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

emit_x64_floating_point: F16C implementation of FPSingleToHalf

Browse source
This commit is contained in:
MerryMage 2019-03-24 15:42:14 +00:00
parent d32d6fe598
commit 13e8b7b516
1 changed files with 12 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
src/backend/x64/emit_x64_floating_point.cpp
View file

@ -1133,8 +1133,19 @@ void EmitX64::EmitFPSingleToDouble(EmitContext& ctx, IR::Inst* inst) {
void EmitX64::EmitFPSingleToHalf(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const auto rounding_mode = static_cast<FP::RoundingMode>(args[1].GetImmediateU8());
const auto round_imm = ConvertRoundingModeToX64Immediate(rounding_mode);
// TODO: F16C implementation requires ForceToDefaultNaN<16> to be implemented.
if (code.DoesCpuSupport(Xbyak::util::Cpu::tF16C) && !ctx.FPCR().AHP() && !ctx.FPCR().FZ16()) {
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
if (ctx.FPCR().DN()) {
ForceToDefaultNaN<32>(code, result);
}
code.vcvtps2ph(result, result, static_cast<u8>(*round_imm));
ctx.reg_alloc.DefineValue(inst, result);
return;
}
ctx.reg_alloc.HostCall(inst, args[0]);
code.mov(code.ABI_PARAM2.cvt32(), ctx.FPCR().Value());