mpz_mul core dump - attempt 2

[Torbjorn Granlund]

Kate Minola <kminola at po0.glue.umd.edu> writes:

  Much much thanks for confirming that a bug exists!  With
  Torbjorn reporting negative results, I was beginning to
  think I had gone crazy.

I suppose it is me that's gone crazy, since now
the bug happens also here...

The crash happens in mpn_sub_n after the FFT code passes
it a negative size.

                                          I do not
  know why the differences occur ... whether this is
  a symptom of the bug or something having to do with
  the fft algorithm on the different architectures.
  
The differences occur because of different gmp-mparam.h.

I had luck provoking the same behaviour on an Opteron by forcing
the use of sparc64/gmp-mparam.h instead of the default one.

I tried to debug this, but this code wasn't written by me,
and it is not enough commented for me to understand it easily.
I'll ask the original author to debug it.

--
Torbjörn