Re: MIOpen package LFS files
On 2024-05-10 04:43, Xuanteng Huang wrote:
> I’ve uploaded the “pure” version of MIOpen without *.bz2 and *.kdb files to a new repo [20].
Thanks, Xuanteng.
> The source tarball is imported by gbp with —-filter=“*.bz2” —-filter=“*.kdb” options, and listing in Files-Excluded in d/copyright as well.
Tip: if you use gbp import-orig --uscan, uscan will automatically repack
the tarball whilst filtering Files-Excluded.
>> Agreed. I think my only remaining question is whether those big assembly files are ok. They contain macros, so they're clearly not plain disassemblies, and I think their size might be exaggerated by loop unrolling. Still... I can't imagine that they were hand-written, so I think we need to know more about them.
>
> One question for DFSG: can we regard the assembly code as a form of “source code”, even if they are generated from the unknown artifact (e.g., some compiled binary objects)?
> Theoretically, the assembly is fully explainable as the AMD GPU ISA manual is public [21].
Unfortunately, no. The source in question must be the "preferred form
for making modifications", so unless upstream indeed wrote these by
hand, they would not be compliant.
(There are corner cases, eg: when the original form no longer exists.)
>> We need to ask upstream. We should find out how they were created and how they are updated (e.g., if bugs are discovered). I would suggest opening an issue to ask the question on the upstream GitHub repository.
>
> Issue submitted [22].
Fantastic, thanks!
Best,
Christian
Reply to: