[llvm] [RISCV] and x (sub -1, y) -> andn x y (PR #129253)

[Gábor Spaits via llvm-commits]

spaits wrote:

> The added tests already produce `andn`.

I get it now.
At first, I did this incorrectly, and matched to `sub 0,y`. And that rightfully didn't work.
After that I have fixed that to `-1`. And it already works.

Sorry for this.

After this I still don't get why the scenario shown in the issue doesn't work.

Do you have any idea for that?

https://github.com/llvm/llvm-project/pull/129253