raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 11:26:56 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: separate legacy FP32 MAD rounding from fused FMA

Browse Source 
Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
nmzik
2026-10-05 03:48:35 +04:00
parent bbb98bd490
commit d8be245c22
7 changed files with 134 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterAlu.cpp
+19
View File
@@ -223,6 +223,25 @@ uint32_t EmitF32ToU32(EmitterState& state, uint32_t src, bool signed_value) {
}
} // namespace
uint32_t EmitFPFma32(EmitterState& state, uint32_t a, uint32_t b, uint32_t c) {
const auto result = EmitGlsl<GLSLstd450Fma, IR::Type::F32>(state, a, b, c);
state.builder.AddAnnotation(spv::OpDecorate, result, spv::DecorationNoContraction);
return result;
}
uint32_t EmitFPMad32(EmitterState& state, uint32_t a, uint32_t b, uint32_t c) {
// Legacy MAD/MAC round and flush between multiply and add, irrespective of SP_DENORM.
a = EmitFlushF32DenormToSignedZero(state, a);
b = EmitFlushF32DenormToSignedZero(state, b);
c = EmitFlushF32DenormToSignedZero(state, c);
const auto product = EmitFPMul32(state, a, b);
state.builder.AddAnnotation(spv::OpDecorate, product, spv::DecorationNoContraction);
const auto sum = EmitFPAdd32(state, EmitFlushF32DenormToSignedZero(state, product), c);
state.builder.AddAnnotation(spv::OpDecorate, sum, spv::DecorationNoContraction);
return EmitFlushF32DenormToSignedZero(state, sum);
}
uint32_t EmitFPMedTri32(EmitterState& state, uint32_t a, uint32_t b, uint32_t c) {
const auto min_ab = EmitMinMaxF32Value(state, a, b, false);
const auto min3 = EmitMinMaxF32Value(state, min_ab, c, false);
src/graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h
+2 -2
View File
@@ -160,8 +160,8 @@ EMIT_NATIVE(FPMul64, OpFMul, F64, uint32_t, uint32_t)
inline constexpr auto EmitFPFma64 =
EmitGlsl<GLSLstd450Fma, IR::Type::F64, uint32_t, uint32_t, uint32_t>;
uint32_t EmitFPRecip64(EmitterState& state, uint32_t arg0);
inline constexpr auto EmitFPFma32 =
EmitGlsl<GLSLstd450Fma, IR::Type::F32, uint32_t, uint32_t, uint32_t>;
uint32_t EmitFPFma32(EmitterState& state, uint32_t arg0, uint32_t arg1, uint32_t arg2);
uint32_t EmitFPMad32(EmitterState& state, uint32_t arg0, uint32_t arg1, uint32_t arg2);
uint32_t EmitFPMin32(EmitterState& state, uint32_t arg0, uint32_t arg1);
uint32_t EmitFPMax32(EmitterState& state, uint32_t arg0, uint32_t arg1);
uint32_t EmitFPMin64(EmitterState& state, uint32_t arg0, uint32_t arg1);