[X86][AVX] Fold concat(extract_subvector(v0,c0), extract_subvector(v1,c1)) -> vperm2x128

Fixes regression exposed by removing bitcasts across logic-ops in D96206.

Differential Revision: https://reviews.llvm.org/D96206

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -49249,6 +49249,29 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
     }
   }
 
+  // concat(extract_subvector(v0,c0), extract_subvector(v1,c1)) -> vperm2x128.
+  // Only concat of subvector high halves which vperm2x128 is best at.
+  // TODO: This should go in combineX86ShufflesRecursively eventually.
+  if (VT.is256BitVector() && Ops.size() == 2) {
+    SDValue Src0 = peekThroughBitcasts(Ops[0]);
+    SDValue Src1 = peekThroughBitcasts(Ops[1]);
+    if (Src0.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
+        Src1.getOpcode() == ISD::EXTRACT_SUBVECTOR) {
+      EVT SrcVT0 = Src0.getOperand(0).getValueType();
+      EVT SrcVT1 = Src1.getOperand(0).getValueType();
+      unsigned NumSrcElts0 = SrcVT0.getVectorNumElements();
+      unsigned NumSrcElts1 = SrcVT1.getVectorNumElements();
+      if (SrcVT0.is256BitVector() && SrcVT1.is256BitVector() &&
+          Src0.getConstantOperandAPInt(1) == (NumSrcElts0 / 2) &&
+          Src1.getConstantOperandAPInt(1) == (NumSrcElts1 / 2)) {
+        return DAG.getNode(X86ISD::VPERM2X128, DL, VT,
+                           DAG.getBitcast(VT, Src0.getOperand(0)),
+                           DAG.getBitcast(VT, Src1.getOperand(0)),
+                           DAG.getTargetConstant(0x31, DL, MVT::i8));
+      }
+    }
+  }
+
   // Repeated opcode.
   // TODO - combineX86ShufflesRecursively should handle shuffle concatenation
   // but it currently struggles with different vector widths.

llvm/test/CodeGen/X86/pr40891.ll

@@ -8,11 +8,9 @@ define <8 x i32> @foo(<8 x i64> %x, <4 x i64> %y) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; CHECK-NEXT:    vandps {{\.LCPI.*}}, %ymm1, %ymm1
-; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
-; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; CHECK-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; CHECK-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; CHECK-NEXT:    vshufps {{.*#+}} ymm0 = ymm2[0,2],ymm0[0,2],ymm2[4,6],ymm0[4,6]
+; CHECK-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; CHECK-NEXT:    retl
   %a = shufflevector <4 x i64> %y, <4 x i64> <i64 12345, i64 67890, i64 13579, i64 24680>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
   %b = and <8 x i64> %x, %a