X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FSelectionDAG%2FLegalizeIntegerTypes.cpp;h=d93b35dfb8784c4cc4cbb55d74b198971702a759;hb=7e2c793a2b5c746344652b6579e958ee42fafdcc;hp=fc9a41ecaa9fa189ae735db549e55670ba2b098a;hpb=17e76bda552d5afa652c4bb45f6dbdad67dc5c20;p=oota-llvm.git diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp index fc9a41ecaa9..d93b35dfb87 100644 --- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp +++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp @@ -20,7 +20,6 @@ #include "LegalizeTypes.h" #include "llvm/DerivedTypes.h" -#include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/raw_ostream.h" using namespace llvm; @@ -57,8 +56,10 @@ void DAGTypeLegalizer::PromoteIntegerResult(SDNode *N, unsigned ResNo) { case ISD::Constant: Res = PromoteIntRes_Constant(N); break; case ISD::CONVERT_RNDSAT: Res = PromoteIntRes_CONVERT_RNDSAT(N); break; + case ISD::CTLZ_ZERO_UNDEF: case ISD::CTLZ: Res = PromoteIntRes_CTLZ(N); break; case ISD::CTPOP: Res = PromoteIntRes_CTPOP(N); break; + case ISD::CTTZ_ZERO_UNDEF: case ISD::CTTZ: Res = PromoteIntRes_CTTZ(N); break; case ISD::EXTRACT_VECTOR_ELT: Res = PromoteIntRes_EXTRACT_VECTOR_ELT(N); break; @@ -211,13 +212,10 @@ SDValue DAGTypeLegalizer::PromoteIntRes_BITCAST(SDNode *N) { DebugLoc dl = N->getDebugLoc(); switch (getTypeAction(InVT)) { - default: - assert(false && "Unknown type action!"); - break; case TargetLowering::TypeLegal: break; case TargetLowering::TypePromoteInteger: - if (NOutVT.bitsEq(NInVT)) + if (NOutVT.bitsEq(NInVT) && !NOutVT.isVector() && !NInVT.isVector()) // The input promotes to the same size. Convert the promoted value. return DAG.getNode(ISD::BITCAST, dl, NOutVT, GetPromotedInteger(InOp)); break; @@ -251,9 +249,11 @@ SDValue DAGTypeLegalizer::PromoteIntRes_BITCAST(SDNode *N) { return DAG.getNode(ISD::BITCAST, dl, NOutVT, InOp); } case TargetLowering::TypeWidenVector: - if (OutVT.bitsEq(NInVT)) - // The input is widened to the same size. Convert to the widened value. - return DAG.getNode(ISD::BITCAST, dl, OutVT, GetWidenedVector(InOp)); + // The input is widened to the same size. Convert to the widened value. + // Make sure that the outgoing value is not a vector, because this would + // make us bitcast between two vectors which are legalized in different ways. + if (NOutVT.bitsEq(NInVT) && !NOutVT.isVector()) + return DAG.getNode(ISD::BITCAST, dl, NOutVT, GetWidenedVector(InOp)); } return DAG.getNode(ISD::ANY_EXTEND, dl, NOutVT, @@ -312,7 +312,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTLZ(SDNode *N) { DebugLoc dl = N->getDebugLoc(); EVT OVT = N->getValueType(0); EVT NVT = Op.getValueType(); - Op = DAG.getNode(ISD::CTLZ, dl, NVT, Op); + Op = DAG.getNode(N->getOpcode(), dl, NVT, Op); // Subtract off the extra leading bits in the bigger type. return DAG.getNode(ISD::SUB, dl, NVT, Op, DAG.getConstant(NVT.getSizeInBits() - @@ -330,13 +330,15 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTTZ(SDNode *N) { EVT OVT = N->getValueType(0); EVT NVT = Op.getValueType(); DebugLoc dl = N->getDebugLoc(); - // The count is the same in the promoted type except if the original - // value was zero. This can be handled by setting the bit just off - // the top of the original type. - APInt TopBit(NVT.getSizeInBits(), 0); - TopBit.setBit(OVT.getSizeInBits()); - Op = DAG.getNode(ISD::OR, dl, NVT, Op, DAG.getConstant(TopBit, NVT)); - return DAG.getNode(ISD::CTTZ, dl, NVT, Op); + if (N->getOpcode() == ISD::CTTZ) { + // The count is the same in the promoted type except if the original + // value was zero. This can be handled by setting the bit just off + // the top of the original type. + APInt TopBit(NVT.getSizeInBits(), 0); + TopBit.setBit(OVT.getSizeInBits()); + Op = DAG.getNode(ISD::OR, dl, NVT, Op, DAG.getConstant(TopBit, NVT)); + } + return DAG.getNode(N->getOpcode(), dl, NVT, Op); } SDValue DAGTypeLegalizer::PromoteIntRes_EXTRACT_VECTOR_ELT(SDNode *N) { @@ -445,7 +447,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SADDSUBO(SDNode *N, unsigned ResNo) { if (ResNo == 1) return PromoteIntRes_Overflow(N); - // The operation overflowed iff the result in the larger type is not the + // The operation overflowed if the result in the larger type is not the // sign extension of its truncation to the original type. SDValue LHS = SExtPromotedInteger(N->getOperand(0)); SDValue RHS = SExtPromotedInteger(N->getOperand(1)); @@ -486,7 +488,11 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SELECT(SDNode *N) { } SDValue DAGTypeLegalizer::PromoteIntRes_VSELECT(SDNode *N) { - SDValue Mask = GetPromotedInteger(N->getOperand(0)); + SDValue Mask = N->getOperand(0); + EVT OpTy = N->getOperand(1).getValueType(); + + // Promote all the way up to the canonical SetCC type. + Mask = PromoteTargetBoolean(Mask, TLI.getSetCCResultType(OpTy)); SDValue LHS = GetPromotedInteger(N->getOperand(1)); SDValue RHS = GetPromotedInteger(N->getOperand(2)); return DAG.getNode(ISD::VSELECT, N->getDebugLoc(), @@ -504,14 +510,12 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SELECT_CC(SDNode *N) { SDValue DAGTypeLegalizer::PromoteIntRes_SETCC(SDNode *N) { EVT SVT = TLI.getSetCCResultType(N->getOperand(0).getValueType()); - // Convert to the expected type. EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0)); // Only use the result of getSetCCResultType if it is legal, // otherwise just use the promoted result type (NVT). - if (getTypeAction(SVT) != TargetLowering::TypeLegal) { - SVT = NVT; - } + if (!TLI.isTypeLegal(SVT)) + SVT = NVT; DebugLoc dl = N->getDebugLoc(); assert(SVT.isVector() == N->getOperand(0).getValueType().isVector() && @@ -522,6 +526,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SETCC(SDNode *N) { N->getOperand(1), N->getOperand(2)); assert(NVT.bitsLE(SVT) && "Integer type overpromoted?"); + // Convert to the expected type. return DAG.getNode(ISD::TRUNCATE, dl, NVT, SetCC); } @@ -583,18 +588,14 @@ SDValue DAGTypeLegalizer::PromoteIntRes_TRUNCATE(SDNode *N) { unsigned NumElts = InVT.getVectorNumElements(); assert(NumElts == NVT.getVectorNumElements() && "Dst and Src must have the same number of elements"); - EVT EltVT = InVT.getScalarType(); assert(isPowerOf2_32(NumElts) && "Promoted vector type must be a power of two"); - EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts/2); + SDValue EOp1, EOp2; + GetSplitVector(InOp, EOp1, EOp2); + EVT HalfNVT = EVT::getVectorVT(*DAG.getContext(), NVT.getScalarType(), NumElts/2); - - SDValue EOp1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, HalfVT, InOp, - DAG.getIntPtrConstant(0)); - SDValue EOp2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, HalfVT, InOp, - DAG.getIntPtrConstant(NumElts/2)); EOp1 = DAG.getNode(ISD::TRUNCATE, dl, HalfNVT, EOp1); EOp2 = DAG.getNode(ISD::TRUNCATE, dl, HalfNVT, EOp2); @@ -609,7 +610,7 @@ SDValue DAGTypeLegalizer::PromoteIntRes_UADDSUBO(SDNode *N, unsigned ResNo) { if (ResNo == 1) return PromoteIntRes_Overflow(N); - // The operation overflowed iff the result in the larger type is not the + // The operation overflowed if the result in the larger type is not the // zero extension of its truncation to the original type. SDValue LHS = ZExtPromotedInteger(N->getOperand(0)); SDValue RHS = ZExtPromotedInteger(N->getOperand(1)); @@ -654,17 +655,17 @@ SDValue DAGTypeLegalizer::PromoteIntRes_XMULO(SDNode *N, unsigned ResNo) { } SDValue Mul = DAG.getNode(ISD::MUL, DL, LHS.getValueType(), LHS, RHS); - // Overflow occurred iff the high part of the result does not + // Overflow occurred if the high part of the result does not // zero/sign-extend the low part. SDValue Overflow; if (N->getOpcode() == ISD::UMULO) { - // Unsigned overflow occurred iff the high part is non-zero. + // Unsigned overflow occurred if the high part is non-zero. SDValue Hi = DAG.getNode(ISD::SRL, DL, Mul.getValueType(), Mul, DAG.getIntPtrConstant(SmallVT.getSizeInBits())); Overflow = DAG.getSetCC(DL, N->getValueType(1), Hi, DAG.getConstant(0, Hi.getValueType()), ISD::SETNE); } else { - // Signed overflow occurred iff the high part does not sign extend the low. + // Signed overflow occurred if the high part does not sign extend the low. SDValue SExt = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, Mul.getValueType(), Mul, DAG.getValueType(SmallVT)); Overflow = DAG.getSetCC(DL, N->getValueType(1), SExt, Mul, ISD::SETNE); @@ -1099,8 +1100,10 @@ void DAGTypeLegalizer::ExpandIntegerResult(SDNode *N, unsigned ResNo) { case ISD::AssertZext: ExpandIntRes_AssertZext(N, Lo, Hi); break; case ISD::BSWAP: ExpandIntRes_BSWAP(N, Lo, Hi); break; case ISD::Constant: ExpandIntRes_Constant(N, Lo, Hi); break; + case ISD::CTLZ_ZERO_UNDEF: case ISD::CTLZ: ExpandIntRes_CTLZ(N, Lo, Hi); break; case ISD::CTPOP: ExpandIntRes_CTPOP(N, Lo, Hi); break; + case ISD::CTTZ_ZERO_UNDEF: case ISD::CTTZ: ExpandIntRes_CTTZ(N, Lo, Hi); break; case ISD::FP_TO_SINT: ExpandIntRes_FP_TO_SINT(N, Lo, Hi); break; case ISD::FP_TO_UINT: ExpandIntRes_FP_TO_UINT(N, Lo, Hi); break; @@ -1172,7 +1175,6 @@ std::pair DAGTypeLegalizer::ExpandAtomic(SDNode *Node) { switch (Opc) { default: llvm_unreachable("Unhandled atomic intrinsic Expand!"); - break; case ISD::ATOMIC_SWAP: switch (VT.SimpleTy) { default: llvm_unreachable("Unexpected value type for atomic!"); @@ -1356,7 +1358,7 @@ ExpandShiftWithKnownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi) { APInt HighBitMask = APInt::getHighBitsSet(ShBits, ShBits - Log2_32(NVTBits)); APInt KnownZero, KnownOne; - DAG.ComputeMaskedBits(N->getOperand(1), HighBitMask, KnownZero, KnownOne); + DAG.ComputeMaskedBits(N->getOperand(1), KnownZero, KnownOne); // If we don't know anything about the high bits, exit. if (((KnownZero|KnownOne) & HighBitMask) == 0) @@ -1391,15 +1393,15 @@ ExpandShiftWithKnownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi) { } } -#if 0 - // FIXME: This code is broken for shifts with a zero amount! // If we know that all of the high bits of the shift amount are zero, then we // can do this as a couple of simple shifts. if ((KnownZero & HighBitMask) == HighBitMask) { - // Compute 32-amt. - SDValue Amt2 = DAG.getNode(ISD::SUB, ShTy, - DAG.getConstant(NVTBits, ShTy), - Amt); + // Calculate 31-x. 31 is used instead of 32 to avoid creating an undefined + // shift if x is zero. We can use XOR here because x is known to be smaller + // than 32. + SDValue Amt2 = DAG.getNode(ISD::XOR, dl, ShTy, Amt, + DAG.getConstant(NVTBits-1, ShTy)); + unsigned Op1, Op2; switch (N->getOpcode()) { default: llvm_unreachable("Unknown shift"); @@ -1408,13 +1410,23 @@ ExpandShiftWithKnownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi) { case ISD::SRA: Op1 = ISD::SRL; Op2 = ISD::SHL; break; } - Lo = DAG.getNode(N->getOpcode(), NVT, InL, Amt); - Hi = DAG.getNode(ISD::OR, NVT, - DAG.getNode(Op1, NVT, InH, Amt), - DAG.getNode(Op2, NVT, InL, Amt2)); + // When shifting right the arithmetic for Lo and Hi is swapped. + if (N->getOpcode() != ISD::SHL) + std::swap(InL, InH); + + // Use a little trick to get the bits that move from Lo to Hi. First + // shift by one bit. + SDValue Sh1 = DAG.getNode(Op2, dl, NVT, InL, DAG.getConstant(1, ShTy)); + // Then compute the remaining shift with amount-1. + SDValue Sh2 = DAG.getNode(Op2, dl, NVT, Sh1, Amt2); + + Lo = DAG.getNode(N->getOpcode(), dl, NVT, InL, Amt); + Hi = DAG.getNode(ISD::OR, dl, NVT, DAG.getNode(Op1, dl, NVT, InH, Amt),Sh2); + + if (N->getOpcode() != ISD::SHL) + std::swap(Hi, Lo); return true; } -#endif return false; } @@ -1494,8 +1506,6 @@ ExpandShiftWithUnknownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi) { Hi = DAG.getNode(ISD::SELECT, dl, NVT, isShort, HiS, HiL); return true; } - - return false; } void DAGTypeLegalizer::ExpandIntRes_ADDSUB(SDNode *N, @@ -1703,8 +1713,8 @@ void DAGTypeLegalizer::ExpandIntRes_CTLZ(SDNode *N, SDValue HiNotZero = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT), Hi, DAG.getConstant(0, NVT), ISD::SETNE); - SDValue LoLZ = DAG.getNode(ISD::CTLZ, dl, NVT, Lo); - SDValue HiLZ = DAG.getNode(ISD::CTLZ, dl, NVT, Hi); + SDValue LoLZ = DAG.getNode(N->getOpcode(), dl, NVT, Lo); + SDValue HiLZ = DAG.getNode(ISD::CTLZ_ZERO_UNDEF, dl, NVT, Hi); Lo = DAG.getNode(ISD::SELECT, dl, NVT, HiNotZero, HiLZ, DAG.getNode(ISD::ADD, dl, NVT, LoLZ, @@ -1733,8 +1743,8 @@ void DAGTypeLegalizer::ExpandIntRes_CTTZ(SDNode *N, SDValue LoNotZero = DAG.getSetCC(dl, TLI.getSetCCResultType(NVT), Lo, DAG.getConstant(0, NVT), ISD::SETNE); - SDValue LoLZ = DAG.getNode(ISD::CTTZ, dl, NVT, Lo); - SDValue HiLZ = DAG.getNode(ISD::CTTZ, dl, NVT, Hi); + SDValue LoLZ = DAG.getNode(ISD::CTTZ_ZERO_UNDEF, dl, NVT, Lo); + SDValue HiLZ = DAG.getNode(N->getOpcode(), dl, NVT, Hi); Lo = DAG.getNode(ISD::SELECT, dl, NVT, LoNotZero, LoLZ, DAG.getNode(ISD::ADD, dl, NVT, HiLZ, @@ -1779,6 +1789,7 @@ void DAGTypeLegalizer::ExpandIntRes_LOAD(LoadSDNode *N, unsigned Alignment = N->getAlignment(); bool isVolatile = N->isVolatile(); bool isNonTemporal = N->isNonTemporal(); + bool isInvariant = N->isInvariant(); DebugLoc dl = N->getDebugLoc(); assert(NVT.isByteSized() && "Expanded type not byte sized!"); @@ -1809,7 +1820,7 @@ void DAGTypeLegalizer::ExpandIntRes_LOAD(LoadSDNode *N, } else if (TLI.isLittleEndian()) { // Little-endian - low bits are at low addresses. Lo = DAG.getLoad(NVT, dl, Ch, Ptr, N->getPointerInfo(), - isVolatile, isNonTemporal, Alignment); + isVolatile, isNonTemporal, isInvariant, Alignment); unsigned ExcessBits = N->getMemoryVT().getSizeInBits() - NVT.getSizeInBits(); @@ -2229,8 +2240,8 @@ void DAGTypeLegalizer::ExpandIntRes_UADDSUBO(SDNode *N, LHS, RHS); SplitInteger(Sum, Lo, Hi); - // Calculate the overflow: addition overflows iff a + b < a, and subtraction - // overflows iff a - b > a. + // Calculate the overflow: addition overflows if a + b < a, and subtraction + // overflows if a - b > a. SDValue Ofl = DAG.getSetCC(dl, N->getValueType(1), Sum, LHS, N->getOpcode () == ISD::UADDO ? ISD::SETULT : ISD::SETUGT); @@ -2242,32 +2253,35 @@ void DAGTypeLegalizer::ExpandIntRes_UADDSUBO(SDNode *N, void DAGTypeLegalizer::ExpandIntRes_XMULO(SDNode *N, SDValue &Lo, SDValue &Hi) { EVT VT = N->getValueType(0); - Type *RetTy = VT.getTypeForEVT(*DAG.getContext()); - EVT PtrVT = TLI.getPointerTy(); - Type *PtrTy = PtrVT.getTypeForEVT(*DAG.getContext()); DebugLoc dl = N->getDebugLoc(); // A divide for UMULO should be faster than a function call. if (N->getOpcode() == ISD::UMULO) { SDValue LHS = N->getOperand(0), RHS = N->getOperand(1); - DebugLoc DL = N->getDebugLoc(); - SDValue MUL = DAG.getNode(ISD::MUL, DL, LHS.getValueType(), LHS, RHS); + SDValue MUL = DAG.getNode(ISD::MUL, dl, LHS.getValueType(), LHS, RHS); SplitInteger(MUL, Lo, Hi); // A divide for UMULO will be faster than a function call. Select to // make sure we aren't using 0. SDValue isZero = DAG.getSetCC(dl, TLI.getSetCCResultType(VT), - RHS, DAG.getConstant(0, VT), ISD::SETNE); + RHS, DAG.getConstant(0, VT), ISD::SETEQ); SDValue NotZero = DAG.getNode(ISD::SELECT, dl, VT, isZero, - DAG.getConstant(1, VT), RHS); - SDValue DIV = DAG.getNode(ISD::UDIV, DL, LHS.getValueType(), MUL, NotZero); - SDValue Overflow; - Overflow = DAG.getSetCC(DL, N->getValueType(1), DIV, LHS, ISD::SETNE); + DAG.getConstant(1, VT), RHS); + SDValue DIV = DAG.getNode(ISD::UDIV, dl, VT, MUL, NotZero); + SDValue Overflow = DAG.getSetCC(dl, N->getValueType(1), DIV, LHS, + ISD::SETNE); + Overflow = DAG.getNode(ISD::SELECT, dl, N->getValueType(1), isZero, + DAG.getConstant(0, N->getValueType(1)), + Overflow); ReplaceValueWith(SDValue(N, 1), Overflow); return; } + Type *RetTy = VT.getTypeForEVT(*DAG.getContext()); + EVT PtrVT = TLI.getPointerTy(); + Type *PtrTy = PtrVT.getTypeForEVT(*DAG.getContext()); + // Replace this with a libcall that will check overflow. RTLIB::Libcall LC = RTLIB::UNKNOWN_LIBCALL; if (VT == MVT::i32) @@ -2281,8 +2295,8 @@ void DAGTypeLegalizer::ExpandIntRes_XMULO(SDNode *N, SDValue Temp = DAG.CreateStackTemporary(PtrVT); // Temporary for the overflow value, default it to zero. SDValue Chain = DAG.getStore(DAG.getEntryNode(), dl, - DAG.getConstant(0, PtrVT), Temp, - MachinePointerInfo(), false, false, 0); + DAG.getConstant(0, PtrVT), Temp, + MachinePointerInfo(), false, false, 0); TargetLowering::ArgListTy Args; TargetLowering::ArgListEntry Entry; @@ -2304,14 +2318,17 @@ void DAGTypeLegalizer::ExpandIntRes_XMULO(SDNode *N, Args.push_back(Entry); SDValue Func = DAG.getExternalSymbol(TLI.getLibcallName(LC), PtrVT); - std::pair CallInfo = - TLI.LowerCallTo(Chain, RetTy, true, false, false, false, - 0, TLI.getLibcallCallingConv(LC), false, - true, Func, Args, DAG, dl); + TargetLowering:: + CallLoweringInfo CLI(Chain, RetTy, true, false, false, false, + 0, TLI.getLibcallCallingConv(LC), + /*isTailCall=*/false, + /*doesNotReturn=*/false, /*isReturnValueUsed=*/true, + Func, Args, DAG, dl); + std::pair