[llvm] [X86] combineINSERT_SUBVECTOR - fold insert_subvector(base,extract_subvector(broadcast)) -> blend shuffle(base,broadcast) (PR #133083)

[Phoebe Wang via llvm-commits]

================
@@ -2239,7 +2239,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
 ; AVX512F-NEXT:    vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
 ; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
 ; AVX512F-NEXT:    vpbroadcastb (%rdi), %ymm1
-; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
----------------
phoebewang wrote:

I was thinking some peephole optimizaion, but I didn't find an existence. X86FixupInstTuning might be the most relevant.

https://github.com/llvm/llvm-project/pull/133083