suyu/dynarmic
28
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

ir: Add opcodes for unsigned saturating left shifts

Browse source
This commit is contained in:
Lioncash 2018-09-18 18:09:47 -04:00 committed by MerryMage
parent ab60720418
commit d426dfe942
5 changed files with 76 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

52
src/backend/x64/emit_x64_vector.cpp
View file

@ -4100,6 +4100,58 @@ void EmitX64::EmitVectorUnsignedSaturatedNarrow64(EmitContext& ctx, IR::Inst* in
});
}
template <typename T, typename S = std::make_signed_t<T>>
static bool VectorUnsignedSaturatedShiftLeft(VectorArray<T>& dst, const VectorArray<T>& data, const VectorArray<T>& shift_values) {
static_assert(std::is_unsigned_v<T>, "T must be an unsigned type.");
bool qc_flag = false;
constexpr size_t bit_size = Common::BitSize<T>();
constexpr S negative_bit_size = -static_cast<S>(bit_size);
for (size_t i = 0; i < dst.size(); i++) {
const T element = data[i];
const S shift = std::clamp(static_cast<S>(Common::SignExtend<8>(shift_values[i] & 0xFF)),
negative_bit_size, std::numeric_limits<S>::max());
if (element == 0 || shift <= negative_bit_size) {
dst[i] = 0;
} else if (shift < 0) {
dst[i] = static_cast<T>(element >> -shift);
} else if (shift >= static_cast<S>(bit_size)) {
dst[i] = std::numeric_limits<T>::max();
qc_flag = true;
} else {
const T shifted = element << shift;
if ((shifted >> shift) != element) {
dst[i] = std::numeric_limits<T>::max();
qc_flag = true;
} else {
dst[i] = shifted;
}
}
}
return qc_flag;
}
void EmitX64::EmitVectorUnsignedSaturatedShiftLeft8(EmitContext& ctx, IR::Inst* inst) {
EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorUnsignedSaturatedShiftLeft<u8>);
}
void EmitX64::EmitVectorUnsignedSaturatedShiftLeft16(EmitContext& ctx, IR::Inst* inst) {
EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorUnsignedSaturatedShiftLeft<u16>);
}
void EmitX64::EmitVectorUnsignedSaturatedShiftLeft32(EmitContext& ctx, IR::Inst* inst) {
EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorUnsignedSaturatedShiftLeft<u32>);
}
void EmitX64::EmitVectorUnsignedSaturatedShiftLeft64(EmitContext& ctx, IR::Inst* inst) {
EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorUnsignedSaturatedShiftLeft<u64>);
}
void EmitX64::EmitVectorZeroExtend8(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);

15
src/frontend/ir/ir_emitter.cpp
View file

@ -1735,6 +1735,21 @@ U128 IREmitter::VectorUnsignedSaturatedNarrow(size_t esize, const U128& a) {
return {};
}
U128 IREmitter::VectorUnsignedSaturatedShiftLeft(size_t esize, const U128& a, const U128& b) {
switch (esize) {
case 8:
return Inst<U128>(Opcode::VectorUnsignedSaturatedShiftLeft8, a, b);
case 16:
return Inst<U128>(Opcode::VectorUnsignedSaturatedShiftLeft16, a, b);
case 32:
return Inst<U128>(Opcode::VectorUnsignedSaturatedShiftLeft32, a, b);
case 64:
return Inst<U128>(Opcode::VectorUnsignedSaturatedShiftLeft64, a, b);
}
UNREACHABLE();
return {};
}
U128 IREmitter::VectorZeroExtend(size_t original_esize, const U128& a) {
switch (original_esize) {
case 8:

1
src/frontend/ir/ir_emitter.h
View file

@ -287,6 +287,7 @@ public:
U128 VectorUnsignedRecipSqrtEstimate(const U128& a);
U128 VectorUnsignedSaturatedAccumulateSigned(size_t esize, const U128& a, const U128& b);
U128 VectorUnsignedSaturatedNarrow(size_t esize, const U128& a);
U128 VectorUnsignedSaturatedShiftLeft(size_t esize, const U128& a, const U128& b);
U128 VectorZeroExtend(size_t original_esize, const U128& a);
U128 VectorZeroUpper(const U128& a);
U128 ZeroVector();

4
src/frontend/ir/microinstruction.cpp
View file

@ -380,6 +380,10 @@ bool Inst::WritesToFPSRCumulativeSaturationBit() const {
case Opcode::VectorUnsignedSaturatedNarrow16:
case Opcode::VectorUnsignedSaturatedNarrow32:
case Opcode::VectorUnsignedSaturatedNarrow64:
case Opcode::VectorUnsignedSaturatedShiftLeft8:
case Opcode::VectorUnsignedSaturatedShiftLeft16:
case Opcode::VectorUnsignedSaturatedShiftLeft32:
case Opcode::VectorUnsignedSaturatedShiftLeft64:
return true;
default:

4
src/frontend/ir/opcodes.inc
View file

@ -446,6 +446,10 @@ OPCODE(VectorUnsignedSaturatedAccumulateSigned64, U128, U128
OPCODE(VectorUnsignedSaturatedNarrow16, U128, U128 )
OPCODE(VectorUnsignedSaturatedNarrow32, U128, U128 )
OPCODE(VectorUnsignedSaturatedNarrow64, U128, U128 )
OPCODE(VectorUnsignedSaturatedShiftLeft8, U128, U128, U128 )
OPCODE(VectorUnsignedSaturatedShiftLeft16, U128, U128, U128 )
OPCODE(VectorUnsignedSaturatedShiftLeft32, U128, U128, U128 )
OPCODE(VectorUnsignedSaturatedShiftLeft64, U128, U128, U128 )
OPCODE(VectorZeroExtend8, U128, U128 )
OPCODE(VectorZeroExtend16, U128, U128 )
OPCODE(VectorZeroExtend32, U128, U128 )