Fix vote and shuffle shader instructions on AMD GPUs (#5540)

* Move shuffle handling out of the backend to a transform pass

* Handle subgroup sizes higher than 32

* Stop using the subgroup size control extension

* Make GenerateShuffleFunction static

* Shader cache version bump
This commit is contained in:
gdkchan 2023-08-16 21:31:07 -03:00 committed by GitHub
parent 64079c034c
commit 6ed613a6e6
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
35 changed files with 445 additions and 265 deletions

View file

@ -9,6 +9,7 @@ using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenFSI;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenMemory;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenPacking;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenShuffle;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenVector;
using static Ryujinx.Graphics.Shader.StructuredIr.InstructionInfo;
@ -174,6 +175,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
case Instruction.PackHalf2x16:
return PackHalf2x16(context, operation);
case Instruction.Shuffle:
return Shuffle(context, operation);
case Instruction.Store:
return Store(context, operation);

View file

@ -13,14 +13,15 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
AggregateType dstType = GetSrcVarType(operation.Inst, 0);
string arg = GetSoureExpr(context, operation.GetSource(0), dstType);
char component = "xyzw"[operation.Index];
if (context.HostCapabilities.SupportsShaderBallot)
{
return $"unpackUint2x32(ballotARB({arg})).x";
return $"unpackUint2x32(ballotARB({arg})).{component}";
}
else
{
return $"subgroupBallot({arg}).x";
return $"subgroupBallot({arg}).{component}";
}
}
}

View file

@ -108,10 +108,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.ShiftLeft, InstType.OpBinary, "<<", 3);
Add(Instruction.ShiftRightS32, InstType.OpBinary, ">>", 3);
Add(Instruction.ShiftRightU32, InstType.OpBinary, ">>", 3);
Add(Instruction.Shuffle, InstType.CallQuaternary, HelperFunctionNames.Shuffle);
Add(Instruction.ShuffleDown, InstType.CallQuaternary, HelperFunctionNames.ShuffleDown);
Add(Instruction.ShuffleUp, InstType.CallQuaternary, HelperFunctionNames.ShuffleUp);
Add(Instruction.ShuffleXor, InstType.CallQuaternary, HelperFunctionNames.ShuffleXor);
Add(Instruction.Shuffle, InstType.Special);
Add(Instruction.ShuffleDown, InstType.CallBinary, "subgroupShuffleDown");
Add(Instruction.ShuffleUp, InstType.CallBinary, "subgroupShuffleUp");
Add(Instruction.ShuffleXor, InstType.CallBinary, "subgroupShuffleXor");
Add(Instruction.Sine, InstType.CallUnary, "sin");
Add(Instruction.SquareRoot, InstType.CallUnary, "sqrt");
Add(Instruction.Store, InstType.Special);

View file

@ -0,0 +1,25 @@
using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
{
static class InstGenShuffle
{
public static string Shuffle(CodeGenContext context, AstOperation operation)
{
string value = GetSoureExpr(context, operation.GetSource(0), AggregateType.FP32);
string index = GetSoureExpr(context, operation.GetSource(1), AggregateType.U32);
if (context.HostCapabilities.SupportsShaderBallot)
{
return $"readInvocationARB({value}, {index})";
}
else
{
return $"subgroupShuffle({value}, {index})";
}
}
}
}