return Base.Reg;
}
void setOffsetReg(unsigned Reg) {
- assert(isRegBase() && "Invalid offset register access!");
OffsetReg = Reg;
}
unsigned getOffsetReg() const {
- assert(isRegBase() && "Invalid offset register access!");
return OffsetReg;
}
void setFI(unsigned FI) {
unsigned emitAdd(MVT RetVT, const Value *LHS, const Value *RHS,
bool SetFlags = false, bool WantResult = true,
bool IsZExt = false);
+ unsigned emitAdd_ri_(MVT VT, unsigned Op0, bool Op0IsKill, int64_t Imm);
unsigned emitSub(MVT RetVT, const Value *LHS, const Value *RHS,
bool SetFlags = false, bool WantResult = true,
bool IsZExt = false);
}
} // end switch
- if (Addr.getReg()) {
- if (!Addr.getOffsetReg()) {
- unsigned Reg = getRegForValue(Obj);
- if (!Reg)
- return false;
- Addr.setOffsetReg(Reg);
- return true;
- }
- return false;
+ if (Addr.isRegBase() && !Addr.getReg()) {
+ unsigned Reg = getRegForValue(Obj);
+ if (!Reg)
+ return false;
+ Addr.setReg(Reg);
+ return true;
}
- unsigned Reg = getRegForValue(Obj);
- if (!Reg)
- return false;
- Addr.setReg(Reg);
- return true;
+ if (!Addr.getOffsetReg()) {
+ unsigned Reg = getRegForValue(Obj);
+ if (!Reg)
+ return false;
+ Addr.setOffsetReg(Reg);
+ return true;
+ }
+
+ return false;
}
bool AArch64FastISel::computeCallAddress(const Value *V, Address &Addr) {
// Cannot encode an offset register and an immediate offset in the same
// instruction. Fold the immediate offset into the load/store instruction and
// emit an additonal add to take care of the offset register.
- if (!ImmediateOffsetNeedsLowering && Addr.getOffset() && Addr.isRegBase() &&
- Addr.getOffsetReg())
+ if (!ImmediateOffsetNeedsLowering && Addr.getOffset() && Addr.getOffsetReg())
RegisterOffsetNeedsLowering = true;
// Cannot encode zero register as base.
// If this is a stack pointer and the offset needs to be simplified then put
// the alloca address into a register, set the base type back to register and
// continue. This should almost never happen.
- if (ImmediateOffsetNeedsLowering && Addr.isFIBase()) {
+ if ((ImmediateOffsetNeedsLowering || Addr.getOffsetReg()) && Addr.isFIBase())
+ {
unsigned ResultReg = createResultReg(&AArch64::GPR64spRegClass);
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(AArch64::ADDXri),
ResultReg)
// reg+offset into a register.
if (ImmediateOffsetNeedsLowering) {
unsigned ResultReg;
- if (Addr.getReg()) {
+ if (Addr.getReg())
// Try to fold the immediate into the add instruction.
- if (Offset < 0)
- ResultReg = emitAddSub_ri(/*UseAdd=*/false, MVT::i64, Addr.getReg(),
- /*IsKill=*/false, -Offset);
- else
- ResultReg = emitAddSub_ri(/*UseAdd=*/true, MVT::i64, Addr.getReg(),
- /*IsKill=*/false, Offset);
- if (!ResultReg) {
- unsigned ImmReg = fastEmit_i(MVT::i64, MVT::i64, ISD::Constant, Offset);
- ResultReg = emitAddSub_rr(/*UseAdd=*/true, MVT::i64, Addr.getReg(),
- /*IsKill=*/false, ImmReg, /*IsKill=*/true);
- }
- } else
+ ResultReg = emitAdd_ri_(MVT::i64, Addr.getReg(), /*IsKill=*/false, Offset);
+ else
ResultReg = fastEmit_i(MVT::i64, MVT::i64, ISD::Constant, Offset);
if (!ResultReg)
MIB.addReg(Addr.getOffsetReg());
MIB.addImm(IsSigned);
MIB.addImm(Addr.getShift() != 0);
- } else {
- MIB.addReg(Addr.getReg());
- MIB.addImm(Offset);
- }
+ } else
+ MIB.addReg(Addr.getReg()).addImm(Offset);
}
if (MMO)
IsZExt);
}
+/// \brief This method is a wrapper to simplify add emission.
+///
+/// First try to emit an add with an immediate operand using emitAddSub_ri. If
+/// that fails, then try to materialize the immediate into a register and use
+/// emitAddSub_rr instead.
