Hey Paul,

On Fri, 29 Oct 2010 16:01:12 -0700, Paul Ivanov <[email protected]> wrote:
> I only figured out that the error was on the Python side as I
> typed the previous email, and should have had more patience to
> just do the above myself.  
> 
> The argument to ceil was inf, because step was 0.0, since on line
> 63 of test_clmath, the step argument was passed as (b-a)/s with
> all a,b,s all ints, and s>(b-a), hence the troubles.  

Me no understand still. test_clmath uses 'from __future__ import
division', and I thought that this automatically makes all the integer
division rubbish go away. Why not here?

> I fixed it here [1] (view in your browser using [2]), and
> also made a one line change in test_clmath for the sizes, two of
> which were both 1024 previously.

Ah, well spotted. :) 

I ended up using a slightly different fix:
http://git.tiker.net/pyopencl.git/commitdiff/e44d00da9e
Thanks for submitting this!

Andreas

Attachment: pgpORAUTzccZc.pgp
Description: PGP signature

_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl

Reply via email to