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

Re: Finishing the ROCm 6 update



Hello,

I looked at the repository, and made a branch based on xuantengh branch
didn'd see or forget there was a MR .. don't want to break anything anyway


I made a patch refresh for 6.1.5
I fixed the warning on libhiprtc.so in the wrong package
(moved to libamdhip64-dev) hope this is correct with https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1088540)

There is an Information about reproducibility that seems to come from
-DHIP_COMMON_DIR=$(shell realpath hip)

I started a salsa-ci to give more visibility

https://salsa.debian.org/rocm-team/rocm-hipamd/-/jobs/7323734
There are 6 remaining warnings and Info that should probably be addressed

W: rocm-opencl-icd: package-name-doesnt-match-sonames libamdocl64
W: libhiprtc-builtins6: shared-library-lacks-prerequisites [usr/lib/x86_64-linux-gnu/libhiprtc-builtins.so.6.1.40093] W: rocm-opencl-icd: shared-library-lacks-version usr/lib/x86_64-linux-gnu/libamdocl64.so libamdocl64.so I: rocm-hipamd source: duplicate-short-description libhiprtc-builtins6 libhiprtc6 [debian/control] I: libhiprtc6: no-symbols-control-file usr/lib/x86_64-linux-gnu/libhiprtc.so.6.1.40093 I: rocm-opencl-icd: no-symbols-control-file usr/lib/x86_64-linux-gnu/libamdocl64.so I: rocm-hipamd source: out-of-date-standards-version 4.7.0 (released 2024-04-07) (current is 4.7.2) I: rocm-hipamd source: patch-not-forwarded-upstream [debian/patches/0008-find-package-minimum-policy.patch]


I would rename rocm-opencl-icd as libamdocl64 keeping a virtual
rocm-opencl-icd
For symbols, i would nee some advice.
And also how to fix the shared-library-lacks-version, move somewhere as recommended, or a reason to override ?

Seems to be a problem with dependency on libamd-comgr2 i don't really understand yet, baut has certainly to be adressed before an upload


Regards

Christian



Le 26/03/2025 à 04:30, Cordell Bloor a écrit :
Hi Christian,

On 2025-03-03 10:53, Christian Kastner wrote:
Uploading to NEW isn't the obstacle, Petter and I (or others) can take
care of that.

It's the review that appears to have stalled, so on the contrary, that
would actually be a good opportunity to insert yourself, if you like.

The review seems the most important part here, as it is a core package
and one must also consider how it ties into/affects the overall ecosystem.

In my final check before an upload to NEW, I myself wouldn't (couldn't)
check this integration part too much, and would entrust that this had
been settled by the active reviewers.

While I appreciate your confidence in me, I do not believe that my input is required. If there are integration issues, we'd find them in experimental. That's what it's for, after all.

Sincerely,
Cory Bloor


Reply to: