A32: Implement ASIMD VSUB (floating-point)
This commit is contained in:
parent
4c939b9d0a
commit
e006f0a205
7 changed files with 16 additions and 9 deletions
|
@ -1401,11 +1401,11 @@ void EmitX64::EmitFPVectorSqrt64(EmitContext& ctx, IR::Inst* inst) {
|
|||
}
|
||||
|
||||
void EmitX64::EmitFPVectorSub32(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitThreeOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::subps);
|
||||
EmitThreeOpVectorOperation<32, DefaultIndexer, FpcrControlledArgument::Present>(code, ctx, inst, &Xbyak::CodeGenerator::subps);
|
||||
}
|
||||
|
||||
void EmitX64::EmitFPVectorSub64(EmitContext& ctx, IR::Inst* inst) {
|
||||
EmitThreeOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::subpd);
|
||||
EmitThreeOpVectorOperation<64, DefaultIndexer, FpcrControlledArgument::Present>(code, ctx, inst, &Xbyak::CodeGenerator::subpd);
|
||||
}
|
||||
|
||||
template<size_t fsize, bool unsigned_>
|
||||
|
|
|
@ -34,7 +34,7 @@ INST(asimd_VMUL, "VMUL", "1111001P0Dzznnnndddd100
|
|||
//INST(asimd_VPADD, "VPADD", "1111001U0-CC--------1011---1----") // ASIMD
|
||||
//INST(asimd_VFMA, "VFMA/VFMS", "111100100-CC--------1100---1----") // ASIMD
|
||||
INST(asimd_VADD_float, "VADD (floating-point)", "111100100D0znnnndddd1101NQM0mmmm") // ASIMD
|
||||
//INST(asimd_VSUB_float, "VSUB (floating-point)", "111100100-1C--------1101---0----") // ASIMD
|
||||
INST(asimd_VSUB_float, "VSUB (floating-point)", "111100100D1znnnndddd1101NQM0mmmm") // ASIMD
|
||||
//INST(asimd_VPADD_float, "VPADD (floating-point)", "111100110-0C--------1101---0----") // ASIMD
|
||||
//INST(asimd_VABD_float, "VABD (floating-point)", "111100110-1C--------1101---0----") // ASIMD
|
||||
//INST(asimd_VMLA_float, "VMLA (floating-point)", "111100100-CC--------1101---1----") // ASIMD
|
||||
|
|
|
@ -362,6 +362,12 @@ bool ArmTranslatorVisitor::asimd_VADD_float(bool D, bool sz, size_t Vn, size_t V
|
|||
});
|
||||
}
|
||||
|
||||
bool ArmTranslatorVisitor::asimd_VSUB_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm) {
|
||||
return FloatingPointInstruction(*this, D, sz, Vn, Vd, N, Q, M, Vm, [this](const auto&, const auto& reg_n, const auto& reg_m) {
|
||||
return ir.FPVectorSub(32, reg_n, reg_m, false);
|
||||
});
|
||||
}
|
||||
|
||||
bool ArmTranslatorVisitor::asimd_VMUL_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm) {
|
||||
return FloatingPointInstruction(*this, D, sz, Vn, Vd, N, Q, M, Vm, [this](const auto&, const auto& reg_n, const auto& reg_m) {
|
||||
return ir.FPVectorMul(32, reg_n, reg_m, false);
|
||||
|
|
|
@ -463,6 +463,7 @@ struct ArmTranslatorVisitor final {
|
|||
bool asimd_VTST(bool D, size_t sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VMUL(bool P, bool D, size_t sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VADD_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VSUB_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VMUL_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VMAX_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
bool asimd_VMIN_float(bool D, bool sz, size_t Vn, size_t Vd, bool N, bool Q, bool M, size_t Vm);
|
||||
|
|
|
@ -2513,12 +2513,12 @@ U128 IREmitter::FPVectorSqrt(size_t esize, const U128& a) {
|
|||
UNREACHABLE();
|
||||
}
|
||||
|
||||
U128 IREmitter::FPVectorSub(size_t esize, const U128& a, const U128& b) {
|
||||
U128 IREmitter::FPVectorSub(size_t esize, const U128& a, const U128& b, bool fpcr_controlled) {
|
||||
switch (esize) {
|
||||
case 32:
|
||||
return Inst<U128>(Opcode::FPVectorSub32, a, b);
|
||||
return Inst<U128>(Opcode::FPVectorSub32, a, b, Imm1(fpcr_controlled));
|
||||
case 64:
|
||||
return Inst<U128>(Opcode::FPVectorSub64, a, b);
|
||||
return Inst<U128>(Opcode::FPVectorSub64, a, b, Imm1(fpcr_controlled));
|
||||
}
|
||||
UNREACHABLE();
|
||||
}
|
||||
|
|
|
@ -366,7 +366,7 @@ public:
|
|||
U128 FPVectorRSqrtEstimate(size_t esize, const U128& a);
|
||||
U128 FPVectorRSqrtStepFused(size_t esize, const U128& a, const U128& b);
|
||||
U128 FPVectorSqrt(size_t esize, const U128& a);
|
||||
U128 FPVectorSub(size_t esize, const U128& a, const U128& b);
|
||||
U128 FPVectorSub(size_t esize, const U128& a, const U128& b, bool fpcr_controlled = true);
|
||||
U128 FPVectorToSignedFixed(size_t esize, const U128& a, size_t fbits, FP::RoundingMode rounding);
|
||||
U128 FPVectorToUnsignedFixed(size_t esize, const U128& a, size_t fbits, FP::RoundingMode rounding);
|
||||
|
||||
|
|
|
@ -630,8 +630,8 @@ OPCODE(FPVectorRSqrtStepFused32, U128, U128
|
|||
OPCODE(FPVectorRSqrtStepFused64, U128, U128, U128 )
|
||||
OPCODE(FPVectorSqrt32, U128, U128 )
|
||||
OPCODE(FPVectorSqrt64, U128, U128 )
|
||||
OPCODE(FPVectorSub32, U128, U128, U128 )
|
||||
OPCODE(FPVectorSub64, U128, U128, U128 )
|
||||
OPCODE(FPVectorSub32, U128, U128, U128, U1 )
|
||||
OPCODE(FPVectorSub64, U128, U128, U128, U1 )
|
||||
OPCODE(FPVectorToSignedFixed16, U128, U128, U8, U8 )
|
||||
OPCODE(FPVectorToSignedFixed32, U128, U128, U8, U8 )
|
||||
OPCODE(FPVectorToSignedFixed64, U128, U128, U8, U8 )
|
||||
|
|
Loading…
Reference in a new issue