if (Subtarget->useCRBits()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
- setOperationAction(ISD::SINT_TO_FP, MVT::i1, Promote);
- AddPromotedToType (ISD::SINT_TO_FP, MVT::i1,
- isPPC64 ? MVT::i64 : MVT::i32);
- setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
- AddPromotedToType (ISD::UINT_TO_FP, MVT::i1,
- isPPC64 ? MVT::i64 : MVT::i32);
+ if (isPPC64 || Subtarget->hasFPCVT()) {
+ setOperationAction(ISD::SINT_TO_FP, MVT::i1, Promote);
+ AddPromotedToType (ISD::SINT_TO_FP, MVT::i1,
+ isPPC64 ? MVT::i64 : MVT::i32);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
+ AddPromotedToType (ISD::UINT_TO_FP, MVT::i1,
+ isPPC64 ? MVT::i64 : MVT::i32);
+ } else {
+ setOperationAction(ISD::SINT_TO_FP, MVT::i1, Custom);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i1, Custom);
+ }
// PowerPC does not support direct load / store of condition registers
setOperationAction(ISD::LOAD, MVT::i1, Custom);
setOperationAction(ISD::MUL, MVT::v4f32, Legal);
setOperationAction(ISD::FMA, MVT::v4f32, Legal);
- if (TM.Options.UnsafeFPMath) {
+ if (TM.Options.UnsafeFPMath || Subtarget->hasVSX()) {
setOperationAction(ISD::FDIV, MVT::v4f32, Legal);
setOperationAction(ISD::FSQRT, MVT::v4f32, Legal);
}
setCondCodeAction(ISD::SETO, MVT::v4f32, Expand);
setCondCodeAction(ISD::SETONE, MVT::v4f32, Expand);
+
+ if (Subtarget->hasVSX()) {
+ setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v2f64, Legal);
+
+ setOperationAction(ISD::FFLOOR, MVT::v2f64, Legal);
+ setOperationAction(ISD::FCEIL, MVT::v2f64, Legal);
+ setOperationAction(ISD::FTRUNC, MVT::v2f64, Legal);
+ setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Legal);
+ setOperationAction(ISD::FROUND, MVT::v2f64, Legal);
+
+ setOperationAction(ISD::FROUND, MVT::v4f32, Legal);
+
+ setOperationAction(ISD::MUL, MVT::v2f64, Legal);
+ setOperationAction(ISD::FMA, MVT::v2f64, Legal);
+
+ setOperationAction(ISD::FDIV, MVT::v2f64, Legal);
+ setOperationAction(ISD::FSQRT, MVT::v2f64, Legal);
+
+ // Share the Altivec comparison restrictions.
+ setCondCodeAction(ISD::SETUO, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETUEQ, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETUGT, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETUGE, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETULT, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETULE, MVT::v2f64, Expand);
+
+ setCondCodeAction(ISD::SETO, MVT::v2f64, Expand);
+ setCondCodeAction(ISD::SETONE, MVT::v2f64, Expand);
+
+ addRegisterClass(MVT::f64, &PPC::VSRCRegClass);
+
+ addRegisterClass(MVT::v4f32, &PPC::VSRCRegClass);
+ addRegisterClass(MVT::v2f64, &PPC::VSRCRegClass);
+ }
}
if (Subtarget->has64BitSupport()) {
setTargetDAGCombine(ISD::BSWAP);
setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
+ setTargetDAGCombine(ISD::SIGN_EXTEND);
+ setTargetDAGCombine(ISD::ZERO_EXTEND);
+ setTargetDAGCombine(ISD::ANY_EXTEND);
+
if (Subtarget->useCRBits()) {
- setTargetDAGCombine(ISD::SIGN_EXTEND);
- setTargetDAGCombine(ISD::ZERO_EXTEND);
- setTargetDAGCombine(ISD::ANY_EXTEND);
setTargetDAGCombine(ISD::TRUNCATE);
setTargetDAGCombine(ISD::SETCC);
setTargetDAGCombine(ISD::SELECT_CC);
case MVT::v8i16:
case MVT::v4i32:
case MVT::v4f32:
+ case MVT::v2f64:
RC = &PPC::VRRCRegClass;
break;
}
// Varargs or 64 bit Altivec parameters are padded to a 16 byte boundary.
if (ObjectVT==MVT::v4f32 || ObjectVT==MVT::v4i32 ||
- ObjectVT==MVT::v8i16 || ObjectVT==MVT::v16i8) {
+ ObjectVT==MVT::v8i16 || ObjectVT==MVT::v16i8 ||
+ ObjectVT==MVT::v2f64) {
if (isVarArg) {
MinReservedArea = ((MinReservedArea+15)/16)*16;
MinReservedArea += CalculateStackSlotSize(ObjectVT,
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
+ case MVT::v2f64:
// Note that vector arguments in registers don't reserve stack space,
// except in varargs functions.
if (VR_idx != Num_VR_Regs) {
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::GPRCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
+
+ if (ObjectVT == MVT::i1)
+ ArgVal = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, ArgVal);
+
++GPR_idx;
} else {
needsLoad = true;
EVT ArgVT = Outs[i].VT;
// Varargs Altivec parameters are padded to a 16 byte boundary.
if (ArgVT==MVT::v4f32 || ArgVT==MVT::v4i32 ||
- ArgVT==MVT::v8i16 || ArgVT==MVT::v16i8) {
+ ArgVT==MVT::v8i16 || ArgVT==MVT::v16i8 ||
+ ArgVT==MVT::v2f64) {
if (!isVarArg && !isPPC64) {
// Non-varargs Altivec parameters go after all the non-Altivec
// parameters; handle those later so we know how much padding we need.
}
if (VA.isRegLoc()) {
+ if (Arg.getValueType() == MVT::i1)
+ Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Arg);
+
seenFloatArg |= VA.getLocVT().isFloatingPoint();
// Put argument in a physical register.
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
+ case MVT::v2f64:
if (isVarArg) {
// These go aligned on the stack, or in the corresponding R registers
// when within range. The Darwin PPC ABI doc claims they also go in
case MVT::i32:
case MVT::i64:
if (GPR_idx != NumGPRs) {
+ if (Arg.getValueType() == MVT::i1)
+ Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, PtrVT, Arg);
+
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Arg));
} else {
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
return SDValue();
+ if (Op.getOperand(0).getValueType() == MVT::i1)
+ return DAG.getNode(ISD::SELECT, dl, Op.getValueType(), Op.getOperand(0),
+ DAG.getConstantFP(1.0, Op.getValueType()),
+ DAG.getConstantFP(0.0, Op.getValueType()));
+
assert((Op.getOpcode() == ISD::SINT_TO_FP || PPCSubTarget.hasFPCVT()) &&
"UINT_TO_FP is supported only with FPCVT");
F->insert(It, loopMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
F->insert(It, loopMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), MBB,
- llvm::next(MachineBasicBlock::iterator(MI)), MBB->end());
+ std::next(MachineBasicBlock::iterator(MI)), MBB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
// Note that the structure of the jmp_buf used here is not compatible
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
// Next, add the true and fallthrough blocks as its successors.
F->insert(It, midMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
// thisMBB:
F->insert(It, midMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
- llvm::next(MachineBasicBlock::iterator(MI)),
- BB->end());
+ std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
if ((VT == MVT::f32 && PPCSubTarget.hasFRES()) ||
(VT == MVT::f64 && PPCSubTarget.hasFRE()) ||
- (VT == MVT::v4f32 && PPCSubTarget.hasAltivec())) {
+ (VT == MVT::v4f32 && PPCSubTarget.hasAltivec()) ||
+ (VT == MVT::v2f64 && PPCSubTarget.hasVSX())) {
// Newton iteration for a function: F(X) is X_{i+1} = X_i - F(X_i)/F'(X_i)
// For the reciprocal, we need to find the zero of the function:
if ((VT == MVT::f32 && PPCSubTarget.hasFRSQRTES()) ||
(VT == MVT::f64 && PPCSubTarget.hasFRSQRTE()) ||
- (VT == MVT::v4f32 && PPCSubTarget.hasAltivec())) {
+ (VT == MVT::v4f32 && PPCSubTarget.hasAltivec()) ||
+ (VT == MVT::v2f64 && PPCSubTarget.hasVSX())) {
// Newton iteration for a function: F(X) is X_{i+1} = X_i - F(X_i)/F'(X_i)
// For the reciprocal sqrt, we need to find the zero of the function:
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
+
+ // Make sure that we're not going to promote the non-output-value
+ // operand(s) or SELECT or SELECT_CC.
+ // FIXME: Although we could sometimes handle this, and it does occur in
+ // practice that one of the condition inputs to the select is also one of
+ // the outputs, we currently can't deal with this.
+ if (User->getOpcode() == ISD::SELECT) {
+ if (User->getOperand(0) == Inputs[i])
+ return SDValue();
+ } else if (User->getOpcode() == ISD::SELECT_CC) {
+ if (User->getOperand(0) == Inputs[i] ||
+ User->getOperand(1) == Inputs[i])
+ return SDValue();
+ }
}
}
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
+
+ // Make sure that we're not going to promote the non-output-value
+ // operand(s) or SELECT or SELECT_CC.
+ // FIXME: Although we could sometimes handle this, and it does occur in
+ // practice that one of the condition inputs to the select is also one of
+ // the outputs, we currently can't deal with this.
+ if (User->getOpcode() == ISD::SELECT) {
+ if (User->getOperand(0) == PromOps[i])
+ return SDValue();
+ } else if (User->getOpcode() == ISD::SELECT_CC) {
+ if (User->getOperand(0) == PromOps[i] ||
+ User->getOperand(1) == PromOps[i])
+ return SDValue();
+ }
}
}
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
- assert(PPCSubTarget.useCRBits() &&
- "Expecting to be tracking CR bits");
// If we're tracking CR bits, we need to be careful that we don't have:
// zext(binary-ops(trunc(x), trunc(y)))
// or
// bits are set as required by the final extension, we still may need to do
// some masking to get the proper behavior.
+ // This same functionality is important on PPC64 when dealing with
+ // 32-to-64-bit extensions; these occur often when 32-bit values are used as
+ // the return values of functions. Because it is so similar, it is handled
+ // here as well.
+
if (N->getValueType(0) != MVT::i32 &&
N->getValueType(0) != MVT::i64)
return SDValue();
- if (N->getOperand(0).getValueType() != MVT::i1)
+ if (!((N->getOperand(0).getValueType() == MVT::i1 &&
+ PPCSubTarget.useCRBits()) ||
+ (N->getOperand(0).getValueType() == MVT::i32 &&
+ PPCSubTarget.isPPC64())))
return SDValue();
if (N->getOperand(0).getOpcode() != ISD::AND &&
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
+
+ // Make sure that we're not going to promote the non-output-value
+ // operand(s) or SELECT or SELECT_CC.
+ // FIXME: Although we could sometimes handle this, and it does occur in
+ // practice that one of the condition inputs to the select is also one of
+ // the outputs, we currently can't deal with this.
+ if (User->getOpcode() == ISD::SELECT) {
+ if (User->getOperand(0) == Inputs[i])
+ return SDValue();
+ } else if (User->getOpcode() == ISD::SELECT_CC) {
+ if (User->getOperand(0) == Inputs[i] ||
+ User->getOperand(1) == Inputs[i])
+ return SDValue();
+ }
}
}
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
+
+ // Make sure that we're not going to promote the non-output-value
+ // operand(s) or SELECT or SELECT_CC.
+ // FIXME: Although we could sometimes handle this, and it does occur in
+ // practice that one of the condition inputs to the select is also one of
+ // the outputs, we currently can't deal with this.
