Summary: | glibc 2.3.4.20040619 (and -r1) fails makecheck for test-float | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Alexandre Rostovtsev (RETIRED) <tetromino> |
Component: | [OLD] Core system | Assignee: | Gentoo Toolchain Maintainers <toolchain> |
Status: | RESOLVED DUPLICATE | ||
Severity: | major | ||
Priority: | High | ||
Version: | unspecified | ||
Hardware: | x86 | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- |
Description
Alexandre Rostovtsev (RETIRED)
2004-08-05 14:18:35 UTC
same problem here, could only avoid it if i dont't use "makecheck" USE-flag or unset my CFLAGS to nothing so that defaults for glibc would be used. i tested this also on a lfs-machine same thing there when running make check. so i played with the flags an noticed that -march= set to pentium4, pentium3 and also athlon-xp make check failed, for other arch i had no chance to test. I think the cause is 64bit vs 80bit double precision of 387 floating point unit. Using -ffloat-store or -mfpmath=sse should make the test pass (both force 64bit), as well as unsetting -march, which disables the use of floating point unit. Actually, I think it's the other way around; the test result data assumes the system gives more accuracy than one should reasonably expect from 32-bit IEEE floating point operations. See bug #87249 where I've been working on this. Original report here had 'USE=hardened', so presumably used hardened compiler which highlights the bug. *** This bug has been marked as a duplicate of 87249 *** |