emit_x64_vector: Emit PMINSD in EmitVectorMinS32 on SSE4.1-capable CPUs
Provides a better alternative to a fallback operation.
This commit is contained in:
parent
75fd4eaaaa
commit
596a8dd1dd
1 changed files with 5 additions and 0 deletions
|
@ -906,6 +906,11 @@ void EmitX64::EmitVectorMinS16(EmitContext& ctx, IR::Inst* inst) {
|
||||||
}
|
}
|
||||||
|
|
||||||
void EmitX64::EmitVectorMinS32(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitVectorMinS32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||||
|
EmitVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::pminsd);
|
||||||
|
return;
|
||||||
|
}
|
||||||
|
|
||||||
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<s32, 4>& result, const std::array<s32, 4>& a, const std::array<s32, 4>& b){
|
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<s32, 4>& result, const std::array<s32, 4>& a, const std::array<s32, 4>& 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