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

Re: rocprim review (Was: rocprim and rocthrust copyright review)



Hi Cory,

Étienne Mollier, on 2022-10-22:
> Étienne Mollier, on 2022-10-13:
> > Cordell Bloor, on 2022-10-13:
> > > I've been working on librocprim2-dev (5.3.0-1). I think it just needs to be
> > > updated with the ITP bug number (#1021695) and go through copyright review,
> > > then it'll be ready for submission. The packaging depends on some of the
> > > fixes for rocm-hipamd that are currently in experimental, so it would
> > > probably go into experimental itself. The librocthrust2-dev (5.3.0-1)
> > > package is still a work-in-progress, but it's very similar to rocprim and
> > > should reach the same state within the next few days.
> > 
> > That's great to read!  I will be busy focusing on release
> > critical bugs during a BSP this weekend, but I'll be happy to
> > help with the review next week while migrating ROCm 5.2.3 to
> > unstable, if no one beats me at one or both.
> 
> I took some time to lookup the rest of the packaging for rocprim
> and here are my observations:
> 
> d/control:
>  * don't hesitate to add yourself as one of the uploaders if you
>    are interested in packaging rocprim an its maintenance;
>  * I would specify that Rules-Requires-Root: no, this is a
>    pedantic lintian flag though.
> 
> The package is in good shape in my opinion.  Please ready the
> d/changelog for upload to unstable instead of UNRELEASED (with
> dch -r), and request for sponsoring here once you're happy with
> the state of the package for upload to New.

I've gotten to run the test suite on gfx1030, but I see test
failures affecting sorting routines:

	 14 - rocprim.block_sort (Failed)
	 28 - rocprim.device_segmented_radix_sort (Failed)
	 44 - rocprim.warp_sort (Failed)

Is it something you reproduce on your end or did I broke
something?  (I vaguely recall having seen test failures on the
gfx803 but initially put that on account of the unsupported
hardware.)  I did two runs to make sure there were no momentary
hiccups[1,2] (note the first was against a personnal build of
hipcc, so I retried against pristine sid).

[1]: https://people.debian.org/~emollier/logs/rocm/rocprim_5.3.0-1_amd64-2022-11-02T21:20:15Z.build
[2]: https://people.debian.org/~emollier/logs/rocm/rocprim_5.3.0-1_amd64-2022-11-03T21:15:14Z.build

Have a nice day,  :)
-- 
Étienne Mollier <emollier@emlwks999.eu>
Fingerprint:  8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
Sent from /dev/pts/2, please excuse my verbosity.
On air: Haven of Echoes - Sirensong

Attachment: signature.asc
Description: PGP signature


Reply to: