suyu/dynarmic
26
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_arm64_vector: Implement Saturated Accumulate

Browse source
This commit is contained in:
Merry 2022-08-06 17:46:05 +01:00 committed by merry
parent c5b45dc5d5
commit 3df0eb30be
1 changed files with 29 additions and 32 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

61
src/dynarmic/backend/arm64/emit_arm64_vector.cpp
View file

@ -198,6 +198,27 @@ static void EmitThreeOpArrangedLower(oaknut::CodeGenerator& code, EmitContext& c
}); });
} }
template<size_t size, typename EmitFn>
static void EmitSaturatedAccumulate(oaknut::CodeGenerator&, EmitContext& ctx, IR::Inst* inst, EmitFn emit) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
auto Qaccumulator = ctx.reg_alloc.ReadWriteQ(args[0], inst);
auto Qoperand = ctx.reg_alloc.ReadQ(args[1]);
RegAlloc::Realize(Qaccumulator, Qoperand);
ctx.fpsr.Load();
if constexpr (size == 8) {
emit(Qaccumulator->B16(), Qoperand->B16());
} else if constexpr (size == 16) {
emit(Qaccumulator->H8(), Qoperand->H8());
} else if constexpr (size == 32) {
emit(Qaccumulator->S4(), Qoperand->S4());
} else if constexpr (size == 64) {
emit(Qaccumulator->D2(), Qoperand->D2());
} else {
static_assert(size == 8 || size == 16 || size == 32 || size == 64);
}
}
template<size_t size, typename EmitFn> template<size_t size, typename EmitFn>
static void EmitImmShift(oaknut::CodeGenerator&, EmitContext& ctx, IR::Inst* inst, EmitFn emit) { static void EmitImmShift(oaknut::CodeGenerator&, EmitContext& ctx, IR::Inst* inst, EmitFn emit) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
@ -1358,34 +1379,22 @@ void EmitIR<IR::Opcode::VectorSignedSaturatedAbs64>(oaknut::CodeGenerator& code,
template<> template<>
void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<8>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.SUQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<16>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.SUQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned32>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned32>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<32>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.SUQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned64>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorSignedSaturatedAccumulateUnsigned64>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<64>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.SUQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
@ -1732,34 +1741,22 @@ void EmitIR<IR::Opcode::VectorUnsignedRecipSqrtEstimate>(oaknut::CodeGenerator&
template<> template<>
void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned8>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<8>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.USQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned16>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<16>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.USQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned32>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned32>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<32>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.USQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>
void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned64>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) { void EmitIR<IR::Opcode::VectorUnsignedSaturatedAccumulateSigned64>(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Inst* inst) {
(void)code; EmitSaturatedAccumulate<64>(code, ctx, inst, [&](auto Vaccumulator, auto Voperand) { code.USQADD(Vaccumulator, Voperand); });
(void)ctx;
(void)inst;
ASSERT_FALSE("Unimplemented");
} }
template<> template<>