1 files changed, 65 insertions, 72 deletions

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index cda64ae..6c58989 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -5104,6 +5104,7 @@ static SDValue lowerShuffleViaVRegSplitting(ShuffleVectorSDNode *SVN,
   SDValue V1 = SVN->getOperand(0);
   SDValue V2 = SVN->getOperand(1);
   ArrayRef<int> Mask = SVN->getMask();
+  unsigned NumElts = VT.getVectorNumElements();
 
   // If we don't know exact data layout, not much we can do.  If this
   // is already m1 or smaller, no point in splitting further.
@@ -5120,70 +5121,58 @@ static SDValue lowerShuffleViaVRegSplitting(ShuffleVectorSDNode *SVN,
 
   MVT ElemVT = VT.getVectorElementType();
   unsigned ElemsPerVReg = *VLen / ElemVT.getFixedSizeInBits();
+  unsigned VRegsPerSrc = NumElts / ElemsPerVReg;
+
+  SmallVector<std::pair<int, SmallVector<int>>>
+    OutMasks(VRegsPerSrc, {-1, {}});
+
+  // Check if our mask can be done as a 1-to-1 mapping from source
+  // to destination registers in the group without needing to
+  // write each destination more than once.
+  for (unsigned DstIdx = 0; DstIdx < Mask.size(); DstIdx++) {
+    int DstVecIdx = DstIdx / ElemsPerVReg;
+    int DstSubIdx = DstIdx % ElemsPerVReg;
+    int SrcIdx = Mask[DstIdx];
+    if (SrcIdx < 0 || (unsigned)SrcIdx >= 2 * NumElts)
+      continue;
+    int SrcVecIdx = SrcIdx / ElemsPerVReg;
+    int SrcSubIdx = SrcIdx % ElemsPerVReg;
+    if (OutMasks[DstVecIdx].first == -1)
+      OutMasks[DstVecIdx].first = SrcVecIdx;
+    if (OutMasks[DstVecIdx].first != SrcVecIdx)
+      // Note: This case could easily be handled by keeping track of a chain
+      // of source values and generating two element shuffles below.  This is
+      // less an implementation question, and more a profitability one.
+      return SDValue();
+
+    OutMasks[DstVecIdx].second.resize(ElemsPerVReg, -1);
+    OutMasks[DstVecIdx].second[DstSubIdx] = SrcSubIdx;
+  }
 
   EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
   MVT OneRegVT = MVT::getVectorVT(ElemVT, ElemsPerVReg);
   MVT M1VT = getContainerForFixedLengthVector(DAG, OneRegVT, Subtarget);
   assert(M1VT == getLMUL1VT(M1VT));
   unsigned NumOpElts = M1VT.getVectorMinNumElements();
-  unsigned NormalizedVF = ContainerVT.getVectorMinNumElements();
-  unsigned NumOfSrcRegs = NormalizedVF / NumOpElts;
-  unsigned NumOfDestRegs = NormalizedVF / NumOpElts;
+  SDValue Vec = DAG.getUNDEF(ContainerVT);
   // The following semantically builds up a fixed length concat_vector
   // of the component shuffle_vectors.  We eagerly lower to scalable here
   // to avoid DAG combining it back to a large shuffle_vector again.
   V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
   V2 = convertToScalableVector(ContainerVT, V2, DAG, Subtarget);
-  SmallVector<SDValue> SubRegs(NumOfDestRegs);
-  unsigned RegCnt = 0;
-  unsigned PrevCnt = 0;
-  processShuffleMasks(
-      Mask, NumOfSrcRegs, NumOfDestRegs, NumOfDestRegs,
-      [&]() {
-        PrevCnt = RegCnt;
-        ++RegCnt;
-      },
-      [&, &DAG = DAG](ArrayRef<int> SrcSubMask, unsigned SrcVecIdx,
-                      unsigned DstVecIdx) {
-        SDValue SrcVec = SrcVecIdx >= NumOfSrcRegs ? V2 : V1;
-        unsigned ExtractIdx = (SrcVecIdx % NumOfSrcRegs) * NumOpElts;
-        SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
-                                     DAG.getVectorIdxConstant(ExtractIdx, DL));
-        SubVec = convertFromScalableVector(OneRegVT, SubVec, DAG, Subtarget);
-        SubVec = DAG.getVectorShuffle(OneRegVT, DL, SubVec, SubVec, SrcSubMask);
-        SubRegs[RegCnt] = convertToScalableVector(M1VT, SubVec, DAG, Subtarget);
-        PrevCnt = RegCnt;
-        ++RegCnt;
-      },
-      [&, &DAG = DAG](ArrayRef<int> SrcSubMask, unsigned Idx1, unsigned Idx2) {
-        if (PrevCnt + 1 == RegCnt)
-          ++RegCnt;
-        SDValue SubVec1 = SubRegs[PrevCnt + 1];
-        if (!SubVec1) {
-          SDValue SrcVec = Idx1 >= NumOfSrcRegs ? V2 : V1;
-          unsigned ExtractIdx = (Idx1 % NumOfSrcRegs) * NumOpElts;
-          SubVec1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
-                                DAG.getVectorIdxConstant(ExtractIdx, DL));
-        }
-        SubVec1 = convertFromScalableVector(OneRegVT, SubVec1, DAG, Subtarget);
-        SDValue SrcVec = Idx2 >= NumOfSrcRegs ? V2 : V1;
-        unsigned ExtractIdx = (Idx2 % NumOfSrcRegs) * NumOpElts;
-        SDValue SubVec2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
-                                      DAG.getVectorIdxConstant(ExtractIdx, DL));
-        SubVec2 = convertFromScalableVector(OneRegVT, SubVec2, DAG, Subtarget);
-        SubVec1 =
-            DAG.getVectorShuffle(OneRegVT, DL, SubVec1, SubVec2, SrcSubMask);
-        SubVec1 = convertToScalableVector(M1VT, SubVec1, DAG, Subtarget);
-        SubRegs[PrevCnt + 1] = SubVec1;
-      });
-  assert(RegCnt == NumOfDestRegs && "Whole vector must be processed");
-  SDValue Vec = DAG.getUNDEF(ContainerVT);
-  for (auto [I, V] : enumerate(SubRegs)) {
-    if (!V)
+  for (unsigned DstVecIdx = 0 ; DstVecIdx < OutMasks.size(); DstVecIdx++) {
+    auto &[SrcVecIdx, SrcSubMask] = OutMasks[DstVecIdx];
+    if (SrcVecIdx == -1)
       continue;
-    unsigned InsertIdx = I * NumOpElts;
-
-    Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, ContainerVT, Vec, V,
+    unsigned ExtractIdx = (SrcVecIdx % VRegsPerSrc) * NumOpElts;
+    SDValue SrcVec = (unsigned)SrcVecIdx >= VRegsPerSrc ? V2 : V1;
+    SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, M1VT, SrcVec,
+                                 DAG.getVectorIdxConstant(ExtractIdx, DL));
+    SubVec = convertFromScalableVector(OneRegVT, SubVec, DAG, Subtarget);
+    SubVec = DAG.getVectorShuffle(OneRegVT, DL, SubVec, SubVec, SrcSubMask);
+    SubVec = convertToScalableVector(M1VT, SubVec, DAG, Subtarget);
+    unsigned InsertIdx = DstVecIdx * NumOpElts;
+    Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, ContainerVT, Vec, SubVec,
                       DAG.getVectorIdxConstant(InsertIdx, DL));
   }
   return convertFromScalableVector(VT, Vec, DAG, Subtarget);
@@ -10165,7 +10154,10 @@ SDValue RISCVTargetLowering::lowerVectorMaskVecReduction(SDValue Op,
   case ISD::VP_REDUCE_AND: {
     // vcpop ~x == 0
     SDValue TrueMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
-    Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL);
+    if (IsVP || VecVT.isFixedLengthVector())
+      Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL);
+    else
+      Vec = DAG.getNode(ISD::XOR, DL, ContainerVT, Vec, TrueMask);
     Vec = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Vec, Mask, VL);
     CC = ISD::SETEQ;
     break;
@@ -12674,8 +12666,7 @@ SDValue RISCVTargetLowering::lowerGET_ROUNDING(SDValue Op,
   const MVT XLenVT = Subtarget.getXLenVT();
   SDLoc DL(Op);
   SDValue Chain = Op->getOperand(0);
-  SDValue SysRegNo = DAG.getTargetConstant(
-      RISCVSysReg::lookupSysRegByName("FRM")->Encoding, DL, XLenVT);
+  SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::frm, DL, XLenVT);
   SDVTList VTs = DAG.getVTList(XLenVT, MVT::Other);
   SDValue RM = DAG.getNode(RISCVISD::READ_CSR, DL, VTs, Chain, SysRegNo);
 
@@ -12706,8 +12697,7 @@ SDValue RISCVTargetLowering::lowerSET_ROUNDING(SDValue Op,
   SDLoc DL(Op);
   SDValue Chain = Op->getOperand(0);
   SDValue RMValue = Op->getOperand(1);
-  SDValue SysRegNo = DAG.getTargetConstant(
-      RISCVSysReg::lookupSysRegByName("FRM")->Encoding, DL, XLenVT);
+  SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::frm, DL, XLenVT);
 
   // Encoding used for rounding mode in RISC-V differs from that used in
   // FLT_ROUNDS. To convert it the C rounding mode is used as an index in
@@ -12910,15 +12900,11 @@ void RISCVTargetLowering::ReplaceNodeResults(SDNode *N,
     SDValue LoCounter, HiCounter;
     MVT XLenVT = Subtarget.getXLenVT();
     if (N->getOpcode() == ISD::READCYCLECOUNTER) {
-      LoCounter = DAG.getTargetConstant(
-          RISCVSysReg::lookupSysRegByName("CYCLE")->Encoding, DL, XLenVT);
-      HiCounter = DAG.getTargetConstant(
-          RISCVSysReg::lookupSysRegByName("CYCLEH")->Encoding, DL, XLenVT);
+      LoCounter = DAG.getTargetConstant(RISCVSysReg::cycle, DL, XLenVT);
+      HiCounter = DAG.getTargetConstant(RISCVSysReg::cycleh, DL, XLenVT);
     } else {
-      LoCounter = DAG.getTargetConstant(
-          RISCVSysReg::lookupSysRegByName("TIME")->Encoding, DL, XLenVT);
-      HiCounter = DAG.getTargetConstant(
-          RISCVSysReg::lookupSysRegByName("TIMEH")->Encoding, DL, XLenVT);
+      LoCounter = DAG.getTargetConstant(RISCVSysReg::time, DL, XLenVT);
+      HiCounter = DAG.getTargetConstant(RISCVSysReg::timeh, DL, XLenVT);
     }
     SDVTList VTs = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
     SDValue RCW = DAG.getNode(RISCVISD::READ_COUNTER_WIDE, DL, VTs,
@@ -18397,6 +18383,15 @@ bool RISCVTargetLowering::isDesirableToCommuteWithShift(
 
     auto *C1 = dyn_cast<ConstantSDNode>(N0->getOperand(1));
     auto *C2 = dyn_cast<ConstantSDNode>(N->getOperand(1));
+
+    // Bail if we might break a sh{1,2,3}add pattern.
+    if (Subtarget.hasStdExtZba() && C2 && C2->getZExtValue() >= 1 &&
+        C2->getZExtValue() <= 3 && N->hasOneUse() &&
+        N->user_begin()->getOpcode() == ISD::ADD &&
+        !isUsedByLdSt(*N->user_begin(), nullptr) &&
+        !isa<ConstantSDNode>(N->user_begin()->getOperand(1)))
+      return false;
+
     if (C1 && C2) {
       const APInt &C1Int = C1->getAPIntValue();
       APInt ShiftedC1Int = C1Int << C2->getAPIntValue();
@@ -20278,13 +20273,11 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
   for (auto &Reg : RegsToPass)
     Ops.push_back(DAG.getRegister(Reg.first, Reg.second.getValueType()));
 
-  if (!IsTailCall) {
-    // Add a register mask operand representing the call-preserved registers.
-    const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
-    const uint32_t *Mask = TRI->getCallPreservedMask(MF, CallConv);
-    assert(Mask && "Missing call preserved mask for calling convention");
-    Ops.push_back(DAG.getRegisterMask(Mask));
-  }
+  // Add a register mask operand representing the call-preserved registers.
+  const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
+  const uint32_t *Mask = TRI->getCallPreservedMask(MF, CallConv);
+  assert(Mask && "Missing call preserved mask for calling convention");
+  Ops.push_back(DAG.getRegisterMask(Mask));
 
   // Glue the call to the argument copies, if any.
   if (Glue.getNode())