6.0.0 crash at high precision mul?

[Torbjörn Granlund]

Viktor Kletzhändler <vkletzhaendler at swissonline.ch> writes:

  No more crashes when using --enable-alloca=malloc-reentrant. Problem solved.
  
Except that this severely hurts GMP performance for most calculations.

  The program uses GCD to run tasks on separate threads. The stack size
  can't be increased as far as I know because the system handles all
  thread management.

I am sure that the thread stack size can be changed.
  
  Performance is not a problem. The binaries of 6.0.0 default config and
  --disable-fft are the same size, and there's hardly any difference in
  my timing test. Actually, the builds are really fast. The mpfr_sin(2)
  test takes about 17.4 secs at 10,000,000 bits. mpfr_pi() returns after
  about 5.2 secs at 10,000,000 bits. Details are in the attachment.
  
The --enable-fft/--disable-fft configure options are actually not
honoured since at least GMP 5.0.  The observed similarities between your
two sets of measurements are therefore unsurprising.


Torbjörn
Please encrypt, key id 0xC8601622