drashna/ryujinx
1
0
Fork 0
Code Activity Actions

Support shared color mask, implement more shader instructions

Browse source 
Support shared color masks (used by Nouveau and maybe the NVIDIA
driver).
Support draw buffers (also required by OpenGL).
Support viewport transform disable (disabled for now as it breaks some
games).
Fix instanced rendering draw being ignored for multi draw.
Fix IADD and IADD3 immediate shader encodings, that was not matching
some ops.
Implement FFMA32I shader instruction.
Implement IMAD shader instruction.
This commit is contained in:
gdkchan 2019-12-06 19:37:00 -03:00 • committed by Thog
parent e25b7c9848
commit cb171f6ebf
32 changed files with 313 additions and 92 deletions
Download patch file Download diff file Expand all files Collapse all files

13
Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
View file

@ -142,6 +142,16 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.AppendLine();
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighS32) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighS32.glsl");
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighU32) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighU32.glsl");
}
if ((info.HelperFunctionsMask & HelperFunctionsMask.Shuffle) != 0)
{
AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/Shuffle.glsl");
@ -170,6 +180,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public static void DeclareLocals(CodeGenContext context, StructuredProgramInfo info)
{
context.AppendLine(GetVarTypeName(VariableType.S32) + " " + DefaultNames.DummyIntName + ";");
context.AppendLine(GetVarTypeName(VariableType.U32) + " " + DefaultNames.DummyUintName + ";");
foreach (AstOperand decl in info.Locals)
{
string name = context.OperandManager.DeclareLocal(decl);

3
Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
View file

@ -22,6 +22,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public const string LocalMemoryName = "local_mem";
public const string SharedMemoryName = "shared_mem";
public const string DummyIntName = "dummyInt";
public const string DummyUintName = "dummyUint";
public const string UndefinedName = "undef";
}
}

3
Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs
View file

@ -2,6 +2,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{
static class HelperFunctionNames
{
public static string MultiplyHighS32 = "Helper_MultiplyHighS32";
public static string MultiplyHighU32 = "Helper_MultiplyHighU32";
public static string Shuffle = "Helper_Shuffle";
public static string ShuffleDown = "Helper_ShuffleDown";
public static string ShuffleUp = "Helper_ShuffleUp";

7
Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighS32.glsl Normal file
View file

@ -0,0 +1,7 @@
int Helper_MultiplyHighS32(int x, int y)
{
int msb;
int lsb;
imulExtended(x, y, msb, lsb);
return msb;
}

7
Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighU32.glsl Normal file
View file

@ -0,0 +1,7 @@
uint Helper_MultiplyHighU32(uint x, uint y)
{
uint msb;
uint lsb;
umulExtended(x, y, msb, lsb);
return msb;
}

2
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
View file

@ -93,6 +93,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.Minimum, InstType.CallBinary, "min");
Add(Instruction.MinimumU32, InstType.CallBinary, "min");
Add(Instruction.Multiply, InstType.OpBinaryCom, "*", 1);
Add(Instruction.MultiplyHighS32, InstType.CallBinary, HelperFunctionNames.MultiplyHighS32);
Add(Instruction.MultiplyHighU32, InstType.CallBinary, HelperFunctionNames.MultiplyHighU32);
Add(Instruction.Negate, InstType.OpUnary, "-", 0);
Add(Instruction.ReciprocalSquareRoot, InstType.CallUnary, "inversesqrt");
Add(Instruction.Return, InstType.OpNullary, "return");

1
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
View file

@ -1,6 +1,5 @@
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation.Optimizations;
using System;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;

9
Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs
View file

@ -57,6 +57,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
Set("0101110001011x", InstEmit.Fadd, typeof(OpCodeFArithReg));
Set("010010011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithCbuf));
Set("0011001x1xxxxx", InstEmit.Ffma, typeof(OpCodeFArithImm));
Set("000011xxxxxxxx", InstEmit.Ffma32i, typeof(OpCodeFArithImm32));
Set("010100011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithRegCbuf));
Set("010110011xxxxx", InstEmit.Ffma, typeof(OpCodeFArithReg));
Set("0100110000110x", InstEmit.Flo, typeof(OpCodeAluCbuf));
@ -102,12 +103,16 @@ namespace Ryujinx.Graphics.Shader.Decoders
Set("0011100x11100x", InstEmit.I2I, typeof(OpCodeAluImm));
Set("0101110011100x", InstEmit.I2I, typeof(OpCodeAluReg));
Set("0100110000010x", InstEmit.Iadd, typeof(OpCodeAluCbuf));
Set("0011100000010x", InstEmit.Iadd, typeof(OpCodeAluImm));
Set("0011100x00010x", InstEmit.Iadd, typeof(OpCodeAluImm));
Set("0001110x0xxxxx", InstEmit.Iadd, typeof(OpCodeAluImm32));
Set("0101110000010x", InstEmit.Iadd, typeof(OpCodeAluReg));
Set("010011001100xx", InstEmit.Iadd3, typeof(OpCodeAluCbuf));
Set("001110001100xx", InstEmit.Iadd3, typeof(OpCodeAluImm));
Set("0011100x1100xx", InstEmit.Iadd3, typeof(OpCodeAluImm));
Set("010111001100xx", InstEmit.Iadd3, typeof(OpCodeAluReg));
Set("010010100xxxxx", InstEmit.Imad, typeof(OpCodeAluCbuf));
Set("0011010x0xxxxx", InstEmit.Imad, typeof(OpCodeAluImm));
Set("010110100xxxxx", InstEmit.Imad, typeof(OpCodeAluReg));
Set("010100100xxxxx", InstEmit.Imad, typeof(OpCodeAluRegCbuf));
Set("0100110000100x", InstEmit.Imnmx, typeof(OpCodeAluCbuf));
Set("0011100x00100x", InstEmit.Imnmx, typeof(OpCodeAluImm));
Set("0101110000100x", InstEmit.Imnmx, typeof(OpCodeAluReg));

44
Ryujinx.Graphics.Shader/Instructions/InstEmitAlu.cs
View file

@ -200,6 +200,50 @@ namespace Ryujinx.Graphics.Shader.Instructions
// TODO: CC, X, corner cases
}
public static void Imad(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;
bool signedA = context.CurrOp.RawOpCode.Extract(48);
bool signedB = context.CurrOp.RawOpCode.Extract(53);
bool high = context.CurrOp.RawOpCode.Extract(54);
Operand srcA = GetSrcA(context);
Operand srcB = GetSrcB(context);
Operand srcC = GetSrcC(context);
Operand res;
if (high)
{
if (signedA && signedB)
{
res = context.MultiplyHighS32(srcA, srcB);
}
else
{
res = context.MultiplyHighU32(srcA, srcB);
if (signedA)
{
res = context.IAdd(res, context.IMultiply(srcB, context.ShiftRightS32(srcA, Const(31))));
}
else if (signedB)
{
res = context.IAdd(res, context.IMultiply(srcA, context.ShiftRightS32(srcB, Const(31))));
}
}
}
else
{
res = context.IMultiply(srcA, srcB);
}
res = context.IAdd(res, srcC);
context.Copy(GetDest(context), res);
}
public static void Imnmx(EmitterContext context)
{
OpCodeAlu op = (OpCodeAlu)context.CurrOp;

20
Ryujinx.Graphics.Shader/Instructions/InstEmitFArith.cs
View file

@ -59,6 +59,26 @@ namespace Ryujinx.Graphics.Shader.Instructions
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Ffma32i(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;
bool saturate = op.RawOpCode.Extract(55);
bool negateA = op.RawOpCode.Extract(56);
bool negateC = op.RawOpCode.Extract(57);
Operand srcA = context.FPNegate(GetSrcA(context), negateA);
Operand srcC = context.FPNegate(GetDest(context), negateC);
Operand srcB = GetSrcB(context);
Operand dest = GetDest(context);
context.Copy(dest, context.FPSaturate(context.FPFusedMultiplyAdd(srcA, srcB, srcC), saturate));
SetFPZnFlags(context, dest, op.SetCondCode);
}
public static void Fmnmx(EmitterContext context)
{
IOpCodeFArith op = (IOpCodeFArith)context.CurrOp;

2
Ryujinx.Graphics.Shader/IntermediateRepresentation/Instruction.cs
View file

@ -84,6 +84,8 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
Minimum,
MinimumU32,
Multiply,
MultiplyHighS32,
MultiplyHighU32,
Negate,
PackDouble2x32,
PackHalf2x16,

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

@ -1,6 +1,8 @@
<Project Sdk="Microsoft.NET.Sdk">
<ItemGroup>
<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" />

8
Ryujinx.Graphics.Shader/ShaderCapabilities.cs
View file

@ -2,14 +2,18 @@ namespace Ryujinx.Graphics.Shader
{
public struct ShaderCapabilities
{
private static readonly ShaderCapabilities _default = new ShaderCapabilities(16);
private static readonly ShaderCapabilities _default = new ShaderCapabilities(32768, 16);
public static ShaderCapabilities Default => _default;
public int MaximumViewportDimensions { get; }
public int StorageBufferOffsetAlignment { get; }
public ShaderCapabilities(int storageBufferOffsetAlignment)
public ShaderCapabilities(
int maximumViewportDimensions,
int storageBufferOffsetAlignment)
{
MaximumViewportDimensions = maximumViewportDimensions;
StorageBufferOffsetAlignment = storageBufferOffsetAlignment;
}
}

12
Ryujinx.Graphics.Shader/StructuredIr/HelperFunctionsMask.cs
View file

@ -5,10 +5,12 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
[Flags]
enum HelperFunctionsMask
{
Shuffle = 1 << 0,
ShuffleDown = 1 << 1,
ShuffleUp = 1 << 2,
ShuffleXor = 1 << 3,
SwizzleAdd = 1 << 4
MultiplyHighS32 = 1 << 0,
MultiplyHighU32 = 1 << 1,
Shuffle = 1 << 2,
ShuffleDown = 1 << 3,
ShuffleUp = 1 << 4,
ShuffleXor = 1 << 5,
SwizzleAdd = 1 << 6
}
}

2
Ryujinx.Graphics.Shader/StructuredIr/InstructionInfo.cs
View file

@ -102,6 +102,8 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Add(Instruction.Minimum, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar);
Add(Instruction.MinimumU32, VariableType.U32, VariableType.U32, VariableType.U32);
Add(Instruction.Multiply, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar);
Add(Instruction.MultiplyHighS32, VariableType.S32, VariableType.S32, VariableType.S32);
Add(Instruction.MultiplyHighU32, VariableType.U32, VariableType.U32, VariableType.U32);
Add(Instruction.Negate, VariableType.Scalar, VariableType.Scalar);
Add(Instruction.PackHalf2x16, VariableType.U32, VariableType.F32, VariableType.F32);
Add(Instruction.ReciprocalSquareRoot, VariableType.Scalar, VariableType.Scalar);

6
Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs
View file

@ -171,6 +171,12 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
// decide which helper functions are needed on the final generated code.
switch (operation.Inst)
{
case Instruction.MultiplyHighS32:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighS32;
break;
case Instruction.MultiplyHighU32:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighU32;
break;
case Instruction.Shuffle:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.Shuffle;
break;

32
Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
View file

@ -11,18 +11,25 @@ namespace Ryujinx.Graphics.Shader.Translation
public Block CurrBlock { get; set; }
public OpCode CurrOp { get; set; }
private ShaderStage _stage;
private ShaderHeader _header;
private ShaderStage _stage;
private ShaderHeader _header;
private ShaderCapabilities _capabilities;
private TranslationFlags _flags;
private List<Operation> _operations;
private Dictionary<ulong, Operand> _labels;
public EmitterContext(ShaderStage stage, ShaderHeader header)
public EmitterContext(
ShaderStage stage,
ShaderHeader header,
ShaderCapabilities capabilities,
TranslationFlags flags)
{
_stage = stage;
_header = header;
_stage = stage;
_header = header;
_capabilities = capabilities;
_flags = flags;
_operations = new List<Operation>();
@ -62,7 +69,18 @@ namespace Ryujinx.Graphics.Shader.Translation
public void PrepareForReturn()
{
if (_stage == ShaderStage.Fragment)
if (_stage == ShaderStage.Vertex)
{
if ((_flags & TranslationFlags.DividePosXY) != 0)
{
Operand posX = Attribute(AttributeConsts.PositionX);
Operand posY = Attribute(AttributeConsts.PositionY);
this.Copy(posX, this.FPDivide(posX, ConstF(_capabilities.MaximumViewportDimensions / 2)));
this.Copy(posY, this.FPDivide(posY, ConstF(_capabilities.MaximumViewportDimensions / 2)));
}
}
else if (_stage == ShaderStage.Fragment)
{
if (_header.OmapDepth)
{

