bug#12350: Composites identified as primes in factor.c (when HAVE_GMP)

[Torbjorn Granlund]

Jim Meyering <address@hidden> writes:

  However, this little command does a lot of I/O, too:
  191M input, 77M output.
  
Sure.  I've never seen significant variance for such stuff, measurable
as CPU time.

I tested with a Sandybridge i3-2120T now.  The range takes 32 s.

In both cases, the systems run GNU/Linux.  The kernel version is 3.2.

The factoring speed varies very much with GCC version.  I particular the
trial division code has a very tight loop, and such loops have more
compiler reliance.  GCC 4.6 an later generate code that executes 4 insns
per (unsuccessful) division.

It is also important to use a 32bit binary.  We should perhaps have
provided better 32-bit code paths to be used for numbers < 2^32 on
32-bit hardware.  Now, Pádraig's example needs about 3x more time for a
32-bit binary on the same hardware.

-- 
Torbjörn