[RISCV][DAGCombiner] Fix potential missed combine in VL->VW extension (#168026)

The previous implementation of `combineOp_VLToVWOp_VL` manually replaced
old
nodes with newly created widened nodes, but only added the new node
itself to
the `DAGCombiner` worklist. Since the users of the new node were not
added,
some combine opportunities could be missed when external `DAGCombiner`
passes
expected those users to be reconsidered.

This patch replaces the custom replacement logic with a call to
`DCI.CombineTo()`, which performs node replacement in a way consistent
with
`DAGCombiner::Run`:
- Replace all uses of the old node.
- Add the new node and its users to the worklist.
- Clean up unused nodes when appropriate.

Using `CombineTo` ensures that `combineOp_VLToVWOp_VL` behaves
consistently with
the standard `DAGCombiner` update model, avoiding discrepancies between
the
private worklist inside this routine and the global worklist managed by
the
combiner.

This resolves missed combine cases involving VL -> VW operator widening.

---------

Co-authored-by: Kai Lin <[email protected]>

Author: OMG-link

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -18396,8 +18396,7 @@ static SDValue combineOp_VLToVWOp_VL(SDNode *N,
     }
   }
   for (std::pair<SDValue, SDValue> OldNewValues : ValuesToReplace) {
-    DAG.ReplaceAllUsesOfValueWith(OldNewValues.first, OldNewValues.second);
-    DCI.AddToWorklist(OldNewValues.second.getNode());
+    DCI.CombineTo(OldNewValues.first.getNode(), OldNewValues.second);
   }
   return InputRootReplacement;
 }

llvm/test/CodeGen/RISCV/rvv/combine-vl-vw-macc.ll

@@ -0,0 +1,54 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define void @matmul_min(ptr %vptr, ptr %scalars, ptr %acc0_ptr, ptr %acc1_ptr) {
+; CHECK-LABEL: matmul_min:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    li a4, 64
+; CHECK-NEXT:    li a5, 32
+; CHECK-NEXT:    vsetvli zero, a5, e8, m2, ta, ma
+; CHECK-NEXT:    vle8.v v16, (a0)
+; CHECK-NEXT:    lb a0, 0(a1)
+; CHECK-NEXT:    lb a1, 1(a1)
+; CHECK-NEXT:    vsetvli zero, a4, e8, m4, ta, ma
+; CHECK-NEXT:    vle8.v v8, (a2)
+; CHECK-NEXT:    vle8.v v12, (a3)
+; CHECK-NEXT:    vsetvli zero, a5, e8, m2, ta, ma
+; CHECK-NEXT:    vwmacc.vx v8, a0, v16
+; CHECK-NEXT:    vwmacc.vx v12, a1, v16
+; CHECK-NEXT:    vsetvli zero, a4, e8, m4, ta, ma
+; CHECK-NEXT:    vse8.v v8, (a2)
+; CHECK-NEXT:    vse8.v v12, (a3)
+; CHECK-NEXT:    ret
+entry:
+  %acc0 = load <32 x i16>, ptr %acc0_ptr, align 1
+  %acc1 = load <32 x i16>, ptr %acc1_ptr, align 1
+
+  %v8 = load <32 x i8>, ptr %vptr, align 1
+  %v16 = sext <32 x i8> %v8 to <32 x i16>
+
+  %s0_ptr = getelementptr i8, ptr %scalars, i32 0
+  %s0_i8 = load i8, ptr %s0_ptr, align 1
+  %s0_i16 = sext i8 %s0_i8 to i16
+  %tmp0 = insertelement <32 x i16> poison, i16 %s0_i16, i32 0
+  %splat0 = shufflevector <32 x i16> %tmp0, <32 x i16> poison, <32 x i32> zeroinitializer
+  %mul0 = mul <32 x i16> %splat0, %v16
+  %add0 = add <32 x i16> %mul0, %acc0
+
+  %s1_ptr = getelementptr i8, ptr %scalars, i32 1
+  %s1_i8 = load i8, ptr %s1_ptr, align 1
+  %s1_i16 = sext i8 %s1_i8 to i16
+  %tmp1 = insertelement <32 x i16> poison, i16 %s1_i16, i32 0
+  %splat1 = shufflevector <32 x i16> %tmp1, <32 x i16> poison, <32 x i32> zeroinitializer
+  %mul1 = mul <32 x i16> %splat1, %v16
+  %add1 = add <32 x i16> %mul1, %acc1
+
+  store <32 x i16> %add0, ptr %acc0_ptr, align 1
+  store <32 x i16> %add1, ptr %acc1_ptr, align 1
+
+  ret void
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; RV32: {{.*}}
+; RV64: {{.*}}