10
Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
View file

@ -476,6 +476,16 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.LoadShared, Local(), a);
}
public static Operand MultiplyHighS32(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.MultiplyHighS32, Local(), a, b);
}
public static Operand MultiplyHighU32(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.MultiplyHighU32, Local(), a, b);
}
public static Operand PackHalf2x16(this EmitterContext context, Operand a, Operand b)
{
return context.Add(Instruction.PackHalf2x16, Local(), a, b);

3
Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs
View file

@ -6,6 +6,7 @@ namespace Ryujinx.Graphics.Shader.Translation
Compute = 1 << 0,
DebugMode = 1 << 1,
Unspecialized = 1 << 2
Unspecialized = 1 << 2,
DividePosXY = 1 << 3
}
}

34
Ryujinx.Graphics.Shader/Translation/Translator.cs
View file

@ -49,15 +49,9 @@ namespace Ryujinx.Graphics.Shader.Translation
public static ShaderProgram Translate(Span<byte> code, ShaderCapabilities capabilities, TranslationFlags flags)
{
bool compute = (flags & TranslationFlags.Compute) != 0;
bool debugMode = (flags & TranslationFlags.DebugMode) != 0;
bool compute = (flags & TranslationFlags.Compute) != 0;
Operation[] ops = DecodeShader(
code,
compute,
debugMode,
out ShaderHeader header,
out int size);
Operation[] ops = DecodeShader(code, capabilities, flags, out ShaderHeader header, out int size);
ShaderStage stage;
@ -94,8 +88,8 @@ namespace Ryujinx.Graphics.Shader.Translation
{
bool debugMode = (flags & TranslationFlags.DebugMode) != 0;
Operation[] vpAOps = DecodeShader(vpACode, compute: false, debugMode, out _, out _);
Operation[] vpBOps = DecodeShader(vpBCode, compute: false, debugMode, out ShaderHeader header, out int sizeB);
Operation[] vpAOps = DecodeShader(vpACode, capabilities, flags, out _, out _);
Operation[] vpBOps = DecodeShader(vpBCode, capabilities, flags, out ShaderHeader header, out int sizeB);
ShaderConfig config = new ShaderConfig(
header.Stage,
@ -142,23 +136,23 @@ namespace Ryujinx.Graphics.Shader.Translation
}
private static Operation[] DecodeShader(
Span<byte> code,
bool compute,
bool debugMode,
out ShaderHeader header,
out int size)
Span<byte> code,
ShaderCapabilities capabilities,
TranslationFlags flags,
out ShaderHeader header,
out int size)
{
Block[] cfg;
EmitterContext context;
if (compute)
if ((flags & TranslationFlags.Compute) != 0)
{
header = null;
cfg = Decoder.Decode(code, 0);
context = new EmitterContext(ShaderStage.Compute, header);
context = new EmitterContext(ShaderStage.Compute, header, capabilities, flags);
}
else
{
@ -166,7 +160,7 @@ namespace Ryujinx.Graphics.Shader.Translation
cfg = Decoder.Decode(code, HeaderSize);
context = new EmitterContext(header.Stage, header);
context = new EmitterContext(header.Stage, header, capabilities, flags);
}
if (cfg == null)
@ -197,7 +191,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
OpCode op = block.OpCodes[opIndex];
if (debugMode)
if ((flags & TranslationFlags.DebugMode) != 0)
{
string instName;
@ -274,7 +268,7 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
size = (int)maxEndAddress + (compute ? 0 : HeaderSize);
size = (int)maxEndAddress + (((flags & TranslationFlags.Compute) != 0) ? 0 : HeaderSize);
return context.GetOperations();
}