kasuga-fj wrote: I'm not trying to emphasize the number of issues raised on GitHub, the real concern is their substance. What I pointed out in the issues represents only a small portion of the overall problem. There are fundamental flaws in DA's implementation, particularly in how it handles wrapping of SCEVs. These issues aren't caught by assertions and can silently lead to incorrect transformations. I don't know how many miscompiles exist, but it would not be a small number.
The situation seems to differ from that of the vectorizer. DA was already broken when this PR was submitted. I believe it's premature to enable any passes that rely on DA by default. This would not directly reflect the severity of the issue, but it's the conclusion I reached after thoroughly examining DA for a month. There are simply too many bugs. Also note that properly fixing these issues will take considerable time, as we (or at least I) still aren't sure what the best approach to solving them is. https://github.com/llvm/llvm-project/pull/140182 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
