[RISCV] Fix corner cases after #170070 (#170438)

There are two fixes:

1. Clear kill flags for `FalseReg` in foldVMergeToMask or we can't
pass the MachineVerifier because of using a killed virtual register.
2. Restrict `lookThruCopies` to only look through COPYs with
one non-debug use.

This was found when backporting #170070 to 21.x branch.

Author: wangpc-pp

llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp

@@ -73,7 +73,7 @@ class RISCVVectorPeephole : public MachineFunctionPass {
   bool isAllOnesMask(const MachineInstr *MaskDef) const;
   std::optional<unsigned> getConstant(const MachineOperand &VL) const;
   bool ensureDominates(const MachineOperand &Use, MachineInstr &Src) const;
-  Register lookThruCopies(Register Reg) const;
+  Register lookThruCopies(Register Reg, bool OneUseOnly = false) const;
 };
 
 } // namespace
@@ -389,13 +389,16 @@ bool RISCVVectorPeephole::convertAllOnesVMergeToVMv(MachineInstr &MI) const {
 
 // If \p Reg is defined by one or more COPYs of virtual registers, traverses
 // the chain and returns the root non-COPY source.
-Register RISCVVectorPeephole::lookThruCopies(Register Reg) const {
+Register RISCVVectorPeephole::lookThruCopies(Register Reg,
+                                             bool OneUseOnly) const {
   while (MachineInstr *Def = MRI->getUniqueVRegDef(Reg)) {
     if (!Def->isFullCopy())
       break;
     Register Src = Def->getOperand(1).getReg();
     if (!Src.isVirtual())
       break;
+    if (OneUseOnly && !MRI->hasOneNonDBGUse(Reg))
+      break;
     Reg = Src;
   }
   return Reg;
@@ -715,7 +718,8 @@ bool RISCVVectorPeephole::foldVMergeToMask(MachineInstr &MI) const {
 
   Register PassthruReg = lookThruCopies(MI.getOperand(1).getReg());
   Register FalseReg = lookThruCopies(MI.getOperand(2).getReg());
-  Register TrueReg = lookThruCopies(MI.getOperand(3).getReg());
+  Register TrueReg =
+      lookThruCopies(MI.getOperand(3).getReg(), /*OneUseOnly=*/true);
   if (!TrueReg.isVirtual() || !MRI->hasOneUse(TrueReg))
     return false;
   MachineInstr &True = *MRI->getUniqueVRegDef(TrueReg);
@@ -834,6 +838,8 @@ bool RISCVVectorPeephole::foldVMergeToMask(MachineInstr &MI) const {
     MRI->constrainRegClass(
         MO.getReg(), True.getRegClassConstraint(MO.getOperandNo(), TII, TRI));
   }
+  // We should clear the IsKill flag since we have a new use now.
+  MRI->clearKillFlags(FalseReg);
   MI.eraseFromParent();
 
   return true;

llvm/test/CodeGen/RISCV/rvv/vmerge-peephole.mir

@@ -136,3 +136,48 @@ body: |
     %y:vrnov0 = COPY %x
     %z:vrnov0 = PseudoVMERGE_VVM_M1 %passthru, %passthru, %y, %mask, %avl, 5 /* e32 */
 ...
+---
+name: copy_is_killed
+body: |
+  bb.0:
+    liveins: $v0, $v8, $v9
+    ; CHECK-LABEL: name: copy_is_killed
+    ; CHECK: liveins: $v0, $v8, $v9
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %x:vr = COPY $v8
+    ; CHECK-NEXT: %y:vr = COPY $v9
+    ; CHECK-NEXT: %mask:vmv0 = COPY $v0
+    ; CHECK-NEXT: %add0:vr = PseudoVADD_VV_M1 $noreg, %x, %y, -1, 5 /* e32 */, 3 /* ta, ma */
+    ; CHECK-NEXT: %add1:vrnov0 = COPY %add:vrnov0
+    ; CHECK-NEXT: %merge:vrnov0 = PseudoVOR_VV_M1_MASK %add:vrnov0, %add1, %y, %mask, -1, 5 /* e32 */, 1 /* ta, mu */
+    %x:vr = COPY $v8
+    %y:vr = COPY $v9
+    %mask:vmv0 = COPY $v0
+    %add0:vr = PseudoVADD_VV_M1 $noreg, %x:vr, %y:vr, -1, 5, 3
+    %add1:vrnov0 = COPY killed %add:vr
+    %or:vrnov0 = PseudoVOR_VV_M1 $noreg, %add1:vrnov0, %y:vr, -1, 5, 3
+    %merge:vrnov0 = PseudoVMERGE_VVM_M1 $noreg, %add1:vrnov0, killed %or:vrnov0, killed %mask:vmv0, -1, 5 /* e32 */
+
+...
+---
+name: copy_multiple_use
+body: |
+  bb.0:
+    liveins: $x8, $v0, $v8
+    ; CHECK-LABEL: name: copy_multiple_use
+    ; CHECK: liveins: $x8, $v0, $v8
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %avl:gprnox0 = COPY $x8
+    ; CHECK-NEXT: %passthru:vrnov0 = COPY $v8
+    ; CHECK-NEXT: %x:vrnov0 = PseudoVLE32_V_M1 $noreg, $noreg, %avl, 5 /* e32 */, 2 /* tu, ma */ :: (load unknown-size, align 1)
+    ; CHECK-NEXT: %copy:vrnov0 = COPY %x
+    ; CHECK-NEXT: %mask:vmv0 = COPY $v0
+    ; CHECK-NEXT: PseudoVSE8_V_M1 %copy, $noreg, %avl, 5 /* e32 */
+    ; CHECK-NEXT: %y:vrnov0 = PseudoVMERGE_VVM_M1 %passthru, %passthru, %copy, %mask, %avl, 5 /* e32 */
+    %avl:gprnox0 = COPY $x8
+    %passthru:vrnov0 = COPY $v8
+    %x:vrnov0 = PseudoVLE32_V_M1 $noreg, $noreg, %avl, 5 /* e32 */, 2 /* tu, ma */ :: (load unknown-size)
+    %copy:vrnov0 = COPY %x
+    %mask:vmv0 = COPY $v0
+    PseudoVSE8_V_M1 %copy, $noreg, %avl, 5 /* e8 */
+    %y:vrnov0 = PseudoVMERGE_VVM_M1 %passthru, %passthru, %copy, %mask, %avl, 5 /* e32 */