memory problem with GMP 5

[Torbjorn Granlund]

Paul Zimmermann <Paul.Zimmermann at loria.fr> writes:

  With the gmp-5.0.0-20100123 snapshot (which still says "GMP 5.0.0" for
  gmp_version) the "RES" usage is first at 1.7g, then goes up to 2.3g with
  the program on gmplib.org/list-archives/gmp-bugs/2010-January/001792.html.
  
  Thus I only see a small improvement with respect to 5.0.0, which took 2.4g max,
  whereas 4.3.2 took only 1.6g.
  
Ouch.  We need to address this before we roll 5.0.1.  I consider it one
of the most important regressions to fix.

  It would be nice if someone else could confirm this, since it is not easy to
  get a precise measure, unless somebody knows how to do without looking at the
  "top" output.
  
I suppose a binary search over "ulimit -Sd nbytes" could do it.

Memory usage is tricky issue.  The number one sees is not just what GMP
allocates, but a measure of fragmentation, which may vary with the
strategies of the underlying malloc/realloc.

-- 
Torbjörn