[Numpy-discussion] how to use numpy-vendor
Tue Aug 14 09:22:17 CDT 2012
On Tue, Aug 14, 2012 at 6:32 AM, Frédéric Bastien <firstname.lastname@example.org> wrote:
> On Tue, Aug 14, 2012 at 6:33 AM, Dag Sverre Seljebotn
> <email@example.com> wrote:
>> On 08/14/2012 06:34 AM, Ondřej Čertík wrote:
>>> How should one use the "vendor" repository (https://github.com/numpy/vendor)
>>> in Wine? Should I put the binaries into .wine/drive_c/Python25/libs/,
>>> or somewhere else?
>>> I've search all mailinglists and I didn't find any information on it.
>>> I vaguely remember
>>> that somebody mentioned it somewhere, but I am not able to find it.
>>> Once I understand it,
>>> I'll send a PR updating the README.
>>> I've played with OpenBlas and managed to compile numpy with it on linux,
>>> following the tutorial  and it works, so at least on linux it's clear to me.
>> One thing to be aware of with OpenBlas is that it is *very* tuned to the
>> CPU at hand. As in, every CPU has hand-coded *assembly* and the makefile
>> more or less probes for which specific CPU generation from which vendor
>> you have and compiles and link the corresponding assembly file. So you
>> may have to take some care that you don't compile and ship a version
>> that breaks if you don't have SSE3 installed etc...
>> (Unless OpenBlas has changed recently. I'm not saying I'm right, I'm
>> saying it should be looked into.)
> OpenBlas as the option to build all version and select at run time the
> right one. MKL do the same.
> But I never tested it. So I don't know how well it work. The other
> option would be to force an older CPU that support only sse2. ATLAS do
> that by default. OpenBLAS select the best one for the computer where
> it is being built by default.
For the record, I don't plan to ship openblas, I was just playing with it
to make sure I understand how things work.
More information about the NumPy-Discussion