emit_x64_vector: Vectorize fallback path of EmitVectorMaxS32()
This commit is contained in:
parent
c692ccdd6d
commit
51e4f1d9db
1 changed files with 13 additions and 3 deletions
|
@ -1026,9 +1026,19 @@ void EmitX64::EmitVectorMaxS32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
|
|
||||||
EmitTwoArgumentFallback(code, ctx, inst, [](std::array<s32, 4>& result, const std::array<s32, 4>& a, const std::array<s32, 4>& b){
|
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||||
std::transform(a.begin(), a.end(), b.begin(), result.begin(), [](auto x, auto y) { return std::max(x, y); });
|
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||||
});
|
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
|
||||||
|
|
||||||
|
const Xbyak::Xmm tmp_b = ctx.reg_alloc.ScratchXmm();
|
||||||
|
code.movdqa(tmp_b, b);
|
||||||
|
|
||||||
|
code.pcmpgtd(tmp_b, a);
|
||||||
|
code.pand(b, tmp_b);
|
||||||
|
code.pandn(tmp_b, a);
|
||||||
|
code.por(tmp_b, b);
|
||||||
|
|
||||||
|
ctx.reg_alloc.DefineValue(inst, tmp_b);
|
||||||
}
|
}
|
||||||
|
|
||||||
void EmitX64::EmitVectorMaxS64(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitVectorMaxS64(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
|
Loading…
Reference in a new issue