Opinions on packaging ATLAS (for the x86 architecture)

[Deji Akingunola]

On Fri, Sep 25, 2009 at 11:30 AM, Jonathan Underwood
<jonathan.underwood at gmail.com> wrote:
> 2009/9/25 Deji Akingunola <dakingun at gmail.com>:
>> On Fri, Sep 25, 2009 at 10:57 AM, Jonathan Underwood
>> <jonathan.underwood at gmail.com> wrote:
>>
>>>
>>> Would it not be best to have the default package using the default
>>> CFLAGS for consistency with the rest of the distribution, and the
>>> subpackages being variants which override the CFLAGS?
>>>
>> This is a different issue altogether. Even if atlas is made to used
>> the default CFLAGS, it will still attempt to build for a particular
>> CPU (either for the archictetural default passed to it at build time
>> or the hardware on which it is being built), that's how the package
>> was designed.
>
> Well, my point was that packages that do respect CFLAGS get built
> currently with certain options, and however Atlas builds, the default
> package should have comparable options. On x86, I see that is
> presently -march=i686 -mtune=atom, and IIRC -march=i686 implies sse

This is actually one of the reasons the atlas package doesn't use the
CFLAGS; those options can be too generic (or too specific), atlas
build procedure already tries to build for particular cpu types.