[llvm] [X86] optimize masked truncated saturating stores (PR #169827)

[Simon Pilgrim via llvm-commits]

================

----------------
RKSimon wrote:

No need to create a SmallVector on the fly as we know the array width:
```
SDValue Ops[]  = {
  Mst->getChain(),
  SatSrc,
  Mst->getBasePtr(),
  Mask
};
```

https://github.com/llvm/llvm-project/pull/169827