raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-25 04:16:55 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: use the V_PERMLANE16 row directly instead of copying it

Browse Source
This commit is contained in:
nmzik
2026-10-25 01:56:36 +02:00
parent 59d8256b36
commit 72903bfce6
2 changed files with 16 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterFlow.cpp
+15 -19
View File
@@ -843,27 +843,23 @@ uint32_t EmitWriteLane(ValueEmitContext& ctx, const IR::Inst& inst) {
}
uint32_t EmitPermlane16U32(ValueEmitContext& ctx, const IR::Inst& inst) {
auto& state = ctx.state;
const auto flags = inst.Flags<IR::PermlaneFlags>();
const auto subid = EmitSubgroupLocalInvocationId(state);
const auto row = state.builder.AllocateId();
const auto row_value = state.builder.AllocateId();
const auto lane = state.builder.AllocateId();
const auto lane8 = state.builder.AllocateId();
const auto shift = state.builder.AllocateId();
const auto upper = state.builder.AllocateId();
const auto selected = state.builder.AllocateId();
const auto shifted = state.builder.AllocateId();
const auto index = state.builder.AllocateId();
const auto target = state.builder.AllocateId();
auto& state = ctx.state;
const auto flags = inst.Flags<IR::PermlaneFlags>();
const auto subid = EmitSubgroupLocalInvocationId(state);
const auto row = state.builder.AllocateId();
const auto lane = state.builder.AllocateId();
const auto lane8 = state.builder.AllocateId();
const auto shift = state.builder.AllocateId();
const auto upper = state.builder.AllocateId();
const auto selected = state.builder.AllocateId();
const auto shifted = state.builder.AllocateId();
const auto index = state.builder.AllocateId();
const auto target = state.builder.AllocateId();
state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row, subid,
ConstantU32(state, 0xfffffff0u));
if (flags.x16) {
state.builder.AddFunction(spv::OpBitwiseXor, TypeU32(state), row_value, row,
ConstantU32(state, 16));
} else {
state.builder.AddFunction(spv::OpCopyObject, TypeU32(state), row_value, row);
}
const auto row_value =
flags.x16 ? Binary(state, spv::OpBitwiseXor, TypeU32(state), row, ConstantU32(state, 16))
: row;
state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
ConstantU32(state, 15));
state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane8, lane,
tests/ShaderRecompilerComputeTests.cpp
+1
View File
@@ -28061,6 +28061,7 @@ TestCase VectorPermlane16InactiveSource(u32 op_sel, u32 wave_size) {
test.opcodes = {O::V_MOV_B32, O::V_ADD_NC_U32, O::S_MOV_B32,
O::V_PERMLANE16_B32, O::V_LSHLREV_B32, O::BUFFER_STORE_DWORD,
O::S_ENDPGM};
test.forbidden_spirv = {"OpCopyObject"};
test.compute_info.wave_size = wave_size;
test.compute_info.threads_num[0] = wave_size;
test.compute_info.threads_num[1] = 1;