[Numpy-discussion] SSEPlus + Framewave

Holger Rapp Rapp@mrt.uka...
Mon Aug 11 05:17:17 CDT 2008


Hi everyone,

I have a question concerning performance of numpy. I'm using it for  
heavy image processing tasks and often need more speed than the stock  
numpy delivers. Especially in numeric calculations with big arrays (in  
my current case shape = (8,4,1200,1600), 'float'). So I often rely on  
self written special modules which do use the IPP (Intel Performance  
Primitives) to speed up simple tasks like dividing a big array by a  
number.

I realize that it was not really feasible to support a proprietary  
library like the IPP in a beautifully crafted Open Source Project, but  
quite recently, AMD came up with two very interesting projects  
(SSEPlus and Framewave, links provided below) which are more or less a  
direct response to intels IPP. And the best: They are OpenSource  
(under a Apache license, afaik). My question is now: Is it intended/Is  
there interest to get this  performance gain into numpy? Are their any  
political restrictions (license/project identity)? Is there already  
work underway?

I for one would consider helping in a effort like that, because it  
would probably safe me time in the long run.

(Sidenote: I'm aware that this optimization would only help INTEL/AMD  
boxes, but hardware acceleration is so common these days that it is a  
shame NOT to use it in a numbercrunching library. Implementing such a  
library on one architecture might make it easier to implement  
something similar on others too with other libs. Maybe we see CUDA  
support in numpy in the future....)

Here are the links:
http://sseplus.sourceforge.net/
http://framewave.sourceforge.net/

Greetings,
Holger


More information about the Numpy-discussion mailing list