raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 03:16:54 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: implement RDNA2 DS forward lane permute

Browse Source 
Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
nmzik
2026-10-04 09:26:44 +04:00
parent 46900d8042
commit c17fa5b936
10 changed files with 147 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.cpp
+1
View File
@@ -294,6 +294,7 @@ Emitter::SpirvRequirements Emitter::AnalyzeProgramRequirements(const IR::Program
break;
}
case IR::ValueOpcode::SwizzleU32:
case IR::ValueOpcode::PermuteU32:
case IR::ValueOpcode::BpermuteU32: {
requirements.subgroup_ballot = true;
requirements.subgroup_shuffle = true;
src/graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h
+1
View File
@@ -236,6 +236,7 @@ uint32_t EmitBufferFloatAtomic(ValueEmitContext& ctx, const IR::Ins
void EmitSharedFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst);
uint32_t EmitSharedIncDec(ValueEmitContext& ctx, const IR::Inst& inst);
uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst);
uint32_t EmitPermuteU32(ValueEmitContext& ctx, const IR::Inst& inst);
uint32_t EmitBpermuteU32(ValueEmitContext& ctx, const IR::Inst& inst);
uint32_t EmitReadConst(ValueEmitContext& ctx, const IR::Inst& inst);
void EmitReadConstBuffer(ValueEmitContext& ctx, const IR::Inst& inst);
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+46
View File
@@ -1421,6 +1421,52 @@ uint32_t EmitSwizzleU32(ValueEmitContext& ctx, const IR::Inst& inst) {
return EmitDsMaskedLaneRead(state, source, target, ctx.Arg(inst, 2));
}
uint32_t EmitPermuteU32(ValueEmitContext& ctx, const IR::Inst& inst) {
auto& state = ctx.state;
auto lane = EmitSubgroupLocalInvocationId(state);
if (state.lane_count == 2) {
lane = Binary(state, spv::OpBitwiseAnd, TypeU32(state), lane, ConstantU32(state, 31));
}
const auto address = ctx.Arg(inst, 1);
const auto word = Binary(state, spv::OpShiftRightLogical, TypeU32(state), lane,
ConstantU32(state, 5));
const auto ballot_word = [&](uint32_t predicate) {
const auto ballot = state.builder.AllocateId();
const auto result = state.builder.AllocateId();
state.builder.AddFunction(spv::OpGroupNonUniformBallot, TypeU32Vector(state, 4), ballot,
ConstantU32(state, spv::ScopeSubgroup), predicate);
state.builder.AddFunction(spv::OpVectorExtractDynamic, TypeU32(state), result, ballot, word);
return result;
};
// RDNA2 permutes independently within each 32-lane half. Intersect the source
// address bit ballots to find this destination's enabled writers without LDS.
auto writers = ballot_word(ctx.Arg(inst, 2));
for (uint32_t bit = 0; bit < 5; ++bit) {
const auto address_bit = Binary(state, spv::OpBitwiseAnd, TypeU32(state), address,
ConstantU32(state, 1u << (bit + 2)));
const auto mask = ballot_word(Binary(state, spv::OpINotEqual, TypeBool(state),
address_bit, ConstantU32(state, 0)));
const auto lane_bit = Binary(state, spv::OpBitwiseAnd, TypeU32(state), lane,
ConstantU32(state, 1u << bit));
const auto selected = Select(
state, TypeU32(state),
Binary(state, spv::OpINotEqual, TypeBool(state), lane_bit, ConstantU32(state, 0)),
mask, Unary(state, spv::OpNot, TypeU32(state), mask));
writers = Binary(state, spv::OpBitwiseAnd, TypeU32(state), writers, selected);
}
const auto active = Binary(state, spv::OpINotEqual, TypeBool(state), writers,
ConstantU32(state, 0));
const auto base = Binary(state, spv::OpBitwiseAnd, TypeU32(state), lane,
ConstantU32(state, ~31u));
const auto source = Select(state, TypeU32(state), active,
Binary(state, spv::OpBitwiseOr, TypeU32(state), base,
EmitFindUMsb32(state, writers)), lane);
const auto result = state.builder.AllocateId();
state.builder.AddFunction(spv::OpGroupNonUniformShuffle, TypeU32(state), result,
ConstantU32(state, spv::ScopeSubgroup), ctx.Arg(inst, 0), source);
return Select(state, TypeU32(state), active, result, ConstantU32(state, 0));
}
uint32_t EmitBpermuteU32(ValueEmitContext& ctx, const IR::Inst& inst) {
auto& state = ctx.state;
const auto source = ctx.Arg(inst, 0);