[llvm-branch-commits] [libcxx] release/18.x: [libcxx] Align `__recommend() + 1` by __endian_factor (#90292) (PR #95264)

via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Mon Jul 1 19:38:47 PDT 2024


tickerguy wrote:

This appears to blow up compilation on the Pi3 for certain source files on FreeBSD 14.1 with a SEGV in 'AArch64O0PreLegalizerCombiner'.

Specifically I have filed this bug thread for FreeBSD: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=280038

I have bisected the source and reverting this commit resolves the crash.  I have a reproducing source file that is only ~400 lines with it being nearly all OpenSSL calls -- and in an attempt to further isolate it, as the FreeBSD bugtracker will not take the reproducer file (too large) as that source file was already split off and very small I took individual functions out into their own source files -- and got it to dump on two of them _but not consistently_; often, if I simply re-type "make" it would complete the second time!

The blow-up _only_ happens with this commit in, and only on the Pi3 -- and I have used both a very old one (no HAT header for POE, for example) and a brand new unit (has the header) with identical results, so a specific hardware fault is excluded.  The same boot media inserted into a Pi4 and booted compiles the code without complaint.   Further, the same code, on the same revision of OS and clang/llvm but built and running on an AMD64 system with the commit in also builds without complaint.

The reproducer files are still too large to upload into FreeBSD's bugtracker, but since I now have a routine that is ~100 lines that does reproduce it, I simply posted that function's source.

https://github.com/llvm/llvm-project/pull/95264


More information about the llvm-branch-commits mailing list