catto
/
shadPS4
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vector_alu: Simplify mbcnt pattern

This commit is contained in:
IndecisiveTurtle 2024-08-02 23:27:34 +03:00
parent b08db4603e
commit a5574bdbf0
1 changed files with 7 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
src/shader_recompiler/frontend/translate/vector_alu.cpp
View File

@ -925,25 +925,13 @@ void Translator::V_FFBL_B32(const GcnInst& inst) {
void Translator::V_MBCNT_U32_B32(bool is_low, const GcnInst& inst) { void Translator::V_MBCNT_U32_B32(bool is_low, const GcnInst& inst) {
const IR::U32 src0{GetSrc(inst.src[0])}; const IR::U32 src0{GetSrc(inst.src[0])};
const IR::U32 src1{GetSrc(inst.src[1])}; const IR::U32 src1{GetSrc(inst.src[1])};
const IR::U32 lane_id = ir.LaneId(); if (!is_low) {
ASSERT(src0.IsImmediate() && src0.U32() == ~0U &&
const auto [warp_half, mask_shift] = [&]() -> std::pair<IR::U32, IR::U32> { src1.IsImmediate() && src1.U32() == 0U);
if (profile.subgroup_size == 32) { return;
const IR::U32 warp_half = ir.BitwiseAnd(ir.WarpId(), ir.Imm32(1)); }
return std::make_pair(warp_half, lane_id); ASSERT(src0.IsImmediate() && src0.U32() == ~0U);
} SetDst(inst.dst[0], ir.LaneId());
const IR::U32 warp_half = ir.ShiftRightLogical(lane_id, ir.Imm32(5));
const IR::U32 mask_shift = ir.BitwiseAnd(lane_id, ir.Imm32(0x1F));
return std::make_pair(warp_half, mask_shift);
}();
const IR::U32 thread_mask = ir.ISub(ir.ShiftLeftLogical(ir.Imm32(1), mask_shift), ir.Imm32(1));
const IR::U1 is_odd_warp = ir.INotEqual(warp_half, ir.Imm32(0));
const IR::U32 mask = IR::U32{ir.Select(is_odd_warp, is_low ? ir.Imm32(~0U) : thread_mask,
is_low ? thread_mask : ir.Imm32(0))};
const IR::U32 masked_value = ir.BitwiseAnd(src0, mask);
const IR::U32 result = ir.IAdd(src1, ir.BitCount(masked_value));
SetDst(inst.dst[0], result);
} }
} // namespace Shader::Gcn } // namespace Shader::Gcn