[X86] Combine concat(palignr,palignr) -> palignr(concat,concat)

combineX86ShufflesRecursively should handle this someday

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 62a13ed..c7c599c 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -46548,6 +46548,24 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
                            DAG.getNode(ISD::CONCAT_VECTORS, DL, SrcVT, RHS));
       }
       break;
+    case X86ISD::PALIGNR:
+      if (!IsSplat &&
+          ((VT.is256BitVector() && Subtarget.hasInt256()) ||
+           (VT.is512BitVector() && Subtarget.useBWIRegs())) &&
+          llvm::all_of(Ops, [Op0](SDValue Op) {
+            return Op0.getOperand(2) == Op.getOperand(2);
+          })) {
+        SmallVector<SDValue, 2> LHS, RHS;
+        for (unsigned i = 0; i != NumOps; ++i) {
+          LHS.push_back(Ops[i].getOperand(0));
+          RHS.push_back(Ops[i].getOperand(1));
+        }
+        return DAG.getNode(Op0.getOpcode(), DL, VT,
+                           DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, LHS),
+                           DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, RHS),
+                           Op0.getOperand(2));
+      }
+      break;
     }
   }
 

diff --git a/llvm/test/CodeGen/X86/x86-interleaved-access.ll b/llvm/test/CodeGen/X86/x86-interleaved-access.ll
index d233e96..c323db1 100644 (file)
--- a/llvm/test/CodeGen/X86/x86-interleaved-access.ll
+++ b/llvm/test/CodeGen/X86/x86-interleaved-access.ll
@@ -1463,13 +1463,8 @@ define <64 x i8> @interleaved_load_vf64_i8_stride3(<192 x i8>* %ptr){
 ; AVX512-NEXT:    vpblendmb %zmm1, %zmm0, %zmm2 {%k1}
 ; AVX512-NEXT:    vpalignr {{.*#+}} zmm1 = zmm3[11,12,13,14,15],zmm1[0,1,2,3,4,5,6,7,8,9,10],zmm3[27,28,29,30,31],zmm1[16,17,18,19,20,21,22,23,24,25,26],zmm3[43,44,45,46,47],zmm1[32,33,34,35,36,37,38,39,40,41,42],zmm3[59,60,61,62,63],zmm1[48,49,50,51,52,53,54,55,56,57,58]
 ; AVX512-NEXT:    vpaddb %zmm1, %zmm2, %zmm1
-; AVX512-NEXT:    vpalignr {{.*#+}} ymm2 = ymm0[11,12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10],ymm0[27,28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26]
-; AVX512-NEXT:    vpalignr {{.*#+}} ymm2 = ymm2[10,11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,26,27,28,29,30,31,16,17,18,19,20,21,22,23,24,25]
-; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512-NEXT:    vextracti64x4 $1, %zmm3, %ymm3
-; AVX512-NEXT:    vpalignr {{.*#+}} ymm0 = ymm0[11,12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10],ymm0[27,28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26]
-; AVX512-NEXT:    vpalignr {{.*#+}} ymm0 = ymm0[10,11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,26,27,28,29,30,31,16,17,18,19,20,21,22,23,24,25]
-; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
+; AVX512-NEXT:    vpalignr {{.*#+}} zmm0 = zmm0[11,12,13,14,15],zmm3[0,1,2,3,4,5,6,7,8,9,10],zmm0[27,28,29,30,31],zmm3[16,17,18,19,20,21,22,23,24,25,26],zmm0[43,44,45,46,47],zmm3[32,33,34,35,36,37,38,39,40,41,42],zmm0[59,60,61,62,63],zmm3[48,49,50,51,52,53,54,55,56,57,58]
+; AVX512-NEXT:    vpalignr {{.*#+}} zmm0 = zmm0[10,11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,26,27,28,29,30,31,16,17,18,19,20,21,22,23,24,25,42,43,44,45,46,47,32,33,34,35,36,37,38,39,40,41,58,59,60,61,62,63,48,49,50,51,52,53,54,55,56,57]
 ; AVX512-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    retq
 %wide.vec = load <192 x i8>, <192 x i8>* %ptr, align 1