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

emit_x64_vector_floating_point: AVX && DN implementation of EmitFPVectorMulX

Browse source
This commit is contained in:
MerryMage 2018-11-18 10:03:13 +00:00
parent e3898e628e
commit 7c8fcaef26
1 changed files with 22 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/backend/x64/emit_x64_vector_floating_point.cpp
View file

@ -960,6 +960,26 @@ static void EmitFPVectorMulX(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
if (ctx.FPSCR_DN() && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm twos = ctx.reg_alloc.ScratchXmm();
FCODE(vcmpunordp)(xmm0, result, operand);
FCODE(vxorp)(twos, result, operand);
FCODE(mulp)(result, operand);
FCODE(andp)(twos, GetNegativeZeroVector<fsize>(code));
FCODE(vcmpunordp)(tmp, result, result);
FCODE(blendvp)(result, GetNaNVector<fsize>(code));
FCODE(orp)(twos, GetVectorOf<fsize, false, 0, 2>(code));
FCODE(andnp)(xmm0, tmp);
FCODE(blendvp)(result, twos);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm(); const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]); const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]); const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);