On Thu, Jan 10, 2008 at 07:15:46PM +0000, Peter Clifton wrote:
> 
> Clearly the "intersect" routine is consuming the most here, with 58% of
> of the profile time in THIS function. Cumulatively, (that is including
> any functions called FROM "intersect", it is 78% of profile time.
> 
> Anyway... looks like there is an obvious place to start hacking.

I disagree for two reasons:  An innermost function like this being
called many, many times is represented disproportionately in profiling
output due to profiling overhead.  Also, the problem is really that it
is being called too many times.  If you fix that, it won't matter how
fast it is individually.

To give one example:  I once profiled a program that was very slow and
the bulk of the time was in a string compare routine.  I made the whole
thing about 10% faster by making it call an optimized one that was already
in the C library.  I made the program 30 *times* faster by realizing that
the btree it was using had no balancing and the entries were being inserted
in order (ie it was a giant linked list).

-- 
Ben Jackson AD7GD
<[EMAIL PROTECTED]>
http://www.ben.com/


_______________________________________________
geda-dev mailing list
[email protected]
http://www.seul.org/cgi-bin/mailman/listinfo/geda-dev

Reply via email to