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

Unsafe Optimization: Extend Unsafe_UnfuseFMA to all FMA-related instructions

Browse source
This commit is contained in:
MerryMage 2020-07-12 12:45:12 +01:00
parent d05d95c132
commit 6bbc53839f
2 changed files with 62 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
src/backend/x64/emit_x64_floating_point.cpp
View file

@ -845,6 +845,21 @@ static void EmitFPRecipStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst*
ctx.reg_alloc.DefineValue(inst, result);
return;
}
if (ctx.HasOptimization(OptimizationFlag::Unsafe_UnfuseFMA)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
code.movaps(result, code.MConst(xword, FP::FPValue<FPT, false, 0, 2>()));
FCODE(muls)(operand1, operand2);
FCODE(subs)(result, operand1);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
}
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
@ -1030,6 +1045,22 @@ static void EmitFPRSqrtStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst*
ctx.reg_alloc.DefineValue(inst, result);
return;
}
if (ctx.HasOptimization(OptimizationFlag::Unsafe_UnfuseFMA)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
code.movaps(result, code.MConst(xword, FP::FPValue<FPT, false, 0, 3>()));
FCODE(muls)(operand1, operand2);
FCODE(subs)(result, operand1);
FCODE(muls)(result, code.MConst(xword, FP::FPValue<FPT, false, -1, 1>()));
ctx.reg_alloc.DefineValue(inst, operand1);
return;
}
}
auto args = ctx.reg_alloc.GetArgumentInfo(inst);

31
src/backend/x64/emit_x64_vector_floating_point.cpp
View file

@ -1267,6 +1267,21 @@ static void EmitRecipStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
ctx.reg_alloc.DefineValue(inst, result);
return;
}
if (ctx.HasOptimization(OptimizationFlag::Unsafe_UnfuseFMA)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
code.movaps(result, GetVectorOf<fsize, false, 0, 2>(code));
FCODE(mulp)(operand1, operand2);
FCODE(subp)(result, operand1);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
}
EmitThreeOpFallback(code, ctx, inst, fallback_fn);
@ -1453,6 +1468,22 @@ static void EmitRSqrtStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
ctx.reg_alloc.DefineValue(inst, result);
return;
}
if (ctx.HasOptimization(OptimizationFlag::Unsafe_UnfuseFMA)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
code.movaps(result, GetVectorOf<fsize, false, 0, 3>(code));
FCODE(mulp)(operand1, operand2);
FCODE(subp)(result, operand1);
FCODE(mulp)(result, GetVectorOf<fsize, false, -1, 1>(code));
ctx.reg_alloc.DefineValue(inst, result);
return;
}
}
EmitThreeOpFallback(code, ctx, inst, fallback_fn);