[PATCH] D116720: [RISCV] Use shift for zext.h when Zbb and Zbp are not enabled

[Ben Shi via Phabricator via llvm-commits]

benshi001 added a comment.

I also concern that the 0xffff has multiple uses, such as

  unsigned short foo(unsigned short a, unsigned short b, int c, int d) {
    return (a >> c) + (b >> d);
  }

And there is only one immediate 0xffff is materialized.

  foo:
          srl     a0, a0, a2
          srl     a1, a1, a3
          add     a0, a0, a1
          lui     a1, 16
          addi    a1, a1, -1
          and     a0, a0, a1
          ret

Using a pass seems to be better, two cases should be excluded,

1. in a loop body
2. 0xffff has multiple uses

What's more, the pass can be generalized to a common immediate optimization pass, which combines several small optimization rules, just like AArch64's AArch64MIPeepholeOpt pass. However we need not make is so complex in current patch, that would be something TODO in the future.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D116720/new/

https://reviews.llvm.org/D116720