[llvm] [SimplifyCFG] Avoid introducing complex phi when removing empty blocks (PR #104887)

[via llvm-commits]

Chengjunp wrote:

Hi, thank you for the review. I think the minor improvement can be expected because this kind of pattern is unusual. I did encounter a very extreme case that by adding this check, it can save more than 20% compile time. 

In term of the heuristic, looking at the total number of entries across all Phis in the successor also makes sense to me. But I'm considering how to determine the threshold in that case. Should the threshold be the same regardless of the number of Phis in the successor?  Or the threshold should be linear to the number of Phis in the Successor?

https://github.com/llvm/llvm-project/pull/104887