AIX/xlc test failures for gmp-5.0.1
Torbjorn Granlund
tg at gmplib.org
Sat Nov 27 23:33:25 CET 2010
Torbjorn Granlund <tg at gmplib.org> writes:
OK, now I am able to see the same error as you do. I have not attempted
to track this down, and I will probably not have time do do that in a
while.
I gave it a go this evening.
Unfortunately, this environment is too buggy for me to make any
progress.
1. A build without optimisation is miscompiled in several places. I
don't think it is possible to use such a build. One problem is that
it in effect folds x & 0 -> x. This is an extremely basic compiler
bug.
2. Debugging the t-toom32 problem with -O is completely hopeless since
the debugging information and gdb does not agree about almost
anything. Single stepping doesn't work, variable printing doesn't
work.
3. Tracking down the problem with t-toom32 will with very high
probability just lead to another compiler bug. The code works on a
large number of systems, and is quite solid. Therefore, I am not
willing to try to fight with print statements or whatnot.
If somebody wants to help isolating this problem, it might be best done
by adding lots of print statements, hoping that they will not make the
compiler work better (which unfortunately is not unlikely). Then
compile the code on a non-problem system (or with gcc on the same
system) and run them side-by-side, comparing the printed results.
I don't plan to work more on this.
--
Torbjörn
More information about the gmp-bugs
mailing list