How to put our new package (for free) into Karmic 9.10

[Christopher James Halse Rogers]

On Mon, 2009-07-27 at 12:10 +0200, Ilgis Ibragimov wrote:
> Hi,
> 
> thank you for your kind answers. Please, see my answer below.
> 
> Max Bowsher wrote: 
> > Jonathan Carter (highvoltage) wrote:
> >   
> > > Hi Ilgis
> > > 
> > > Ilgis Ibragimov wrote:
> > >     
> > > > One small question regarding to our package. We have it right now as
> > > > sources, and we can provide as sources. However, it is highly desirable
> > > > to build many (about 10) different libraries for different platforms,
> > > > for example:
> > > > 1. CPU AMD 32bit,
> > > > 2. CPU AMD 64bit,
> > > > 3. CPU Intel 32bit,
> > > > 4. CPU Inter 64bit,
> > > > 5. GPU NVIDIA 8xxx,
> > > > 6. GPU NVIDIA 9xxx,
> > > > 7. GPU NVIDIA 2xxx,
> > > > and probably some other minor things depending on amount of Cores in CPU.
> > > > 
> > > > We can provide binaries as well. Would you please, to tell me is the
> > > > REVU is the right place where I get an information how to organize the
> > > > package so that it provide only the best suitable version for user?
> > > >       
> > > I believe you will only need to upload the source package to REVU, from
> > > there the build server(s) should be able to build the the binary
> > > packages.
> > >     
> > 
> > REVU is a tool for publishing proposed source packages for review by
> > MOTUs and the community - no automatic builds are performed on REVU uploads.
> > 
> > REVU is definitely the right place to be publishing a new source package
> > you would like advice and comments on, but asking specific questions
> > either on this mailing list or on the #ubuntu-motu IRC channel is also
> > advisable.
> > 
> > In particular, why does your package need so many platform-variant
> > builds? Mostly Ubuntu only differentiates between x86 and x86_64 CPU
> > architectures.
> >   
> Our package is very processor dependent. It means we have several
> different algorithms depending on amount of CPU cores, or processor
> core and on the amount of L1/L2/L3 cache. It is really highly
> commercially tuned. We decide to share it in Ubuntu community, so, it
> is the main reason why we should build several versions depending on
> hardware. We may install CPU detector in sources, but it will force
> user to compile this package, that might be not very convenient.

I presume this detection isn't done at run-time for performance reasons?
I would naively expect the library to perform this detection once at
initialisation and from then on call into the appropriate
system-specific code, but there's probably a reason you aren't already
doing this.

As long as all these different build configurations can be manually
configured (ie: don't automatically tune for the system they're being
built on) it would be possible to build all these different binary
packages; it'll just be a bit of a pain.  Both for the packaging and for
the users.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 489 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/ubuntu-motu/attachments/20090727/ea95ee3c/attachment.pgp