[Numpy-discussion] [Fwd: Re: [SAGEdev] faster random number generation]

Travis Oliphant oliphant at ee.byu.edu
Mon Mar 13 13:29:03 CST 2006


Fernando Perez wrote:

> Hi all,
>
> I'm wondering if one of our resident experts who is more familiar than 
> me with the details of the current build process could help us here a 
> little.  This is a message from the SAGE development list, where I am 
> encouraging them to integrate more tightly with numpy.
>
> SAGE is a very interesting python-based system for mathematics:
>
> http://modular.ucsd.edu/sage/
>
> After my comment below on numpy NOT needing a fortran compiler, 
> William came back with that g77 error message.  Now I'm a little 
> confused, and before I confuse him even more, I'd prefere to clarify 
> things with those who actually konw what's going on.
>
> Any help will be greatly appreciated.


Pearu is the expert on this.    The _dotblas.c function is optimized dot 
using ATLAS and/or BLAS for the system.  I'm not sure why g77 is being 
used as the linker unless it is in their configuration somehow.

For example, on my system, this is the command that links _dotblas.so

cc -pthread -shared 
build/temp.linux-i686-2.4/numpy/core/blasdot/_dotblas.o -L/usr/lib/atlas 
-lptf77blas -lptcblas -latlas -o 
build/lib.linux-i686-2.4/numpy/core/_dotblas.so

Please have them post to the list to figure out their issues.  I want to 
squash any idea that numpy *requires* a  Fortran compiler as this was a 
major design decision not to.   It's always possible that configuration 
issues need to be improved for certain platforms.


-Travis





More information about the Numpy-discussion mailing list