fastmath library?

[Michael Hudson]

"Alex Martelli" <aleaxit at yahoo.com> writes:

> "Michael Hudson" <mwh21 at cam.ac.uk> wrote in message
> news:m3zoizr5nx.fsf at atrus.jesus.cam.ac.uk...
> > "Pete Shinners" <pete at visionart.com> writes:
> >
> > > is there a "fast" math library available?
> > >
> > > one that uses high-speed lookup tables and estimates for
> > > low quality results? i'd mainly like one that can handle
> > > things like square-root, cosine, etc, etc
> > >
> > > i figure there's got to be something like this already
> > > available for python?
> >
> > Wouldn't have though there'd be much point - the faffing around Python
> > does in calling a function surely dwarfs the cost of the actual
> 
> Indeed, *measuring* the proposed 'memoize' solution shows that:

[snip]

> I.e., the lookup-table is slowing things down
> by a factor of 4 or more (while the actual operation
> of math.sin accounts for just slightly more than
> 50% of the call-cost -- the comparison with the
> dummy-function call tells us that).

Hmm... you didn't post your dummy function!  Anyway, I bet actually
computing the sin of the argument is still dwarfed by the function
call mechanism, so even if you have a fast C memoization/lookup thingy
then you aren't likely to save even 50% of the call cost.  I could be
wrong, though (and am more likely to be on machines without an FPU).

Cheers,
M.

-- 
  It's a measure of how much I love Python that I moved to VA, where
  if things don't work out Guido will buy a plantation and put us to
  work harvesting peanuts instead.     -- Tim Peters, comp.lang.python