[RISCV] Handle more cases when combining (vfmv.s.f (extract_subvector X, 0)) (#154175)

Previously, we fold `(vfmv.s.f (extract_subvector X, 0))` into X when
X's type is the same as `vfmv.s.f`'s result type. This patch generalizes
it by folding it into insert_subvector when X is narrower and
extract_subvector when X is wider.

Co-authored-by: Craig Topper <[email protected]>

Author: mshockwave

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -20759,12 +20759,22 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
         isNullConstant(Src.getOperand(1)) &&
         Src.getOperand(0).getValueType().isScalableVector()) {
       EVT VT = N->getValueType(0);
-      EVT SrcVT = Src.getOperand(0).getValueType();
-      assert(SrcVT.getVectorElementType() == VT.getVectorElementType());
+      SDValue EVSrc = Src.getOperand(0);
+      EVT EVSrcVT = EVSrc.getValueType();
+      assert(EVSrcVT.getVectorElementType() == VT.getVectorElementType());
       // Widths match, just return the original vector.
-      if (SrcVT == VT)
-        return Src.getOperand(0);
-      // TODO: Use insert_subvector/extract_subvector to change widen/narrow?
+      if (EVSrcVT == VT)
+        return EVSrc;
+      SDLoc DL(N);
+      // Width is narrower, using insert_subvector.
+      if (EVSrcVT.getVectorMinNumElements() < VT.getVectorMinNumElements()) {
+        return DAG.getNode(ISD::INSERT_SUBVECTOR, DL, VT, DAG.getUNDEF(VT),
+                           EVSrc,
+                           DAG.getConstant(0, DL, Subtarget.getXLenVT()));
+      }
+      // Width is wider, using extract_subvector.
+      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, VT, EVSrc,
+                         DAG.getConstant(0, DL, Subtarget.getXLenVT()));
     }
     [[fallthrough]];
   }

llvm/test/CodeGen/RISCV/rvv/redundant-vfmvsf.ll

@@ -0,0 +1,26 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=riscv64 -mattr='+v,+zvl512b' < %s | FileCheck %s
+
+define <2 x float> @redundant_vfmv(<2 x float> %arg0, <64 x float> %arg1, <64 x float> %arg2) {
+; CHECK-LABEL: redundant_vfmv:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li a0, 64
+; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, ma
+; CHECK-NEXT:    vfredusum.vs v9, v12, v8
+; CHECK-NEXT:    vsetivli zero, 1, e32, mf2, ta, ma
+; CHECK-NEXT:    vslidedown.vi v8, v8, 1
+; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, ma
+; CHECK-NEXT:    vfredusum.vs v8, v16, v8
+; CHECK-NEXT:    vfmv.f.s fa5, v8
+; CHECK-NEXT:    vsetivli zero, 2, e32, mf2, ta, ma
+; CHECK-NEXT:    vrgather.vi v8, v9, 0
+; CHECK-NEXT:    vfslide1down.vf v8, v8, fa5
+; CHECK-NEXT:    ret
+  %s0 = extractelement <2 x float> %arg0, i64 0
+  %r0 = tail call reassoc float @llvm.vector.reduce.fadd.v64f32(float %s0, <64 x float> %arg1)
+  %a0 = insertelement <2 x float> poison, float %r0, i64 0
+  %s1 = extractelement <2 x float> %arg0, i64 1
+  %r1 = tail call reassoc float @llvm.vector.reduce.fadd.v64f32(float %s1, <64 x float> %arg2)
+  %a1 = insertelement <2 x float> %a0, float %r1, i64 1
+  ret <2 x float> %a1
+}