frontend/ir/ir_emitter: Add opcodes for floating point reciprocal exponents

This commit is contained in:
Lioncash 2019-03-02 20:18:59 -05:00 committed by MerryMage
parent b168c2a9f9
commit 9cf3c25811
4 changed files with 27 additions and 0 deletions

View file

@ -719,6 +719,23 @@ void EmitX64::EmitFPRecipEstimate64(EmitContext& ctx, IR::Inst* inst) {
EmitFPRecipEstimate<u64>(code, ctx, inst);
}
template <typename FPT>
static void EmitFPRecipExponent(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
ctx.reg_alloc.HostCall(inst, args[0]);
code.mov(code.ABI_PARAM2.cvt32(), ctx.FPCR());
code.lea(code.ABI_PARAM3, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
code.CallFunction(&FP::FPRecipExponent<FPT>);
}
void EmitX64::EmitFPRecipExponent32(EmitContext& ctx, IR::Inst* inst) {
EmitFPRecipExponent<u32>(code, ctx, inst);
}
void EmitX64::EmitFPRecipExponent64(EmitContext& ctx, IR::Inst* inst) {
EmitFPRecipExponent<u64>(code, ctx, inst);
}
template<size_t fsize>
static void EmitFPRecipStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;

View file

@ -1895,6 +1895,13 @@ U32U64 IREmitter::FPRecipEstimate(const U32U64& a) {
return Inst<U64>(Opcode::FPRecipEstimate64, a);
}
U32U64 IREmitter::FPRecipExponent(const U32U64& a) {
if (a.GetType() == Type::U32) {
return Inst<U32>(Opcode::FPRecipExponent32, a);
}
return Inst<U64>(Opcode::FPRecipExponent64, a);
}
U32U64 IREmitter::FPRecipStepFused(const U32U64& a, const U32U64& b) {
if (a.GetType() == Type::U32) {
return Inst<U32>(Opcode::FPRecipStepFused32, a, b);

View file

@ -305,6 +305,7 @@ public:
U32U64 FPMulX(const U32U64& a, const U32U64& b);
U32U64 FPNeg(const U32U64& a);
U32U64 FPRecipEstimate(const U32U64& a);
U32U64 FPRecipExponent(const U32U64& a);
U32U64 FPRecipStepFused(const U32U64& a, const U32U64& b);
U32U64 FPRoundInt(const U32U64& a, FP::RoundingMode rounding, bool exact);
U32U64 FPRSqrtEstimate(const U32U64& a);

View file

@ -484,6 +484,8 @@ OPCODE(FPNeg32, U32, U32
OPCODE(FPNeg64, U64, U64 )
OPCODE(FPRecipEstimate32, U32, U32 )
OPCODE(FPRecipEstimate64, U64, U64 )
OPCODE(FPRecipExponent32, U32, U32 )
OPCODE(FPRecipExponent64, U64, U64 )
OPCODE(FPRecipStepFused32, U32, U32, U32 )
OPCODE(FPRecipStepFused64, U64, U64, U64 )
OPCODE(FPRoundInt32, U32, U32, U8, U1 )