[SciPy-dev] Fwd: scipy/lapack

Charles سمير Doutriaux doutriaux1@llnl....
Wed Dec 3 16:25:21 CST 2008


Just forwarding, maybe a more general way could be implemented? or at  
least an option for lapack3.2

Here's what's happening

We have no lapack/blas on the system
we built numpy (1.2.1 i believe)

now we're trying to build scipy using LAPACK_SRC / BLAS_SRC

But we downloaded LAPACK 3.2

it seems as the name of the source files is hardcoded somewhere at the  
numpy level and in 3.2 there's a lot of new source files.

Any clues on this? (in the mean time we're building with 3.1)

C.


Begin forwarded message:

> From: Bob Drach <drach@llnl.gov>
> Date: December 3, 2008 2:04:35 PM PST
> To: Bob Drach <drach@llnl.gov>
> Cc: Charles Doutriaux <doutriaux1@llnl.gov>
> Subject: Re: scipy/lapack
>
> numpy.distutils.system_info is hardwired for lapack 3.1. It has a  
> list of the routines to build which differ from 3.2. This module is  
> used by scipy to figure out which sources to build. There's no way  
> it would build with lapack 3.2 that I can see. I'm rebuilding from  
> scratch with blas and lapack enabled to see the current repository  
> version works.
>
> lapack-lite has the same functions as lapack, without the html and  
> man pages. So I'm happy with that if it works.
>
> Bob
>
>
>
> On Dec 2, 2008, at 7:10 PM, Bob Drach wrote:
>
>> Hi Charles,
>>
>> Do you know a straightforward way to build a usable scipy.linalg  
>> module? I don't care about having optimized Atlas libraries, etc.
>>
>> Bob
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/scipy-dev/attachments/20081203/f6d53324/attachment.html 


More information about the Scipy-dev mailing list