[llvm] [X86] Generate `kmov` for masking integers (PR #120593)

[Abhishek Kaushik via llvm-commits]

abhishek-kaushik22 wrote:

> @abhishek-kaushik22 Can you confirm if this will fix #72803 please?

Yes, for 
```llvm
define void @example(ptr noalias nocapture noundef writeonly sret([8 x i32]) align 4 dereferenceable(32) %_0, ptr noalias nocapture noundef readonly align 4 dereferenceable(32) %a, i64 noundef %m, ptr noalias nocapture noundef readonly align 4 dereferenceable(32) %b) {
start:
  %0 = insertelement <8 x i64> poison, i64 %m, i64 0
  %1 = shufflevector <8 x i64> %0, <8 x i64> poison, <8 x i32> zeroinitializer
  %2 = and <8 x i64> %1, <i64 1, i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128>
  %3 = icmp eq <8 x i64> %2, zeroinitializer
  %4 = load <8 x i32>, ptr %b, align 4
  %5 = select <8 x i1> %3, <8 x i32> zeroinitializer, <8 x i32> %4
  %6 = load <8 x i32>, ptr %a, align 4
  %7 = add <8 x i32> %6, %5
  store <8 x i32> %7, ptr %_0, align 4
  ret void
}
```
this generates 
```asm
        movq	%rdi, %rax
	kmovd	%edx, %k1
	vmovdqu	(%rsi), %ymm0
	vpaddd	(%rcx), %ymm0, %ymm0 {%k1}
	vmovdqu	%ymm0, (%rdi)
	vzeroupper
	retq
```

https://github.com/llvm/llvm-project/pull/120593