The program I wish to package uses SSE and SSE2: can it be included in rpmfusion?

[Andrea Musuruane]

Hi,

On Mon, Sep 3, 2012 at 6:51 PM, Nicolas Chauvet <kwizart at gmail.com> wrote:

> 2012/9/3 Giles <gbirchley at blueyonder.co.uk>:
> > Hi,
> >
> > I'm trying to make a package pcsx2, a playstation2 emulator, for Fedora
> to
> > be included on the rpmfusion repository. (Bug 2455 - Review request:
> pcsx2 -
> > A Sony Playstation2 emulator).
> > I have run into problems as I have removed the SSE and SSE2 cflags and
> now
> > the package does not build. I am being given some advice by another
> > packager, who thinks this may disqualify the package from rpmfusion. Is
> > there any way to include a package with such dependencies in the
> repository?
>
> I'm the one who suggested Giles to write here.


> I don't know any guideline that would prevent this ?
> But someone could point it to me.
>

Not all i686 arch supported by Fedora have SSE or SSE2 capabilities. AFAIK
some of the %{optflags} were selected to match these architectures. Hence
someone could install a package that won't work on his PC if you enable SSE
and SSE2 flags.


> Using runtime detection would be expected in nowadays code, but if the
> default implementation is sse/sse2 this couldn't be expected by
> default on i686, whereas this is always present on x86_64 (and maybe
> ia64 if relevant).
>

ATM the package in question won't even build without SSE and SSE2 flags and
it can't be compiled on x86_64.


> I don't see any point to use ExclusiveArch: x86_64 that would prevent
> some i686 users because of that.
>

Since it cannot be compiled on x86_64, the packages is  ExclusiveArch: i686


> Also I would compare with 3D card, you will likely hit a hardware
> limitation somewhere if you load a huge 3D CAD model on an Intel IGP
> with an Atom CPU, this will crash early somewhere.
> So in your case, if you disable sse2, you may lower the level of such
> hardware limitation, because then sse2 capable CPU may not be able to
> run the software appropriately with sse1 only CPU already too short
> for this emulation anyway.
>
> So I would prefer to have a simple check on i686 for sse2 capability
> as an entry point instead of disabling sse2 and have sse2 capable CPU
> to behave like sse1 only one and not been able to run the software
> whereas they should have.
>

I agree. A check with upstream can be enlightening :-)

Bye,

Andrea.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.rpmfusion.org/pipermail/rpmfusion-developers/attachments/20120903/cc5f1f55/attachment.html>