emit_x64_vector: Emit VPMINSQ and VPMINUQ for 64-bit vector min operations if AVX-512VL is available
This commit is contained in:
parent
b97b71b8aa
commit
3d00dd63b4
1 changed files with 10 additions and 0 deletions
|
@ -1160,6 +1160,11 @@ void EmitX64::EmitVectorMinS32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
}
|
}
|
||||||
|
|
||||||
void EmitX64::EmitVectorMinS64(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitVectorMinS64(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL)) {
|
||||||
|
EmitAVXVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::vpminsq);
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<s64, 2>& result, const std::array<s64, 2>& a, const std::array<s64, 2>& b){
|
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<s64, 2>& result, const std::array<s64, 2>& a, const std::array<s64, 2>& b){
|
||||||
std::transform(a.begin(), a.end(), b.begin(), result.begin(), [](auto x, auto y) { return std::min(x, y); });
|
std::transform(a.begin(), a.end(), b.begin(), result.begin(), [](auto x, auto y) { return std::min(x, y); });
|
||||||
});
|
});
|
||||||
|
@ -1218,6 +1223,11 @@ void EmitX64::EmitVectorMinU32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
}
|
}
|
||||||
|
|
||||||
void EmitX64::EmitVectorMinU64(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitVectorMinU64(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL)) {
|
||||||
|
EmitAVXVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::vpminuq);
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<u64, 2>& result, const std::array<u64, 2>& a, const std::array<u64, 2>& b){
|
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<u64, 2>& result, const std::array<u64, 2>& a, const std::array<u64, 2>& b){
|
||||||
std::transform(a.begin(), a.end(), b.begin(), result.begin(), [](auto x, auto y) { return std::min(x, y); });
|
std::transform(a.begin(), a.end(), b.begin(), result.begin(), [](auto x, auto y) { return std::min(x, y); });
|
||||||
});
|
});
|
||||||
|
|
Loading…
Reference in a new issue