[llvm] [X86] combineINSERT_SUBVECTOR - attempt to widen mask predicate as scalar integer instead of inserting into a zero vXi1 vector (PR #192699)

[Phoebe Wang via llvm-commits]

================
@@ -18,10 +18,8 @@ define <8 x half> @maskload_v8f16_byref(ptr %p, i8 %k, ptr %mem_addr) {
 ; CHECK-LABEL: maskload_v8f16_byref:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
-; CHECK-NEXT:    andb $1, %sil
-; CHECK-NEXT:    kmovd %esi, %k0
-; CHECK-NEXT:    kshiftld $24, %k0, %k0
-; CHECK-NEXT:    kshiftrd $24, %k0, %k1
----------------
phoebewang wrote:

Seems we can use kmovb previously.

https://github.com/llvm/llvm-project/pull/192699