[llvm] [X86] shouldReduceLoadWidth - don't split loads if ANY uses are a extract+store or a full width legal binop (PR #129695)

[Simon Pilgrim via llvm-commits]

================
@@ -11,7 +11,7 @@ define <8 x i16> @pr25080(<8 x i32> %a) nounwind {
 ; AVX1-LABEL: pr25080:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}+16(%rip), %xmm0, %xmm0
----------------
RKSimon wrote:

In this case, whats happened is that the extract has now loaded from the upper half of a single 256-bit constant pool (the other fold is loading the lower half). In the old codegen, we have 2 separate 128-bit constant pool entries.

https://github.com/llvm/llvm-project/pull/129695