+unsigned AArch64FastISel::emitAdd_ri_(MVT VT, unsigned Op0, bool Op0IsKill,
+ int64_t Imm) {
+ unsigned ResultReg;
+ if (Imm < 0)
+ ResultReg = emitAddSub_ri(false, VT, Op0, Op0IsKill, -Imm);
+ else
+ ResultReg = emitAddSub_ri(true, VT, Op0, Op0IsKill, Imm);
+
+ if (ResultReg)
+ return ResultReg;
+
+ unsigned CReg = fastEmit_i(VT, VT, ISD::Constant, Imm);
+ if (!CReg)
+ return 0;
+
+ ResultReg = emitAddSub_rr(true, VT, Op0, Op0IsKill, CReg, true);
+ return ResultReg;
+}
+
unsigned AArch64FastISel::emitSub(MVT RetVT, const Value *LHS, const Value *RHS,
bool SetFlags, bool WantResult, bool IsZExt) {
return emitAddSub(/*UseAdd=*/false, RetVT, LHS, RHS, SetFlags, WantResult,
return false;
if (const auto *AI = dyn_cast<BinaryOperator>(LHS))
- if (AI->getOpcode() == Instruction::And) {
+ if (AI->getOpcode() == Instruction::And && isValueAvailable(AI)) {
const Value *AndLHS = AI->getOperand(0);
const Value *AndRHS = AI->getOperand(1);
bool Is64Bit = BW == 64;
if (TestBit < 32 && TestBit >= 0)
Is64Bit = false;
-
+
unsigned Opc = OpcTable[IsBitTest][IsCmpNE][Is64Bit];
const MCInstrDesc &II = TII.get(Opc);
SrcReg = fastEmitInst_extractsubreg(MVT::i32, SrcReg, SrcIsKill,
AArch64::sub_32);
+ if ((BW < 32) && !IsBitTest) {
+ EVT CmpEVT = TLI.getValueType(Ty, true);
+ SrcReg =
+ emitIntExt(CmpEVT.getSimpleVT(), SrcReg, MVT::i32, /*isZExt*/ true);
+ }
+
// Emit the combined compare and branch instruction.
SrcReg = constrainOperandRegClass(II, SrcReg, II.getNumDefs());
MachineInstrBuilder MIB =
return true;
}
- unsigned Pow2MinusOne = (1 << Lg2) - 1;
- unsigned AddReg = emitAddSub_ri(/*UseAdd=*/true, VT, Src0Reg,
- /*IsKill=*/false, Pow2MinusOne);
+ int64_t Pow2MinusOne = (1ULL << Lg2) - 1;
+ unsigned AddReg = emitAdd_ri_(VT, Src0Reg, /*IsKill=*/false, Pow2MinusOne);
if (!AddReg)
return false;
return true;
}
+/// This is mostly a copy of the existing FastISel GEP code, but we have to
+/// duplicate it for AArch64, because otherwise we would bail out even for
+/// simple cases. This is because the standard fastEmit functions don't cover
+/// MUL at all and ADD is lowered very inefficientily.
bool AArch64FastISel::selectGetElementPtr(const Instruction *I) {
unsigned N = getRegForValue(I->getOperand(0));
if (!N)
continue;
}
if (TotalOffs) {
- N = emitAddSub_ri(/*UseAdd=*/true, VT, N, NIsKill, TotalOffs);
- if (!N) {
- unsigned C = fastEmit_i(VT, VT, ISD::Constant, TotalOffs);
- if (!C)
- return false;
- N = emitAddSub_rr(/*UseAdd=*/true, VT, N, NIsKill, C, true);
- if (!N)
- return false;
- }
+ N = emitAdd_ri_(VT, N, NIsKill, TotalOffs);
+ if (!N)
+ return false;
NIsKill = true;
TotalOffs = 0;
}
}
}
if (TotalOffs) {
- N = emitAddSub_ri(/*UseAdd=*/true, VT, N, NIsKill, TotalOffs);
- if (!N) {
- unsigned C = fastEmit_i(VT, VT, ISD::Constant, TotalOffs);
- if (!C)
- return false;
- N = emitAddSub_rr(/*UseAdd=*/true, VT, N, NIsKill, C, true);
- if (!N)
- return false;
- }
+ N = emitAdd_ri_(VT, N, NIsKill, TotalOffs);
+ if (!N)
+ return false;
}
-
updateValueMap(I, N);
return true;
}
projects / oota-llvm.git / blobdiff