CallInfo = TLI.LowerCallTo(CLI); SplitInteger(CallInfo.first, Lo, Hi); SDValue Temp2 = DAG.getLoad(PtrVT, dl, CallInfo.second, Temp, - MachinePointerInfo(), false, false, 0); + MachinePointerInfo(), false, false, false, 0); SDValue Ofl = DAG.getSetCC(dl, N->getValueType(1), Temp2, DAG.getConstant(0, PtrVT), ISD::SETNE); @@ -2782,7 +2799,7 @@ SDValue DAGTypeLegalizer::ExpandIntOp_UINT_TO_FP(SDNode *N) { else if (SrcVT == MVT::i128) FF = APInt(32, F32TwoE128); else - assert(false && "Unsupported UINT_TO_FP!"); + llvm_unreachable("Unsupported UINT_TO_FP!"); // Check whether the sign bit is set. SDValue Lo, Hi; @@ -2927,38 +2944,28 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SCALAR_TO_VECTOR(SDNode *N) { SDValue DAGTypeLegalizer::PromoteIntRes_CONCAT_VECTORS(SDNode *N) { DebugLoc dl = N->getDebugLoc(); - SDValue Op0 = N->getOperand(1); - SDValue Op1 = N->getOperand(1); - assert(Op0.getValueType() == Op1.getValueType() && - "Invalid input vector types"); - EVT OutVT = N->getValueType(0); EVT NOutVT = TLI.getTypeToTransformTo(*DAG.getContext(), OutVT); assert(NOutVT.isVector() && "This type must be promoted to a vector type"); + EVT InElemTy = OutVT.getVectorElementType(); EVT OutElemTy = NOutVT.getVectorElementType(); - unsigned NumElem0 = Op0.getValueType().getVectorNumElements(); - unsigned NumElem1 = Op1.getValueType().getVectorNumElements(); + unsigned NumElem = N->getOperand(0).getValueType().getVectorNumElements(); unsigned NumOutElem = NOutVT.getVectorNumElements(); - assert(NumElem0 + NumElem1 == NumOutElem && - "Invalid number of incoming elements"); + unsigned NumOperands = N->getNumOperands(); + assert(NumElem * NumOperands == NumOutElem && + "Unexpected number of elements"); // Take the elements from the first vector. SmallVector Ops(NumOutElem); - for (unsigned i = 0; i < NumElem0; ++i) { - SDValue Ext = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, - Op0.getValueType().getScalarType(), Op0, - DAG.getIntPtrConstant(i)); - Ops[i] = DAG.getNode(ISD::ANY_EXTEND, dl, OutElemTy, Ext); - } - - // Take the elements from the second vector - for (unsigned i = 0; i < NumElem1; ++i) { - SDValue Ext = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, - Op1.getValueType().getScalarType(), Op1, - DAG.getIntPtrConstant(i)); - Ops[i + NumElem0] = DAG.getNode(ISD::ANY_EXTEND, dl, OutElemTy, Ext); + for (unsigned i = 0; i < NumOperands; ++i) { + SDValue Op = N->getOperand(i); + for (unsigned j = 0; j < NumElem; ++j) { + SDValue Ext = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, + InElemTy, Op, DAG.getIntPtrConstant(j)); + Ops[i * NumElem + j] = DAG.getNode(ISD::ANY_EXTEND, dl, OutElemTy, Ext); + } } return DAG.getNode(ISD::BUILD_VECTOR, dl, NOutVT, &Ops[0], Ops.size()); @@ -2987,8 +2994,10 @@ SDValue DAGTypeLegalizer::PromoteIntOp_EXTRACT_VECTOR_ELT(SDNode *N) { SDValue Ext = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, V0->getValueType(0).getScalarType(), V0, V1); - return DAG.getNode(ISD::TRUNCATE, dl, N->getValueType(0), Ext); - + // EXTRACT_VECTOR_ELT can return types which are wider than the incoming + // element types. If this is the case then we need to expand the outgoing + // value and not truncate it. + return DAG.getAnyExtOrTrunc(Ext, dl, N->getValueType(0)); } SDValue DAGTypeLegalizer::PromoteIntOp_CONCAT_VECTORS(SDNode *N) {