Code Submition ( Improvement for mpn_mul_fft )

[Torbjorn Granlund]

"Jeunder Yu" <gis91542 at cis.nctu.edu.tw> writes:

  > How does the speed compare between the old and new code?  In the
  > GMP world, speed is all that counts, of course.  :-)
  
  CPU: Pentium 4 2.6GHz
  RAM: 512MB
  
  Two operands are 2^n limbs with random value.
  
  size    old     new
  ====================
  n=13   0.013   0.011
  n=14   0.031   0.027
  n=15   0.072   0.064
  n=16   0.160   0.145
  n=17   0.374   0.348
  n=18   0.824   0.767
  n=19   1.682   1.441
  n=20   3.398   2.845
  n=21   7.252   6.053
  n=22  16.625  13.957
  n=23  34.203  29.497
  n=24 408.061 350.596
  (virtual memory swap in/out for n>=24)
  
  Is this comparison enough?
  
Quite nice improvements!  

Paul Zimmermann improved our present code a while ago, but I
think your code is faster than even the improved code.

-- 
Torbjörn