[llvm] [LoongArch] Optimize extractelement containing variable index for lsx (PR #156792)

[via llvm-commits]

tangaac wrote:

> > Very little regression. [tangaac/loong-opt-cov-ts at 9805f6b](https://github.com/tangaac/loong-opt-cov-ts/commit/9805f6b186621f58d05a1f699ffe62503c343abb)
> 
> So do we need to handle the result differently depending on whether it goes into a GPR or a VR?

I think it's still beneficial for i8/i16.
May be we should turn it down for i64 and check the results again.

https://github.com/llvm/llvm-project/pull/156792