https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121623
--- Comment #27 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Frank Scheiner from comment #26) > (In reply to Richard Biener from comment #25) > Thanks for your analysis, Richard. > > Well, I don't think I'll find the root cause and be able to fix it any time > soon. I believe here are other people in our community far better suited for > such a task. > > But let's also not overestimate the problem at hand. To summarize, we have: > > * a compare difference between stage 2 (w/o debug info) and 3 (w/debug info), > > * due to two independent "st8" instructions in the same MMI that are > differently ordered between stages (see above), > > * hence not affecting runtime behaviour IIUC, > > * and that in one single C++ source file (gcc/trans-mem.cc) over the whole > of the GCC for languages c and c++ (and also fortran and objc in the past). > > Yes, I think this should be observed in the future for any new > miscomparisons and possible runtime issues emerging. I believe [the current > patch](https://gcc.gnu.org/bugzilla/attachment.cgi?id=62166) is good enough > for now to spot these and allow for later analysis. And it also allows a > bootstrap build (which takes nearly 5.5 hours on an rx2800 i2 w/1 x Itanium > 2 9320) with such error(s) to be checked later. You can achieve the same result a less intrusive way by specifying --without-build-config (to override the default bootstrap-debug config).
