Re: [RFC] multiarch and virtual packages
Le 03/10/2013 13:28, Simon McVittie a écrit :
> On 03/10/13 10:54, Vincent Danjean wrote:
>> The only "problem" is that it is not currently possible to install
>> an ICD Loader:i386 from one vendor and an ICD Loader:amd64 from another
> Is there a valid reason to do that, other than "because I can"? If
> nobody would actually want to do that, then it isn't a problem :-)
> Indeed, if the ICD loaders all provide the same ABI and functionality,
> does there need to be anything beyond "the best one" in Debian? We only
> need alternative implementations if there are significant reasons to
> prefer A over B, and also significant reasons to prefer B over A.
Some might want to be able to use the ICD Loader of the vendor of the
ICD they use, so that this vendor accepts bug reports for its ICD.
And, until recently, the code of most (all?) vendors comes from a
closed source repository of Khronos. It means that the free implementation
(ocl-icd) had to do some reverse engineering at each new version
(mostly to get the correct order in a table of functions). Now,
the Khonos code can be seen (the code is still non-free) in a tarball.
We can hope that future versions will also be available but it is not
sure. Ie, we might need to use vendor ICD Loader during a few time
when new OpenCL versions are released.
That said, the first remark is a good one. I'm not convince at all
that we must fight in order to be able to install ICD Loaders from
different vendors for different architectures in multiarch mode.
Vincent Danjean GPG key ID 0x9D025E87 email@example.com
GPG key fingerprint: FC95 08A6 854D DB48 4B9A 8A94 0BF7 7867 9D02 5E87
Unofficial pkgs: http://moais.imag.fr/membres/vincent.danjean/deb.html
APT repo: deb http://people.debian.org/~vdanjean/debian unstable main