Bug found in nightbuilds

[Torbjorn Granlund]

bodrato at mail.dm.unipi.it writes:

  Code says:
    if (d & GMP_NUMB_HIGHBIT)
      { /* Normalized case */
  
        uh = up[--n]; /* Here n goes to 0 */
  ...
        if (BELOW_THRESHOLD (n, DIV_QR_1_NORM_THRESHOLD))
         {
            while (n > 0)
              udiv_qrnnd (...);
            return uh;
         }
        return mpn_div_qr_1n_pi1 (qp, up, n, uh, d, dinv);
  }
  
  The error arises when DIV_QR_1_NORM_THRESHOLD is zero, and the
  BELOW_THRESHOLD is unconditionally false, so that the code doesn't simply
  return uh, but calls mpn_div_qr_1n_pi1 with n = 0...
  
  I disallowed zero DIV_QR_1_NORM_THRESHOLD with this patch:
  
  https://gmplib.org/repo/gmp/rev/83258bea6f8f
  
  and the  GMP_CHECK_RANDOMIZE=3526906869 test passes.
  
  I believe that we should, for the same reasons, disallow zero 
  DIV_QR_1_UNNORM_THRESHOLD in the /* Unnormalized case */ branch.
  
Since we control these thresholds at the source level, if it is
desirable to keep DIV_QR_1_UNNORM_THRESHOLD >= 1, that should be done in
tuneup.

Your patch adds some overhead when tuning, since then
DIV_QR_1_UNNORM_THRESHOLD is not really constant.

Also, keeping DIV_QR_1_UNNORM_THRESHOLD >= 1 surely does the job, but it
puts another test it the critical path.  Perhaps that's unavoidable, I
haven't checked.


Torbjörn