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

Re: Question about how to handle HIP vs hipamd



Hi Cory,

Cordell Bloor, on 2022-05-15:
> I've been having trouble reproducing your build. I got further once I switch
> to gbp, but I'm still getting stuck before it actually begins to build the
> package. This is my full process:
[…]
> dpkg-source: info: local changes detected, the modified files are:
>  rocm-hipamd/opencl/tests/ocltst/module/perf/oclperf.exclude
> dpkg-source: error: aborting due to unexpected upstream changes, see
> /tmp/rocm-hipamd_5.0.0-1~exp1.diff.qe8gdD
[…]
> I've tried a few things to try and fix this (including dpkg-source --commit
> and quilt refresh), but thus far I have failed to advance any further. I
> will need some help to reach the point at which I can debug the failures you
> describe.

I followed your footsteps and reproduced your issue.  According
to cat -v of the patch file in /tmp/rocm-hipamd_5.0.0-1~exp1.*,
the file oclperf.exclude is converted somehow from LF to CRLF.
I could work around the issue with dos2unix on oclperf.exclude
while patches are unapplied, and commiting that change (and not
pushing it back to salsa).  You should the be able to reproduce
the __managed__ issue once this is done.  If there are further
interferences, the source code may be rewrapped to exclude .git*
files.

I didn't caught that on my end because I configured gbp to
trigger sbuild when building a package.  sbuild will trigger the
build in yet another layer of isolation through schroot.

> I notice that Maxime reported an issue related to oclperf.exclude [1], which
> I have now opened an internal ticket for. The low-level ROCm components use
> an entirely different set of tools and processes for making code
> contributions as compared to the ROCm math libraries I'm familiar with, but
> Jeremy recently taught me how it's done. I'll submit a patch to correct that
> issue upstream.
> 
> [1]: https://github.com/RadeonOpenCompute/ROCm-OpenCL-Runtime/issues/140

That definitely sounds related.

Thank you for your time investigating this issue!

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: Steve Morse Band - More To The Point

Attachment: signature.asc
Description: PGP signature


Reply to: