Merge branch 'main' into users/mingmingl-llvm/samplefdo-profile-formatusers/mingmingl-llvm/samplefdo-profile-format

1 files changed, 22 insertions, 2 deletions

diff --git a/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp b/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp
index 050de3d58a2f..62651185137c 100644
--- a/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp
@@ -745,12 +745,24 @@ bool RISCVVectorPeephole::foldVMergeToMask(MachineInstr &MI) const {
   if (PassthruReg && !isKnownSameDefs(PassthruReg, FalseReg))
     return false;
 
+  std::optional<std::pair<unsigned, unsigned>> NeedsCommute;
+
   // If True has a passthru operand then it needs to be the same as vmerge's
   // False, since False will be used for the result's passthru operand.
   Register TruePassthru = True.getOperand(True.getNumExplicitDefs()).getReg();
   if (RISCVII::isFirstDefTiedToFirstUse(True.getDesc()) && TruePassthru &&
-      !isKnownSameDefs(TruePassthru, FalseReg))
-    return false;
+      !isKnownSameDefs(TruePassthru, FalseReg)) {
+    // If True's passthru != False, check if it uses False in another operand
+    // and try to commute it.
+    int OtherIdx = True.findRegisterUseOperandIdx(FalseReg, TRI);
+    if (OtherIdx == -1)
+      return false;
+    unsigned OpIdx1 = OtherIdx;
+    unsigned OpIdx2 = True.getNumExplicitDefs();
+    if (!TII->findCommutedOpIndices(True, OpIdx1, OpIdx2))
+      return false;
+    NeedsCommute = {OpIdx1, OpIdx2};
+  }
 
   // Make sure it doesn't raise any observable fp exceptions, since changing the
   // active elements will affect how fflags is set.
@@ -796,6 +808,14 @@ bool RISCVVectorPeephole::foldVMergeToMask(MachineInstr &MI) const {
   if (!ensureDominates(MaskOp, True))
     return false;
 
+  if (NeedsCommute) {
+    auto [OpIdx1, OpIdx2] = *NeedsCommute;
+    [[maybe_unused]] bool Commuted =
+        TII->commuteInstruction(True, /*NewMI=*/false, OpIdx1, OpIdx2);
+    assert(Commuted && "Failed to commute True?");
+    Info = RISCV::lookupMaskedIntrinsicByUnmasked(True.getOpcode());
+  }
+
   True.setDesc(TII->get(Info->MaskedPseudo));
 
   // Insert the mask operand.