vitalybuka added a reviewer: vitalybuka. vitalybuka added a comment. In D134410#3817190 <https://reviews.llvm.org/D134410#3817190>, @vitalybuka wrote:
>> Also I applied this patch with D134698 <https://reviews.llvm.org/D134698> >> and used on our large test set, expecting significant number of pre-existing >> reports. To my surprise I see not many of them. > > Something wrong with my msan experiment, I'll re-evaluate size and reports > tomorrow. Finally I had a time to fix my msan experiment D134698 <https://reviews.llvm.org/D134698> It's about 5.5% .text savings for Msan, and 10% for msan with "track origins", which is pretty good. For context, for msan it's usually cheaper to report uninitialized ASAP, then propagating and report it later. With this metadata it will happen immediately after load. However cleanup looks scary. Msan reports maybe 20% of unique tests on our code base. Many a of them share root cause, but still many unique root causes. On quick looks I see no false report. A lot of stuff like this https://stackoverflow.com/questions/60112841/copying-structs-with-uninitialized-members which is technically is UB. I assume with this patch landed, many such cases may change code behavior. So we will need to update msan to have a tool to detect cases like this anyway. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D134410/new/ https://reviews.llvm.org/D134410 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
