aespipe testsuite failure - help
Hi hppa porters,
I'm seeing an interesting failure of the aespipe testsuite on hppa
when built with gcc-4.0.
I'm a bit at loss to debug this further since the generated code is
quite different from earlier builds with gcc3. Your help would be
greatly appreciated.
aespipe includes a C implementation of SHA256/384/512 that is tested
after the build. SHA384 and SHA512 currently generate wrong results on
arm and hppa, whereas SHA256 works as expected on all archs.
I can reproduce the failure in paer's sid chroot, but only with the
combination of gcc-4.0 and -O2 optimisation. The gcc-3.4 and gcc-4.0
builds including preprocessed sha512.c and generated .s are up on
<http://people.debian.org/~xam/aespipe/hppa/>.
What I found out so far:
- No problem with -O0 or -O1
- No runtime problem, old binary passes the testsuite
- No problem with gcc-3.3 or gcc-3.4
- No difference in preprocessed sha512.c between gcc-3.4 and gcc-4.0
Any suggestions how to proceed to debug/resolve this?
I've seen some packages use lower optimisation levels on certain archs,
which could be done for aespipe. I would prefer to find the source of
the problem, but then I don't actually know how and where to start.
cheers,
Max
Reply to: