[llvm] [ConstraintElim] Drop invalid rows instead of failing the elimination (PR #76299)

[Yingwei Zheng via llvm-commits]

dtcxzyw wrote:

> I am curious how you use this to spot regressions 
 
I find these regressions by reviewing the IR diffs of real-world applications, which are generated by different version of `opt`. TBH it is a time-consuming job :( But I think this new system is better than [my old CI](https://github.com/dtcxzyw/llvm-ci) because the old system creates a bunch of false positive regression reports.

> and if there's anything to automate?

The system will build the trunk version of LLVM and run the benchmark every hour. 

BTW, this regression was also detected by the pre-commit test https://github.com/dtcxzyw/llvm-opt-benchmark/pull/30#discussion_r1435335183.
If you would like to test your patch before merging it, feel free to ping me and I will create a pull request in llvm-opt-benchmark :)


https://github.com/llvm/llvm-project/pull/76299