after above passes.
RunOptimizationPasses(blocks);
RunOptimizationPasses(blocks, config);
}
private static void RunOptimizationPasses(BasicBlock[] blocks)
private static void RunOptimizationPasses(BasicBlock[] blocks, ShaderConfig config)
{
bool modified;
@ -73,7 +73,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
continue;
}
ConstantFolding.RunPass(operation);
ConstantFolding.RunPass(config, operation);
Simplification.RunPass(operation);
if (DestIsLocalVar(operation))

View file

@ -0,0 +1,126 @@
using Ryujinx.Graphics.Shader.StructuredIr;
using System;
using System.Collections.Generic;
using System.Globalization;
namespace Ryujinx.Graphics.Shader.Translation
{
class ResourceManager
{
private static readonly string[] _stagePrefixes = new string[] { "cp", "vp", "tcp", "tep", "gp", "fp" };
private readonly IGpuAccessor _gpuAccessor;
private readonly ShaderProperties _properties;
private readonly string _stagePrefix;
private readonly int[] _cbSlotToBindingMap;
private readonly HashSet<int> _usedConstantBufferBindings;
public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor, ShaderProperties properties)
{
_gpuAccessor = gpuAccessor;
_properties = properties;
_stagePrefix = GetShaderStagePrefix(stage);
_cbSlotToBindingMap = new int[18];
_cbSlotToBindingMap.AsSpan().Fill(-1);
_usedConstantBufferBindings = new HashSet<int>();
properties.AddConstantBuffer(0, new BufferDefinition(BufferLayout.Std140, 0, 0, "support_buffer", SupportBuffer.GetStructureType()));
}
public int GetConstantBufferBinding(int slot)
{
int binding = _cbSlotToBindingMap[slot];
if (binding < 0)
{
binding = _gpuAccessor.QueryBindingConstantBuffer(slot);
_cbSlotToBindingMap[slot] = binding;
string slotNumber = slot.ToString(CultureInfo.InvariantCulture);
AddNewConstantBuffer(binding, $"{_stagePrefix}_c{slotNumber}");
}
return binding;
}
public bool TryGetConstantBufferSlot(int binding, out int slot)
{
for (slot = 0; slot < _cbSlotToBindingMap.Length; slot++)
{
if (_cbSlotToBindingMap[slot] == binding)
{
return true;
}
}
slot = 0;
return false;
}
public void SetUsedConstantBufferBinding(int binding)
{
_usedConstantBufferBindings.Add(binding);
}
public BufferDescriptor[] GetConstantBufferDescriptors()
{
var descriptors = new BufferDescriptor[_usedConstantBufferBindings.Count];
int descriptorIndex = 0;
for (int slot = 0; slot < _cbSlotToBindingMap.Length; slot++)
{
int binding = _cbSlotToBindingMap[slot];
if (binding >= 0 && _usedConstantBufferBindings.Contains(binding))
{
descriptors[descriptorIndex++] = new BufferDescriptor(binding, slot);
}
}
if (descriptors.Length != descriptorIndex)
{
Array.Resize(ref descriptors, descriptorIndex);
}
return descriptors;
}
private void AddNewConstantBuffer(int binding, string name)
{
StructureType type = new StructureType(new[]
{
new StructureField(AggregateType.Array | AggregateType.Vector4 | AggregateType.FP32, "data", Constants.ConstantBufferSize / 16)
});
_properties.AddConstantBuffer(binding, new BufferDefinition(BufferLayout.Std140, 0, binding, name, type));
}
public void InheritFrom(ResourceManager other)
{
for (int i = 0; i < other._cbSlotToBindingMap.Length; i++)
{
int binding = other._cbSlotToBindingMap[i];
if (binding >= 0)
{
_cbSlotToBindingMap[i] = binding;
}
}
}
public static string GetShaderStagePrefix(ShaderStage stage)
{
uint index = (uint)stage;
if (index >= _stagePrefixes.Length)
{
return "invalid";
}
return _stagePrefixes[index];
}
}
}

View file

@ -1,7 +1,6 @@
using Ryujinx.Graphics.Shader.Decoders;
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Numerics;
@ -16,6 +15,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
bool isVertexShader = config.Stage == ShaderStage.Vertex;
bool hasConstantBufferDrawParameters = config.GpuAccessor.QueryHasConstantBufferDrawParameters();
bool hasVectorIndexingBug = config.GpuAccessor.QueryHostHasVectorIndexingBug();
bool supportsSnormBufferTextureFormat = config.GpuAccessor.QueryHostSupportsSnormBufferTextureFormat();
for (int blkIndex = 0; blkIndex < blocks.Length; blkIndex++)
@ -45,6 +45,11 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
if (hasVectorIndexingBug)
{
InsertVectorComponentSelect(node, config);
}
LinkedListNode<INode> nextNode = node.Next;
if (operation is TextureOperation texOp)
@ -71,6 +76,84 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
private static void InsertVectorComponentSelect(LinkedListNode<INode> node, ShaderConfig config)
{
Operation operation = (Operation)node.Value;
if (operation.Inst != Instruction.Load ||
operation.StorageKind != StorageKind.ConstantBuffer ||
operation.SourcesCount < 3)
{
return;
}
Operand bindingIndex = operation.GetSource(0);
Operand fieldIndex = operation.GetSource(1);
Operand elemIndex = operation.GetSource(operation.SourcesCount - 1);
if (bindingIndex.Type != OperandType.Constant ||
fieldIndex.Type != OperandType.Constant ||
elemIndex.Type == OperandType.Constant)
{
return;
}
BufferDefinition buffer = config.Properties.ConstantBuffers[bindingIndex.Value];
StructureField field = buffer.Type.Fields[fieldIndex.Value];
int elemCount = (field.Type & AggregateType.ElementCountMask) switch
{
AggregateType.Vector2 => 2,
AggregateType.Vector3 => 3,
AggregateType.Vector4 => 4,
_ => 1
};
if (elemCount == 1)
{
return;
}
Operand result = null;
for (int i = 0; i < elemCount; i++)
{
Operand value = Local();
Operand[] inputs = new Operand[operation.SourcesCount];
for (int srcIndex = 0; srcIndex < inputs.Length - 1; srcIndex++)
{
inputs[srcIndex] = operation.GetSource(srcIndex);
}
inputs[inputs.Length - 1] = Const(i);
Operation loadOp = new Operation(Instruction.Load, StorageKind.ConstantBuffer, value, inputs);
node.List.AddBefore(node, loadOp);
if (i == 0)
{
result = value;
}
else
{
Operand isCurrentIndex = Local();
Operand selection = Local();
Operation compareOp = new Operation(Instruction.CompareEqual, isCurrentIndex, new Operand[] { elemIndex, Const(i) });
Operation selectOp = new Operation(Instruction.ConditionalSelect, selection, new Operand[] { isCurrentIndex, value, result });
node.List.AddBefore(node, compareOp);
node.List.AddBefore(node, selectOp);
result = selection;
}
}
operation.TurnIntoCopy(result);
}
private static LinkedListNode<INode> RewriteGlobalAccess(LinkedListNode<INode> node, ShaderConfig config)
{
Operation operation = (Operation)node.Value;
@ -90,6 +173,15 @@ namespace Ryujinx.Graphics.Shader.Translation
return local;
}
Operand PrependStorageOperation(Instruction inst, StorageKind storageKind, params Operand[] sources)
{
Operand local = Local();
node.List.AddBefore(node, new Operation(inst, storageKind, local, sources));
return local;
}
Operand PrependExistingOperation(Operation operation)
{
Operand local = Local();
@ -204,8 +296,6 @@ namespace Ryujinx.Graphics.Shader.Translation
cbeUseMask &= ~(1 << slot);
config.SetUsedConstantBuffer(cbSlot);
Operand previousResult = PrependExistingOperation(storageOp);
int cbOffset = GetConstantUbeOffset(slot);
@ -216,18 +306,17 @@ namespace Ryujinx.Graphics.Shader.Translation
Operand byteOffsetConst = PrependOperation(Instruction.Subtract, addrLow, baseAddrTruncConst);
Operand cbIndex = PrependOperation(Instruction.ShiftRightU32, byteOffsetConst, Const(2));
Operand vecIndex = PrependOperation(Instruction.ShiftRightU32, cbIndex, Const(2));
Operand elemIndex = PrependOperation(Instruction.BitwiseAnd, cbIndex, Const(3));
Operand[] sourcesCb = new Operand[operation.SourcesCount];
Operand[] sourcesCb = new Operand[4];
sourcesCb[0] = Const(cbSlot);
sourcesCb[1] = cbIndex;
sourcesCb[0] = Const(config.ResourceManager.GetConstantBufferBinding(cbSlot));
sourcesCb[1] = Const(0);
sourcesCb[2] = vecIndex;
sourcesCb[3] = elemIndex;
for (int index = 2; index < operation.SourcesCount; index++)
{
sourcesCb[index] = operation.GetSource(index);
}
Operand ldcResult = PrependOperation(Instruction.LoadConstant, sourcesCb);
Operand ldcResult = PrependStorageOperation(Instruction.Load, StorageKind.ConstantBuffer, sourcesCb);
storageOp = new Operation(Instruction.ConditionalSelect, operation.Dest, inRange, ldcResult, previousResult);
}

View file

@ -39,6 +39,10 @@ namespace Ryujinx.Graphics.Shader.Translation
public TranslationOptions Options { get; }
public ShaderProperties Properties { get; }
public ResourceManager ResourceManager { get; }
public bool TransformFeedbackEnabled { get; }
private TransformFeedbackOutput[] _transformFeedbackOutputs;
@ -109,7 +113,6 @@ namespace Ryujinx.Graphics.Shader.Translation
public int AccessibleStorageBuffersMask { get; private set; }
public int AccessibleConstantBuffersMask { get; private set; }
private int _usedConstantBuffers;
private int _usedStorageBuffers;
private int _usedStorageBuffersWrite;
@ -128,20 +131,17 @@ namespace Ryujinx.Graphics.Shader.Translation
private readonly Dictionary<int, int> _sbSlots;
private readonly Dictionary<int, int> _sbSlotsReverse;
private BufferDescriptor[] _cachedConstantBufferDescriptors;
private BufferDescriptor[] _cachedStorageBufferDescriptors;
private TextureDescriptor[] _cachedTextureDescriptors;
private TextureDescriptor[] _cachedImageDescriptors;
private int _firstConstantBufferBinding;
private int _firstStorageBufferBinding;
public int FirstConstantBufferBinding => _firstConstantBufferBinding;
public int FirstStorageBufferBinding => _firstStorageBufferBinding;
public ShaderConfig(IGpuAccessor gpuAccessor, TranslationOptions options)
public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options)
{
Stage = ShaderStage.Compute;
Stage = stage;
GpuAccessor = gpuAccessor;
Options = options;
@ -158,6 +158,9 @@ namespace Ryujinx.Graphics.Shader.Translation
_sbSlots = new Dictionary<int, int>();
_sbSlotsReverse = new Dictionary<int, int>();
Properties = new ShaderProperties();
ResourceManager = new ResourceManager(stage, gpuAccessor, Properties);
}
public ShaderConfig(
@ -165,9 +168,8 @@ namespace Ryujinx.Graphics.Shader.Translation
OutputTopology outputTopology,
int maxOutputVertices,
IGpuAccessor gpuAccessor,
TranslationOptions options) : this(gpuAccessor, options)
TranslationOptions options) : this(stage, gpuAccessor, options)
{
Stage = stage;
ThreadsPerInputPrimitive = 1;
OutputTopology = outputTopology;
MaxOutputVertices = maxOutputVertices;
@ -179,9 +181,8 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(gpuAccessor, options)
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(header.Stage, gpuAccessor, options)
{
Stage = header.Stage;
GpPassthrough = header.Stage == ShaderStage.Geometry && header.GpPassthrough;
ThreadsPerInputPrimitive = header.ThreadsPerInputPrimitive;
OutputTopology = header.OutputTopology;
@ -428,12 +429,13 @@ namespace Ryujinx.Graphics.Shader.Translation
public void InheritFrom(ShaderConfig other)
{
ResourceManager.InheritFrom(other.ResourceManager);
ClipDistancesWritten |= other.ClipDistancesWritten;
UsedFeatures |= other.UsedFeatures;
UsedInputAttributes |= other.UsedInputAttributes;
UsedOutputAttributes |= other.UsedOutputAttributes;
