[llvm] [X86] getFauxShuffleMask - insert_subvector - skip undemanded subvectors (PR #129042)

[Simon Pilgrim via llvm-commits]

================
@@ -270,11 +270,10 @@ define <16 x i16> @insert_i16_firstelts(<16 x i16> %x, i16 %s) {
 ; AVX2-LABEL: insert_i16_firstelts:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpinsrw $0, %edi, %xmm0, %xmm1
+; AVX2-NEXT:    vmovd %edi, %xmm2
+; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
----------------
RKSimon wrote:

LowerINSERT_VECTOR_ELT is very limited on when it uses the broadcast+blend path - I'll see if there's a way to improve it.

https://github.com/llvm/llvm-project/pull/129042