[AArch64] Fold the ushll2 + ushll into zip2 + zip1

[vfdff]

• test: https://gcc.godbolt.org/z/TsoWzMsdx

  void zip_noalias (int *b, unsigned short *a, int n)
  {
  #pragma clang loop vectorize(assume_safety)
  for (int i = 0; i< (n & -4); i ++)
  {
      b[i] = a[i];
  }
  }

• gcc use zip with short latency

  .L5:
      ldr     q30, [x4], 16
      zip1    v29.8h, v30.8h, v31.8h
      zip2    v30.8h, v30.8h, v31.8h
      stp     q29, q30, [x3], 32
      cmp     x4, x5

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Allen (vfdff)

* test: https://gcc.godbolt.org/z/TsoWzMsdx ``` void zip_noalias (int *b, unsigned short *a, int n) { #pragma clang loop vectorize(assume_safety) for (int i = 0; i< (n & -4); i ++) { b[i] = a[i]; } } ``` * gcc use zip with short latency ``` .L5: ldr q30, [x4], 16 zip1 v29.8h, v30.8h, v31.8h zip2 v30.8h, v30.8h, v31.8h stp q29, q30, [x3], 32 cmp x4, x5 ```

[davemgreen]

They likely do this as the zip's have a higher throughput. We already do the same for truncates (https://gcc.godbolt.org/z/hdqbGvcGY) from https://reviews.llvm.org/D115435.