//===----------------------------------------------------------------------===//
#include "LegalizeTypes.h"
-#include "llvm/Target/TargetData.h"
+#include "llvm/IR/DataLayout.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
+#define DEBUG_TYPE "legalize-types"
+
//===----------------------------------------------------------------------===//
// Result Vector Scalarization: <1 x ty> -> ty.
//===----------------------------------------------------------------------===//
case ISD::MERGE_VALUES: R = ScalarizeVecRes_MERGE_VALUES(N, ResNo);break;
case ISD::BITCAST: R = ScalarizeVecRes_BITCAST(N); break;
- case ISD::BUILD_VECTOR: R = N->getOperand(0); break;
+ case ISD::BUILD_VECTOR: R = ScalarizeVecRes_BUILD_VECTOR(N); break;
case ISD::CONVERT_RNDSAT: R = ScalarizeVecRes_CONVERT_RNDSAT(N); break;
case ISD::EXTRACT_SUBVECTOR: R = ScalarizeVecRes_EXTRACT_SUBVECTOR(N); break;
case ISD::FP_ROUND: R = ScalarizeVecRes_FP_ROUND(N); break;
case ISD::LOAD: R = ScalarizeVecRes_LOAD(cast<LoadSDNode>(N));break;
case ISD::SCALAR_TO_VECTOR: R = ScalarizeVecRes_SCALAR_TO_VECTOR(N); break;
case ISD::SIGN_EXTEND_INREG: R = ScalarizeVecRes_InregOp(N); break;
+ case ISD::VSELECT: R = ScalarizeVecRes_VSELECT(N); break;
case ISD::SELECT: R = ScalarizeVecRes_SELECT(N); break;
case ISD::SELECT_CC: R = ScalarizeVecRes_SELECT_CC(N); break;
case ISD::SETCC: R = ScalarizeVecRes_SETCC(N); break;
case ISD::UNDEF: R = ScalarizeVecRes_UNDEF(N); break;
case ISD::VECTOR_SHUFFLE: R = ScalarizeVecRes_VECTOR_SHUFFLE(N); break;
case ISD::ANY_EXTEND:
+ case ISD::BSWAP:
case ISD::CTLZ:
+ case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTPOP:
case ISD::CTTZ:
+ case ISD::CTTZ_ZERO_UNDEF:
case ISD::FABS:
case ISD::FCEIL:
case ISD::FCOS:
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
case ISD::FRINT:
+ case ISD::FROUND:
case ISD::FSIN:
case ISD::FSQRT:
case ISD::FTRUNC:
case ISD::ADD:
case ISD::AND:
case ISD::FADD:
+ case ISD::FCOPYSIGN:
case ISD::FDIV:
case ISD::FMUL:
+ case ISD::FMINNUM:
+ case ISD::FMAXNUM:
+
case ISD::FPOW:
case ISD::FREM:
case ISD::FSUB:
case ISD::SRL:
R = ScalarizeVecRes_BinOp(N);
break;
+ case ISD::FMA:
+ R = ScalarizeVecRes_TernaryOp(N);
+ break;
}
// If R is null, the sub-method took care of registering the result.
SDValue DAGTypeLegalizer::ScalarizeVecRes_BinOp(SDNode *N) {
SDValue LHS = GetScalarizedVector(N->getOperand(0));
SDValue RHS = GetScalarizedVector(N->getOperand(1));
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(),
+ return DAG.getNode(N->getOpcode(), SDLoc(N),
LHS.getValueType(), LHS, RHS);
}
+SDValue DAGTypeLegalizer::ScalarizeVecRes_TernaryOp(SDNode *N) {
+ SDValue Op0 = GetScalarizedVector(N->getOperand(0));
+ SDValue Op1 = GetScalarizedVector(N->getOperand(1));
+ SDValue Op2 = GetScalarizedVector(N->getOperand(2));
+ return DAG.getNode(N->getOpcode(), SDLoc(N),
+ Op0.getValueType(), Op0, Op1, Op2);
+}
+
SDValue DAGTypeLegalizer::ScalarizeVecRes_MERGE_VALUES(SDNode *N,
unsigned ResNo) {
SDValue Op = DisintegrateMERGE_VALUES(N, ResNo);
SDValue DAGTypeLegalizer::ScalarizeVecRes_BITCAST(SDNode *N) {
EVT NewVT = N->getValueType(0).getVectorElementType();
- return DAG.getNode(ISD::BITCAST, N->getDebugLoc(),
+ return DAG.getNode(ISD::BITCAST, SDLoc(N),
NewVT, N->getOperand(0));
}
+SDValue DAGTypeLegalizer::ScalarizeVecRes_BUILD_VECTOR(SDNode *N) {
+ EVT EltVT = N->getValueType(0).getVectorElementType();
+ SDValue InOp = N->getOperand(0);
+ // The BUILD_VECTOR operands may be of wider element types and
+ // we may need to truncate them back to the requested return type.
+ if (EltVT.isInteger())
+ return DAG.getNode(ISD::TRUNCATE, SDLoc(N), EltVT, InOp);
+ return InOp;
+}
+
SDValue DAGTypeLegalizer::ScalarizeVecRes_CONVERT_RNDSAT(SDNode *N) {
EVT NewVT = N->getValueType(0).getVectorElementType();
SDValue Op0 = GetScalarizedVector(N->getOperand(0));
- return DAG.getConvertRndSat(NewVT, N->getDebugLoc(),
+ return DAG.getConvertRndSat(NewVT, SDLoc(N),
Op0, DAG.getValueType(NewVT),
DAG.getValueType(Op0.getValueType()),
N->getOperand(3),
}
SDValue DAGTypeLegalizer::ScalarizeVecRes_EXTRACT_SUBVECTOR(SDNode *N) {
- return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, N->getDebugLoc(),
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N),
N->getValueType(0).getVectorElementType(),
N->getOperand(0), N->getOperand(1));
}
SDValue DAGTypeLegalizer::ScalarizeVecRes_FP_ROUND(SDNode *N) {
EVT NewVT = N->getValueType(0).getVectorElementType();
SDValue Op = GetScalarizedVector(N->getOperand(0));
- return DAG.getNode(ISD::FP_ROUND, N->getDebugLoc(),
+ return DAG.getNode(ISD::FP_ROUND, SDLoc(N),
NewVT, Op, N->getOperand(1));
}
SDValue DAGTypeLegalizer::ScalarizeVecRes_FPOWI(SDNode *N) {
SDValue Op = GetScalarizedVector(N->getOperand(0));
- return DAG.getNode(ISD::FPOWI, N->getDebugLoc(),
+ return DAG.getNode(ISD::FPOWI, SDLoc(N),
Op.getValueType(), Op, N->getOperand(1));
}
EVT EltVT = N->getValueType(0).getVectorElementType();
if (Op.getValueType() != EltVT)
// FIXME: Can this happen for floating point types?
- Op = DAG.getNode(ISD::TRUNCATE, N->getDebugLoc(), EltVT, Op);
+ Op = DAG.getNode(ISD::TRUNCATE, SDLoc(N), EltVT, Op);
return Op;
}
SDValue Result = DAG.getLoad(ISD::UNINDEXED,
N->getExtensionType(),
N->getValueType(0).getVectorElementType(),
- N->getDebugLoc(),
+ SDLoc(N),
N->getChain(), N->getBasePtr(),
DAG.getUNDEF(N->getBasePtr().getValueType()),
N->getPointerInfo(),
N->getMemoryVT().getVectorElementType(),
N->isVolatile(), N->isNonTemporal(),
- N->isInvariant(), N->getOriginalAlignment());
+ N->isInvariant(), N->getOriginalAlignment(),
+ N->getAAInfo());
// Legalized the chain result - switch anything that used the old chain to
// use the new one.
SDValue DAGTypeLegalizer::ScalarizeVecRes_UnaryOp(SDNode *N) {
// Get the dest type - it doesn't always match the input type, e.g. int_to_fp.
EVT DestVT = N->getValueType(0).getVectorElementType();
- SDValue Op = GetScalarizedVector(N->getOperand(0));
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(), DestVT, Op);
+ SDValue Op = N->getOperand(0);
+ EVT OpVT = Op.getValueType();
+ SDLoc DL(N);
+ // The result needs scalarizing, but it's not a given that the source does.
+ // This is a workaround for targets where it's impossible to scalarize the
+ // result of a conversion, because the source type is legal.
+ // For instance, this happens on AArch64: v1i1 is illegal but v1i{8,16,32}
+ // are widened to v8i8, v4i16, and v2i32, which is legal, because v1i64 is
+ // legal and was not scalarized.
+ // See the similar logic in ScalarizeVecRes_VSETCC
+ if (getTypeAction(OpVT) == TargetLowering::TypeScalarizeVector) {
+ Op = GetScalarizedVector(Op);
+ } else {
+ EVT VT = OpVT.getVectorElementType();
+ Op = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, Op,
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
+ }
+ return DAG.getNode(N->getOpcode(), SDLoc(N), DestVT, Op);
}
SDValue DAGTypeLegalizer::ScalarizeVecRes_InregOp(SDNode *N) {
EVT EltVT = N->getValueType(0).getVectorElementType();
EVT ExtVT = cast<VTSDNode>(N->getOperand(1))->getVT().getVectorElementType();
SDValue LHS = GetScalarizedVector(N->getOperand(0));
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(), EltVT,
+ return DAG.getNode(N->getOpcode(), SDLoc(N), EltVT,
LHS, DAG.getValueType(ExtVT));
}
EVT EltVT = N->getValueType(0).getVectorElementType();
SDValue InOp = N->getOperand(0);
if (InOp.getValueType() != EltVT)
- return DAG.getNode(ISD::TRUNCATE, N->getDebugLoc(), EltVT, InOp);
+ return DAG.getNode(ISD::TRUNCATE, SDLoc(N), EltVT, InOp);
return InOp;
}
+SDValue DAGTypeLegalizer::ScalarizeVecRes_VSELECT(SDNode *N) {
+ SDValue Cond = GetScalarizedVector(N->getOperand(0));
+ SDValue LHS = GetScalarizedVector(N->getOperand(1));
+ TargetLowering::BooleanContent ScalarBool =
+ TLI.getBooleanContents(false, false);
+ TargetLowering::BooleanContent VecBool = TLI.getBooleanContents(true, false);
+
+ // If integer and float booleans have different contents then we can't
+ // reliably optimize in all cases. There is a full explanation for this in
+ // DAGCombiner::visitSELECT() where the same issue affects folding
+ // (select C, 0, 1) to (xor C, 1).
+ if (TLI.getBooleanContents(false, false) !=
+ TLI.getBooleanContents(false, true)) {
+ // At least try the common case where the boolean is generated by a
+ // comparison.
+ if (Cond->getOpcode() == ISD::SETCC) {
+ EVT OpVT = Cond->getOperand(0)->getValueType(0);
+ ScalarBool = TLI.getBooleanContents(OpVT.getScalarType());
+ VecBool = TLI.getBooleanContents(OpVT);
+ } else
+ ScalarBool = TargetLowering::UndefinedBooleanContent;
+ }
+
+ if (ScalarBool != VecBool) {
+ EVT CondVT = Cond.getValueType();
+ switch (ScalarBool) {
+ case TargetLowering::UndefinedBooleanContent:
+ break;
+ case TargetLowering::ZeroOrOneBooleanContent:
+ assert(VecBool == TargetLowering::UndefinedBooleanContent ||
+ VecBool == TargetLowering::ZeroOrNegativeOneBooleanContent);
+ // Vector read from all ones, scalar expects a single 1 so mask.
+ Cond = DAG.getNode(ISD::AND, SDLoc(N), CondVT,
+ Cond, DAG.getConstant(1, CondVT));
+ break;
+ case TargetLowering::ZeroOrNegativeOneBooleanContent:
+ assert(VecBool == TargetLowering::UndefinedBooleanContent ||
+ VecBool == TargetLowering::ZeroOrOneBooleanContent);
+ // Vector reads from a one, scalar from all ones so sign extend.
+ Cond = DAG.getNode(ISD::SIGN_EXTEND_INREG, SDLoc(N), CondVT,
+ Cond, DAG.getValueType(MVT::i1));
+ break;
+ }
+ }
+
+ return DAG.getSelect(SDLoc(N),
+ LHS.getValueType(), Cond, LHS,
+ GetScalarizedVector(N->getOperand(2)));
+}
+
SDValue DAGTypeLegalizer::ScalarizeVecRes_SELECT(SDNode *N) {
SDValue LHS = GetScalarizedVector(N->getOperand(1));
- return DAG.getNode(ISD::SELECT, N->getDebugLoc(),
- LHS.getValueType(), N->getOperand(0), LHS,
- GetScalarizedVector(N->getOperand(2)));
+ return DAG.getSelect(SDLoc(N),
+ LHS.getValueType(), N->getOperand(0), LHS,
+ GetScalarizedVector(N->getOperand(2)));
}
SDValue DAGTypeLegalizer::ScalarizeVecRes_SELECT_CC(SDNode *N) {
SDValue LHS = GetScalarizedVector(N->getOperand(2));
- return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(), LHS.getValueType(),
+ return DAG.getNode(ISD::SELECT_CC, SDLoc(N), LHS.getValueType(),
N->getOperand(0), N->getOperand(1),
LHS, GetScalarizedVector(N->getOperand(3)),
N->getOperand(4));
SDValue LHS = GetScalarizedVector(N->getOperand(0));
SDValue RHS = GetScalarizedVector(N->getOperand(1));
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
// Turn it into a scalar SETCC.
return DAG.getNode(ISD::SETCC, DL, MVT::i1, LHS, RHS, N->getOperand(2));
assert(N->getValueType(0).isVector() &&
N->getOperand(0).getValueType().isVector() &&
"Operand types must be vectors");
-
- SDValue LHS = GetScalarizedVector(N->getOperand(0));
- SDValue RHS = GetScalarizedVector(N->getOperand(1));
+ SDValue LHS = N->getOperand(0);
+ SDValue RHS = N->getOperand(1);
+ EVT OpVT = LHS.getValueType();
EVT NVT = N->getValueType(0).getVectorElementType();
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
+
+ // The result needs scalarizing, but it's not a given that the source does.
+ if (getTypeAction(OpVT) == TargetLowering::TypeScalarizeVector) {
+ LHS = GetScalarizedVector(LHS);
+ RHS = GetScalarizedVector(RHS);
+ } else {
+ EVT VT = OpVT.getVectorElementType();
+ LHS = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, LHS,
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
+ RHS = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, RHS,
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
+ }
// Turn it into a scalar SETCC.
SDValue Res = DAG.getNode(ISD::SETCC, DL, MVT::i1, LHS, RHS,
// Vectors may have a different boolean contents to scalars. Promote the
// value appropriately.
ISD::NodeType ExtendCode =
- TargetLowering::getExtendForContent(TLI.getBooleanContents(true));
+ TargetLowering::getExtendForContent(TLI.getBooleanContents(OpVT));
return DAG.getNode(ExtendCode, DL, NVT, Res);
}
dbgs() << "\n");
SDValue Res = SDValue();
- if (Res.getNode() == 0) {
+ if (!Res.getNode()) {
switch (N->getOpcode()) {
default:
#ifndef NDEBUG
case ISD::BITCAST:
Res = ScalarizeVecOp_BITCAST(N);
break;
+ case ISD::ANY_EXTEND:
+ case ISD::ZERO_EXTEND:
+ case ISD::SIGN_EXTEND:
+ case ISD::TRUNCATE:
+ case ISD::FP_TO_SINT:
+ case ISD::FP_TO_UINT:
+ case ISD::SINT_TO_FP:
+ case ISD::UINT_TO_FP:
+ Res = ScalarizeVecOp_UnaryOp(N);
+ break;
case ISD::CONCAT_VECTORS:
Res = ScalarizeVecOp_CONCAT_VECTORS(N);
break;
case ISD::EXTRACT_VECTOR_ELT:
Res = ScalarizeVecOp_EXTRACT_VECTOR_ELT(N);
break;
+ case ISD::VSELECT:
+ Res = ScalarizeVecOp_VSELECT(N);
+ break;
case ISD::STORE:
Res = ScalarizeVecOp_STORE(cast<StoreSDNode>(N), OpNo);
break;
+ case ISD::FP_ROUND:
+ Res = ScalarizeVecOp_FP_ROUND(N, OpNo);
+ break;
}
}
/// to be scalarized, it must be <1 x ty>. Convert the element instead.
SDValue DAGTypeLegalizer::ScalarizeVecOp_BITCAST(SDNode *N) {
SDValue Elt = GetScalarizedVector(N->getOperand(0));
- return DAG.getNode(ISD::BITCAST, N->getDebugLoc(),
+ return DAG.getNode(ISD::BITCAST, SDLoc(N),
N->getValueType(0), Elt);
}
+/// ScalarizeVecOp_UnaryOp - If the input is a vector that needs to be
+/// scalarized, it must be <1 x ty>. Do the operation on the element instead.
+SDValue DAGTypeLegalizer::ScalarizeVecOp_UnaryOp(SDNode *N) {
+ assert(N->getValueType(0).getVectorNumElements() == 1 &&
+ "Unexpected vector type!");
+ SDValue Elt = GetScalarizedVector(N->getOperand(0));
+ SDValue Op = DAG.getNode(N->getOpcode(), SDLoc(N),
+ N->getValueType(0).getScalarType(), Elt);
+ // Revectorize the result so the types line up with what the uses of this
+ // expression expect.
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), N->getValueType(0), Op);
+}
+
/// ScalarizeVecOp_CONCAT_VECTORS - The vectors to concatenate have length one -
/// use a BUILD_VECTOR instead.
SDValue DAGTypeLegalizer::ScalarizeVecOp_CONCAT_VECTORS(SDNode *N) {
SmallVector<SDValue, 8> Ops(N->getNumOperands());
for (unsigned i = 0, e = N->getNumOperands(); i < e; ++i)
Ops[i] = GetScalarizedVector(N->getOperand(i));
- return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(), N->getValueType(0),
- &Ops[0], Ops.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), N->getValueType(0), Ops);
}
/// ScalarizeVecOp_EXTRACT_VECTOR_ELT - If the input is a vector that needs to
SDValue DAGTypeLegalizer::ScalarizeVecOp_EXTRACT_VECTOR_ELT(SDNode *N) {
SDValue Res = GetScalarizedVector(N->getOperand(0));
if (Res.getValueType() != N->getValueType(0))
- Res = DAG.getNode(ISD::ANY_EXTEND, N->getDebugLoc(), N->getValueType(0),
+ Res = DAG.getNode(ISD::ANY_EXTEND, SDLoc(N), N->getValueType(0),
Res);
return Res;
}
+
+/// ScalarizeVecOp_VSELECT - If the input condition is a vector that needs to be
+/// scalarized, it must be <1 x i1>, so just convert to a normal ISD::SELECT
+/// (still with vector output type since that was acceptable if we got here).
+SDValue DAGTypeLegalizer::ScalarizeVecOp_VSELECT(SDNode *N) {
+ SDValue ScalarCond = GetScalarizedVector(N->getOperand(0));
+ EVT VT = N->getValueType(0);
+
+ return DAG.getNode(ISD::SELECT, SDLoc(N), VT, ScalarCond, N->getOperand(1),
+ N->getOperand(2));
+}
+
/// ScalarizeVecOp_STORE - If the value to store is a vector that needs to be
/// scalarized, it must be <1 x ty>. Just store the element.
SDValue DAGTypeLegalizer::ScalarizeVecOp_STORE(StoreSDNode *N, unsigned OpNo){
assert(N->isUnindexed() && "Indexed store of one-element vector?");
assert(OpNo == 1 && "Do not know how to scalarize this operand!");
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
if (N->isTruncatingStore())
return DAG.getTruncStore(N->getChain(), dl,
N->getBasePtr(), N->getPointerInfo(),
N->getMemoryVT().getVectorElementType(),
N->isVolatile(), N->isNonTemporal(),
- N->getAlignment());
+ N->getAlignment(), N->getAAInfo());
return DAG.getStore(N->getChain(), dl, GetScalarizedVector(N->getOperand(1)),
N->getBasePtr(), N->getPointerInfo(),
N->isVolatile(), N->isNonTemporal(),
- N->getOriginalAlignment());
+ N->getOriginalAlignment(), N->getAAInfo());
}
+/// ScalarizeVecOp_FP_ROUND - If the value to round is a vector that needs
+/// to be scalarized, it must be <1 x ty>. Convert the element instead.
+SDValue DAGTypeLegalizer::ScalarizeVecOp_FP_ROUND(SDNode *N, unsigned OpNo) {
+ SDValue Elt = GetScalarizedVector(N->getOperand(0));
+ SDValue Res = DAG.getNode(ISD::FP_ROUND, SDLoc(N),
+ N->getValueType(0).getVectorElementType(), Elt,
+ N->getOperand(1));
+ return DAG.getNode(ISD::SCALAR_TO_VECTOR, SDLoc(N), N->getValueType(0), Res);
+}
//===----------------------------------------------------------------------===//
// Result Vector Splitting
dbgs() << "\n");
SDValue Lo, Hi;
+ // See if the target wants to custom expand this node.
+ if (CustomLowerNode(N, N->getValueType(ResNo), true))
+ return;
+
switch (N->getOpcode()) {
default:
#ifndef NDEBUG
N->dump(&DAG);
dbgs() << "\n";
#endif
- llvm_unreachable("Do not know how to split the result of this operator!");
+ report_fatal_error("Do not know how to split the result of this "
+ "operator!\n");
case ISD::MERGE_VALUES: SplitRes_MERGE_VALUES(N, ResNo, Lo, Hi); break;
case ISD::VSELECT:
case ISD::BUILD_VECTOR: SplitVecRes_BUILD_VECTOR(N, Lo, Hi); break;
case ISD::CONCAT_VECTORS: SplitVecRes_CONCAT_VECTORS(N, Lo, Hi); break;
case ISD::EXTRACT_SUBVECTOR: SplitVecRes_EXTRACT_SUBVECTOR(N, Lo, Hi); break;
+ case ISD::INSERT_SUBVECTOR: SplitVecRes_INSERT_SUBVECTOR(N, Lo, Hi); break;
case ISD::FP_ROUND_INREG: SplitVecRes_InregOp(N, Lo, Hi); break;
case ISD::FPOWI: SplitVecRes_FPOWI(N, Lo, Hi); break;
case ISD::INSERT_VECTOR_ELT: SplitVecRes_INSERT_VECTOR_ELT(N, Lo, Hi); break;
case ISD::LOAD:
SplitVecRes_LOAD(cast<LoadSDNode>(N), Lo, Hi);
break;
+ case ISD::MLOAD:
+ SplitVecRes_MLOAD(cast<MaskedLoadSDNode>(N), Lo, Hi);
+ break;
case ISD::SETCC:
SplitVecRes_SETCC(N, Lo, Hi);
break;
SplitVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N), Lo, Hi);
break;
- case ISD::ANY_EXTEND:
+ case ISD::BSWAP:
case ISD::CONVERT_RNDSAT:
case ISD::CTLZ:
- case ISD::CTPOP:
case ISD::CTTZ:
+ case ISD::CTLZ_ZERO_UNDEF:
+ case ISD::CTTZ_ZERO_UNDEF:
+ case ISD::CTPOP:
case ISD::FABS:
case ISD::FCEIL:
case ISD::FCOS:
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
case ISD::FRINT:
+ case ISD::FROUND:
case ISD::FSIN:
case ISD::FSQRT:
case ISD::FTRUNC:
- case ISD::SIGN_EXTEND:
case ISD::SINT_TO_FP:
case ISD::TRUNCATE:
case ISD::UINT_TO_FP:
- case ISD::ZERO_EXTEND:
SplitVecRes_UnaryOp(N, Lo, Hi);
break;
+ case ISD::ANY_EXTEND:
+ case ISD::SIGN_EXTEND:
+ case ISD::ZERO_EXTEND:
+ SplitVecRes_ExtendOp(N, Lo, Hi);
+ break;
+
case ISD::ADD:
case ISD::SUB:
case ISD::MUL:
case ISD::FADD:
+ case ISD::FCOPYSIGN:
case ISD::FSUB:
case ISD::FMUL:
+ case ISD::FMINNUM:
+ case ISD::FMAXNUM:
case ISD::SDIV:
case ISD::UDIV:
case ISD::FDIV:
case ISD::FREM:
SplitVecRes_BinOp(N, Lo, Hi);
break;
+ case ISD::FMA:
+ SplitVecRes_TernaryOp(N, Lo, Hi);
+ break;
}
// If Lo/Hi is null, the sub-method took care of registering results etc.
GetSplitVector(N->getOperand(0), LHSLo, LHSHi);
SDValue RHSLo, RHSHi;
GetSplitVector(N->getOperand(1), RHSLo, RHSHi);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
Lo = DAG.getNode(N->getOpcode(), dl, LHSLo.getValueType(), LHSLo, RHSLo);
Hi = DAG.getNode(N->getOpcode(), dl, LHSHi.getValueType(), LHSHi, RHSHi);
}
+void DAGTypeLegalizer::SplitVecRes_TernaryOp(SDNode *N, SDValue &Lo,
+ SDValue &Hi) {
+ SDValue Op0Lo, Op0Hi;
+ GetSplitVector(N->getOperand(0), Op0Lo, Op0Hi);
+ SDValue Op1Lo, Op1Hi;
+ GetSplitVector(N->getOperand(1), Op1Lo, Op1Hi);
+ SDValue Op2Lo, Op2Hi;
+ GetSplitVector(N->getOperand(2), Op2Lo, Op2Hi);
+ SDLoc dl(N);
+
+ Lo = DAG.getNode(N->getOpcode(), dl, Op0Lo.getValueType(),
+ Op0Lo, Op1Lo, Op2Lo);
+ Hi = DAG.getNode(N->getOpcode(), dl, Op0Hi.getValueType(),
+ Op0Hi, Op1Hi, Op2Hi);
+}
+
void DAGTypeLegalizer::SplitVecRes_BITCAST(SDNode *N, SDValue &Lo,
SDValue &Hi) {
// We know the result is a vector. The input may be either a vector or a
// scalar value.
EVT LoVT, HiVT;
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
- DebugLoc dl = N->getDebugLoc();
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
+ SDLoc dl(N);
SDValue InOp = N->getOperand(0);
EVT InVT = InOp.getValueType();
void DAGTypeLegalizer::SplitVecRes_BUILD_VECTOR(SDNode *N, SDValue &Lo,
SDValue &Hi) {
EVT LoVT, HiVT;
- DebugLoc dl = N->getDebugLoc();
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ SDLoc dl(N);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
unsigned LoNumElts = LoVT.getVectorNumElements();
SmallVector<SDValue, 8> LoOps(N->op_begin(), N->op_begin()+LoNumElts);
- Lo = DAG.getNode(ISD::BUILD_VECTOR, dl, LoVT, &LoOps[0], LoOps.size());
+ Lo = DAG.getNode(ISD::BUILD_VECTOR, dl, LoVT, LoOps);
SmallVector<SDValue, 8> HiOps(N->op_begin()+LoNumElts, N->op_end());
- Hi = DAG.getNode(ISD::BUILD_VECTOR, dl, HiVT, &HiOps[0], HiOps.size());
+ Hi = DAG.getNode(ISD::BUILD_VECTOR, dl, HiVT, HiOps);
}
void DAGTypeLegalizer::SplitVecRes_CONCAT_VECTORS(SDNode *N, SDValue &Lo,
SDValue &Hi) {
assert(!(N->getNumOperands() & 1) && "Unsupported CONCAT_VECTORS");
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
unsigned NumSubvectors = N->getNumOperands() / 2;
if (NumSubvectors == 1) {
Lo = N->getOperand(0);
}
EVT LoVT, HiVT;
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
SmallVector<SDValue, 8> LoOps(N->op_begin(), N->op_begin()+NumSubvectors);
- Lo = DAG.getNode(ISD::CONCAT_VECTORS, dl, LoVT, &LoOps[0], LoOps.size());
+ Lo = DAG.getNode(ISD::CONCAT_VECTORS, dl, LoVT, LoOps);
SmallVector<SDValue, 8> HiOps(N->op_begin()+NumSubvectors, N->op_end());
- Hi = DAG.getNode(ISD::CONCAT_VECTORS, dl, HiVT, &HiOps[0], HiOps.size());
+ Hi = DAG.getNode(ISD::CONCAT_VECTORS, dl, HiVT, HiOps);
}
void DAGTypeLegalizer::SplitVecRes_EXTRACT_SUBVECTOR(SDNode *N, SDValue &Lo,
SDValue &Hi) {
SDValue Vec = N->getOperand(0);
SDValue Idx = N->getOperand(1);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
EVT LoVT, HiVT;
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, LoVT, Vec, Idx);
uint64_t IdxVal = cast<ConstantSDNode>(Idx)->getZExtValue();
Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, HiVT, Vec,
- DAG.getIntPtrConstant(IdxVal + LoVT.getVectorNumElements()));
+ DAG.getConstant(IdxVal + LoVT.getVectorNumElements(),
+ TLI.getVectorIdxTy()));
+}
+
+void DAGTypeLegalizer::SplitVecRes_INSERT_SUBVECTOR(SDNode *N, SDValue &Lo,
+ SDValue &Hi) {
+ SDValue Vec = N->getOperand(0);
+ SDValue SubVec = N->getOperand(1);
+ SDValue Idx = N->getOperand(2);
+ SDLoc dl(N);
+ GetSplitVector(Vec, Lo, Hi);
+
+ // Spill the vector to the stack.
+ EVT VecVT = Vec.getValueType();
+ EVT SubVecVT = VecVT.getVectorElementType();
+ SDValue StackPtr = DAG.CreateStackTemporary(VecVT);
+ SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr,
+ MachinePointerInfo(), false, false, 0);
+
+ // Store the new subvector into the specified index.
+ SDValue SubVecPtr = GetVectorElementPointer(StackPtr, SubVecVT, Idx);
+ Type *VecType = VecVT.getTypeForEVT(*DAG.getContext());
+ unsigned Alignment = TLI.getDataLayout()->getPrefTypeAlignment(VecType);
+ Store = DAG.getStore(Store, dl, SubVec, SubVecPtr, MachinePointerInfo(),
+ false, false, 0);
+
+ // Load the Lo part from the stack slot.
+ Lo = DAG.getLoad(Lo.getValueType(), dl, Store, StackPtr, MachinePointerInfo(),
+ false, false, false, 0);
+
+ // Increment the pointer to the other part.
+ unsigned IncrementSize = Lo.getValueType().getSizeInBits() / 8;
+ StackPtr =
+ DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
+ DAG.getConstant(IncrementSize, StackPtr.getValueType()));
+
+ // Load the Hi part from the stack slot.
+ Hi = DAG.getLoad(Hi.getValueType(), dl, Store, StackPtr, MachinePointerInfo(),
+ false, false, false, MinAlign(Alignment, IncrementSize));
}
void DAGTypeLegalizer::SplitVecRes_FPOWI(SDNode *N, SDValue &Lo,
SDValue &Hi) {
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
GetSplitVector(N->getOperand(0), Lo, Hi);
Lo = DAG.getNode(ISD::FPOWI, dl, Lo.getValueType(), Lo, N->getOperand(1));
Hi = DAG.getNode(ISD::FPOWI, dl, Hi.getValueType(), Hi, N->getOperand(1));
SDValue &Hi) {
SDValue LHSLo, LHSHi;
GetSplitVector(N->getOperand(0), LHSLo, LHSHi);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
EVT LoVT, HiVT;
- GetSplitDestVTs(cast<VTSDNode>(N->getOperand(1))->getVT(), LoVT, HiVT);
+ std::tie(LoVT, HiVT) =
+ DAG.GetSplitDestVTs(cast<VTSDNode>(N->getOperand(1))->getVT());
Lo = DAG.getNode(N->getOpcode(), dl, LHSLo.getValueType(), LHSLo,
DAG.getValueType(LoVT));
SDValue Vec = N->getOperand(0);
SDValue Elt = N->getOperand(1);
SDValue Idx = N->getOperand(2);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
GetSplitVector(Vec, Lo, Hi);
if (ConstantSDNode *CIdx = dyn_cast<ConstantSDNode>(Idx)) {
Lo.getValueType(), Lo, Elt, Idx);
else
Hi = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, Hi.getValueType(), Hi, Elt,
- DAG.getIntPtrConstant(IdxVal - LoNumElts));
+ DAG.getConstant(IdxVal - LoNumElts,
+ TLI.getVectorIdxTy()));
return;
}
+ // See if the target wants to custom expand this node.
+ if (CustomLowerNode(N, N->getValueType(0), true))
+ return;
+
// Spill the vector to the stack.
EVT VecVT = Vec.getValueType();
EVT EltVT = VecVT.getVectorElementType();
SDValue EltPtr = GetVectorElementPointer(StackPtr, EltVT, Idx);
Type *VecType = VecVT.getTypeForEVT(*DAG.getContext());
unsigned Alignment =
- TLI.getTargetData()->getPrefTypeAlignment(VecType);
+ TLI.getDataLayout()->getPrefTypeAlignment(VecType);
Store = DAG.getTruncStore(Store, dl, Elt, EltPtr, MachinePointerInfo(), EltVT,
false, false, 0);
// Increment the pointer to the other part.
unsigned IncrementSize = Lo.getValueType().getSizeInBits() / 8;
StackPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
- DAG.getIntPtrConstant(IncrementSize));
+ DAG.getConstant(IncrementSize, StackPtr.getValueType()));
// Load the Hi part from the stack slot.
Hi = DAG.getLoad(Hi.getValueType(), dl, Store, StackPtr, MachinePointerInfo(),
void DAGTypeLegalizer::SplitVecRes_SCALAR_TO_VECTOR(SDNode *N, SDValue &Lo,
SDValue &Hi) {
EVT LoVT, HiVT;
- DebugLoc dl = N->getDebugLoc();
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ SDLoc dl(N);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
Lo = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, LoVT, N->getOperand(0));
Hi = DAG.getUNDEF(HiVT);
}
SDValue &Hi) {
assert(ISD::isUNINDEXEDLoad(LD) && "Indexed load during type legalization!");
EVT LoVT, HiVT;
- DebugLoc dl = LD->getDebugLoc();
- GetSplitDestVTs(LD->getValueType(0), LoVT, HiVT);
+ SDLoc dl(LD);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(LD->getValueType(0));
ISD::LoadExtType ExtType = LD->getExtensionType();
SDValue Ch = LD->getChain();
bool isVolatile = LD->isVolatile();
bool isNonTemporal = LD->isNonTemporal();
bool isInvariant = LD->isInvariant();
+ AAMDNodes AAInfo = LD->getAAInfo();
EVT LoMemVT, HiMemVT;
- GetSplitDestVTs(MemoryVT, LoMemVT, HiMemVT);
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
Lo = DAG.getLoad(ISD::UNINDEXED, ExtType, LoVT, dl, Ch, Ptr, Offset,
LD->getPointerInfo(), LoMemVT, isVolatile, isNonTemporal,
- isInvariant, Alignment);
+ isInvariant, Alignment, AAInfo);
unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
- DAG.getIntPtrConstant(IncrementSize));
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
Hi = DAG.getLoad(ISD::UNINDEXED, ExtType, HiVT, dl, Ch, Ptr, Offset,
LD->getPointerInfo().getWithOffset(IncrementSize),
- HiMemVT, isVolatile, isNonTemporal, isInvariant, Alignment);
+ HiMemVT, isVolatile, isNonTemporal, isInvariant, Alignment,
+ AAInfo);
// Build a factor node to remember that this load is independent of the
// other one.
ReplaceValueWith(SDValue(LD, 1), Ch);
}
+void DAGTypeLegalizer::SplitVecRes_MLOAD(MaskedLoadSDNode *MLD,
+ SDValue &Lo, SDValue &Hi) {
+ EVT LoVT, HiVT;
+ SDLoc dl(MLD);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(MLD->getValueType(0));
+
+ SDValue Ch = MLD->getChain();
+ SDValue Ptr = MLD->getBasePtr();
+ SDValue Mask = MLD->getMask();
+ unsigned Alignment = MLD->getOriginalAlignment();
+ ISD::LoadExtType ExtType = MLD->getExtensionType();
+
+ // if Alignment is equal to the vector size,
+ // take the half of it for the second part
+ unsigned SecondHalfAlignment =
+ (Alignment == MLD->getValueType(0).getSizeInBits()/8) ?
+ Alignment/2 : Alignment;
+
+ SDValue MaskLo, MaskHi;
+ std::tie(MaskLo, MaskHi) = DAG.SplitVector(Mask, dl);
+
+ EVT MemoryVT = MLD->getMemoryVT();
+ EVT LoMemVT, HiMemVT;
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
+
+ SDValue Src0 = MLD->getSrc0();
+ SDValue Src0Lo, Src0Hi;
+ std::tie(Src0Lo, Src0Hi) = DAG.SplitVector(Src0, dl);
+
+ MachineMemOperand *MMO = DAG.getMachineFunction().
+ getMachineMemOperand(MLD->getPointerInfo(),
+ MachineMemOperand::MOLoad, LoMemVT.getStoreSize(),
+ Alignment, MLD->getAAInfo(), MLD->getRanges());
+
+ Lo = DAG.getMaskedLoad(LoVT, dl, Ch, Ptr, MaskLo, Src0Lo, LoMemVT, MMO,
+ ExtType);
+
+ unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
+ Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr,
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
+
+ MMO = DAG.getMachineFunction().
+ getMachineMemOperand(MLD->getPointerInfo(),
+ MachineMemOperand::MOLoad, HiMemVT.getStoreSize(),
+ SecondHalfAlignment, MLD->getAAInfo(), MLD->getRanges());
+
+ Hi = DAG.getMaskedLoad(HiVT, dl, Ch, Ptr, MaskHi, Src0Hi, HiMemVT, MMO,
+ ExtType);
+
+
+ // Build a factor node to remember that this load is independent of the
+ // other one.
+ Ch = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Lo.getValue(1),
+ Hi.getValue(1));
+
+ // Legalized the chain result - switch anything that used the old chain to
+ // use the new one.
+ ReplaceValueWith(SDValue(MLD, 1), Ch);
+
+}
+
void DAGTypeLegalizer::SplitVecRes_SETCC(SDNode *N, SDValue &Lo, SDValue &Hi) {
assert(N->getValueType(0).isVector() &&
N->getOperand(0).getValueType().isVector() &&
"Operand types must be vectors");
EVT LoVT, HiVT;
- DebugLoc DL = N->getDebugLoc();
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ SDLoc DL(N);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
// Split the input.
- EVT InVT = N->getOperand(0).getValueType();
SDValue LL, LH, RL, RH;
- EVT InNVT = EVT::getVectorVT(*DAG.getContext(), InVT.getVectorElementType(),
- LoVT.getVectorNumElements());
- LL = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, N->getOperand(0),
- DAG.getIntPtrConstant(0));
- LH = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, N->getOperand(0),
- DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
-
- RL = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, N->getOperand(1),
- DAG.getIntPtrConstant(0));
- RH = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, N->getOperand(1),
- DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
+ std::tie(LL, LH) = DAG.SplitVectorOperand(N, 0);
+ std::tie(RL, RH) = DAG.SplitVectorOperand(N, 1);
Lo = DAG.getNode(N->getOpcode(), DL, LoVT, LL, RL, N->getOperand(2));
Hi = DAG.getNode(N->getOpcode(), DL, HiVT, LH, RH, N->getOperand(2));
SDValue &Hi) {
// Get the dest types - they may not match the input types, e.g. int_to_fp.
EVT LoVT, HiVT;
- DebugLoc dl = N->getDebugLoc();
- GetSplitDestVTs(N->getValueType(0), LoVT, HiVT);
+ SDLoc dl(N);
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
// If the input also splits, handle it directly for a compile time speedup.
// Otherwise split it by hand.
EVT InVT = N->getOperand(0).getValueType();
- if (getTypeAction(InVT) == TargetLowering::TypeSplitVector) {
+ if (getTypeAction(InVT) == TargetLowering::TypeSplitVector)
GetSplitVector(N->getOperand(0), Lo, Hi);
- } else {
- EVT InNVT = EVT::getVectorVT(*DAG.getContext(), InVT.getVectorElementType(),
- LoVT.getVectorNumElements());
- Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, InNVT, N->getOperand(0),
- DAG.getIntPtrConstant(0));
- Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, InNVT, N->getOperand(0),
- DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
- }
+ else
+ std::tie(Lo, Hi) = DAG.SplitVectorOperand(N, 0);
if (N->getOpcode() == ISD::FP_ROUND) {
Lo = DAG.getNode(N->getOpcode(), dl, LoVT, Lo, N->getOperand(1));
}
}
+void DAGTypeLegalizer::SplitVecRes_ExtendOp(SDNode *N, SDValue &Lo,
+ SDValue &Hi) {
+ SDLoc dl(N);
+ EVT SrcVT = N->getOperand(0).getValueType();
+ EVT DestVT = N->getValueType(0);
+ EVT LoVT, HiVT;
+ std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(DestVT);
+
+ // We can do better than a generic split operation if the extend is doing
+ // more than just doubling the width of the elements and the following are
+ // true:
+ // - The number of vector elements is even,
+ // - the source type is legal,
+ // - the type of a split source is illegal,
+ // - the type of an extended (by doubling element size) source is legal, and
+ // - the type of that extended source when split is legal.
+ //
+ // This won't necessarily completely legalize the operation, but it will
+ // more effectively move in the right direction and prevent falling down
+ // to scalarization in many cases due to the input vector being split too
+ // far.
+ unsigned NumElements = SrcVT.getVectorNumElements();
+ if ((NumElements & 1) == 0 &&
+ SrcVT.getSizeInBits() * 2 < DestVT.getSizeInBits()) {
+ LLVMContext &Ctx = *DAG.getContext();
+ EVT NewSrcVT = EVT::getVectorVT(
+ Ctx, EVT::getIntegerVT(
+ Ctx, SrcVT.getVectorElementType().getSizeInBits() * 2),
+ NumElements);
+ EVT SplitSrcVT =
+ EVT::getVectorVT(Ctx, SrcVT.getVectorElementType(), NumElements / 2);
+ EVT SplitLoVT, SplitHiVT;
+ std::tie(SplitLoVT, SplitHiVT) = DAG.GetSplitDestVTs(NewSrcVT);
+ if (TLI.isTypeLegal(SrcVT) && !TLI.isTypeLegal(SplitSrcVT) &&
+ TLI.isTypeLegal(NewSrcVT) && TLI.isTypeLegal(SplitLoVT)) {
+ DEBUG(dbgs() << "Split vector extend via incremental extend:";
+ N->dump(&DAG); dbgs() << "\n");
+ // Extend the source vector by one step.
+ SDValue NewSrc =
+ DAG.getNode(N->getOpcode(), dl, NewSrcVT, N->getOperand(0));
+ // Get the low and high halves of the new, extended one step, vector.
+ std::tie(Lo, Hi) = DAG.SplitVector(NewSrc, dl);
+ // Extend those vector halves the rest of the way.
+ Lo = DAG.getNode(N->getOpcode(), dl, LoVT, Lo);
+ Hi = DAG.getNode(N->getOpcode(), dl, HiVT, Hi);
+ return;
+ }
+ }
+ // Fall back to the generic unary operator splitting otherwise.
+ SplitVecRes_UnaryOp(N, Lo, Hi);
+}
+
void DAGTypeLegalizer::SplitVecRes_VECTOR_SHUFFLE(ShuffleVectorSDNode *N,
SDValue &Lo, SDValue &Hi) {
// The low and high parts of the original input give four input vectors.
SDValue Inputs[4];
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
GetSplitVector(N->getOperand(0), Inputs[0], Inputs[1]);
GetSplitVector(N->getOperand(1), Inputs[2], Inputs[3]);
EVT NewVT = Inputs[0].getValueType();
// Extract the vector element by hand.
SVOps.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT,
- Inputs[Input], DAG.getIntPtrConstant(Idx)));
+ Inputs[Input], DAG.getConstant(Idx,
+ TLI.getVectorIdxTy())));
}
// Construct the Lo/Hi output using a BUILD_VECTOR.
- Output = DAG.getNode(ISD::BUILD_VECTOR,dl,NewVT, &SVOps[0], SVOps.size());
+ Output = DAG.getNode(ISD::BUILD_VECTOR, dl, NewVT, SVOps);
} else if (InputUsed[0] == -1U) {
// No input vectors were used! The result is undefined.
Output = DAG.getUNDEF(NewVT);
dbgs() << "\n");
SDValue Res = SDValue();
- if (Res.getNode() == 0) {
+ // See if the target wants to custom split this node.
+ if (CustomLowerNode(N, N->getOperand(OpNo).getValueType(), false))
+ return false;
+
+ if (!Res.getNode()) {
switch (N->getOpcode()) {
default:
#ifndef NDEBUG
N->dump(&DAG);
dbgs() << "\n";
#endif
- llvm_unreachable("Do not know how to split this operator's operand!");
+ report_fatal_error("Do not know how to split this operator's "
+ "operand!\n");
+
case ISD::SETCC: Res = SplitVecOp_VSETCC(N); break;
case ISD::BITCAST: Res = SplitVecOp_BITCAST(N); break;
case ISD::EXTRACT_SUBVECTOR: Res = SplitVecOp_EXTRACT_SUBVECTOR(N); break;
case ISD::EXTRACT_VECTOR_ELT:Res = SplitVecOp_EXTRACT_VECTOR_ELT(N); break;
case ISD::CONCAT_VECTORS: Res = SplitVecOp_CONCAT_VECTORS(N); break;
+ case ISD::TRUNCATE: Res = SplitVecOp_TRUNCATE(N); break;
case ISD::FP_ROUND: Res = SplitVecOp_FP_ROUND(N); break;
case ISD::STORE:
Res = SplitVecOp_STORE(cast<StoreSDNode>(N), OpNo);
break;
-
+ case ISD::MSTORE:
+ Res = SplitVecOp_MSTORE(cast<MaskedStoreSDNode>(N), OpNo);
+ break;
+ case ISD::VSELECT:
+ Res = SplitVecOp_VSELECT(N, OpNo);
+ break;
case ISD::CTTZ:
case ISD::CTLZ:
case ISD::CTPOP:
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
case ISD::FTRUNC:
- case ISD::TRUNCATE:
case ISD::SIGN_EXTEND:
case ISD::ZERO_EXTEND:
case ISD::ANY_EXTEND:
return false;
}
+SDValue DAGTypeLegalizer::SplitVecOp_VSELECT(SDNode *N, unsigned OpNo) {
+ // The only possibility for an illegal operand is the mask, since result type
+ // legalization would have handled this node already otherwise.
+ assert(OpNo == 0 && "Illegal operand must be mask");
+
+ SDValue Mask = N->getOperand(0);
+ SDValue Src0 = N->getOperand(1);
+ SDValue Src1 = N->getOperand(2);
+ EVT Src0VT = Src0.getValueType();
+ SDLoc DL(N);
+ assert(Mask.getValueType().isVector() && "VSELECT without a vector mask?");
+
+ SDValue Lo, Hi;
+ GetSplitVector(N->getOperand(0), Lo, Hi);
+ assert(Lo.getValueType() == Hi.getValueType() &&
+ "Lo and Hi have differing types");
+
+ EVT LoOpVT, HiOpVT;
+ std::tie(LoOpVT, HiOpVT) = DAG.GetSplitDestVTs(Src0VT);
+ assert(LoOpVT == HiOpVT && "Asymmetric vector split?");
+
+ SDValue LoOp0, HiOp0, LoOp1, HiOp1, LoMask, HiMask;
+ std::tie(LoOp0, HiOp0) = DAG.SplitVector(Src0, DL);
+ std::tie(LoOp1, HiOp1) = DAG.SplitVector(Src1, DL);
+ std::tie(LoMask, HiMask) = DAG.SplitVector(Mask, DL);
+
+ SDValue LoSelect =
+ DAG.getNode(ISD::VSELECT, DL, LoOpVT, LoMask, LoOp0, LoOp1);
+ SDValue HiSelect =
+ DAG.getNode(ISD::VSELECT, DL, HiOpVT, HiMask, HiOp0, HiOp1);
+
+ return DAG.getNode(ISD::CONCAT_VECTORS, DL, Src0VT, LoSelect, HiSelect);
+}
+
SDValue DAGTypeLegalizer::SplitVecOp_UnaryOp(SDNode *N) {
// The result has a legal vector type, but the input needs splitting.
EVT ResVT = N->getValueType(0);
SDValue Lo, Hi;
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
GetSplitVector(N->getOperand(0), Lo, Hi);
EVT InVT = Lo.getValueType();
if (TLI.isBigEndian())
std::swap(Lo, Hi);
- return DAG.getNode(ISD::BITCAST, N->getDebugLoc(), N->getValueType(0),
+ return DAG.getNode(ISD::BITCAST, SDLoc(N), N->getValueType(0),
JoinIntegers(Lo, Hi));
}
// We know that the extracted result type is legal.
EVT SubVT = N->getValueType(0);
SDValue Idx = N->getOperand(1);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
SDValue Lo, Hi;
GetSplitVector(N->getOperand(0), Lo, Hi);
Idx.getValueType())), 0);
}
+ // See if the target wants to custom expand this node.
+ if (CustomLowerNode(N, N->getValueType(0), true))
+ return SDValue();
+
// Store the vector to the stack.
EVT EltVT = VecVT.getVectorElementType();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
SDValue StackPtr = DAG.CreateStackTemporary(VecVT);
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr,
MachinePointerInfo(), false, false, 0);
// Load back the required element.
StackPtr = GetVectorElementPointer(StackPtr, EltVT, Idx);
return DAG.getExtLoad(ISD::EXTLOAD, dl, N->getValueType(0), Store, StackPtr,
- MachinePointerInfo(), EltVT, false, false, 0);
+ MachinePointerInfo(), EltVT, false, false, false, 0);
+}
+
+SDValue DAGTypeLegalizer::SplitVecOp_MSTORE(MaskedStoreSDNode *N,
+ unsigned OpNo) {
+ SDValue Ch = N->getChain();
+ SDValue Ptr = N->getBasePtr();
+ SDValue Mask = N->getMask();
+ SDValue Data = N->getValue();
+ EVT MemoryVT = N->getMemoryVT();
+ unsigned Alignment = N->getOriginalAlignment();
+ SDLoc DL(N);
+
+ EVT LoMemVT, HiMemVT;
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
+
+ SDValue DataLo, DataHi;
+ GetSplitVector(Data, DataLo, DataHi);
+ SDValue MaskLo, MaskHi;
+ GetSplitVector(Mask, MaskLo, MaskHi);
+
+ // if Alignment is equal to the vector size,
+ // take the half of it for the second part
+ unsigned SecondHalfAlignment =
+ (Alignment == Data->getValueType(0).getSizeInBits()/8) ?
+ Alignment/2 : Alignment;
+
+ SDValue Lo, Hi;
+ MachineMemOperand *MMO = DAG.getMachineFunction().
+ getMachineMemOperand(N->getPointerInfo(),
+ MachineMemOperand::MOStore, LoMemVT.getStoreSize(),
+ Alignment, N->getAAInfo(), N->getRanges());
+
+ Lo = DAG.getMaskedStore(Ch, DL, DataLo, Ptr, MaskLo, LoMemVT, MMO,
+ N->isTruncatingStore());
+
+ unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
+ Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
+
+ MMO = DAG.getMachineFunction().
+ getMachineMemOperand(N->getPointerInfo(),
+ MachineMemOperand::MOStore, HiMemVT.getStoreSize(),
+ SecondHalfAlignment, N->getAAInfo(), N->getRanges());
+
+ Hi = DAG.getMaskedStore(Ch, DL, DataHi, Ptr, MaskHi, HiMemVT, MMO,
+ N->isTruncatingStore());
+
+
+ // Build a factor node to remember that this store is independent of the
+ // other one.
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
+
}
SDValue DAGTypeLegalizer::SplitVecOp_STORE(StoreSDNode *N, unsigned OpNo) {
assert(N->isUnindexed() && "Indexed store of vector?");
assert(OpNo == 1 && "Can only split the stored value");
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
bool isTruncating = N->isTruncatingStore();
SDValue Ch = N->getChain();
unsigned Alignment = N->getOriginalAlignment();
bool isVol = N->isVolatile();
bool isNT = N->isNonTemporal();
+ AAMDNodes AAInfo = N->getAAInfo();
SDValue Lo, Hi;
GetSplitVector(N->getOperand(1), Lo, Hi);
EVT LoMemVT, HiMemVT;
- GetSplitDestVTs(MemoryVT, LoMemVT, HiMemVT);
+ std::tie(LoMemVT, HiMemVT) = DAG.GetSplitDestVTs(MemoryVT);
unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
if (isTruncating)
Lo = DAG.getTruncStore(Ch, DL, Lo, Ptr, N->getPointerInfo(),
- LoMemVT, isVol, isNT, Alignment);
+ LoMemVT, isVol, isNT, Alignment, AAInfo);
else
Lo = DAG.getStore(Ch, DL, Lo, Ptr, N->getPointerInfo(),
- isVol, isNT, Alignment);
+ isVol, isNT, Alignment, AAInfo);
// Increment the pointer to the other half.
Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
- DAG.getIntPtrConstant(IncrementSize));
+ DAG.getConstant(IncrementSize, Ptr.getValueType()));
if (isTruncating)
Hi = DAG.getTruncStore(Ch, DL, Hi, Ptr,
N->getPointerInfo().getWithOffset(IncrementSize),
- HiMemVT, isVol, isNT, Alignment);
+ HiMemVT, isVol, isNT, Alignment, AAInfo);
else
Hi = DAG.getStore(Ch, DL, Hi, Ptr,
N->getPointerInfo().getWithOffset(IncrementSize),
- isVol, isNT, Alignment);
+ isVol, isNT, Alignment, AAInfo);
return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
}
SDValue DAGTypeLegalizer::SplitVecOp_CONCAT_VECTORS(SDNode *N) {
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
- // The input operands all must have the same type, and we know the result the
- // result type is valid. Convert this to a buildvector which extracts all the
+ // The input operands all must have the same type, and we know the result
+ // type is valid. Convert this to a buildvector which extracts all the
// input elements.
// TODO: If the input elements are power-two vectors, we could convert this to
// a new CONCAT_VECTORS node with elements that are half-wide.
for (unsigned i = 0, e = Op.getValueType().getVectorNumElements();
i != e; ++i) {
Elts.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT,
- Op, DAG.getIntPtrConstant(i)));
+ Op, DAG.getConstant(i, TLI.getVectorIdxTy())));
}
}
- return DAG.getNode(ISD::BUILD_VECTOR, DL, N->getValueType(0),
- &Elts[0], Elts.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, N->getValueType(0), Elts);
+}
+
+SDValue DAGTypeLegalizer::SplitVecOp_TRUNCATE(SDNode *N) {
+ // The result type is legal, but the input type is illegal. If splitting
+ // ends up with the result type of each half still being legal, just
+ // do that. If, however, that would result in an illegal result type,
+ // we can try to get more clever with power-two vectors. Specifically,
+ // split the input type, but also widen the result element size, then
+ // concatenate the halves and truncate again. For example, consider a target
+ // where v8i8 is legal and v8i32 is not (ARM, which doesn't have 256-bit
+ // vectors). To perform a "%res = v8i8 trunc v8i32 %in" we do:
+ // %inlo = v4i32 extract_subvector %in, 0
+ // %inhi = v4i32 extract_subvector %in, 4
+ // %lo16 = v4i16 trunc v4i32 %inlo
+ // %hi16 = v4i16 trunc v4i32 %inhi
+ // %in16 = v8i16 concat_vectors v4i16 %lo16, v4i16 %hi16
+ // %res = v8i8 trunc v8i16 %in16
+ //
+ // Without this transform, the original truncate would end up being
+ // scalarized, which is pretty much always a last resort.
+ SDValue InVec = N->getOperand(0);
+ EVT InVT = InVec->getValueType(0);
+ EVT OutVT = N->getValueType(0);
+ unsigned NumElements = OutVT.getVectorNumElements();
+ // Widening should have already made sure this is a power-two vector
+ // if we're trying to split it at all. assert() that's true, just in case.
+ assert(!(NumElements & 1) && "Splitting vector, but not in half!");
+
+ unsigned InElementSize = InVT.getVectorElementType().getSizeInBits();
+ unsigned OutElementSize = OutVT.getVectorElementType().getSizeInBits();
+
+ // If the input elements are only 1/2 the width of the result elements,
+ // just use the normal splitting. Our trick only work if there's room
+ // to split more than once.
+ if (InElementSize <= OutElementSize * 2)
+ return SplitVecOp_UnaryOp(N);
+ SDLoc DL(N);
+
+ // Extract the halves of the input via extract_subvector.
+ SDValue InLoVec, InHiVec;
+ std::tie(InLoVec, InHiVec) = DAG.SplitVector(InVec, DL);
+ // Truncate them to 1/2 the element size.
+ EVT HalfElementVT = EVT::getIntegerVT(*DAG.getContext(), InElementSize/2);
+ EVT HalfVT = EVT::getVectorVT(*DAG.getContext(), HalfElementVT,
+ NumElements/2);
+ SDValue HalfLo = DAG.getNode(ISD::TRUNCATE, DL, HalfVT, InLoVec);
+ SDValue HalfHi = DAG.getNode(ISD::TRUNCATE, DL, HalfVT, InHiVec);
+ // Concatenate them to get the full intermediate truncation result.
+ EVT InterVT = EVT::getVectorVT(*DAG.getContext(), HalfElementVT, NumElements);
+ SDValue InterVec = DAG.getNode(ISD::CONCAT_VECTORS, DL, InterVT, HalfLo,
+ HalfHi);
+ // Now finish up by truncating all the way down to the original result
+ // type. This should normally be something that ends up being legal directly,
+ // but in theory if a target has very wide vectors and an annoyingly
+ // restricted set of legal types, this split can chain to build things up.
+ return DAG.getNode(ISD::TRUNCATE, DL, OutVT, InterVec);
}
SDValue DAGTypeLegalizer::SplitVecOp_VSETCC(SDNode *N) {
"Operand types must be vectors");
// The result has a legal vector type, but the input needs splitting.
SDValue Lo0, Hi0, Lo1, Hi1, LoRes, HiRes;
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
GetSplitVector(N->getOperand(0), Lo0, Hi0);
GetSplitVector(N->getOperand(1), Lo1, Hi1);
unsigned PartElements = Lo0.getValueType().getVectorNumElements();
// The result has a legal vector type, but the input needs splitting.
EVT ResVT = N->getValueType(0);
SDValue Lo, Hi;
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
GetSplitVector(N->getOperand(0), Lo, Hi);
EVT InVT = Lo.getValueType();
-
+
EVT OutVT = EVT::getVectorVT(*DAG.getContext(), ResVT.getVectorElementType(),
InVT.getVectorNumElements());
-
+
Lo = DAG.getNode(ISD::FP_ROUND, DL, OutVT, Lo, N->getOperand(1));
Hi = DAG.getNode(ISD::FP_ROUND, DL, OutVT, Hi, N->getOperand(1));
-
+
return DAG.getNode(ISD::CONCAT_VECTORS, DL, ResVT, Lo, Hi);
-}
+}
case ISD::VECTOR_SHUFFLE:
Res = WidenVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N));
break;
+ case ISD::MLOAD:
+ Res = WidenVecRes_MLOAD(cast<MaskedLoadSDNode>(N));
+ break;
+
case ISD::ADD:
case ISD::AND:
- case ISD::BSWAP:
+ case ISD::MUL:
+ case ISD::MULHS:
+ case ISD::MULHU:
+ case ISD::OR:
+ case ISD::SUB:
+ case ISD::XOR:
+ case ISD::FMINNUM:
+ case ISD::FMAXNUM:
+ Res = WidenVecRes_Binary(N);
+ break;
+
case ISD::FADD:
case ISD::FCOPYSIGN:
- case ISD::FDIV:
case ISD::FMUL:
case ISD::FPOW:
- case ISD::FREM:
case ISD::FSUB:
- case ISD::MUL:
- case ISD::MULHS:
- case ISD::MULHU:
- case ISD::OR:
+ case ISD::FDIV:
+ case ISD::FREM:
case ISD::SDIV:
- case ISD::SREM:
case ISD::UDIV:
+ case ISD::SREM:
case ISD::UREM:
- case ISD::SUB:
- case ISD::XOR:
- Res = WidenVecRes_Binary(N);
+ Res = WidenVecRes_BinaryCanTrap(N);
break;
case ISD::FPOWI:
Res = WidenVecRes_Convert(N);
break;
+ case ISD::BSWAP:
case ISD::CTLZ:
case ISD::CTPOP:
case ISD::CTTZ:
case ISD::FNEARBYINT:
case ISD::FNEG:
case ISD::FRINT:
+ case ISD::FROUND:
case ISD::FSIN:
case ISD::FSQRT:
case ISD::FTRUNC:
Res = WidenVecRes_Unary(N);
break;
+ case ISD::FMA:
+ Res = WidenVecRes_Ternary(N);
+ break;
}
// If Res is null, the sub-method took care of registering the result.
SetWidenedVector(SDValue(N, ResNo), Res);
}
+SDValue DAGTypeLegalizer::WidenVecRes_Ternary(SDNode *N) {
+ // Ternary op widening.
+ SDLoc dl(N);
+ EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+ SDValue InOp1 = GetWidenedVector(N->getOperand(0));
+ SDValue InOp2 = GetWidenedVector(N->getOperand(1));
+ SDValue InOp3 = GetWidenedVector(N->getOperand(2));
+ return DAG.getNode(N->getOpcode(), dl, WidenVT, InOp1, InOp2, InOp3);
+}
+
SDValue DAGTypeLegalizer::WidenVecRes_Binary(SDNode *N) {
// Binary op widening.
+ SDLoc dl(N);
+ EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+ SDValue InOp1 = GetWidenedVector(N->getOperand(0));
+ SDValue InOp2 = GetWidenedVector(N->getOperand(1));
+ return DAG.getNode(N->getOpcode(), dl, WidenVT, InOp1, InOp2);
+}
+
+SDValue DAGTypeLegalizer::WidenVecRes_BinaryCanTrap(SDNode *N) {
+ // Binary op widening for operations that can trap.
unsigned Opcode = N->getOpcode();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
EVT WidenEltVT = WidenVT.getVectorElementType();
EVT VT = WidenVT;
while (CurNumElts != 0) {
while (CurNumElts >= NumElts) {
SDValue EOp1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, InOp1,
- DAG.getIntPtrConstant(Idx));
+ DAG.getConstant(Idx, TLI.getVectorIdxTy()));
SDValue EOp2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, InOp2,
- DAG.getIntPtrConstant(Idx));
+ DAG.getConstant(Idx, TLI.getVectorIdxTy()));
ConcatOps[ConcatEnd++] = DAG.getNode(Opcode, dl, VT, EOp1, EOp2);
Idx += NumElts;
CurNumElts -= NumElts;
if (NumElts == 1) {
for (unsigned i = 0; i != CurNumElts; ++i, ++Idx) {
SDValue EOp1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, WidenEltVT,
- InOp1, DAG.getIntPtrConstant(Idx));
+ InOp1, DAG.getConstant(Idx,
+ TLI.getVectorIdxTy()));
SDValue EOp2 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, WidenEltVT,
- InOp2, DAG.getIntPtrConstant(Idx));
+ InOp2, DAG.getConstant(Idx,
+ TLI.getVectorIdxTy()));
ConcatOps[ConcatEnd++] = DAG.getNode(Opcode, dl, WidenEltVT,
EOp1, EOp2);
}
unsigned NumToInsert = ConcatEnd - Idx - 1;
for (unsigned i = 0, OpIdx = Idx+1; i < NumToInsert; i++, OpIdx++) {
VecOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, NextVT, VecOp,
- ConcatOps[OpIdx], DAG.getIntPtrConstant(i));
+ ConcatOps[OpIdx], DAG.getConstant(i,
+ TLI.getVectorIdxTy()));
}
ConcatOps[Idx+1] = VecOp;
ConcatEnd = Idx + 2;
while (SubConcatEnd < OpsToConcat)
SubConcatOps[SubConcatEnd++] = undefVec;
ConcatOps[SubConcatIdx] = DAG.getNode(ISD::CONCAT_VECTORS, dl,
- NextVT, &SubConcatOps[0],
- OpsToConcat);
+ NextVT, SubConcatOps);
ConcatEnd = SubConcatIdx + 1;
}
}
for (unsigned j = ConcatEnd; j < NumOps; ++j)
ConcatOps[j] = UndefVal;
}
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, &ConcatOps[0], NumOps);
+ return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT,
+ makeArrayRef(ConcatOps.data(), NumOps));
}
SDValue DAGTypeLegalizer::WidenVecRes_Convert(SDNode *N) {
SDValue InOp = N->getOperand(0);
- DebugLoc DL = N->getDebugLoc();
+ SDLoc DL(N);
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
unsigned WidenNumElts = WidenVT.getVectorNumElements();
SDValue UndefVal = DAG.getUNDEF(InVT);
for (unsigned i = 1; i != NumConcat; ++i)
Ops[i] = UndefVal;
- SDValue InVec = DAG.getNode(ISD::CONCAT_VECTORS, DL, InWidenVT,
- &Ops[0], NumConcat);
+ SDValue InVec = DAG.getNode(ISD::CONCAT_VECTORS, DL, InWidenVT, Ops);
if (N->getNumOperands() == 1)
return DAG.getNode(Opcode, DL, WidenVT, InVec);
return DAG.getNode(Opcode, DL, WidenVT, InVec, N->getOperand(1));
if (InVTNumElts % WidenNumElts == 0) {
SDValue InVal = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InWidenVT,
- InOp, DAG.getIntPtrConstant(0));
+ InOp, DAG.getConstant(0,
+ TLI.getVectorIdxTy()));
// Extract the input and convert the shorten input vector.
if (N->getNumOperands() == 1)
return DAG.getNode(Opcode, DL, WidenVT, InVal);
unsigned i;
for (i=0; i < MinElts; ++i) {
SDValue Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, InEltVT, InOp,
- DAG.getIntPtrConstant(i));
+ DAG.getConstant(i, TLI.getVectorIdxTy()));
if (N->getNumOperands() == 1)
Ops[i] = DAG.getNode(Opcode, DL, EltVT, Val);
else
for (; i < WidenNumElts; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, DL, WidenVT, &Ops[0], WidenNumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, DL, WidenVT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecRes_POWI(SDNode *N) {
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
SDValue InOp = GetWidenedVector(N->getOperand(0));
SDValue ShOp = N->getOperand(1);
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(), WidenVT, InOp, ShOp);
+ return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT, InOp, ShOp);
}
SDValue DAGTypeLegalizer::WidenVecRes_Shift(SDNode *N) {
if (ShVT != ShWidenVT)
ShOp = ModifyToType(ShOp, ShWidenVT);
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(), WidenVT, InOp, ShOp);
+ return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT, InOp, ShOp);
}
SDValue DAGTypeLegalizer::WidenVecRes_Unary(SDNode *N) {
// Unary op widening.
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
SDValue InOp = GetWidenedVector(N->getOperand(0));
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(), WidenVT, InOp);
+ return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT, InOp);
}
SDValue DAGTypeLegalizer::WidenVecRes_InregOp(SDNode *N) {
.getVectorElementType(),
WidenVT.getVectorNumElements());
SDValue WidenLHS = GetWidenedVector(N->getOperand(0));
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(),
+ return DAG.getNode(N->getOpcode(), SDLoc(N),
WidenVT, WidenLHS, DAG.getValueType(ExtVT));
}
EVT InVT = InOp.getValueType();
EVT VT = N->getValueType(0);
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
switch (getTypeAction(InVT)) {
- default:
- assert(false && "Unknown type action!");
- break;
case TargetLowering::TypeLegal:
break;
case TargetLowering::TypePromoteInteger:
+ // If the incoming type is a vector that is being promoted, then
+ // we know that the elements are arranged differently and that we
+ // must perform the conversion using a stack slot.
+ if (InVT.isVector())
+ break;
+
// If the InOp is promoted to the same size, convert it. Otherwise,
// fall out of the switch and widen the promoted input.
InOp = GetPromotedInteger(InOp);
SDValue NewVec;
if (InVT.isVector())
- NewVec = DAG.getNode(ISD::CONCAT_VECTORS, dl,
- NewInVT, &Ops[0], NewNumElts);
+ NewVec = DAG.getNode(ISD::CONCAT_VECTORS, dl, NewInVT, Ops);
else
- NewVec = DAG.getNode(ISD::BUILD_VECTOR, dl,
- NewInVT, &Ops[0], NewNumElts);
+ NewVec = DAG.getNode(ISD::BUILD_VECTOR, dl, NewInVT, Ops);
return DAG.getNode(ISD::BITCAST, dl, WidenVT, NewVec);
}
}
}
SDValue DAGTypeLegalizer::WidenVecRes_BUILD_VECTOR(SDNode *N) {
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
// Build a vector with undefined for the new nodes.
EVT VT = N->getValueType(0);
- EVT EltVT = VT.getVectorElementType();
+
+ // Integer BUILD_VECTOR operands may be larger than the node's vector element
+ // type. The UNDEFs need to have the same type as the existing operands.
+ EVT EltVT = N->getOperand(0).getValueType();
unsigned NumElts = VT.getVectorNumElements();
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
unsigned WidenNumElts = WidenVT.getVectorNumElements();
SmallVector<SDValue, 16> NewOps(N->op_begin(), N->op_end());
- NewOps.reserve(WidenNumElts);
- for (unsigned i = NumElts; i < WidenNumElts; ++i)
- NewOps.push_back(DAG.getUNDEF(EltVT));
+ assert(WidenNumElts >= NumElts && "Shrinking vector instead of widening!");
+ NewOps.append(WidenNumElts - NumElts, DAG.getUNDEF(EltVT));
- return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, &NewOps[0], NewOps.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, NewOps);
}
SDValue DAGTypeLegalizer::WidenVecRes_CONCAT_VECTORS(SDNode *N) {
EVT InVT = N->getOperand(0).getValueType();
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
unsigned WidenNumElts = WidenVT.getVectorNumElements();
unsigned NumInElts = InVT.getVectorNumElements();
unsigned NumOperands = N->getNumOperands();
Ops[i] = N->getOperand(i);
for (unsigned i = NumOperands; i != NumConcat; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, &Ops[0], NumConcat);
+ return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, Ops);
}
} else {
InputWidened = true;
if (InputWidened)
InOp = GetWidenedVector(InOp);
for (unsigned j=0; j < NumInElts; ++j)
- Ops[Idx++] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, InOp,
- DAG.getIntPtrConstant(j));
+ Ops[Idx++] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, InOp,
+ DAG.getConstant(j, TLI.getVectorIdxTy()));
}
SDValue UndefVal = DAG.getUNDEF(EltVT);
for (; Idx < WidenNumElts; ++Idx)
Ops[Idx] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, &Ops[0], WidenNumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecRes_CONVERT_RNDSAT(SDNode *N) {
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
SDValue InOp = N->getOperand(0);
SDValue RndOp = N->getOperand(3);
SDValue SatOp = N->getOperand(4);
for (unsigned i = 1; i != NumConcat; ++i)
Ops[i] = UndefVal;
- InOp = DAG.getNode(ISD::CONCAT_VECTORS, dl, InWidenVT, &Ops[0],NumConcat);
+ InOp = DAG.getNode(ISD::CONCAT_VECTORS, dl, InWidenVT, Ops);
return DAG.getConvertRndSat(WidenVT, dl, InOp, DTyOp, STyOp, RndOp,
SatOp, CvtCode);
}
if (InVTNumElts % WidenNumElts == 0) {
// Extract the input and convert the shorten input vector.
InOp = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, InWidenVT, InOp,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
return DAG.getConvertRndSat(WidenVT, dl, InOp, DTyOp, STyOp, RndOp,
- SatOp, CvtCode);
+ SatOp, CvtCode);
}
}
unsigned i;
for (i=0; i < MinElts; ++i) {
SDValue ExtVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, InEltVT, InOp,
- DAG.getIntPtrConstant(i));
+ DAG.getConstant(i, TLI.getVectorIdxTy()));
Ops[i] = DAG.getConvertRndSat(WidenVT, dl, ExtVal, DTyOp, STyOp, RndOp,
- SatOp, CvtCode);
+ SatOp, CvtCode);
}
SDValue UndefVal = DAG.getUNDEF(EltVT);
for (; i < WidenNumElts; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, &Ops[0], WidenNumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecRes_EXTRACT_SUBVECTOR(SDNode *N) {
unsigned WidenNumElts = WidenVT.getVectorNumElements();
SDValue InOp = N->getOperand(0);
SDValue Idx = N->getOperand(1);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
if (getTypeAction(InOp.getValueType()) == TargetLowering::TypeWidenVector)
InOp = GetWidenedVector(InOp);
unsigned i;
for (i=0; i < NumElts; ++i)
Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, InOp,
- DAG.getIntPtrConstant(IdxVal+i));
+ DAG.getConstant(IdxVal+i, TLI.getVectorIdxTy()));
SDValue UndefVal = DAG.getUNDEF(EltVT);
for (; i < WidenNumElts; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, &Ops[0], WidenNumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecRes_INSERT_VECTOR_ELT(SDNode *N) {
SDValue InOp = GetWidenedVector(N->getOperand(0));
- return DAG.getNode(ISD::INSERT_VECTOR_ELT, N->getDebugLoc(),
+ return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N),
InOp.getValueType(), InOp,
N->getOperand(1), N->getOperand(2));
}
if (LdChain.size() == 1)
NewChain = LdChain[0];
else
- NewChain = DAG.getNode(ISD::TokenFactor, LD->getDebugLoc(), MVT::Other,
- &LdChain[0], LdChain.size());
+ NewChain = DAG.getNode(ISD::TokenFactor, SDLoc(LD), MVT::Other, LdChain);
// Modified the chain - switch anything that used the old chain to use
// the new one.
return Result;
}
+SDValue DAGTypeLegalizer::WidenVecRes_MLOAD(MaskedLoadSDNode *N) {
+
+ EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(),N->getValueType(0));
+ SDValue Mask = N->getMask();
+ EVT MaskVT = Mask.getValueType();
+ SDValue Src0 = GetWidenedVector(N->getSrc0());
+ ISD::LoadExtType ExtType = N->getExtensionType();
+ SDLoc dl(N);
+
+ if (getTypeAction(MaskVT) == TargetLowering::TypeWidenVector)
+ Mask = GetWidenedVector(Mask);
+ else {
+ EVT BoolVT = getSetCCResultType(WidenVT);
+
+ // We can't use ModifyToType() because we should fill the mask with
+ // zeroes
+ unsigned WidenNumElts = BoolVT.getVectorNumElements();
+ unsigned MaskNumElts = MaskVT.getVectorNumElements();
+
+ unsigned NumConcat = WidenNumElts / MaskNumElts;
+ SmallVector<SDValue, 16> Ops(NumConcat);
+ SDValue ZeroVal = DAG.getConstant(0, MaskVT);
+ Ops[0] = Mask;
+ for (unsigned i = 1; i != NumConcat; ++i)
+ Ops[i] = ZeroVal;
+
+ Mask = DAG.getNode(ISD::CONCAT_VECTORS, dl, BoolVT, Ops);
+ }
+
+ SDValue Res = DAG.getMaskedLoad(WidenVT, dl, N->getChain(), N->getBasePtr(),
+ Mask, Src0, N->getMemoryVT(),
+ N->getMemOperand(), ExtType);
+ // Legalized the chain result - switch anything that used the old chain to
+ // use the new one.
+ ReplaceValueWith(SDValue(N, 1), Res.getValue(1));
+ return Res;
+}
+
SDValue DAGTypeLegalizer::WidenVecRes_SCALAR_TO_VECTOR(SDNode *N) {
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
- return DAG.getNode(ISD::SCALAR_TO_VECTOR, N->getDebugLoc(),
+ return DAG.getNode(ISD::SCALAR_TO_VECTOR, SDLoc(N),
WidenVT, N->getOperand(0));
}
if (getTypeAction(CondVT) == TargetLowering::TypeWidenVector)
Cond1 = GetWidenedVector(Cond1);
+ // If we have to split the condition there is no point in widening the
+ // select. This would result in an cycle of widening the select ->
+ // widening the condition operand -> splitting the condition operand ->
+ // splitting the select -> widening the select. Instead split this select
+ // further and widen the resulting type.
+ if (getTypeAction(CondVT) == TargetLowering::TypeSplitVector) {
+ SDValue SplitSelect = SplitVecOp_VSELECT(N, 0);
+ SDValue Res = ModifyToType(SplitSelect, WidenVT);
+ return Res;
+ }
+
if (Cond1.getValueType() != CondWidenVT)
- Cond1 = ModifyToType(Cond1, CondWidenVT);
+ Cond1 = ModifyToType(Cond1, CondWidenVT);
}
SDValue InOp1 = GetWidenedVector(N->getOperand(1));
SDValue InOp2 = GetWidenedVector(N->getOperand(2));
assert(InOp1.getValueType() == WidenVT && InOp2.getValueType() == WidenVT);
- return DAG.getNode(N->getOpcode(), N->getDebugLoc(),
+ return DAG.getNode(N->getOpcode(), SDLoc(N),
WidenVT, Cond1, InOp1, InOp2);
}
SDValue DAGTypeLegalizer::WidenVecRes_SELECT_CC(SDNode *N) {
SDValue InOp1 = GetWidenedVector(N->getOperand(2));
SDValue InOp2 = GetWidenedVector(N->getOperand(3));
- return DAG.getNode(ISD::SELECT_CC, N->getDebugLoc(),
+ return DAG.getNode(ISD::SELECT_CC, SDLoc(N),
InOp1.getValueType(), N->getOperand(0),
N->getOperand(1), InOp1, InOp2, N->getOperand(4));
}
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
SDValue InOp1 = GetWidenedVector(N->getOperand(0));
SDValue InOp2 = GetWidenedVector(N->getOperand(1));
- return DAG.getNode(ISD::SETCC, N->getDebugLoc(), WidenVT,
+ return DAG.getNode(ISD::SETCC, SDLoc(N), WidenVT,
InOp1, InOp2, N->getOperand(2));
}
SDValue DAGTypeLegalizer::WidenVecRes_VECTOR_SHUFFLE(ShuffleVectorSDNode *N) {
EVT VT = N->getValueType(0);
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
unsigned NumElts = VT.getVectorNumElements();
SDValue InOp1 = N->getOperand(0);
EVT InVT = InOp1.getValueType();
- assert(InVT.isVector() && "can not widen non vector type");
+ assert(InVT.isVector() && "can not widen non-vector type");
EVT WidenInVT = EVT::getVectorVT(*DAG.getContext(),
InVT.getVectorElementType(), WidenNumElts);
InOp1 = GetWidenedVector(InOp1);
InOp2.getValueType() == WidenInVT &&
"Input not widened to expected type!");
(void)WidenInVT;
- return DAG.getNode(ISD::SETCC, N->getDebugLoc(),
+ return DAG.getNode(ISD::SETCC, SDLoc(N),
WidenVT, InOp1, InOp2, N->getOperand(2));
}
//===----------------------------------------------------------------------===//
// Widen Vector Operand
//===----------------------------------------------------------------------===//
-bool DAGTypeLegalizer::WidenVectorOperand(SDNode *N, unsigned ResNo) {
- DEBUG(dbgs() << "Widen node operand " << ResNo << ": ";
+bool DAGTypeLegalizer::WidenVectorOperand(SDNode *N, unsigned OpNo) {
+ DEBUG(dbgs() << "Widen node operand " << OpNo << ": ";
N->dump(&DAG);
dbgs() << "\n");
SDValue Res = SDValue();
+ // See if the target wants to custom widen this node.
+ if (CustomLowerNode(N, N->getOperand(OpNo).getValueType(), false))
+ return false;
+
switch (N->getOpcode()) {
default:
#ifndef NDEBUG
- dbgs() << "WidenVectorOperand op #" << ResNo << ": ";
+ dbgs() << "WidenVectorOperand op #" << OpNo << ": ";
N->dump(&DAG);
dbgs() << "\n";
#endif
case ISD::EXTRACT_SUBVECTOR: Res = WidenVecOp_EXTRACT_SUBVECTOR(N); break;
case ISD::EXTRACT_VECTOR_ELT: Res = WidenVecOp_EXTRACT_VECTOR_ELT(N); break;
case ISD::STORE: Res = WidenVecOp_STORE(N); break;
+ case ISD::MSTORE: Res = WidenVecOp_MSTORE(N, OpNo); break;
case ISD::SETCC: Res = WidenVecOp_SETCC(N); break;
+ case ISD::ANY_EXTEND:
+ case ISD::SIGN_EXTEND:
+ case ISD::ZERO_EXTEND:
+ Res = WidenVecOp_EXTEND(N);
+ break;
+
case ISD::FP_EXTEND:
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
case ISD::TRUNCATE:
- case ISD::SIGN_EXTEND:
- case ISD::ZERO_EXTEND:
- case ISD::ANY_EXTEND:
Res = WidenVecOp_Convert(N);
break;
}
return false;
}
+SDValue DAGTypeLegalizer::WidenVecOp_EXTEND(SDNode *N) {
+ SDLoc DL(N);
+ EVT VT = N->getValueType(0);
+
+ SDValue InOp = N->getOperand(0);
+ // If some legalization strategy other than widening is used on the operand,
+ // we can't safely assume that just extending the low lanes is the correct
+ // transformation.
+ if (getTypeAction(InOp.getValueType()) != TargetLowering::TypeWidenVector)
+ return WidenVecOp_Convert(N);
+ InOp = GetWidenedVector(InOp);
+ assert(VT.getVectorNumElements() <
+ InOp.getValueType().getVectorNumElements() &&
+ "Input wasn't widened!");
+
+ // We may need to further widen the operand until it has the same total
+ // vector size as the result.
+ EVT InVT = InOp.getValueType();
+ if (InVT.getSizeInBits() != VT.getSizeInBits()) {
+ EVT InEltVT = InVT.getVectorElementType();
+ for (int i = MVT::FIRST_VECTOR_VALUETYPE, e = MVT::LAST_VECTOR_VALUETYPE; i < e; ++i) {
+ EVT FixedVT = (MVT::SimpleValueType)i;
+ EVT FixedEltVT = FixedVT.getVectorElementType();
+ if (TLI.isTypeLegal(FixedVT) &&
+ FixedVT.getSizeInBits() == VT.getSizeInBits() &&
+ FixedEltVT == InEltVT) {
+ assert(FixedVT.getVectorNumElements() >= VT.getVectorNumElements() &&
+ "Not enough elements in the fixed type for the operand!");
+ assert(FixedVT.getVectorNumElements() != InVT.getVectorNumElements() &&
+ "We can't have the same type as we started with!");
+ if (FixedVT.getVectorNumElements() > InVT.getVectorNumElements())
+ InOp = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, FixedVT,
+ DAG.getUNDEF(FixedVT), InOp,
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
+ else
+ InOp = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, FixedVT, InOp,
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
+ break;
+ }
+ }
+ InVT = InOp.getValueType();
+ if (InVT.getSizeInBits() != VT.getSizeInBits())
+ // We couldn't find a legal vector type that was a widening of the input
+ // and could be extended in-register to the result type, so we have to
+ // scalarize.
+ return WidenVecOp_Convert(N);
+ }
+
+ // Use special DAG nodes to represent the operation of extending the
+ // low lanes.
+ switch (N->getOpcode()) {
+ default:
+ llvm_unreachable("Extend legalization on on extend operation!");
+ case ISD::ANY_EXTEND:
+ return DAG.getAnyExtendVectorInReg(InOp, DL, VT);
+ case ISD::SIGN_EXTEND:
+ return DAG.getSignExtendVectorInReg(InOp, DL, VT);
+ case ISD::ZERO_EXTEND:
+ return DAG.getZeroExtendVectorInReg(InOp, DL, VT);
+ }
+}
+
SDValue DAGTypeLegalizer::WidenVecOp_Convert(SDNode *N) {
// Since the result is legal and the input is illegal, it is unlikely
// that we can fix the input to a legal type so unroll the convert
// into some scalar code and create a nasty build vector.
EVT VT = N->getValueType(0);
EVT EltVT = VT.getVectorElementType();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
unsigned NumElts = VT.getVectorNumElements();
SDValue InOp = N->getOperand(0);
if (getTypeAction(InOp.getValueType()) == TargetLowering::TypeWidenVector)
for (unsigned i=0; i < NumElts; ++i)
Ops[i] = DAG.getNode(Opcode, dl, EltVT,
DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, InEltVT, InOp,
- DAG.getIntPtrConstant(i)));
+ DAG.getConstant(i, TLI.getVectorIdxTy())));
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], NumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecOp_BITCAST(SDNode *N) {
EVT VT = N->getValueType(0);
SDValue InOp = GetWidenedVector(N->getOperand(0));
EVT InWidenVT = InOp.getValueType();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
// Check if we can convert between two legal vector types and extract.
unsigned InWidenSize = InWidenVT.getSizeInBits();
if (TLI.isTypeLegal(NewVT)) {
SDValue BitOp = DAG.getNode(ISD::BITCAST, dl, NewVT, InOp);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, BitOp,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
}
}
// nasty build vector.
EVT VT = N->getValueType(0);
EVT EltVT = VT.getVectorElementType();
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
unsigned NumElts = VT.getVectorNumElements();
SmallVector<SDValue, 16> Ops(NumElts);
InOp = GetWidenedVector(InOp);
for (unsigned j=0; j < NumInElts; ++j)
Ops[Idx++] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, InOp,
- DAG.getIntPtrConstant(j));
+ DAG.getConstant(j, TLI.getVectorIdxTy()));
}
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], NumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}
SDValue DAGTypeLegalizer::WidenVecOp_EXTRACT_SUBVECTOR(SDNode *N) {
SDValue InOp = GetWidenedVector(N->getOperand(0));
- return DAG.getNode(ISD::EXTRACT_SUBVECTOR, N->getDebugLoc(),
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(N),
N->getValueType(0), InOp, N->getOperand(1));
}
SDValue DAGTypeLegalizer::WidenVecOp_EXTRACT_VECTOR_ELT(SDNode *N) {
SDValue InOp = GetWidenedVector(N->getOperand(0));
- return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, N->getDebugLoc(),
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N),
N->getValueType(0), InOp, N->getOperand(1));
}
if (StChain.size() == 1)
return StChain[0];
else
- return DAG.getNode(ISD::TokenFactor, ST->getDebugLoc(),
- MVT::Other,&StChain[0],StChain.size());
+ return DAG.getNode(ISD::TokenFactor, SDLoc(ST), MVT::Other, StChain);
+}
+
+SDValue DAGTypeLegalizer::WidenVecOp_MSTORE(SDNode *N, unsigned OpNo) {
+ MaskedStoreSDNode *MST = cast<MaskedStoreSDNode>(N);
+ SDValue Mask = MST->getMask();
+ EVT MaskVT = Mask.getValueType();
+ SDValue StVal = MST->getValue();
+ // Widen the value
+ SDValue WideVal = GetWidenedVector(StVal);
+ SDLoc dl(N);
+
+ if (OpNo == 2 || getTypeAction(MaskVT) == TargetLowering::TypeWidenVector)
+ Mask = GetWidenedVector(Mask);
+ else {
+ // The mask should be widened as well
+ EVT BoolVT = getSetCCResultType(WideVal.getValueType());
+ // We can't use ModifyToType() because we should fill the mask with
+ // zeroes
+ unsigned WidenNumElts = BoolVT.getVectorNumElements();
+ unsigned MaskNumElts = MaskVT.getVectorNumElements();
+
+ unsigned NumConcat = WidenNumElts / MaskNumElts;
+ SmallVector<SDValue, 16> Ops(NumConcat);
+ SDValue ZeroVal = DAG.getConstant(0, MaskVT);
+ Ops[0] = Mask;
+ for (unsigned i = 1; i != NumConcat; ++i)
+ Ops[i] = ZeroVal;
+
+ Mask = DAG.getNode(ISD::CONCAT_VECTORS, dl, BoolVT, Ops);
+ }
+ assert(Mask.getValueType().getVectorNumElements() ==
+ WideVal.getValueType().getVectorNumElements() &&
+ "Mask and data vectors should have the same number of elements");
+ return DAG.getMaskedStore(MST->getChain(), dl, WideVal, MST->getBasePtr(),
+ Mask, MST->getMemoryVT(), MST->getMemOperand(),
+ false);
}
SDValue DAGTypeLegalizer::WidenVecOp_SETCC(SDNode *N) {
SDValue InOp0 = GetWidenedVector(N->getOperand(0));
SDValue InOp1 = GetWidenedVector(N->getOperand(1));
- DebugLoc dl = N->getDebugLoc();
+ SDLoc dl(N);
// WARNING: In this code we widen the compare instruction with garbage.
// This garbage may contain denormal floats which may be slow. Is this a real
// Get a new SETCC node to compare the newly widened operands.
// Only some of the compared elements are legal.
- EVT SVT = TLI.getSetCCResultType(InOp0.getValueType());
- SDValue WideSETCC = DAG.getNode(ISD::SETCC, N->getDebugLoc(),
+ EVT SVT = TLI.getSetCCResultType(*DAG.getContext(), InOp0.getValueType());
+ SDValue WideSETCC = DAG.getNode(ISD::SETCC, SDLoc(N),
SVT, InOp0, InOp1, N->getOperand(2));
// Extract the needed results from the result vector.
SVT.getVectorElementType(),
N->getValueType(0).getVectorNumElements());
SDValue CC = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl,
- ResVT, WideSETCC, DAG.getIntPtrConstant(0));
+ ResVT, WideSETCC, DAG.getConstant(0,
+ TLI.getVectorIdxTy()));
- return PromoteTargetBoolean(CC, N->getValueType(0));
+ return PromoteTargetBoolean(CC, N->getValueType(0));
}
// LDOps: Load operators to build a vector type
// [Start,End) the list of loads to use.
static SDValue BuildVectorFromScalar(SelectionDAG& DAG, EVT VecTy,
- SmallVector<SDValue, 16>& LdOps,
+ SmallVectorImpl<SDValue> &LdOps,
unsigned Start, unsigned End) {
- DebugLoc dl = LdOps[Start].getDebugLoc();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ SDLoc dl(LdOps[Start]);
EVT LdTy = LdOps[Start].getValueType();
unsigned Width = VecTy.getSizeInBits();
unsigned NumElts = Width / LdTy.getSizeInBits();
LdTy = NewLdTy;
}
VecOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, NewVecVT, VecOp, LdOps[i],
- DAG.getIntPtrConstant(Idx++));
+ DAG.getConstant(Idx++, TLI.getVectorIdxTy()));
}
return DAG.getNode(ISD::BITCAST, dl, VecTy, VecOp);
}
-SDValue DAGTypeLegalizer::GenWidenVectorLoads(SmallVector<SDValue, 16> &LdChain,
+SDValue DAGTypeLegalizer::GenWidenVectorLoads(SmallVectorImpl<SDValue> &LdChain,
LoadSDNode *LD) {
// The strategy assumes that we can efficiently load powers of two widths.
