[Numpy-discussion] Intel MKL - was: parallel numpy - any info?

David Cournapeau david@ar.media.kyoto-u.ac...
Thu Jan 10 04:42:36 CST 2008


Albert Strasheim wrote:
> Hello
>
> On Jan 10, 2008 6:56 AM, David Cournapeau <david@ar.media.kyoto-u.ac.jp> wrote:
>   
>> The one thing which I am not sure about is:  say one MKL binary does not
>> work, and say I (or anyone outside your company) build numpy with the
>> MKL ro debug it, can I redistribute a new binary, even if it is just for
>> testing purpose ?
>>     
>
> Let's say Ray's company buys one copy of Intel MKL. This gives them
> the Intel MKL DLL and link libraries (.libs).
>
> Now they compile NumPy and link it against Intel MKL. They can then
> (as I understand the license agreement, IANAL, etc.) distribute this
> binary and the Intel MKL DLLs.
>
> They may *not* distribute the link libraries. This means that there is
> no easy way for anyone else to build a new executable that is linked
> against MKL.
>   
Ah yes, I forgot about this windows idiosyncrasy of import libraries. In 
this case, you have a clear technical difference between a shared 
library for distribution and for developers. But anyway, I now realize 
that the MKL is not free on windows anyway, so this is a non problem.

cheers,

David


More information about the Numpy-discussion mailing list