Bill Allombert on Thu, 10 Mar 2016 16:01:54 +0100


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: Port of pari/gp 2.7.5 on AIX


On Thu, Mar 10, 2016 at 02:45:44PM +0000, REIX, Tony wrote:
> Hi,
> 
> We are porting pari/gp 2.7.5 on AIX.
> We are now using gcc (4.8.4). We plan to use xlc (v12).
> 
> We'd like to compare benchmark results with "official" benchmark results.
> However, on Pari/gp web-site, we only can find page :
> <http://pari.math.u-bordeaux.fr/buildlog.html>
> http://pari.math.u-bordeaux.fr/buildlog.html  and we do not understand
> why some "bench" result are very small, like :
> 
> 
> Mar 10
> 11:19<http://pari.math.u-bordeaux.fr/buildlogs/27f8423a-d25b-4e93-bafb-e3fb1cc7c46c>
> 131      0      2.7.5    93f8512   amd64    linux   x86-64/GMP    g++
> 5.3.1 +++ Total bench for gp-sta is 128 +++ Total bench for gp-dyn is
> 131
> 
> Mar 10
> 11:18<http://pari.math.u-bordeaux.fr/buildlogs/b1178ac6-3883-4963-850e-2c217512c498>
> 187327   0      2.7.5    93f8512   amd64    linux   x86-64/GMP    gcc
> 5.3.1 +++ Total bench for gp-sta is 178934 +++ Total bench for gp-dyn
> is 187327
> 
> 
> Looking at details, the list of tests that were run is either very
> short or very long. Why ?

See the color code at the top of buildlog.html:

<green>  nnn: bench for gp-sta
<cyan>   nnn: test-all for gp-sta

There are two benchs:

The short one (green) is 
  make bench
The long one  (cyan) is
  make statest-all

the second one require the optional packages to be installed and is much
longer.

But unfortunately, neither of them is a good indicator of performances.

Cheers,
Bill