// The routines chops the vector into the largest vector loads with the same
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(),LD->getValueType(0));
unsigned WidenWidth = WidenVT.getSizeInBits();
EVT LdVT = LD->getMemoryVT();
- DebugLoc dl = LD->getDebugLoc();
+ SDLoc dl(LD);
assert(LdVT.isVector() && WidenVT.isVector());
assert(LdVT.getVectorElementType() == WidenVT.getVectorElementType());
bool isVolatile = LD->isVolatile();
bool isNonTemporal = LD->isNonTemporal();
bool isInvariant = LD->isInvariant();
+ AAMDNodes AAInfo = LD->getAAInfo();
int LdWidth = LdVT.getSizeInBits();
int WidthDiff = WidenWidth - LdWidth; // Difference
EVT NewVT = FindMemType(DAG, TLI, LdWidth, WidenVT, LdAlign, WidthDiff);
int NewVTWidth = NewVT.getSizeInBits();
SDValue LdOp = DAG.getLoad(NewVT, dl, Chain, BasePtr, LD->getPointerInfo(),
- isVolatile, isNonTemporal, isInvariant, Align);
+ isVolatile, isNonTemporal, isInvariant, Align,
+ AAInfo);
LdChain.push_back(LdOp.getValue(1));
// Check if we can load the element with one instruction
ConcatOps[0] = LdOp;
for (unsigned i = 1; i != NumConcat; ++i)
ConcatOps[i] = UndefVal;
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, &ConcatOps[0],
- NumConcat);
+ return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, ConcatOps);
}
// Load vector by using multiple loads from largest vector to scalar
unsigned Increment = NewVTWidth / 8;
Offset += Increment;
BasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
- DAG.getIntPtrConstant(Increment));
+ DAG.getConstant(Increment, BasePtr.getValueType()));
+ SDValue L;
if (LdWidth < NewVTWidth) {
// Our current type we are using is too large, find a better size
NewVT = FindMemType(DAG, TLI, LdWidth, WidenVT, LdAlign, WidthDiff);
NewVTWidth = NewVT.getSizeInBits();
+ L = DAG.getLoad(NewVT, dl, Chain, BasePtr,
+ LD->getPointerInfo().getWithOffset(Offset), isVolatile,
+ isNonTemporal, isInvariant, MinAlign(Align, Increment),
+ AAInfo);
+ LdChain.push_back(L.getValue(1));
+ if (L->getValueType(0).isVector()) {
+ SmallVector<SDValue, 16> Loads;
+ Loads.push_back(L);
+ unsigned size = L->getValueSizeInBits(0);
+ while (size < LdOp->getValueSizeInBits(0)) {
+ Loads.push_back(DAG.getUNDEF(L->getValueType(0)));
+ size += L->getValueSizeInBits(0);
+ }
+ L = DAG.getNode(ISD::CONCAT_VECTORS, dl, LdOp->getValueType(0), Loads);
+ }
+ } else {
+ L = DAG.getLoad(NewVT, dl, Chain, BasePtr,
+ LD->getPointerInfo().getWithOffset(Offset), isVolatile,
+ isNonTemporal, isInvariant, MinAlign(Align, Increment),
+ AAInfo);
+ LdChain.push_back(L.getValue(1));
}
- SDValue LdOp = DAG.getLoad(NewVT, dl, Chain, BasePtr,
- LD->getPointerInfo().getWithOffset(Offset),
- isVolatile,
- isNonTemporal, isInvariant,
- MinAlign(Align, Increment));
- LdChain.push_back(LdOp.getValue(1));
- LdOps.push_back(LdOp);
+ LdOps.push_back(L);
+
LdWidth -= NewVTWidth;
}
if (NewLdTy != LdTy) {
// Create a larger vector
ConcatOps[End-1] = DAG.getNode(ISD::CONCAT_VECTORS, dl, NewLdTy,
- &ConcatOps[Idx], End - Idx);
+ makeArrayRef(&ConcatOps[Idx], End - Idx));
Idx = End - 1;
LdTy = NewLdTy;
}
if (WidenWidth == LdTy.getSizeInBits()*(End - Idx))
return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT,
- &ConcatOps[Idx], End - Idx);
+ makeArrayRef(&ConcatOps[Idx], End - Idx));
// We need to fill the rest with undefs to build the vector
unsigned NumOps = WidenWidth / LdTy.getSizeInBits();
for (; i != NumOps; ++i)
WidenOps[i] = UndefVal;
}
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, &WidenOps[0],NumOps);
+ return DAG.getNode(ISD::CONCAT_VECTORS, dl, WidenVT, WidenOps);
}
SDValue
-DAGTypeLegalizer::GenWidenVectorExtLoads(SmallVector<SDValue, 16>& LdChain,
- LoadSDNode * LD,
+DAGTypeLegalizer::GenWidenVectorExtLoads(SmallVectorImpl<SDValue> &LdChain,
+ LoadSDNode *LD,
ISD::LoadExtType ExtType) {
// For extension loads, it may not be more efficient to chop up the vector
// and then extended it. Instead, we unroll the load and build a new vector.
EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(),LD->getValueType(0));
EVT LdVT = LD->getMemoryVT();
- DebugLoc dl = LD->getDebugLoc();
+ SDLoc dl(LD);
assert(LdVT.isVector() && WidenVT.isVector());
// Load information
unsigned Align = LD->getAlignment();
bool isVolatile = LD->isVolatile();
bool isNonTemporal = LD->isNonTemporal();
+ bool isInvariant = LD->isInvariant();
+ AAMDNodes AAInfo = LD->getAAInfo();
EVT EltVT = WidenVT.getVectorElementType();
EVT LdEltVT = LdVT.getVectorElementType();
unsigned Increment = LdEltVT.getSizeInBits() / 8;
Ops[0] = DAG.getExtLoad(ExtType, dl, EltVT, Chain, BasePtr,
LD->getPointerInfo(),
- LdEltVT, isVolatile, isNonTemporal, Align);
+ LdEltVT, isVolatile, isNonTemporal, isInvariant,
+ Align, AAInfo);
LdChain.push_back(Ops[0].getValue(1));
unsigned i = 0, Offset = Increment;
for (i=1; i < NumElts; ++i, Offset += Increment) {
SDValue NewBasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(),
- BasePtr, DAG.getIntPtrConstant(Offset));
+ BasePtr,
+ DAG.getConstant(Offset,
+ BasePtr.getValueType()));
Ops[i] = DAG.getExtLoad(ExtType, dl, EltVT, Chain, NewBasePtr,
LD->getPointerInfo().getWithOffset(Offset), LdEltVT,
- isVolatile, isNonTemporal, Align);
+ isVolatile, isNonTemporal, isInvariant, Align,
+ AAInfo);
LdChain.push_back(Ops[i].getValue(1));
}
for (; i != WidenNumElts; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, &Ops[0], Ops.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, WidenVT, Ops);
}
-void DAGTypeLegalizer::GenWidenVectorStores(SmallVector<SDValue, 16>& StChain,
+void DAGTypeLegalizer::GenWidenVectorStores(SmallVectorImpl<SDValue> &StChain,
StoreSDNode *ST) {
// The strategy assumes that we can efficiently store powers of two widths.
// The routines chops the vector into the largest vector stores with the same
unsigned Align = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
+ AAMDNodes AAInfo = ST->getAAInfo();
SDValue ValOp = GetWidenedVector(ST->getValue());
- DebugLoc dl = ST->getDebugLoc();
+ SDLoc dl(ST);
EVT StVT = ST->getMemoryVT();
unsigned StWidth = StVT.getSizeInBits();
unsigned NumVTElts = NewVT.getVectorNumElements();
do {
SDValue EOp = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, NewVT, ValOp,
- DAG.getIntPtrConstant(Idx));
+ DAG.getConstant(Idx, TLI.getVectorIdxTy()));
StChain.push_back(DAG.getStore(Chain, dl, EOp, BasePtr,
ST->getPointerInfo().getWithOffset(Offset),
isVolatile, isNonTemporal,
- MinAlign(Align, Offset)));
+ MinAlign(Align, Offset), AAInfo));
StWidth -= NewVTWidth;
Offset += Increment;
Idx += NumVTElts;
BasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
- DAG.getIntPtrConstant(Increment));
+ DAG.getConstant(Increment, BasePtr.getValueType()));
} while (StWidth != 0 && StWidth >= NewVTWidth);
} else {
// Cast the vector to the scalar type we can store
Idx = Idx * ValEltWidth / NewVTWidth;
do {
SDValue EOp = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, NewVT, VecOp,
- DAG.getIntPtrConstant(Idx++));
+ DAG.getConstant(Idx++, TLI.getVectorIdxTy()));
StChain.push_back(DAG.getStore(Chain, dl, EOp, BasePtr,
ST->getPointerInfo().getWithOffset(Offset),
isVolatile, isNonTemporal,
- MinAlign(Align, Offset)));
+ MinAlign(Align, Offset), AAInfo));
StWidth -= NewVTWidth;
Offset += Increment;
BasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
- DAG.getIntPtrConstant(Increment));
- } while (StWidth != 0 && StWidth >= NewVTWidth);
+ DAG.getConstant(Increment, BasePtr.getValueType()));
+ } while (StWidth != 0 && StWidth >= NewVTWidth);
// Restore index back to be relative to the original widen element type
Idx = Idx * NewVTWidth / ValEltWidth;
}
}
void
-DAGTypeLegalizer::GenWidenVectorTruncStores(SmallVector<SDValue, 16>& StChain,
+DAGTypeLegalizer::GenWidenVectorTruncStores(SmallVectorImpl<SDValue> &StChain,
StoreSDNode *ST) {
// For extension loads, it may not be more efficient to truncate the vector
// and then store it. Instead, we extract each element and then store it.
unsigned Align = ST->getAlignment();
bool isVolatile = ST->isVolatile();
bool isNonTemporal = ST->isNonTemporal();
+ AAMDNodes AAInfo = ST->getAAInfo();
SDValue ValOp = GetWidenedVector(ST->getValue());
- DebugLoc dl = ST->getDebugLoc();
+ SDLoc dl(ST);
EVT StVT = ST->getMemoryVT();
EVT ValVT = ValOp.getValueType();
unsigned Increment = ValEltVT.getSizeInBits() / 8;
unsigned NumElts = StVT.getVectorNumElements();
SDValue EOp = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ValEltVT, ValOp,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
StChain.push_back(DAG.getTruncStore(Chain, dl, EOp, BasePtr,
ST->getPointerInfo(), StEltVT,
- isVolatile, isNonTemporal, Align));
+ isVolatile, isNonTemporal, Align,
+ AAInfo));
unsigned Offset = Increment;
for (unsigned i=1; i < NumElts; ++i, Offset += Increment) {
SDValue NewBasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(),
- BasePtr, DAG.getIntPtrConstant(Offset));
+ BasePtr, DAG.getConstant(Offset,
+ BasePtr.getValueType()));
SDValue EOp = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ValEltVT, ValOp,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
StChain.push_back(DAG.getTruncStore(Chain, dl, EOp, NewBasePtr,
ST->getPointerInfo().getWithOffset(Offset),
StEltVT, isVolatile, isNonTemporal,
- MinAlign(Align, Offset)));
+ MinAlign(Align, Offset), AAInfo));
}
}
EVT InVT = InOp.getValueType();
assert(InVT.getVectorElementType() == NVT.getVectorElementType() &&
"input and widen element type must match");
- DebugLoc dl = InOp.getDebugLoc();
+ SDLoc dl(InOp);
// Check if InOp already has the right width.
if (InVT == NVT)
for (unsigned i = 1; i != NumConcat; ++i)
Ops[i] = UndefVal;
- return DAG.getNode(ISD::CONCAT_VECTORS, dl, NVT, &Ops[0], NumConcat);
+ return DAG.getNode(ISD::CONCAT_VECTORS, dl, NVT, Ops);
}
if (WidenNumElts < InNumElts && InNumElts % WidenNumElts)
return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, NVT, InOp,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, TLI.getVectorIdxTy()));
// Fall back to extract and build.
SmallVector<SDValue, 16> Ops(WidenNumElts);
unsigned Idx;
for (Idx = 0; Idx < MinNumElts; ++Idx)
Ops[Idx] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, InOp,
- DAG.getIntPtrConstant(Idx));
+ DAG.getConstant(Idx, TLI.getVectorIdxTy()));
SDValue UndefVal = DAG.getUNDEF(EltVT);
for ( ; Idx < WidenNumElts; ++Idx)
Ops[Idx] = UndefVal;
- return DAG.getNode(ISD::BUILD_VECTOR, dl, NVT, &Ops[0], WidenNumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, NVT, Ops);
}