Re: [Chicken-users] performance of bignums

[Peter Bex]

On Thu, Jun 25, 2015 at 10:09:19PM +0200, Peter Bex wrote:
> I already removed the use of the "format" egg (the code contains a
> commented-out version that relies only on "display"), and even
> completely disable the println call.  It's hard to be sure, but it
> *looks* like the majority of the time is being spent in the division
> procedures.

So far it seems my implementation of Burnikel/Ziegler division is rather
unstable, performance-wise.  If I disable burnikel/ziegler so it falls
back to the traditional gradebook method, the benchmark finishes in
a quarter of the time it takes to run with BZ enabled.

It's not just that the BZ implementation is really bad: it does improve
performance on Chudnovsky's "digits of pi" calculation benchmark by a
factor of four or so.  I will do further experimentation.  I also noticed
a bug in the gradebook method, which I will have to address as well.

Cheers,
Peter

signature.asc
Description: Digital signature