[llvm] [X86] X86FixupVectorConstants - load+sign-extend vector constants that can be stored in a truncated form (PR #79815)

[Phoebe Wang via llvm-commits]

================
@@ -750,7 +750,7 @@ define void @vec128_i16_widen_to_i32_factor2_broadcast_to_v4i32_factor4(ptr %in.
 ; AVX512BW-SLOW-LABEL: vec128_i16_widen_to_i32_factor2_broadcast_to_v4i32_factor4:
 ; AVX512BW-SLOW:       # %bb.0:
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,9,0,11,0,13,0,15]
+; AVX512BW-SLOW-NEXT:    vpmovsxbw {{.*#+}} xmm1 = [0,9,0,11,0,13,0,15]
----------------
phoebewang wrote:

Can small memory size beat defects in micro arch, e.g.

- vpmovsxbw uses one more port
- vmovdqa can be eliminated by uarch

https://github.com/llvm/llvm-project/pull/79815