Cool_Tools/Ryujinx-uplift
1
0
Fork 0
mirror of https://github.com/GreemDev/Ryujinx.git synced 2024-11-28 03:40:50 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Force reciprocal operation with value biased by constant to be precise on macOS (#5110)

Browse Source 
* Force operations to be precise in some cases on SPIR-V

* Make it a bit more strict, add comments

* Shader cache version bump
This commit is contained in:
gdkchan 2023-05-26 15:19:37 -03:00 committed by GitHub
parent e6658c133c
commit 3b375525fb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 67 additions and 14 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 = 5102;
private const uint CodeGenVersion = 5110;
private const string SharedTocFileName = "shared.toc";
private const string SharedDataFileName = "shared.data";

8
src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs
View File

@ -2245,7 +2245,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
var result = emitF(context.TypeFP64(), context.GetFP64(src1), context.GetFP64(src2));
if (!context.Config.GpuAccessor.QueryHostReducedPrecision())
if (!context.Config.GpuAccessor.QueryHostReducedPrecision() || operation.ForcePrecise)
{
context.Decorate(result, Decoration.NoContraction);
}
@ -2256,7 +2256,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
var result = emitF(context.TypeFP32(), context.GetFP32(src1), context.GetFP32(src2));
if (!context.Config.GpuAccessor.QueryHostReducedPrecision())
if (!context.Config.GpuAccessor.QueryHostReducedPrecision() || operation.ForcePrecise)
{
context.Decorate(result, Decoration.NoContraction);
}
@ -2316,7 +2316,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
var result = emitF(context.TypeFP64(), context.GetFP64(src1), context.GetFP64(src2), context.GetFP64(src3));
if (!context.Config.GpuAccessor.QueryHostReducedPrecision())
if (!context.Config.GpuAccessor.QueryHostReducedPrecision() || operation.ForcePrecise)
{
context.Decorate(result, Decoration.NoContraction);
}
@ -2327,7 +2327,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
{
var result = emitF(context.TypeFP32(), context.GetFP32(src1), context.GetFP32(src2), context.GetFP32(src3));
if (!context.Config.GpuAccessor.QueryHostReducedPrecision())
if (!context.Config.GpuAccessor.QueryHostReducedPrecision() || operation.ForcePrecise)
{
context.Decorate(result, Decoration.NoContraction);
}

2
src/Ryujinx.Graphics.Shader/IntermediateRepresentation/Operation.cs
View File

@ -8,6 +8,8 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
public Instruction Inst { get; private set; }
public StorageKind StorageKind { get; }
public bool ForcePrecise { get; set; }
private Operand[] _dests;
public Operand Dest

14
src/Ryujinx.Graphics.Shader/StructuredIr/AstOperation.cs
View File

@ -10,6 +10,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{
public Instruction Inst { get; }
public StorageKind StorageKind { get; }
public bool ForcePrecise { get; }
public int Index { get; }
@ -17,10 +18,11 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
public int SourcesCount => _sources.Length;
public AstOperation(Instruction inst, StorageKind storageKind, IAstNode[] sources, int sourcesCount)
public AstOperation(Instruction inst, StorageKind storageKind, bool forcePrecise, IAstNode[] sources, int sourcesCount)
{
Inst = inst;
StorageKind = storageKind;
ForcePrecise = forcePrecise;
_sources = sources;
for (int index = 0; index < sources.Length; index++)
@ -38,12 +40,18 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Index = 0;
}
public AstOperation(Instruction inst, StorageKind storageKind, int index, IAstNode[] sources, int sourcesCount) : this(inst, storageKind, sources, sourcesCount)
public AstOperation(
Instruction inst,
StorageKind storageKind,
bool forcePrecise,
int index,
IAstNode[] sources,
int sourcesCount) : this(inst, storageKind, forcePrecise, sources, sourcesCount)
{
Index = index;
}
public AstOperation(Instruction inst, params IAstNode[] sources) : this(inst, StorageKind.None, sources, sources.Length)
public AstOperation(Instruction inst, params IAstNode[] sources) : this(inst, StorageKind.None, false, sources, sources.Length)
{
}

2
src/Ryujinx.Graphics.Shader/StructuredIr/AstTextureOperation.cs
View File

@ -19,7 +19,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
int cbufSlot,
int handle,
int index,
params IAstNode[] sources) : base(inst, StorageKind.None, index, sources, sources.Length)
params IAstNode[] sources) : base(inst, StorageKind.None, false, index, sources, sources.Length)
{
Type = type;
Format = format;

26
src/Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs
View File

@ -156,7 +156,13 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
}
else
{
source = new AstOperation(inst, operation.StorageKind, operation.Index, sources, operation.SourcesCount);
source = new AstOperation(
inst,
operation.StorageKind,
operation.ForcePrecise,
operation.Index,
sources,
operation.SourcesCount);
}
AggregateType destElemType = destType;
@ -179,7 +185,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
dest.VarType = destElemType;
context.AddNode(new AstAssignment(dest, new AstOperation(Instruction.VectorExtract, StorageKind.None, new[] { destVec, index }, 2)));
context.AddNode(new AstAssignment(dest, new AstOperation(Instruction.VectorExtract, StorageKind.None, false, new[] { destVec, index }, 2)));
}
}
else if (operation.Dest != null)
@ -227,7 +233,13 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
}
else if (!isCopy)
{
source = new AstOperation(inst, operation.StorageKind, operation.Index, sources, operation.SourcesCount);
source = new AstOperation(
inst,
operation.StorageKind,
operation.ForcePrecise,
operation.Index,
sources,
operation.SourcesCount);
}
else
{
@ -248,7 +260,13 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
}
else
{
context.AddNode(new AstOperation(inst, operation.StorageKind, operation.Index, sources, operation.SourcesCount));
context.AddNode(new AstOperation(
inst,
operation.StorageKind,
operation.ForcePrecise,
operation.Index,
sources,
operation.SourcesCount));
}
// Those instructions needs to be emulated by using helper functions,

2
src/Ryujinx.Graphics.Shader/StructuredIr/StructuredProgramContext.cs
View File

@ -319,7 +319,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
new AstOperand(OperandType.Constant, elemIndex)
};
return new AstOperation(Instruction.Load, StorageKind.ConstantBuffer, sources, sources.Length);
return new AstOperation(Instruction.Load, StorageKind.ConstantBuffer, false, sources, sources.Length);
}
return GetOperand(operand);

25
src/Ryujinx.Graphics.Shader/Translation/Rewriter.cs
View File

@ -14,6 +14,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public static void RunPass(HelperFunctionManager hfm, BasicBlock[] blocks, ShaderConfig config)
{
bool isVertexShader = config.Stage == ShaderStage.Vertex;
bool isImpreciseFragmentShader = config.Stage == ShaderStage.Fragment && config.GpuAccessor.QueryHostReducedPrecision();
bool hasConstantBufferDrawParameters = config.GpuAccessor.QueryHasConstantBufferDrawParameters();
bool hasVectorIndexingBug = config.GpuAccessor.QueryHostHasVectorIndexingBug();
bool supportsSnormBufferTextureFormat = config.GpuAccessor.QueryHostSupportsSnormBufferTextureFormat();
@ -45,6 +46,11 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
if (isImpreciseFragmentShader)
{
EnableForcePreciseIfNeeded(operation);
}
if (hasVectorIndexingBug)
{
InsertVectorComponentSelect(node, config);
@ -81,6 +87,25 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
private static void EnableForcePreciseIfNeeded(Operation operation)
{
// There are some cases where a small bias is added to values to prevent division by zero.
// When operating with reduced precision, it is possible for this bias to get rounded to 0
// and cause a division by zero.
// To prevent that, we force those operations to be precise even if the host wants
// imprecise operations for performance.
if (operation.Inst == (Instruction.FP32 | Instruction.Divide) &&
operation.GetSource(0).Type == OperandType.Constant &&
operation.GetSource(0).AsFloat() == 1f &&
operation.GetSource(1).AsgOp is Operation addOp &&
addOp.Inst == (Instruction.FP32 | Instruction.Add) &&
addOp.GetSource(1).Type == OperandType.Constant)
{
addOp.ForcePrecise = true;
}
}
private static void InsertVectorComponentSelect(LinkedListNode<INode> node, ShaderConfig config)
{
Operation operation = (Operation)node.Value;