[llvm] [AMDGPU] Add getCmpSelInstrCost override to re-enable SimplifyCFG speculation for vector types (PR #208043)

[Frederik Harwath via llvm-commits]

================

----------------
frederik-h wrote:

I used the lambda because I wanted to avoid the recursive call (for clarity ...) just for getting the scalar cost. I have removed the lambda again and now I am calling the function recursively. The corresponding part of the code now looks like in the base function and it also contains the same comment explaining that this does "multiply by cost", but it assumes the cost is 1 in this case.

https://github.com/llvm/llvm-project/pull/208043