tblah wrote: Hi @vzakhari, currently there is a serious performance regression in spec2017 exchange2. With the TBAA tags we produce slightly better code (less repeated loads, hoisting some code out of loops, etc). Unfortunately, this change to the generated code causes LLVM's function specialization pass to incorrectly judge that specialization will not be profitable, leading to the large performance regression.
@Leporacanthicus is working on this now. If you're happy, I think we can enable the tbaa tags by default after that. https://github.com/llvm/llvm-project/pull/68597 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
