[Ecls-list] Slow bignums

[Juan Jose Garcia-Ripoll]

On Mon, Jan 25, 2010 at 4:26 AM, Waldek Hebisch <hebisch at math.uni.wroc.pl>wrote:

> I have noticed that ecl seem to be quite slow on some bignum
> operations, namely bignum addition and multiplication of
> bignum by a fixnum.  [...]
> clisp and Closure CL cons comparably to ecl (gcl does not report
> consing).


ECL's consing numbers are not very accurate: they are an estimate produced
by the garbage collector.

What I have done is profiling ECL with your code and looking for performance
sinks. The result is quite surprising: bignum computations are just 10% of
the total time, so ECL could be 10 times faster, but the problem is garbage
collection, which takes the 90% remaining time.

Your code is thus not actually testing bignum computations, but rather the
ability of the garbage collector to allocate bignums which are increasing in
size and quickly discarding it.

I expect these numbers to get better depending on the platform and the
version number of the garbage collector, but I am unsure on how to further
improve this.

Juanjo

-- 
Instituto de Física Fundamental, CSIC
c/ Serrano, 113b, Madrid 28006 (Spain)
http://juanjose.garciaripoll.googlepages.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.common-lisp.net/pipermail/ecl-devel/attachments/20100127/a3d9d947/attachment.html>