raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-22 11:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: read lane 0 in V_READFIRSTLANE when EXEC is zero instead of an undefined lane

Browse Source
This commit is contained in:
Nawid3333
2026-10-22 01:17:31 +02:00
committed by nmzik
parent a52d1ed4e2
commit 919fac13d0
2 changed files with 53 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterProgram.cpp
+17 -22
View File
@@ -493,29 +493,24 @@ uint32_t ValueEmitContext::Ballot(IR::Value predicate) {
}
uint32_t ValueEmitContext::FirstLane(uint32_t ballot) {
if (other_half == nullptr) {
const auto result = state.builder.AllocateId();
state.builder.AddFunction(spv::OpGroupNonUniformBallotFindLSB, TypeU32(state), result,
ConstantU32(state, spv::ScopeSubgroup), ballot);
return result;
// An empty ballot (EXEC zero) gives lane 0, which V_READFIRSTLANE reads on hardware. FindILsb
// returns -1 for zero, unlike BallotFindLSB, whose result is then undefined.
const auto first = [&](uint32_t word) {
const auto bits = state.builder.AllocateId();
state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), bits, ballot, word);
return std::pair {bits, EmitGlsl<GLSLstd450FindILsb, IR::Type::U32>(state, bits)};
};
const auto [low, low_first] = first(0);
if (state.program.wave_size == 32u) {
return EmitGlsl<GLSLstd450SMax, IR::Type::U32>(state, low_first, ConstantU32(state, 0));
}
const auto low = state.builder.AllocateId();
const auto high = state.builder.AllocateId();
const auto low_first = state.builder.AllocateId();
const auto high_first = state.builder.AllocateId();
const auto low_active = state.builder.AllocateId();
const auto result = state.builder.AllocateId();
state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low, ballot, 0);
state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), high, ballot, 1);
state.builder.AddFunction(spv::OpExtInst, TypeU32(state), low_first, GlslStd450(state),
GLSLstd450FindILsb, low);
state.builder.AddFunction(spv::OpExtInst, TypeU32(state), high_first, GlslStd450(state),
GLSLstd450FindILsb, high);
state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), low_active, low,
ConstantU32(state, 0));
state.builder.AddFunction(spv::OpSelect, TypeU32(state), result, low_active, low_first,
EmitAddU32(state, high_first, ConstantU32(state, 32)));
return result;
const auto high_first = first(1).second;
const auto high_lane = EmitGlsl<GLSLstd450SMax, IR::Type::U32>(
state, EmitBinaryU32(state, spv::OpBitwiseOr, high_first, ConstantU32(state, 32)),
ConstantU32(state, 0));
return Select(state, TypeU32(state),
Binary(state, spv::OpINotEqual, TypeBool(state), low, ConstantU32(state, 0)),
low_first, high_lane);
}
uint32_t ValueEmitContext::Shuffle(const IR::Inst& inst, size_t index, uint32_t lane) {