kmclaughlin-arm wrote: > Things to try: > > * Change the patch to insert the new diagnostics somewhere else in the > file. > > * Diff the assembly to see where there are non-trivial changes.
Thank you for the suggestions, @efriedma. I tried moving the diagnostics to the end of DiagnosticSemaKinds.td and comparing the assembly before & after, but could not find any reason for the difference still. It looks like the way I was attempting to reproduce the compile-time regression locally was not giving accurate results. After some discussion with @c-rhodes, we started investigating the changes to AArch64InstrAtomics.td again and found that lowering the AddedComplexity value I used on the atomic hint patterns improves the compile time. I've created a PR for this which improves the CTMark geomean by -0.12% on aarch64-O0-g: https://github.com/llvm/llvm-project/pull/230478 https://github.com/llvm/llvm-project/pull/198316 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
