[PATCH] D145614: [AARCH64] Enable STORE of v4i8 to help more vectorization opportunities

Fri Mar 10 11:34:21 PST 2023

Carrot added a comment.

I tried isTruncStoreLegalOrCustom and it works, generates the same code as my patch. The final code is

  ldr s0, [x1]
  ldr s1, [x0]
  ushll   v0.8h, v0.8b, #0       // *
  ushll   v1.8h, v1.8b, #0       // *
  orr v0.8b, v1.8b, v0.8b
  xtn v0.8b, v0.8h               // *
  str s0, [x0]
  ret

But I still prefer my patch. In the final code you can see some redundant instructions, these are generated by SelectionDAG, which promotes v4i8 to v4i16, processes the data, and truncs it to v4i8 before store. If we widen the vector v4i8 to v8i8, all the extra instructions can be deleted. I will work on it in a separate patch.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D145614/new/

https://reviews.llvm.org/D145614