Failure in the GMP test suite (ARM processor)

Torbjorn Granlund tege at swox.com
Tue Sep 19 11:19:27 CEST 2006


"Jim Francis" <francis at pleiadescom.com> writes:

  Problem:  One of the tests, mpz/t-pow, is intermittently failing.  I'm
  attaching a sample output file from a run which attempts this test 20 times
  and experiences 5 failures.
  
  Other info:  Of the 136 tests in the test suite (including the 14 under the
  cxx subdirectory), this is the only one which has failed.  Repeated runs of
  this test seem to indicate a 20-25% failure rate.  Because the failures are
  intermittent, I'm more inclined to believe the problem concerns the GMP
  library's interaction with the ARM chip rather than the GCC compiler suite.

And I am more inclined to believe that your hardware is broken.

I can assure you that there is nothing non-deterministic in how
GMP executes its tests.  Yes, it uses pseudo random numbers, but
don't miss the pseudo in there.

(It is possible to seed the "make check" target, making the
pseudo random number sequence vary from run to run, but since you
don't say you've figured out how to do that, I assume you've not
done it.)

-- 
Torbjörn


More information about the gmp-bugs mailing list