[llvm-branch-commits] [llvm] [LoongArch] Initial implementation for `enableMemCmpExpansion` hook (PR #166526)

[via llvm-branch-commits]

zhaoqi5 wrote:

I have just tested the `MicroBenchmarks/MemFunctions` in test-suite, using `O3+lsx`, `O2+nolsx` or `Os+lsx` options. All get the similar results:

- About 208 memcmp in the benchmark are expanded after this commit.
- Among its 168 tests, 144 have significant improvement (about 2 to 10 times).

Detail results of `O3+lsx`: [before.txt](https://github.com/user-attachments/files/23390441/before.txt), [after.txt](https://github.com/user-attachments/files/23390451/after.txt).

This benchmark is specifically designed to test `memcmp`, so it can directly show the benefits of this commit. Is it necessary to run spec cpu for more general results?

https://github.com/llvm/llvm-project/pull/166526