[llvm] [AMDGPU] Try to reuse in v_cndmask register with constant from compare. (PR #131146)

[Daniil Fukalov via llvm-commits]

dfukalov wrote:

> Here is a test case that your patch will break because it does not handle +0/-0 correctly.
> 
Yes, you're  right, thanks!
Now it checks for +0/-0 float point imm and doesn't use its reg now.

https://github.com/llvm/llvm-project/pull/131146