The approach and results look promising. It looks like a quite reliable way to test new optimizations and I would love to see if it might be a viable part for testing llvm patches.
Though it would also be interesting to see how this method would be applicable to other compilers, like GCC.
Meta: this is the highest I’ve seen something voted on the front page with 0 comments. Maybe ‘High-throughput’, ‘Formal Methods’, ‘Fuzzing’, and ‘LLVM’ is enough buzzwords to get everyone’s partial attention but not enough to really read!
John Regehr does excellent work in general.
Oh I know, I just mean that this is so edge engineering most can only look in awe (those are multiple very cool but complex subjects)