[llvm] 94a62b3 - [X86] Add test case showing its not always beneficial to fold concat(pshufb(),pshufb()) -> pshufb(concat(),concat())

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2025-03-10T12:09:32Z
New Revision: 94a62b3085e35e1a7781c46bccee236255e1ca28

URL: https://github.com/llvm/llvm-project/commit/94a62b3085e35e1a7781c46bccee236255e1ca28
DIFF: https://github.com/llvm/llvm-project/commit/94a62b3085e35e1a7781c46bccee236255e1ca28.diff

LOG: [X86] Add test case showing its not always beneficial to fold concat(pshufb(),pshufb()) -> pshufb(concat(),concat())

Added: 
    

Modified: 
    llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index 8e9f5c335ac44..65358c3ac7a3c 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -806,6 +806,22 @@ define <32 x i8> @concat_packsr_unnecessary(<8 x i16> %a0, <8 x i16> %a1, <8 x i
 }
 declare <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16>, <8 x i16>)
 
+; TODO: Not beneficial to concatenate both inputs just to create a 256-bit pshufb
+define <32 x i8> @concat_pshufb_unnecessary(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> %a2) nounwind {
+; CHECK-LABEL: concat_pshufb_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:    vpshufb %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %lo = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> %a1)
+  %hi = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> %a2)
+  %res = shufflevector <16 x i8> %lo, <16 x i8> %hi, <32 x i32> <i32 0, i32 1, i32 2, 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, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+  ret <32 x i8> %res
+}
+
 define <8 x i32> @constant_fold_permd() {
 ; AVX2-LABEL: constant_fold_permd:
 ; AVX2:       # %bb.0: