[hashcash] Re: libfastmint update 20040619

Previous best on 3.06Ghz P4:

-O3 -funroll-loops -march=pentium4 -mmmx -msse -msse2
  1841293 ANSI Compact 1-pipe
  1506512 ANSI Standard 1-pipe
  2109118 ANSI Compact 2-pipe
  1611131 ANSI Standard 2-pipe
  3052671 AMD64/x86 MMX Compact 1x2-pipe
  3866716 AMD64/x86 MMX Standard 1x2-pipe *

and new code:

-O3 -funroll-loops -march=pentium4 -mmmx -msse -msse2
  1993854 ANSI Compact 1-pipe
  1715641 ANSI Standard 1-pipe
   905185 ANSI Ultra-Compact 1-pipe
  2379761 ANSI Compact 2-pipe
  2202167 ANSI Standard 2-pipe
  3278782 AMD64/x86 MMX Compact 1x2-pipe
  4339564 AMD64/x86 MMX Standard 1x2-pipe *

so 12% faster on P4.

Previous best on AMD k6-2:

   434462 AMD64/x86 MMX Compact 1x2-pipe

and new code:

   245091 ANSI Compact 1-pipe
   245499 ANSI Standard 1-pipe
   110108 ANSI Ultra-Compact 1-pipe
   222206 ANSI Compact 2-pipe
   235695 ANSI Standard 2-pipe
   474421 AMD64/x86 MMX Compact 1x2-pipe
   466915 AMD64/x86 MMX Standard 1x2-pipe *

so 9% faster on AMD k6-2.

Adam

On Sat, Jun 19, 2004 at 10:36:22PM +0100, Jonathan Morton wrote:
> Here, finally, is a version of libfastmint that incorporates the 
> work-reduction techniques we discussed earlier.  It also builds on 
> MacOS Carbon/Classic, which it didn't before.  Performance should be up 
> by about 10-15% on all minter cores, usually in the upper end of that 
> range.
> 
> I can now get 3-million per second on my 667MHz 7450, and 3.6-million 
> on an Athlon-XP 2500+ (slightly faster than the 1600MHz model I was 
> using before).  If I overclocked to slightly over 2GHz, I would break 4 
> million with that machine.

Other related posts: