[DAG] Fold insert_subvector undef, (extract_subvector X, 0), 0 with non-matching types

We have an existing DAG combine for when an insert/extract subvector pair is entirely a nop, but we hadn't handled the case where the net result was either an insert or an extract (but not both). The transform is restricted to index = 0 to avoid having to adjust indices after the transform.

Differential Revision: https://reviews.llvm.org/D158201

Author: preames

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -25597,10 +25597,25 @@ SDValue DAGCombiner::visitINSERT_SUBVECTOR(SDNode *N) {
     return N0;
 
   // If this is an insert of an extracted vector into an undef vector, we can
-  // just use the input to the extract.
+  // just use the input to the extract if the types match, and can simplify
+  // in some cases even if they don't.
   if (N0.isUndef() && N1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
-      N1.getOperand(1) == N2 && N1.getOperand(0).getValueType() == VT)
-    return N1.getOperand(0);
+      N1.getOperand(1) == N2) {
+    EVT SrcVT = N1.getOperand(0).getValueType();
+    if (SrcVT == VT)
+      return N1.getOperand(0);
+    // TODO: To remove the zero check, need to adjust the offset to
+    // a multiple of the new src type.
+    if (isNullConstant(N2) &&
+        VT.isScalableVector() == SrcVT.isScalableVector()) {
+      if (VT.getVectorMinNumElements() >= SrcVT.getVectorMinNumElements())
+        return DAG.getNode(ISD::INSERT_SUBVECTOR, SDLoc(N),
+                           VT, N0, N1.getOperand(0), N2);
+      else
+        return DAG.getNode(ISD::EXTRACT_SUBVECTOR, SDLoc(N),
+                           VT, N1.getOperand(0), N2);
+    }
+  }
 
   // Simplify scalar inserts into an undef vector:
   // insert_subvector undef, (splat X), N2 -> splat X

llvm/test/CodeGen/RISCV/rvv/active_lane_mask.ll

@@ -103,15 +103,15 @@ define <8 x i1> @fv8(ptr %p, i64 %index, i64 %tc) {
 define <32 x i1> @fv32(ptr %p, i64 %index, i64 %tc) {
 ; CHECK-LABEL: fv32:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    lui a0, %hi(.LCPI8_0)
 ; CHECK-NEXT:    addi a0, a0, %lo(.LCPI8_0)
-; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    vle64.v v8, (a0)
+; CHECK-NEXT:    vid.v v16
+; CHECK-NEXT:    vsaddu.vx v16, v16, a1
+; CHECK-NEXT:    vmsltu.vx v0, v16, a2
 ; CHECK-NEXT:    vsaddu.vx v8, v8, a1
 ; CHECK-NEXT:    vmsltu.vx v16, v8, a2
-; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    vsaddu.vx v8, v8, a1
-; CHECK-NEXT:    vmsltu.vx v0, v8, a2
 ; CHECK-NEXT:    vsetivli zero, 4, e8, mf4, ta, ma
 ; CHECK-NEXT:    vslideup.vi v0, v16, 2
 ; CHECK-NEXT:    ret
@@ -122,15 +122,15 @@ define <32 x i1> @fv32(ptr %p, i64 %index, i64 %tc) {
 define <64 x i1> @fv64(ptr %p, i64 %index, i64 %tc) {
 ; CHECK-LABEL: fv64:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    lui a0, %hi(.LCPI9_0)
 ; CHECK-NEXT:    addi a0, a0, %lo(.LCPI9_0)
-; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    vle64.v v8, (a0)
+; CHECK-NEXT:    vid.v v16
+; CHECK-NEXT:    vsaddu.vx v16, v16, a1
+; CHECK-NEXT:    vmsltu.vx v0, v16, a2
 ; CHECK-NEXT:    vsaddu.vx v8, v8, a1
 ; CHECK-NEXT:    vmsltu.vx v16, v8, a2
-; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    vsaddu.vx v8, v8, a1
-; CHECK-NEXT:    vmsltu.vx v0, v8, a2
 ; CHECK-NEXT:    vsetivli zero, 4, e8, mf2, tu, ma
 ; CHECK-NEXT:    vslideup.vi v0, v16, 2
 ; CHECK-NEXT:    lui a0, %hi(.LCPI9_1)
@@ -157,15 +157,15 @@ define <64 x i1> @fv64(ptr %p, i64 %index, i64 %tc) {
 define <128 x i1> @fv128(ptr %p, i64 %index, i64 %tc) {
 ; CHECK-LABEL: fv128:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    lui a0, %hi(.LCPI10_0)
 ; CHECK-NEXT:    addi a0, a0, %lo(.LCPI10_0)
-; CHECK-NEXT:    vsetivli zero, 16, e64, m8, ta, ma
 ; CHECK-NEXT:    vle64.v v8, (a0)
+; CHECK-NEXT:    vid.v v16
+; CHECK-NEXT:    vsaddu.vx v16, v16, a1
+; CHECK-NEXT:    vmsltu.vx v0, v16, a2
 ; CHECK-NEXT:    vsaddu.vx v8, v8, a1
 ; CHECK-NEXT:    vmsltu.vx v16, v8, a2
-; CHECK-NEXT:    vid.v v8
-; CHECK-NEXT:    vsaddu.vx v8, v8, a1
-; CHECK-NEXT:    vmsltu.vx v0, v8, a2
 ; CHECK-NEXT:    vsetivli zero, 4, e8, m1, tu, ma
 ; CHECK-NEXT:    vslideup.vi v0, v16, 2
 ; CHECK-NEXT:    lui a0, %hi(.LCPI10_1)

llvm/test/CodeGen/RISCV/rvv/extract-subvector.ll

@@ -469,13 +469,14 @@ define <vscale x 6 x half> @extract_nxv6f16_nxv12f16_6(<vscale x 12 x half> %in)
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    csrr a0, vlenb
 ; CHECK-NEXT:    srli a0, a0, 2
+; CHECK-NEXT:    vsetvli zero, a0, e16, m1, ta, ma
+; CHECK-NEXT:    vslidedown.vx v13, v10, a0
 ; CHECK-NEXT:    vsetvli a1, zero, e16, m1, ta, ma
-; CHECK-NEXT:    vslidedown.vx v8, v9, a0
+; CHECK-NEXT:    vslidedown.vx v12, v9, a0
 ; CHECK-NEXT:    add a1, a0, a0
 ; CHECK-NEXT:    vsetvli zero, a1, e16, m1, tu, ma
-; CHECK-NEXT:    vslideup.vx v8, v10, a0
-; CHECK-NEXT:    vsetvli zero, a0, e16, m1, tu, ma
-; CHECK-NEXT:    vslidedown.vx v9, v10, a0
+; CHECK-NEXT:    vslideup.vx v12, v10, a0
+; CHECK-NEXT:    vmv2r.v v8, v12
 ; CHECK-NEXT:    ret
   %res = call <vscale x 6 x half> @llvm.vector.extract.nxv6f16.nxv12f16(<vscale x 12 x half> %in, i64 6)
   ret <vscale x 6 x half> %res