A32: Implement ASIMD VQRSHRUN
This commit is contained in:
parent
e009d99924
commit
589d717af5
3 changed files with 70 additions and 19 deletions
|
@ -74,7 +74,7 @@ INST(asimd_VSLI, "VSLI", "111100111Diiiiiidddd010
|
||||||
INST(asimd_VSHRN, "VSHRN", "111100101Diiiiiidddd100000M1mmmm") // ASIMD
|
INST(asimd_VSHRN, "VSHRN", "111100101Diiiiiidddd100000M1mmmm") // ASIMD
|
||||||
//INST(asimd_VRSHRN, "VRSHRN", "111100101-vvv-------100001-1----") // ASIMD
|
//INST(asimd_VRSHRN, "VRSHRN", "111100101-vvv-------100001-1----") // ASIMD
|
||||||
//INST(asimd_VQSHRUN, "VQSHRUN", "111100111-vvv-------100000-1----") // ASIMD
|
//INST(asimd_VQSHRUN, "VQSHRUN", "111100111-vvv-------100000-1----") // ASIMD
|
||||||
//INST(asimd_VQRSHRUN, "VQRSHRUN", "111100111-vvv-------100001-1----") // ASIMD
|
INST(asimd_VQRSHRUN, "VQRSHRUN", "111100111Diiiiiidddd100001M1mmmm") // ASIMD
|
||||||
//INST(asimd_VQSHRN, "VQSHRN", "1111001U1-vvv-------100100-1----") // ASIMD
|
//INST(asimd_VQSHRN, "VQSHRN", "1111001U1-vvv-------100100-1----") // ASIMD
|
||||||
//INST(asimd_VQRSHRN, "VQRSHRN", "1111001U1-vvv-------100101-1----") // ASIMD
|
//INST(asimd_VQRSHRN, "VQRSHRN", "1111001U1-vvv-------100101-1----") // ASIMD
|
||||||
//INST(asimd_SHLL, "SHLL", "1111001U1-vvv-------101000-1----") // ASIMD
|
//INST(asimd_SHLL, "SHLL", "1111001U1-vvv-------101000-1----") // ASIMD
|
||||||
|
|
|
@ -20,6 +20,17 @@ enum class Rounding {
|
||||||
Round,
|
Round,
|
||||||
};
|
};
|
||||||
|
|
||||||
|
enum class Narrowing {
|
||||||
|
Truncation,
|
||||||
|
SaturateToUnsigned,
|
||||||
|
SaturateToSigned,
|
||||||
|
};
|
||||||
|
|
||||||
|
enum class Signedness {
|
||||||
|
Signed,
|
||||||
|
Unsigned
|
||||||
|
};
|
||||||
|
|
||||||
IR::U128 PerformRoundingCorrection(ArmTranslatorVisitor& v, size_t esize, u64 round_value, IR::U128 original, IR::U128 shifted) {
|
IR::U128 PerformRoundingCorrection(ArmTranslatorVisitor& v, size_t esize, u64 round_value, IR::U128 original, IR::U128 shifted) {
|
||||||
const auto round_const = v.ir.VectorBroadcast(esize, v.I(esize, round_value));
|
const auto round_const = v.ir.VectorBroadcast(esize, v.I(esize, round_value));
|
||||||
const auto round_correction = v.ir.VectorEqual(esize, v.ir.VectorAnd(original, round_const), round_const);
|
const auto round_correction = v.ir.VectorEqual(esize, v.ir.VectorAnd(original, round_const), round_const);
|
||||||
|
@ -78,6 +89,57 @@ bool ShiftRight(ArmTranslatorVisitor& v, bool U, bool D, size_t imm6, size_t Vd,
|
||||||
v.ir.SetVector(d, result);
|
v.ir.SetVector(d, result);
|
||||||
return true;
|
return true;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool ShiftRightNarrowing(ArmTranslatorVisitor& v, bool D, size_t imm6, size_t Vd, bool M, size_t Vm,
|
||||||
|
Rounding rounding, Narrowing narrowing, Signedness signedness) {
|
||||||
|
if (Common::Bits<3, 5>(imm6) == 0) {
|
||||||
|
// TODO: Decode error
|
||||||
|
return v.UndefinedInstruction();
|
||||||
|
}
|
||||||
|
|
||||||
|
if (Common::Bit<0>(Vm)) {
|
||||||
|
return v.UndefinedInstruction();
|
||||||
|
}
|
||||||
|
|
||||||
|
const auto [esize, shift_amount_] = ElementSizeAndShiftAmount(true, false, imm6);
|
||||||
|
const auto source_esize = 2 * esize;
|
||||||
|
const auto shift_amount = static_cast<u8>(shift_amount_);
|
||||||
|
|
||||||
|
const auto d = ToVector(false, Vd, D);
|
||||||
|
const auto m = ToVector(true, Vm, M);
|
||||||
|
|
||||||
|
const auto reg_m = v.ir.GetVector(m);
|
||||||
|
auto wide_result = [&] {
|
||||||
|
if (signedness == Signedness::Signed) {
|
||||||
|
return v.ir.VectorArithmeticShiftRight(source_esize, reg_m, shift_amount);
|
||||||
|
}
|
||||||
|
return v.ir.VectorLogicalShiftRight(source_esize, reg_m, shift_amount);
|
||||||
|
}();
|
||||||
|
|
||||||
|
if (rounding == Rounding::Round) {
|
||||||
|
const u64 round_value = 1ULL << (shift_amount - 1);
|
||||||
|
wide_result = PerformRoundingCorrection(v, source_esize, round_value, reg_m, wide_result);
|
||||||
|
}
|
||||||
|
|
||||||
|
const auto result = [&] {
|
||||||
|
switch (narrowing) {
|
||||||
|
case Narrowing::Truncation:
|
||||||
|
return v.ir.VectorNarrow(source_esize, wide_result);
|
||||||
|
case Narrowing::SaturateToUnsigned:
|
||||||
|
if (signedness == Signedness::Signed) {
|
||||||
|
return v.ir.VectorSignedSaturatedNarrowToUnsigned(source_esize, wide_result);
|
||||||
|
}
|
||||||
|
return v.ir.VectorUnsignedSaturatedNarrow(source_esize, wide_result);
|
||||||
|
case Narrowing::SaturateToSigned:
|
||||||
|
ASSERT(signedness == Signedness::Signed);
|
||||||
|
return v.ir.VectorSignedSaturatedNarrowToSigned(source_esize, wide_result);
|
||||||
|
}
|
||||||
|
UNREACHABLE();
|
||||||
|
}();
|
||||||
|
|
||||||
|
v.ir.SetVector(d, result);
|
||||||
|
return true;
|
||||||
|
}
|
||||||
} // Anonymous namespace
|
} // Anonymous namespace
|
||||||
|
|
||||||
bool ArmTranslatorVisitor::asimd_SHR(bool U, bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm) {
|
bool ArmTranslatorVisitor::asimd_SHR(bool U, bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm) {
|
||||||
|
@ -176,25 +238,13 @@ bool ArmTranslatorVisitor::asimd_VSHL(bool D, size_t imm6, size_t Vd, bool L, bo
|
||||||
}
|
}
|
||||||
|
|
||||||
bool ArmTranslatorVisitor::asimd_VSHRN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm) {
|
bool ArmTranslatorVisitor::asimd_VSHRN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm) {
|
||||||
if (Common::Bits<3, 5>(imm6) == 0) {
|
return ShiftRightNarrowing(*this, D, imm6, Vd, M, Vm,
|
||||||
// TODO: Decode error
|
Rounding::None, Narrowing::Truncation, Signedness::Unsigned);
|
||||||
return UndefinedInstruction();
|
}
|
||||||
}
|
|
||||||
|
|
||||||
if (Common::Bit<0>(Vm)) {
|
bool ArmTranslatorVisitor::asimd_VQRSHRUN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm) {
|
||||||
return UndefinedInstruction();
|
return ShiftRightNarrowing(*this, D, imm6, Vd, M, Vm,
|
||||||
}
|
Rounding::Round, Narrowing::SaturateToUnsigned, Signedness::Signed);
|
||||||
|
|
||||||
const auto [esize, shift_amount] = ElementSizeAndShiftAmount(true, false, imm6);
|
|
||||||
const auto d = ToVector(false, Vd, D);
|
|
||||||
const auto m = ToVector(true, Vm, M);
|
|
||||||
|
|
||||||
const auto reg_m = ir.GetVector(m);
|
|
||||||
const auto wide_result = ir.VectorLogicalShiftRight(2 * esize, reg_m, shift_amount);
|
|
||||||
const auto result = ir.VectorNarrow(2 * esize, wide_result);
|
|
||||||
|
|
||||||
ir.SetVector(d, result);
|
|
||||||
return true;
|
|
||||||
}
|
}
|
||||||
|
|
||||||
} // namespace Dynarmic::A32
|
} // namespace Dynarmic::A32
|
||||||
|
|
|
@ -509,6 +509,7 @@ struct ArmTranslatorVisitor final {
|
||||||
bool asimd_VSHL(bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm);
|
bool asimd_VSHL(bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm);
|
||||||
bool asimd_VSLI(bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm);
|
bool asimd_VSLI(bool D, size_t imm6, size_t Vd, bool L, bool Q, bool M, size_t Vm);
|
||||||
bool asimd_VSHRN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm);
|
bool asimd_VSHRN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm);
|
||||||
|
bool asimd_VQRSHRUN(bool D, size_t imm6, size_t Vd, bool M, size_t Vm);
|
||||||
|
|
||||||
// Advanced SIMD two register, miscellaneous
|
// Advanced SIMD two register, miscellaneous
|
||||||
bool asimd_VREV(bool D, size_t sz, size_t Vd, size_t op, bool Q, bool M, size_t Vm);
|
bool asimd_VREV(bool D, size_t sz, size_t Vd, size_t op, bool Q, bool M, size_t Vm);
|
||||||
|
|
Loading…
Reference in a new issue