MaskRay wrote: > > > > We need to calculate StackSafetyGlobalInfo before inserting > > > > `__asan_memcpy` > > > > > > Yes, but this is currently not guaranteed. This patch will ensure this > > property and fix the issue. > > Then the correct way to fix this issue to avoid lazy getInfo(). It should be > calculated at SSI = &MAM.getResult(M); Having analysis calculated before > modifications is quite important here.
We can set `-mllvm -stack-safety-run` to force `getInfo`, but it seems that the current lazy `getInfo` works as this patch does. > > > And I would expect that neither > > > [8ed1d81](https://github.com/llvm/llvm-project/commit/8ed1d8196bef89c3099be4ce4aa65f613ab819cc) > > > or StackSafety should apply to -O0 > > > > > > @AnnaZaks for > > [8ed1d81](https://github.com/llvm/llvm-project/commit/8ed1d8196bef89c3099be4ce4aa65f613ab819cc) > > `[asan] Skip promotable allocas to improve performance at -O0` > > Turning off isAllocaPromotable is LGTM, I would probably recommend just > switch opt<> default to false, and keep it just in case. > > I don't see @AnnaZaks active here recently. @yln @rsundahl instead Let me rename `ClSkipPromotableAllocas` (misleading option name even before this patch) to `ClSkipUninterestingAllocas`. There are some cases that `ClSkipUninterestingAllocas=0` ignores AllocaInsts that `ClSkipUninterestingAllocas=1` doesn't. FWIW I've built a stage-2 -O0 asan clang and a stage-2 -O3 asan clang. The size does not change with or without this patch. https://github.com/llvm/llvm-project/pull/77221 _______________________________________________ llvm-branch-commits mailing list llvm-branch-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
