[llvm] [X86][AVX512] Better lowering for `_mm512_maskz_shuffle_epi32` (PR #121147)

[Simon Pilgrim via llvm-commits]

================
@@ -990,3 +990,23 @@ bb:
   ret void
 }
 
+define <8 x i64> @pr121147(<8 x i64> %a) {
----------------
RKSimon wrote:

drop the <8 x i64> src/dst bitcasts and just use <16 x i32> directly

https://github.com/llvm/llvm-project/pull/121147