IR: Implement FPRSqrtStepFused
This commit is contained in:
parent
6eb069e80d
commit
506e544bfe
4 changed files with 27 additions and 0 deletions
|
@ -912,6 +912,23 @@ void EmitX64::EmitFPRSqrtEstimate64(EmitContext& ctx, IR::Inst* inst) {
|
|||
EmitFPRSqrtEstimate<u64>(code, ctx, inst);
|
||||
}
|
||||
|
||||
template<typename FPT>
|
||||
static void EmitFPRSqrtStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
ctx.reg_alloc.HostCall(inst, args[0], args[1]);
|
||||
code.mov(code.ABI_PARAM3.cvt32(), ctx.FPCR());
|
||||
code.lea(code.ABI_PARAM4, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
|
||||
code.CallFunction(&FP::FPRSqrtStepFused<FPT>);
|
||||
}
|
||||
|
||||
void EmitX64::EmitFPRSqrtStepFused32(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitFPRSqrtStepFused<u32>(code, ctx, inst);
|
||||
}
|
||||
|
||||
void EmitX64::EmitFPRSqrtStepFused64(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitFPRSqrtStepFused<u64>(code, ctx, inst);
|
||||
}
|
||||
|
||||
void EmitX64::EmitFPSqrt32(EmitContext& ctx, IR::Inst* inst) {
|
||||
FPTwoOp32(code, ctx, inst, &Xbyak::CodeGenerator::sqrtss);
|
||||
}
|
||||
|
|
|
@ -1463,6 +1463,13 @@ U32U64 IREmitter::FPRSqrtEstimate(const U32U64& a) {
|
|||
return Inst<U64>(Opcode::FPRSqrtEstimate64, a);
|
||||
}
|
||||
|
||||
U32U64 IREmitter::FPRSqrtStepFused(const U32U64& a, const U32U64& b) {
|
||||
if (a.GetType() == Type::U32) {
|
||||
return Inst<U32>(Opcode::FPRSqrtStepFused32, a, b);
|
||||
}
|
||||
return Inst<U64>(Opcode::FPRSqrtStepFused64, a, b);
|
||||
}
|
||||
|
||||
U32U64 IREmitter::FPSqrt(const U32U64& a) {
|
||||
if (a.GetType() == Type::U32) {
|
||||
return Inst<U32>(Opcode::FPSqrt32, a);
|
||||
|
|
|
@ -268,6 +268,7 @@ public:
|
|||
U32U64 FPNeg(const U32U64& a);
|
||||
U32U64 FPRoundInt(const U32U64& a, FP::RoundingMode rounding, bool exact);
|
||||
U32U64 FPRSqrtEstimate(const U32U64& a);
|
||||
U32U64 FPRSqrtStepFused(const U32U64& a, const U32U64& b);
|
||||
U32U64 FPSqrt(const U32U64& a);
|
||||
U32U64 FPSub(const U32U64& a, const U32U64& b, bool fpscr_controlled);
|
||||
U32 FPDoubleToSingle(const U64& a, bool fpscr_controlled);
|
||||
|
|
|
@ -388,6 +388,8 @@ OPCODE(FPRoundInt32, T::U32, T::U32, T::U
|
|||
OPCODE(FPRoundInt64, T::U64, T::U64, T::U8, T::U1 )
|
||||
OPCODE(FPRSqrtEstimate32, T::U32, T::U32 )
|
||||
OPCODE(FPRSqrtEstimate64, T::U64, T::U64 )
|
||||
OPCODE(FPRSqrtStepFused32, T::U32, T::U32, T::U32 )
|
||||
OPCODE(FPRSqrtStepFused64, T::U64, T::U64, T::U64 )
|
||||
OPCODE(FPSqrt32, T::U32, T::U32 )
|
||||
OPCODE(FPSqrt64, T::U64, T::U64 )
|
||||
OPCODE(FPSub32, T::U32, T::U32, T::U32 )
|
||||
|
|
Loading…
Reference in a new issue