+ if (User->getOpcode() == ISD::SELECT) {
+ if (User->getOperand(0) == PromOps[i])
+ return SDValue();
+ } else if (User->getOpcode() == ISD::SELECT_CC) {
+ if (User->getOperand(0) == PromOps[i] ||
+ User->getOperand(1) == PromOps[i])
+ return SDValue();
+ }
}
}
+ unsigned PromBits = N->getOperand(0).getValueSizeInBits();
bool ReallyNeedsExt = false;
if (N->getOpcode() != ISD::ANY_EXTEND) {
// If all of the inputs are not already sign/zero extended, then
unsigned OpBits =
Inputs[i].getOperand(0).getValueSizeInBits();
+ assert(PromBits < OpBits && "Truncation not to a smaller bit count?");
+
if ((N->getOpcode() == ISD::ZERO_EXTEND &&
!DAG.MaskedValueIsZero(Inputs[i].getOperand(0),
- APInt::getHighBitsSet(OpBits,
- OpBits-1))) ||
+ APInt::getHighBitsSet(OpBits,
+ OpBits-PromBits))) ||
(N->getOpcode() == ISD::SIGN_EXTEND &&
- DAG.ComputeNumSignBits(Inputs[i].getOperand(0)) != OpBits)) {
+ DAG.ComputeNumSignBits(Inputs[i].getOperand(0)) <
+ (OpBits-(PromBits-1)))) {
ReallyNeedsExt = true;
break;
}
if (!ReallyNeedsExt)
return N->getOperand(0);
- // To zero extend, just mask off everything except for the first bit.
+ // To zero extend, just mask off everything except for the first bit (in the
+ // i1 case).
if (N->getOpcode() == ISD::ZERO_EXTEND)
return DAG.getNode(ISD::AND, dl, N->getValueType(0), N->getOperand(0),
- DAG.getConstant(1, N->getValueType(0)));
+ DAG.getConstant(APInt::getLowBitsSet(
+ N->getValueSizeInBits(0), PromBits),
+ N->getValueType(0)));
assert(N->getOpcode() == ISD::SIGN_EXTEND &&
"Invalid extension type");
EVT ShiftAmountTy = getShiftAmountTy(N->getValueType(0));
SDValue ShiftCst =
- DAG.getConstant(N->getValueSizeInBits(0)-1, ShiftAmountTy);
+ DAG.getConstant(N->getValueSizeInBits(0)-PromBits, ShiftAmountTy);
return DAG.getNode(ISD::SRA, dl, N->getValueType(0),
DAG.getNode(ISD::SHL, dl, N->getValueType(0),
N->getOperand(0), ShiftCst), ShiftCst);
unsigned ABIAlignment = getDataLayout()->getABITypeAlignment(Ty);
if (ISD::isNON_EXTLoad(N) && VT.isVector() &&
TM.getSubtarget<PPCSubtarget>().hasAltivec() &&
+ // FIXME: Update this for VSX!
(VT == MVT::v16i8 || VT == MVT::v8i16 ||
VT == MVT::v4i32 || VT == MVT::v4f32) &&
LD->getAlignment() < ABIAlignment) {
// suboptimal.
return C_Memory;
}
+ } else if (Constraint == "wc") { // individual CR bits.
+ return C_RegisterClass;
+ } else if (Constraint == "wa" || Constraint == "wd" ||
+ Constraint == "wf" || Constraint == "ws") {
+ return C_RegisterClass; // VSX registers.
}
return TargetLowering::getConstraintType(Constraint);
}
if (CallOperandVal == NULL)
return CW_Default;
Type *type = CallOperandVal->getType();
+
// Look at the constraint type.
+ if (StringRef(constraint) == "wc" && type->isIntegerTy(1))
+ return CW_Register; // an individual CR bit.
+ else if ((StringRef(constraint) == "wa" ||
+ StringRef(constraint) == "wd" ||
+ StringRef(constraint) == "wf") &&
+ type->isVectorTy())
+ return CW_Register;
+ else if (StringRef(constraint) == "ws" && type->isDoubleTy())
+ return CW_Register;
+
switch (*constraint) {
default:
weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
case 'y': // crrc
return std::make_pair(0U, &PPC::CRRCRegClass);
}
+ } else if (Constraint == "wc") { // an individual CR bit.
+ return std::make_pair(0U, &PPC::CRBITRCRegClass);
+ } else if (Constraint == "wa" || Constraint == "wd" ||
+ Constraint == "wf" || Constraint == "ws") {
+ return std::make_pair(0U, &PPC::VSRCRegClass);
}
std::pair<unsigned, const TargetRegisterClass*> R =
projects / oota-llvm.git / blobdiff