Bill Allombert on Fri, 7 Feb 2003 14:01:59 +0100


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

Re: GMP performance


On Thu, Feb 06, 2003 at 07:16:43PM -0500, Igor Schein wrote:
> Hi,
> 
> One immediate advantage of GMP kernel that I see is on 64bit
> platforms, on which so far ( with an exception of Alpha ) only
> portable kernel existed.  I compared GMP/portable kernel with regular
> portable on 64bit HPUX/HPPA, and I saw 20% performance improvement.
> It is a noticeable improvement, but I guess I kinda expected more.
> Was it unreasonable of me?

First, PARI still relies on the level 0 kernel for operations on one word
operands. This avoids some GMP over-head with a fast level 0 kernel
but may be a bad idea with the portable kernel.

The best way to solve that I thing is to implement a level 0 kernel
using the longlong.h interface of GMP. This is something to do, not
essentially difficult in fact.

Please note that GMP has not the same level of assembly support for
all CPU. 

Cheers,
Bill