ir: Add opcode for floating-point GE and GT comparisons

The rest of the comparisons can be implemented in terms of these two
This commit is contained in:
Lioncash 2018-06-03 19:39:21 -04:00 committed by MerryMage
parent d86fea0d28
commit c695da1cf3
4 changed files with 68 additions and 0 deletions

View file

@ -209,6 +209,46 @@ void EmitX64::EmitFPVectorEqual64(EmitContext& ctx, IR::Inst* inst) {
ctx.reg_alloc.DefineValue(inst, a); ctx.reg_alloc.DefineValue(inst, a);
} }
void EmitX64::EmitFPVectorGreater32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpltps(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreater64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpltpd(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreaterEqual32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpleps(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreaterEqual64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmplepd(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorMul32(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitFPVectorMul32(EmitContext& ctx, IR::Inst* inst) {
EmitVectorOperation32(code, ctx, inst, &Xbyak::CodeGenerator::mulps); EmitVectorOperation32(code, ctx, inst, &Xbyak::CodeGenerator::mulps);
} }

View file

@ -1514,6 +1514,28 @@ U128 IREmitter::FPVectorEqual(size_t esize, const U128& a, const U128& b) {
return {}; return {};
} }
U128 IREmitter::FPVectorGreater(size_t esize, const U128& a, const U128& b) {
switch (esize) {
case 32:
return Inst<U128>(Opcode::FPVectorGreater32, a, b);
case 64:
return Inst<U128>(Opcode::FPVectorGreater64, a, b);
}
UNREACHABLE();
return {};
}
U128 IREmitter::FPVectorGreaterEqual(size_t esize, const U128& a, const U128& b) {
switch (esize) {
case 32:
return Inst<U128>(Opcode::FPVectorGreaterEqual32, a, b);
case 64:
return Inst<U128>(Opcode::FPVectorGreaterEqual64, a, b);
}
UNREACHABLE();
return {};
}
U128 IREmitter::FPVectorMul(size_t esize, const U128& a, const U128& b) { U128 IREmitter::FPVectorMul(size_t esize, const U128& a, const U128& b) {
switch (esize) { switch (esize) {
case 32: case 32:

View file

@ -275,6 +275,8 @@ public:
U128 FPVectorAdd(size_t esize, const U128& a, const U128& b); U128 FPVectorAdd(size_t esize, const U128& a, const U128& b);
U128 FPVectorDiv(size_t esize, const U128& a, const U128& b); U128 FPVectorDiv(size_t esize, const U128& a, const U128& b);
U128 FPVectorEqual(size_t esize, const U128& a, const U128& b); U128 FPVectorEqual(size_t esize, const U128& a, const U128& b);
U128 FPVectorGreater(size_t esize, const U128& a, const U128& b);
U128 FPVectorGreaterEqual(size_t esize, const U128& a, const U128& b);
U128 FPVectorMul(size_t esize, const U128& a, const U128& b); U128 FPVectorMul(size_t esize, const U128& a, const U128& b);
U128 FPVectorSub(size_t esize, const U128& a, const U128& b); U128 FPVectorSub(size_t esize, const U128& a, const U128& b);
U128 FPVectorS32ToSingle(const U128& a); U128 FPVectorS32ToSingle(const U128& a);

View file

@ -400,6 +400,10 @@ OPCODE(FPVectorDiv32, T::U128, T::U128, T::U
OPCODE(FPVectorDiv64, T::U128, T::U128, T::U128 ) OPCODE(FPVectorDiv64, T::U128, T::U128, T::U128 )
OPCODE(FPVectorEqual32, T::U128, T::U128, T::U128 ) OPCODE(FPVectorEqual32, T::U128, T::U128, T::U128 )
OPCODE(FPVectorEqual64, T::U128, T::U128, T::U128 ) OPCODE(FPVectorEqual64, T::U128, T::U128, T::U128 )
OPCODE(FPVectorGreater32, T::U128, T::U128, T::U128 )
OPCODE(FPVectorGreater64, T::U128, T::U128, T::U128 )
OPCODE(FPVectorGreaterEqual32, T::U128, T::U128, T::U128 )
OPCODE(FPVectorGreaterEqual64, T::U128, T::U128, T::U128 )
OPCODE(FPVectorMul32, T::U128, T::U128, T::U128 ) OPCODE(FPVectorMul32, T::U128, T::U128, T::U128 )
OPCODE(FPVectorMul64, T::U128, T::U128, T::U128 ) OPCODE(FPVectorMul64, T::U128, T::U128, T::U128 )
OPCODE(FPVectorS32ToSingle, T::U128, T::U128 ) OPCODE(FPVectorS32ToSingle, T::U128, T::U128 )