[AArch64] Add target DAG combine for UUNPKHI/LO

[david-arm]

Summary

When created a UUNPKLO/HI node with an undef input then the
output should also be undef. I've added a target DAG combine
function to ensure we avoid creating an unnecessary uunpklo/hi
instruction.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

static SDValue performSpliceCombine(SDNode *N, SelectionDAG &DAG) {
 
   // splice(pg, op1, undef) -> op1
   if (N->getOperand(2).isUndef())
     return N->getOperand(1);
 
   return SDValue();
 }
 
+static SDValue performUnpackCombine(SDNode *N, SelectionDAG &DAG) {
+  assert((N->getOpcode() == AArch64ISD::UUNPKHI ||
+          N->getOpcode() == AArch64ISD::UUNPKLO) &&
+         "Unexpected Opcode!");
+
+  // uunpklo/hi undef -> undef
+  if (N->getOperand(0).isUndef())
+    return DAG.getUNDEF(N->getValueType(0));
+
+  return SDValue();
+}
+
 static SDValue performUzpCombine(SDNode *N, SelectionDAG &DAG) {
   SDLoc DL(N);
   SDValue Op0 = N->getOperand(0);
   SDValue Op1 = N->getOperand(1);
   EVT ResVT = N->getValueType(0);
 
   // uzp1(unpklo(uzp1(x, y)), z) => uzp1(x, z)
   if (Op0.getOpcode() == AArch64ISD::UUNPKLO) {
SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
   case AArch64ISD::CSEL:
     return performCSELCombine(N, DCI, DAG);
   case AArch64ISD::DUP:
     return performPostLD1Combine(N, DCI, false);
   case AArch64ISD::NVCAST:
     return performNVCASTCombine(N);
   case AArch64ISD::SPLICE:
     return performSpliceCombine(N, DAG);
+  case AArch64ISD::UUNPKLO:
+  case AArch64ISD::UUNPKHI:
+    return performUnpackCombine(N, DAG);
   case AArch64ISD::UZP1:
     return performUzpCombine(N, DAG);
   case AArch64ISD::SETCC_MERGE_ZERO:
     return performSetccMergeZeroCombine(N, DAG);
   case AArch64ISD::GLD1_MERGE_ZERO:
   case AArch64ISD::GLD1_SCALED_MERGE_ZERO:
   case AArch64ISD::GLD1_UXTW_MERGE_ZERO:
   case AArch64ISD::GLD1_SXTW_MERGE_ZERO:

llvm/test/CodeGen/AArch64/sve-insert-vector.ll

 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;  Upacked types that need result widening
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 define <vscale x 3 x i32> @insert_nxv3i32_nxv2i32(<vscale x 2 x i32> %sv0) {
 ; CHECK-LABEL: insert_nxv3i32_nxv2i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    uunpkhi z1.d, z0.s
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z1.s
+; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
 ; CHECK-NEXT:    ret
   %v0 = call <vscale x 3 x i32> @llvm.experimental.vector.insert.nxv3i32.nxv2i32(<vscale x 3 x i32> undef, <vscale x 2 x i32> %sv0, i64 0)
  ret <vscale x 3 x i32> %v0
 }
 
 ;; Check that the Subvector is not widen so it does not crash.
 define <vscale x 3 x i32> @insert_nxv3i32_nxv2i32_2(<vscale x 3 x i32> %sv0, <vscale x 2 x i32> %sv1) {
 ; CHECK-LABEL: insert_nxv3i32_nxv2i32_2:

  ret <vscale x 3 x float> %v0
 }
 
 define <vscale x 6 x i32>  @insert_nxv6i32_nxv2i32(<vscale x 2 x i32> %sv0, <vscale x 2 x i32> %sv1) nounwind {
 ; CHECK-LABEL: insert_nxv6i32_nxv2i32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
 ; CHECK-NEXT:    addvl sp, sp, #-2
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z1.s
 ; CHECK-NEXT:    ptrue p0.s
-; CHECK-NEXT:    uunpklo z1.d, z0.s
-; CHECK-NEXT:    ptrue p1.d
+; CHECK-NEXT:    uzp1 z0.s, z0.s, z1.s
 ; CHECK-NEXT:    st1w { z0.s }, p0, [sp]
-; CHECK-NEXT:    st1w { z1.d }, p1, [sp, #2, mul vl]
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [sp]
 ; CHECK-NEXT:    ld1w { z1.s }, p0/z, [sp, #1, mul vl]
 ; CHECK-NEXT:    addvl sp, sp, #2
 ; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
 ; CHECK-NEXT:    ret
   %v0 = call <vscale x 6 x i32> @llvm.experimental.vector.insert.nxv6i32.nxv2i32(<vscale x 6 x i32> undef, <vscale x 2 x i32> %sv0, i64 0)
   %v1 = call <vscale x 6 x i32> @llvm.experimental.vector.insert.nxv6i32.nxv2i32(<vscale x 6 x i32> %v0, <vscale x 2 x i32> %sv1, i64 2)
   ret <vscale x 6 x i32> %v1
 }