frontend/ir_emitter: Add half-precision opcode for FPRecipEstimate
This commit is contained in:
parent
af2e5afed6
commit
2184d24e8f
5 changed files with 17 additions and 4 deletions
|
@ -754,6 +754,10 @@ static void EmitFPRecipEstimate(BlockOfCode& code, EmitContext& ctx, IR::Inst* i
|
||||||
code.CallFunction(&FP::FPRecipEstimate<FPT>);
|
code.CallFunction(&FP::FPRecipEstimate<FPT>);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
void EmitX64::EmitFPRecipEstimate16(EmitContext& ctx, IR::Inst* inst) {
|
||||||
|
EmitFPRecipEstimate<u16>(code, ctx, inst);
|
||||||
|
}
|
||||||
|
|
||||||
void EmitX64::EmitFPRecipEstimate32(EmitContext& ctx, IR::Inst* inst) {
|
void EmitX64::EmitFPRecipEstimate32(EmitContext& ctx, IR::Inst* inst) {
|
||||||
EmitFPRecipEstimate<u32>(code, ctx, inst);
|
EmitFPRecipEstimate<u32>(code, ctx, inst);
|
||||||
}
|
}
|
||||||
|
|
|
@ -1922,11 +1922,18 @@ U16U32U64 IREmitter::FPNeg(const U16U32U64& a) {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
U32U64 IREmitter::FPRecipEstimate(const U32U64& a) {
|
U16U32U64 IREmitter::FPRecipEstimate(const U16U32U64& a) {
|
||||||
if (a.GetType() == Type::U32) {
|
switch (a.GetType()) {
|
||||||
|
case Type::U16:
|
||||||
|
return Inst<U16>(Opcode::FPRecipEstimate16, a);
|
||||||
|
case Type::U32:
|
||||||
return Inst<U32>(Opcode::FPRecipEstimate32, a);
|
return Inst<U32>(Opcode::FPRecipEstimate32, a);
|
||||||
|
case Type::U64:
|
||||||
|
return Inst<U64>(Opcode::FPRecipEstimate64, a);
|
||||||
|
default:
|
||||||
|
UNREACHABLE();
|
||||||
|
return U16U32U64{};
|
||||||
}
|
}
|
||||||
return Inst<U64>(Opcode::FPRecipEstimate64, a);
|
|
||||||
}
|
}
|
||||||
|
|
||||||
U16U32U64 IREmitter::FPRecipExponent(const U16U32U64& a) {
|
U16U32U64 IREmitter::FPRecipExponent(const U16U32U64& a) {
|
||||||
|
|
|
@ -305,7 +305,7 @@ public:
|
||||||
U16U32U64 FPMulAdd(const U16U32U64& addend, const U16U32U64& op1, const U16U32U64& op2, bool fpcr_controlled);
|
U16U32U64 FPMulAdd(const U16U32U64& addend, const U16U32U64& op1, const U16U32U64& op2, bool fpcr_controlled);
|
||||||
U32U64 FPMulX(const U32U64& a, const U32U64& b);
|
U32U64 FPMulX(const U32U64& a, const U32U64& b);
|
||||||
U16U32U64 FPNeg(const U16U32U64& a);
|
U16U32U64 FPNeg(const U16U32U64& a);
|
||||||
U32U64 FPRecipEstimate(const U32U64& a);
|
U16U32U64 FPRecipEstimate(const U16U32U64& a);
|
||||||
U16U32U64 FPRecipExponent(const U16U32U64& a);
|
U16U32U64 FPRecipExponent(const U16U32U64& a);
|
||||||
U16U32U64 FPRecipStepFused(const U16U32U64& a, const U16U32U64& b);
|
U16U32U64 FPRecipStepFused(const U16U32U64& a, const U16U32U64& b);
|
||||||
U16U32U64 FPRoundInt(const U16U32U64& a, FP::RoundingMode rounding, bool exact);
|
U16U32U64 FPRoundInt(const U16U32U64& a, FP::RoundingMode rounding, bool exact);
|
||||||
|
|
|
@ -272,6 +272,7 @@ bool Inst::ReadsFromAndWritesToFPSRCumulativeExceptionBits() const {
|
||||||
case Opcode::FPMulAdd16:
|
case Opcode::FPMulAdd16:
|
||||||
case Opcode::FPMulAdd32:
|
case Opcode::FPMulAdd32:
|
||||||
case Opcode::FPMulAdd64:
|
case Opcode::FPMulAdd64:
|
||||||
|
case Opcode::FPRecipEstimate16:
|
||||||
case Opcode::FPRecipEstimate32:
|
case Opcode::FPRecipEstimate32:
|
||||||
case Opcode::FPRecipEstimate64:
|
case Opcode::FPRecipEstimate64:
|
||||||
case Opcode::FPRecipExponent16:
|
case Opcode::FPRecipExponent16:
|
||||||
|
|
|
@ -491,6 +491,7 @@ OPCODE(FPMulX64, U64, U64,
|
||||||
OPCODE(FPNeg16, U16, U16 )
|
OPCODE(FPNeg16, U16, U16 )
|
||||||
OPCODE(FPNeg32, U32, U32 )
|
OPCODE(FPNeg32, U32, U32 )
|
||||||
OPCODE(FPNeg64, U64, U64 )
|
OPCODE(FPNeg64, U64, U64 )
|
||||||
|
OPCODE(FPRecipEstimate16, U16, U16 )
|
||||||
OPCODE(FPRecipEstimate32, U32, U32 )
|
OPCODE(FPRecipEstimate32, U32, U32 )
|
||||||
OPCODE(FPRecipEstimate64, U64, U64 )
|
OPCODE(FPRecipEstimate64, U64, U64 )
|
||||||
OPCODE(FPRecipExponent16, U16, U16 )
|
OPCODE(FPRecipExponent16, U16, U16 )
|
||||||
|
|
Loading…
Reference in a new issue