[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ROCm RFS - workflow and pipeline

On 2/3/22 23:45, M. Zhou wrote:
> rocr-runtime still has build dependency in NEW queue.
> If we upload that it will immediately receive FTBFS bug
> due to missing dependency.
> Let me know if any of you truly want to see the package
> in experimental with its build dependencies being absent.
> Once B-D's are set I'll take care of it.

If NEW is the bottleneck for the whole Debian project, as we currently
see it being discussed in debian-devel@l.d.o., I would tend to think
that pushing interdependent packages simultaneously in the queue would
be beneficial to the ROCm-Debian packaging effort.

If we push sequentially without masked time (parallel);
if on average it takes one month to pass NEW for one package;
if hipblas depends on hipsolver depends on rocsolver depends on rocblas
  depends on rocrand depends on hipamd depends on rocr-runtime
  depends on rocm-device-libs;
it would take around 8 months for hipblas to reach experimental!
And then there are the barriers exp => sid => testing => stable...

Being ready to announce an _unstable_ packaging of the ROCm stack
2-3 months after ROCm 5.0.1 releases (bound to happen in the following
weeks), would be good for the mindshare of ROCm in my opinion.

> On Thu, 2022-02-03 at 22:55 +0100, Étienne Mollier wrote:
>> I'm hopeful to have time to make a final review and upload of
>> the rocm-smi-lib in the weekend, I'm also fine if someone else
>> beats me at it in the meantime.  I don't claim rocr-runtime as I
>> believe Mo Zhou has more experience on this one, and leave
>> rocminfo open for claims too for the moment.
>> Thank you for your contributions!

Thank you Etienne, no big pressure - in that it is not a matter
of days/weeks is it!

Best regards, Maxime

Reply to: