[llvm] [VectorCombine][X86] Use updated getVectorInstrCost hook (PR #137823)

[Florian Mayer via llvm-commits]

fmayer wrote:

IR diff looks like this (in various places). `-` is before this patch, `+` after

```
   call void @__msan_unpoison(ptr noundef nonnull %add.ptr88, i64 noundef 4) #55, !dbg !242357
   %30 = load i32, ptr %add.ptr88, align 4, !dbg !242362, !alias.scope !242364, !noalias !242371
   %vecinit3.i.i.i.i = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %30, i64 0, !dbg !242373
-  %cmp.i.i.i54.i.i = icmp ult <4 x i32> %vecinit3.i.i.i.i, splat (i32 2), !dbg !242374
+  %cmp.i.i.i54.i.i.scalar = icmp ult i32 %30, 2, !dbg !242374
+  %cmp.i.i.i54.i.i = insertelement <4 x i1> <i1 poison, i1 true, i1 true, i1 true>, i1 %cmp.i.i.i54.i.i.scalar, i64 0, !dbg !242374
   %31 = bitcast <4 x i1> %cmp.i.i.i54.i.i to i4, !dbg !242381
   %32 = and i4 %31, 1, !dbg !242384
   %cmp.i32.i.i.not = icmp eq i4 %32, 0, !dbg !242386
```

https://github.com/llvm/llvm-project/pull/137823