TranslateArm: Implement UQSUB8.

This commit is contained in:
bunnei 2016-08-12 11:53:16 -04:00 committed by MerryMage
parent 1029fd27ce
commit 86fe29c6d2
7 changed files with 49 additions and 2 deletions

View file

@ -1055,6 +1055,24 @@ void EmitX64::EmitByteReverseDual(IR::Block&, IR::Inst* inst) {
code->BSWAP(64, result); code->BSWAP(64, result);
} }
void EmitX64::EmitPackedSaturatedSubU8(IR::Block& block, IR::Inst* inst) {
IR::Value a = inst->GetArg(0);
IR::Value b = inst->GetArg(1);
X64Reg result = reg_alloc.UseDefRegister(a, inst, any_gpr);
OpArg op_arg = reg_alloc.UseOpArg(b, any_gpr);
X64Reg xmm_scratch_a = reg_alloc.ScratchRegister(any_xmm);
X64Reg xmm_scratch_b = reg_alloc.ScratchRegister(any_xmm);
code->MOVD_xmm(xmm_scratch_a, R(result));
code->MOVD_xmm(xmm_scratch_b, op_arg);
code->PSUBUSB(xmm_scratch_a, R(xmm_scratch_b));
code->MOVD_xmm(R(result), xmm_scratch_a);
}
static void DenormalsAreZero32(BlockOfCode* code, X64Reg xmm_value, X64Reg gpr_scratch) { static void DenormalsAreZero32(BlockOfCode* code, X64Reg xmm_value, X64Reg gpr_scratch) {
// We need to report back whether we've found a denormal on input. // We need to report back whether we've found a denormal on input.
// SSE doesn't do this for us when SSE's DAZ is enabled. // SSE doesn't do this for us when SSE's DAZ is enabled.

View file

@ -673,7 +673,9 @@ public:
std::string arm_UQADD16(Cond cond, Reg n, Reg d, Reg m) { return "ice"; } std::string arm_UQADD16(Cond cond, Reg n, Reg d, Reg m) { return "ice"; }
std::string arm_UQASX(Cond cond, Reg n, Reg d, Reg m) { return "ice"; } std::string arm_UQASX(Cond cond, Reg n, Reg d, Reg m) { return "ice"; }
std::string arm_UQSAX(Cond cond, Reg n, Reg d, Reg m) { return "ice"; } std::string arm_UQSAX(Cond cond, Reg n, Reg d, Reg m) { return "ice"; }
std::string arm_UQSUB8(Cond cond, Reg n, Reg d, Reg m) { return "ice"; } std::string arm_UQSUB8(Cond cond, Reg n, Reg d, Reg m) {
return Common::StringFromFormat("uqsub8%s %s, %s, %s", CondToString(cond), RegToString(d), RegToString(n), RegToString(m));
}
std::string arm_UQSUB16(Cond cond, Reg n, Reg d, Reg m) { return "ice"; } std::string arm_UQSUB16(Cond cond, Reg n, Reg d, Reg m) { return "ice"; }
// Parallel Add/Subtract (Halving) instructions // Parallel Add/Subtract (Halving) instructions

View file

@ -282,6 +282,10 @@ IR::Value IREmitter::ByteReverseDual(const IR::Value& a) {
return Inst(IR::Opcode::ByteReverseDual, {a}); return Inst(IR::Opcode::ByteReverseDual, {a});
} }
IR::Value IREmitter::PackedSaturatedSubU8(const IR::Value& a, const IR::Value& b) {
return Inst(IR::Opcode::PackedSaturatedSubU8, {a, b});
}
IR::Value IREmitter::TransferToFP32(const IR::Value& a) { IR::Value IREmitter::TransferToFP32(const IR::Value& a) {
return Inst(IR::Opcode::TransferToFP32, {a}); return Inst(IR::Opcode::TransferToFP32, {a});
} }

View file

@ -98,6 +98,7 @@ public:
IR::Value ByteReverseWord(const IR::Value& a); IR::Value ByteReverseWord(const IR::Value& a);
IR::Value ByteReverseHalf(const IR::Value& a); IR::Value ByteReverseHalf(const IR::Value& a);
IR::Value ByteReverseDual(const IR::Value& a); IR::Value ByteReverseDual(const IR::Value& a);
IR::Value PackedSaturatedSubU8(const IR::Value& a, const IR::Value& b);
IR::Value TransferToFP32(const IR::Value& a); IR::Value TransferToFP32(const IR::Value& a);
IR::Value TransferToFP64(const IR::Value& a); IR::Value TransferToFP64(const IR::Value& a);

View file

@ -60,6 +60,7 @@ OPCODE(ZeroExtendByteToWord, T::U32, T::U8
OPCODE(ByteReverseWord, T::U32, T::U32 ) OPCODE(ByteReverseWord, T::U32, T::U32 )
OPCODE(ByteReverseHalf, T::U16, T::U16 ) OPCODE(ByteReverseHalf, T::U16, T::U16 )
OPCODE(ByteReverseDual, T::U64, T::U64 ) OPCODE(ByteReverseDual, T::U64, T::U64 )
OPCODE(PackedSaturatedSubU8, T::U32, T::U32, T::U32 )
// Floating-point // Floating-point
OPCODE(TransferToFP32, T::F32, T::U32 ) OPCODE(TransferToFP32, T::F32, T::U32 )

View file

@ -101,7 +101,11 @@ bool ArmTranslatorVisitor::arm_UQSAX(Cond cond, Reg n, Reg d, Reg m) {
} }
bool ArmTranslatorVisitor::arm_UQSUB8(Cond cond, Reg n, Reg d, Reg m) { bool ArmTranslatorVisitor::arm_UQSUB8(Cond cond, Reg n, Reg d, Reg m) {
return InterpretThisInstruction(); if (ConditionPassed(cond)) {
auto result = ir.PackedSaturatedSubU8(ir.GetRegister(n), ir.GetRegister(m));
ir.SetRegister(d, result);
}
return true;
} }
bool ArmTranslatorVisitor::arm_UQSUB16(Cond cond, Reg n, Reg d, Reg m) { bool ArmTranslatorVisitor::arm_UQSUB16(Cond cond, Reg n, Reg d, Reg m) {

View file

@ -822,3 +822,20 @@ TEST_CASE("Fuzz ARM multiply instructions", "[JitX64]") {
}); });
} }
} }
TEST_CASE("Fuzz ARM parallel instructions", "[JitX64]") {
const auto is_valid = [](u32 instr) -> bool {
// R15 as Rd, Rn, or Rm is UNPREDICTABLE
return Bits<0, 3>(instr) != 0b1111 && Bits<12, 15>(instr) != 0b1111 && Bits<16, 19>(instr) != 0b1111;
};
const std::array<InstructionGenerator, 1> saturating_instructions = {{
InstructionGenerator("cccc01100110nnnndddd11111111mmmm", is_valid), // UQSUB8
}};
SECTION("Parallel Add/Subtract (Saturating)") {
FuzzJitArm(1, 1, 10000, [&saturating_instructions]() -> u32 {
return saturating_instructions[RandInt<size_t>(0, saturating_instructions.size() - 1)].Generate();
});
}
}