_usedConstantBuffers |= other._usedConstantBuffers;
_usedStorageBuffers |= other._usedStorageBuffers;
_usedStorageBuffersWrite |= other._usedStorageBuffersWrite;
@ -641,11 +643,6 @@ namespace Ryujinx.Graphics.Shader.Translation
AccessibleConstantBuffersMask = ubeMask;
}
public void SetUsedConstantBuffer(int slot)
{
_usedConstantBuffers |= 1 << slot;
}
public void SetUsedStorageBuffer(int slot, bool write)
{
int mask = 1 << slot;
@ -762,27 +759,6 @@ namespace Ryujinx.Graphics.Shader.Translation
return meta;
}
public BufferDescriptor[] GetConstantBufferDescriptors()
{
if (_cachedConstantBufferDescriptors != null)
{
return _cachedConstantBufferDescriptors;
}
int usedMask = _usedConstantBuffers;
if (UsedFeatures.HasFlag(FeatureFlags.CbIndexing))
{
usedMask |= (int)GpuAccessor.QueryConstantBufferUse();
}
return _cachedConstantBufferDescriptors = GetUniformBufferDescriptors(
usedMask,
UsedFeatures.HasFlag(FeatureFlags.CbIndexing),
out _firstConstantBufferBinding,
GpuAccessor.QueryBindingConstantBuffer);
}
public BufferDescriptor[] GetStorageBufferDescriptors()
{
if (_cachedStorageBufferDescriptors != null)
@ -798,47 +774,6 @@ namespace Ryujinx.Graphics.Shader.Translation
GpuAccessor.QueryBindingStorageBuffer);
}
private static BufferDescriptor[] GetUniformBufferDescriptors(int usedMask, bool isArray, out int firstBinding, Func<int, int> getBindingCallback)
{
firstBinding = 0;
int lastSlot = -1;
bool hasFirstBinding = false;
var descriptors = new BufferDescriptor[BitOperations.PopCount((uint)usedMask)];
for (int i = 0; i < descriptors.Length; i++)
{
int slot = BitOperations.TrailingZeroCount(usedMask);
if (isArray)
{
// The next array entries also consumes bindings, even if they are unused.
for (int j = lastSlot + 1; j < slot; j++)
{
int binding = getBindingCallback(j);
if (!hasFirstBinding)
{
firstBinding = binding;
hasFirstBinding = true;
}
}
}
lastSlot = slot;
descriptors[i] = new BufferDescriptor(getBindingCallback(slot), slot);
if (!hasFirstBinding)
{
firstBinding = descriptors[i].Binding;
hasFirstBinding = true;
}
usedMask &= ~(1 << slot);
}
return descriptors;
}
private BufferDescriptor[] GetStorageBufferDescriptors(
int usedMask,
int writtenMask,
@ -1009,7 +944,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public ShaderProgramInfo CreateProgramInfo(ShaderIdentification identification = ShaderIdentification.None)
{
return new ShaderProgramInfo(
GetConstantBufferDescriptors(),
ResourceManager.GetConstantBufferDescriptors(),
GetStorageBufferDescriptors(),
GetTextureDescriptors(),
GetImageDescriptors(),

View file

@ -99,7 +99,7 @@ namespace Ryujinx.Graphics.Shader.Translation
if (options.Flags.HasFlag(TranslationFlags.Compute))
{
config = new ShaderConfig(gpuAccessor, options);
config = new ShaderConfig(ShaderStage.Compute, gpuAccessor, options);
program = Decoder.Decode(config, address);
}