mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_x64_vector: AVX512 implementation of EmitVectorLogicalVShiftS16

Browse source
This commit is contained in:
MerryMage 2018-09-23 10:41:41 +01:00
parent bd47f2ca8f
commit b51dae790d
1 changed files with 26 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

26
src/backend/x64/emit_x64_vector.cpp
View file

@ -1371,6 +1371,32 @@ void EmitX64::EmitVectorLogicalVShiftS8(EmitContext& ctx, IR::Inst* inst) {
}
void EmitX64::EmitVectorLogicalVShiftS16(EmitContext& ctx, IR::Inst* inst) {
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512BW)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm left_shift = ctx.reg_alloc.UseScratchXmm(args[1]);
const Xbyak::Xmm right_shift = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.vmovdqa(tmp, code.MConst(xword, 0x00FF00FF00FF00FF, 0x00FF00FF00FF00FF));
code.vpxor(right_shift, right_shift, right_shift);
code.vpsubw(right_shift, right_shift, left_shift);
code.vpsllw(xmm0, left_shift, 8);
code.vpsraw(xmm0, xmm0, 15);
code.vpand(right_shift, right_shift, tmp);
code.vpand(left_shift, left_shift, tmp);
code.vpsravw(tmp, result, right_shift);
code.vpsllvw(result, result, left_shift);
code.pblendvb(result, tmp);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
EmitTwoArgumentFallback(code, ctx, inst, [](VectorArray<s16>& result, const VectorArray<s16>& a, const VectorArray<s16>& b) {
std::transform(a.begin(), a.end(), b.begin(), result.begin(), LogicalVShift<s16>);
});