mirror of
https://github.com/KytyPS5/KytyPS5.git
synced 2026-10-28 00:12:40 +00:00
shader: lower wave32 whole-wave UMin and UMax reductions to subgroup reductions
Co-authored-by: Senaxx <michelbeld@gmail.com> Co-authored-by: Mac <me@itsmemac.com>
This commit is contained in:
1 parent
be1c84f4d7
commit
6357740fa1
7 files changed
+102
-60
No files matched your search
@@ -821,18 +821,28 @@ uint32_t EmitReadLane(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
return ctx.Shuffle(inst, 0, ctx.Arg(inst, 1));
|
||||
}
|
||||
|
||||
uint32_t EmitSubgroupUMinU32(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
template <spv::Op Reduce, auto Combine>
|
||||
static uint32_t EmitSubgroupReduceU32(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
auto& state = ctx.state;
|
||||
auto value = ctx.Arg(inst, 0);
|
||||
if (ctx.other_half != nullptr) {
|
||||
value = EmitUMin32(state, ctx.HalfArg(inst, 0, 0), ctx.HalfArg(inst, 0, 1));
|
||||
value = Combine(state, ctx.HalfArg(inst, 0, 0), ctx.HalfArg(inst, 0, 1));
|
||||
}
|
||||
const auto result = state.builder.AllocateId();
|
||||
state.builder.AddFunction(spv::OpGroupNonUniformUMin, TypeU32(state), result,
|
||||
ConstantU32(state, spv::ScopeSubgroup), spv::GroupOperationReduce, value);
|
||||
state.builder.AddFunction(Reduce, TypeU32(state), result,
|
||||
ConstantU32(state, spv::ScopeSubgroup), spv::GroupOperationReduce,
|
||||
value);
|
||||
return result;
|
||||
}
|
||||
|
||||
uint32_t EmitSubgroupUMinU32(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
return EmitSubgroupReduceU32<spv::OpGroupNonUniformUMin, EmitUMin32>(ctx, inst);
|
||||
}
|
||||
|
||||
uint32_t EmitSubgroupUMaxU32(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
return EmitSubgroupReduceU32<spv::OpGroupNonUniformUMax, EmitUMax32>(ctx, inst);
|
||||
}
|
||||
|
||||
uint32_t EmitWriteLane(ValueEmitContext& ctx, const IR::Inst& inst) {
|
||||
auto& state = ctx.state;
|
||||
const auto hit = state.builder.AllocateId();
|
||||
|
||||
Reference in new issue
Block a user