raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: evaluate scalar branch decisions across the complete guest wave

Browse Source
This commit is contained in:
Claxten
2026-10-07 19:18:50 +04:00
committed by nmzik
parent b96871e2bf
commit a8beac190f
4 changed files with 68 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.cpp
+4
View File
@@ -280,6 +280,10 @@ Emitter::SpirvRequirements Emitter::AnalyzeProgramRequirements(const IR::Program
switch (inst.GetOpcode()) {
case IR::ValueOpcode::StoreCompletion: requirements.subgroup_barrier = true; break;
case IR::ValueOpcode::BvhIntersect: requirements.bvh = true; break;
case IR::ValueOpcode::ConditionRef:
requirements.subgroup_ballot |=
inst.Flags<CFG::BranchCondition>() != CFG::BranchCondition::ScalarInstruction;
break;
case IR::ValueOpcode::Ballot: requirements.subgroup_ballot = true; break;
case IR::ValueOpcode::DppMoveU32:
case IR::ValueOpcode::ReadFirstLane:
src/graphics/shader/recompiler/backend/spirv/spirvEmitterFlow.cpp
+8 -16
View File
@@ -713,26 +713,18 @@ uint32_t EmitDppUpdateU32(ValueEmitContext& ctx, const IR::Inst& inst) {
}
uint32_t EmitConditionRef(ValueEmitContext& ctx, const IR::Inst& inst) {
if (ctx.other_half == nullptr) return ctx.Arg(inst, 0);
// A native scalar branch makes one decision for both emulated wave halves.
if (ctx.half != 0) return ctx.other_half->Def(IR::Value(&inst));
// Scalar branches include lanes disabled by EXEC, which the branch body may restore.
if (ctx.other_half != nullptr && ctx.half != 0) return ctx.other_half->Def(IR::Value(&inst));
const auto kind = inst.Flags<CFG::BranchCondition>();
if (kind == CFG::BranchCondition::ScalarInstruction) return ctx.Arg(inst, 0);
const auto ballot = ctx.Ballot(inst.Arg(0));
const auto low = ctx.state.builder.AllocateId();
const auto high = ctx.state.builder.AllocateId();
const auto combined = ctx.state.builder.AllocateId();
const auto result = ctx.state.builder.AllocateId();
ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), low, ballot, 0);
ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), high, ballot, 1);
const bool zero = kind == CFG::BranchCondition::ExecZero ||
kind == CFG::BranchCondition::VccZero || kind == CFG::BranchCondition::SccZero;
ctx.state.builder.AddFunction(zero ? spv::OpBitwiseAnd : spv::OpBitwiseOr,
TypeU32(ctx.state), combined, low, high);
ctx.state.builder.AddFunction(zero ? spv::OpIEqual : spv::OpINotEqual,
TypeBool(ctx.state), result, combined,
ConstantU32(ctx.state, zero ? ~0u : 0u));
return result;
const auto ballot = ctx.Ballot(inst.Arg(0));
const auto reference = zero ? ctx.Ballot(IR::Value(true))
: ConstantU32CompositeZero(ctx.state, 4);
const auto compare = Binary(ctx.state, zero ? spv::OpIEqual : spv::OpINotEqual,
TypeBoolVector(ctx.state, 4), ballot, reference);
return Unary(ctx.state, zero ? spv::OpAll : spv::OpAny, TypeBool(ctx.state), compare);
}
uint32_t EmitBallot(ValueEmitContext& ctx, IR::Value predicate) {