[RISCV] Fold vmv.v.v into vleNff.v (#143981)

We currently already fold vmerge.vvm into vleNff.v via
RISCVDAGToDAGISel::performCombineVMergeAndVOps, so this teaches
RISCVVectorPeephole::foldVMV_V_V to do the same.

Author: lukel97

llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp

@@ -611,7 +611,7 @@ bool RISCVVectorPeephole::foldVMV_V_V(MachineInstr &MI) {
 
   MachineInstr *Src = MRI->getVRegDef(MI.getOperand(2).getReg());
   if (!Src || Src->hasUnmodeledSideEffects() ||
-      Src->getParent() != MI.getParent() || Src->getNumDefs() != 1 ||
+      Src->getParent() != MI.getParent() ||
       !RISCVII::isFirstDefTiedToFirstUse(Src->getDesc()) ||
       !RISCVII::hasVLOp(Src->getDesc().TSFlags) ||
       !RISCVII::hasVecPolicyOp(Src->getDesc().TSFlags))
@@ -622,7 +622,7 @@ bool RISCVVectorPeephole::foldVMV_V_V(MachineInstr &MI) {
     return false;
 
   // Src needs to have the same passthru as VMV_V_V
-  MachineOperand &SrcPassthru = Src->getOperand(1);
+  MachineOperand &SrcPassthru = Src->getOperand(Src->getNumExplicitDefs());
   if (SrcPassthru.getReg() != RISCV::NoRegister &&
       SrcPassthru.getReg() != Passthru.getReg())
     return false;
@@ -643,7 +643,8 @@ bool RISCVVectorPeephole::foldVMV_V_V(MachineInstr &MI) {
     // If Src is masked then its passthru needs to be in VRNoV0.
     if (Passthru.getReg() != RISCV::NoRegister)
       MRI->constrainRegClass(Passthru.getReg(),
-                             TII->getRegClass(Src->getDesc(), 1, TRI,
+                             TII->getRegClass(Src->getDesc(),
+                                              SrcPassthru.getOperandNo(), TRI,
                                               *Src->getParent()->getParent()));
   }
 

llvm/test/CodeGen/RISCV/rvv/vmv.v.v-peephole.ll

@@ -206,3 +206,19 @@ define <vscale x 1 x i64> @undef_passthru(<vscale x 1 x i64> %passthru, <vscale
   %b = call <vscale x 1 x i64> @llvm.riscv.vmv.v.v.nxv1i64(<vscale x 1 x i64> undef, <vscale x 1 x i64> %a, iXLen %avl)
   ret <vscale x 1 x i64> %b
 }
+
+; Check that we can fold into vle64ff.v even if we need to move it past the
+; passthru and it's safe.
+define <vscale x 1 x i64> @vleff_move_past_passthru(ptr %p, ptr %q, iXLen %avl) {
+; CHECK-LABEL: vleff_move_past_passthru:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vl1re64.v v8, (a1)
+; CHECK-NEXT:    vsetvli zero, a2, e64, m1, tu, ma
+; CHECK-NEXT:    vle64ff.v v8, (a0)
+; CHECK-NEXT:    ret
+  %a = call { <vscale x 1 x i64>, iXLen } @llvm.riscv.vleff(<vscale x 1 x i64> poison, ptr %p, iXLen %avl)
+  %vec = extractvalue { <vscale x 1 x i64>, iXLen } %a, 0
+  %passthru = load <vscale x 1 x i64>, ptr %q
+  %b = call <vscale x 1 x i64> @llvm.riscv.vmv.v.v.nxv1i64(<vscale x 1 x i64> %passthru, <vscale x 1 x i64> %vec, iXLen %avl)
+  ret <vscale x 1 x i64> %b
+}

llvm/test/CodeGen/RISCV/rvv/vmv.v.v-peephole.mir

@@ -135,3 +135,20 @@ body: |
     %3:vrnov0 = PseudoVMV_V_V_MF2 $noreg, %2, 0, 5 /* e32 */, 0 /* tu, mu */
     %7:vmv0 = COPY $v8
     %6:vrnov0 = PseudoVLSE32_V_MF2_MASK %3, $noreg, $noreg, %7, 0, 5 /* e32 */, 0 /* tu, mu */ :: (load unknown-size, align 4)
+...
+---
+name: move_vleff
+body: |
+  bb.0:
+    liveins: $v8
+    ; CHECK-LABEL: name: move_vleff
+    ; CHECK: liveins: $v8
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: %passthru:vr = COPY $v8
+    ; CHECK-NEXT: %x:vr, %vl:gpr = PseudoVLE32FF_V_M1 %passthru, $noreg, 4, 5 /* e32 */, 0 /* tu, mu */ :: (load unknown-size, align 1)
+    ; CHECK-NEXT: %y:gpr = ADDI $x0, 1
+    %x:vr, %vl:gpr = PseudoVLE32FF_V_M1 $noreg, $noreg, 4, 5 /* e32 */, 0 /* tu, mu */ :: (load unknown-size)
+    %passthru:vr = COPY $v8
+    %y:gpr = ADDI $x0, 1
+    %z:vr = PseudoVMV_V_V_M1 %passthru, %x, 4, 5 /* e32 */, 0 /* tu, mu */
+...