[X86] combineConcatVectorOps - convert X86ISD::PALIGNR concatenation to use combineConcatVectorOps recursion

[RKSimon]

Only concatenate X86ISD::PALIGNR nodes if at least one operand is beneficial to concatenate

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

Only concatenate X86ISD::PALIGNR nodes if at least one operand is beneficial to concatenate

---

Full diff: https://github.com/llvm/llvm-project/pull/130572.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+7-3)
• (modified) llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll (+4-6)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d83033d24bdbb..ccd7f2418fcd1 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -58443,9 +58443,13 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
           llvm::all_of(Ops, [Op0](SDValue Op) {
             return Op0.getOperand(2) == Op.getOperand(2);
           })) {
-        return DAG.getNode(Op0.getOpcode(), DL, VT,
-                           ConcatSubOperand(VT, Ops, 0),
-                           ConcatSubOperand(VT, Ops, 1), Op0.getOperand(2));
+        SDValue Concat0 = CombineSubOperand(VT, Ops, 0);
+        SDValue Concat1 = CombineSubOperand(VT, Ops, 1);
+        if (Concat0 || Concat1)
+          return DAG.getNode(Op0.getOpcode(), DL, VT,
+                             Concat0 ? Concat0 : ConcatSubOperand(VT, Ops, 0),
+                             Concat1 ? Concat1 : ConcatSubOperand(VT, Ops, 1),
+                             Op0.getOperand(2));
       }
       break;
     case X86ISD::BLENDI:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index a6e66af77dd20..399f137676fb4 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -775,15 +775,13 @@ define <32 x i8> @combine_pshufb_pshufb_or_pshufb(<32 x i8> %a0) {
   ret <32 x i8> %4
 }
 
-; TODO: Not beneficial to concatenate both inputs just to create a 256-bit palignr
+; Not beneficial to concatenate both inputs just to create a 256-bit palignr
 define <32 x i8> @concat_alignr_unnecessary(<16 x i8> %a0, <16 x i8> noundef %a1, <16 x i8> %a2) nounwind {
 ; CHECK-LABEL: concat_alignr_unnecessary:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; CHECK-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; CHECK-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; CHECK-NEXT:    vpalignr {{.*#+}} ymm0 = ymm1[3,4,5,6,7,8,9,10,11,12,13,14,15],ymm0[0,1,2],ymm1[19,20,21,22,23,24,25,26,27,28,29,30,31],ymm0[16,17,18]
+; CHECK-NEXT:    vpalignr {{.*#+}} xmm1 = xmm1[3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2]
+; CHECK-NEXT:    vpalignr {{.*#+}} xmm0 = xmm2[3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2]
+; CHECK-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %lo = shufflevector <16 x i8> %a1, <16 x i8> %a0, <16 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18>
   %hi = shufflevector <16 x i8> %a2, <16 x i8> %a0, <16 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18>