Hello, čt 28. 9. 2023 v 17:50 odesílatel John Paul Adrian Glaubitz <glaub...@physik.fu-berlin.de> napsal: > > Well, I talked to Adhemerval from glibc upstream regarding this and he said > there > is no chance that these issues are going to be fixed because that would > require a > lot of work due to a potential ABI breakage. You would need to find a glibc > expert > to fix the math failures. >
Giving an update on this. The math issues are reported in glibc Bugzilla, see https://sourceware.org/bugzilla/show_bug.cgi?id=10163. Me and Frank are currently working on them, they are mostly missing/wrongly implemented functionality in errno handling. After applying the patch by Aurelien from 2009 and some more tweaking, we managed to fix 30 from 201 tests failing in total, and fixes for more are under way. The in-progress patch can be accessed in t2-trunk branch of T2 SDE SVN repository at http://svn.exactcode.de/t2/trunk/package/base/glibc/math-fpu-error-partial-fix.patch.ia64. Parallel to that is another effort to debug an issue with gcc's CSE algorithm at https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111425. We managed to bisect the commit causing the issue, as well as analyze where the segmentation fault is coming from. Tomas