SDValue visitFP_EXTEND(SDNode *N);
SDValue visitFNEG(SDNode *N);
SDValue visitFABS(SDNode *N);
+ SDValue visitFCEIL(SDNode *N);
+ SDValue visitFTRUNC(SDNode *N);
+ SDValue visitFFLOOR(SDNode *N);
SDValue visitBRCOND(SDNode *N);
SDValue visitBR_CC(SDNode *N);
SDValue visitLOAD(SDNode *N);
case ISD::FP_EXTEND: return visitFP_EXTEND(N);
case ISD::FNEG: return visitFNEG(N);
case ISD::FABS: return visitFABS(N);
+ case ISD::FFLOOR: return visitFFLOOR(N);
+ case ISD::FCEIL: return visitFCEIL(N);
+ case ISD::FTRUNC: return visitFTRUNC(N);
case ISD::BRCOND: return visitBRCOND(N);
case ISD::BR_CC: return visitBR_CC(N);
case ISD::LOAD: return visitLOAD(N);
if ((DAG.getTarget().Options.AllowFPOpFusion == FPOpFusion::Fast ||
DAG.getTarget().Options.UnsafeFPMath) &&
DAG.getTarget().getTargetLowering()->isFMAFasterThanMulAndAdd(VT) &&
- TLI.isOperationLegal(ISD::FMA, VT)) {
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) {
// fold (fadd (fmul x, y), z) -> (fma x, y, z)
if (N0.getOpcode() == ISD::FMUL && N0->hasOneUse()) {
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
EVT VT = N->getValueType(0);
+ DebugLoc dl = N->getDebugLoc();
// fold vector ops
if (VT.isVector()) {
if (isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options))
return GetNegatedExpression(N1, DAG, LegalOperations);
if (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))
- return DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT, N1);
+ return DAG.getNode(ISD::FNEG, dl, VT, N1);
}
// fold (fsub A, (fneg B)) -> (fadd A, B)
if (isNegatibleForFree(N1, LegalOperations, TLI, &DAG.getTarget().Options))
- return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N0,
+ return DAG.getNode(ISD::FADD, dl, VT, N0,
GetNegatedExpression(N1, DAG, LegalOperations));
// If 'unsafe math' is enabled, fold
if ((DAG.getTarget().Options.AllowFPOpFusion == FPOpFusion::Fast ||
DAG.getTarget().Options.UnsafeFPMath) &&
DAG.getTarget().getTargetLowering()->isFMAFasterThanMulAndAdd(VT) &&
- TLI.isOperationLegal(ISD::FMA, VT)) {
+ TLI.isOperationLegalOrCustom(ISD::FMA, VT)) {
// fold (fsub (fmul x, y), z) -> (fma x, y, (fneg z))
if (N0.getOpcode() == ISD::FMUL && N0->hasOneUse()) {
- return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
+ return DAG.getNode(ISD::FMA, dl, VT,
N0.getOperand(0), N0.getOperand(1),
- DAG.getNode(ISD::FNEG, N1->getDebugLoc(), VT, N1));
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
}
// fold (fsub x, (fmul y, z)) -> (fma (fneg y), z, x)
// Note: Commutes FSUB operands.
if (N1.getOpcode() == ISD::FMUL && N1->hasOneUse()) {
- return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
- DAG.getNode(ISD::FNEG, N1->getDebugLoc(), VT,
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT,
N1.getOperand(0)),
N1.getOperand(1), N0);
}
+
+ // fold (fsub (-(fmul, x, y)), z) -> (fma (fneg x), y, (fneg z))
+ if (N0.getOpcode() == ISD::FNEG &&
+ N0.getOperand(0).getOpcode() == ISD::FMUL &&
+ N0->hasOneUse() && N0.getOperand(0).hasOneUse()) {
+ SDValue N00 = N0.getOperand(0).getOperand(0);
+ SDValue N01 = N0.getOperand(0).getOperand(1);
+ return DAG.getNode(ISD::FMA, dl, VT,
+ DAG.getNode(ISD::FNEG, dl, VT, N00), N01,
+ DAG.getNode(ISD::FNEG, dl, VT, N1));
+ }
}
return SDValue();
return SDValue();
}
+SDValue DAGCombiner::visitFCEIL(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (fceil c1) -> fceil(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FCEIL, N->getDebugLoc(), VT, N0);
+
+ return SDValue();
+}
+
+SDValue DAGCombiner::visitFTRUNC(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (ftrunc c1) -> ftrunc(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FTRUNC, N->getDebugLoc(), VT, N0);
+
+ return SDValue();
+}
+
+SDValue DAGCombiner::visitFFLOOR(SDNode *N) {
+ SDValue N0 = N->getOperand(0);
+ ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
+ EVT VT = N->getValueType(0);
+
+ // fold (ffloor c1) -> ffloor(c1)
+ if (N0CFP && VT != MVT::ppcf128)
+ return DAG.getNode(ISD::FFLOOR, N->getDebugLoc(), VT, N0);
+
+ return SDValue();
+}
+
SDValue DAGCombiner::visitFABS(SDNode *N) {
SDValue N0 = N->getOperand(0);
ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
if (VecIn1.getValueType().getSizeInBits()*2 != VT.getSizeInBits())
return SDValue();
- // Widen the input vector by adding undef values.
- VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, N->getDebugLoc(), VT,
- VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
+ // If the element type of the input vector is not the same as
+ // the output element type, make concat_vectors based on input element
+ // type and then bitcast it to the output vector type.
+ //
+ // In another words avoid nodes like this:
+ // <NODE> v16i8 = concat_vectors v4i16 v4i16
+ // Replace it with this one:
+ // <NODE0> v8i16 = concat_vectors v4i16 v4i16
+ // <NODE1> v16i8 = bitcast NODE0
+ EVT ItemType = VecIn1.getValueType().getVectorElementType();
+ if (ItemType != VT.getVectorElementType()) {
+ EVT ConcatVT = EVT::getVectorVT(*DAG.getContext(),
+ ItemType,
+ VecIn1.getValueType().getVectorNumElements()*2);
+ // Widen the input vector by adding undef values.
+ VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, dl, ConcatVT,
+ VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
+ VecIn1 = DAG.getNode(ISD::BITCAST, dl, VT, VecIn1);
+ } else
+ // Widen the input vector by adding undef values.
+ VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, dl, VT,
+ VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
+
}
// If VecIn2 is unused then change it to undef.