Consider to drop -Werror for tests/configs pulling external APIs/frameworks about dbcsr HOT 12 OPEN

@alazzaro , could you point me to all the functions that DBCSR is using but is deprecated now? Is there an error report from CI that I could refer to?

For that, please see @hfp issue of today (#810)

from dbcsr.

Oh, oh, I didn't pay much attention... #810 is related to CUDA...
Then, for rocm, the deprecated functions are:

https://github.com/cp2k/dbcsr/actions/runs/9636368793/job/26574254231

Search for "deprecated and may"...

from dbcsr.

Are your comments meant for this issue?

yes, but it seems I omitted some part: by passing the include dirs for external API/frameworks with -isystem, at least gcc should not be emitting warnings coming from those headers.

from dbcsr.

So, the problem here is that we use FROM rocm/dev-ubuntu-22.04:latest which moved from ROCM 6.0.1 to 6.1.0 and now some HIP functions are deprecated...
I wonder if we should set a tag for ROCM (up to 6.0.2 for example). I don't think we should remove -Werror...
ROCM 6.1 introduced a long list of deprecated functions, however I'm not sure how we should update them... Maybe @gsitaram can help?

@hfp I would say, for the moment, please ignore this problem.

from dbcsr.

@hfp , you can also add -Wno-error=deprecated-declarations for the moment...

from dbcsr.

I'll take a look and get back.

from dbcsr.

Hi @alazzaro, @hfp, I see the following message in the ROCm docs:

On the AMD platform, context management APIs are deprecated as there are better alternate interfaces, such as using hipSetDevice and stream APIs to achieve the required functionality.

On the NVIDIA platform, CUDA supports the driver API that defines "Context" and "Devices" as separate entities. Each context contains a single device, which can theoretically have multiple contexts. HIP initially added limited support for these APIs to facilitate easy porting from existing driver codes.

These APIs are only for equivalent driver APIs on the NVIDIA platform.

Are these functions necessary in DBCSR? Or could we replace them with equivalent hipDevice* functionality?

from dbcsr.

Thanks @gsitaram I have no idea how to answer to your question...
I spent some time to understand how to replace them, but I cannot get it from the AMD documentation...

from dbcsr.

@alazzaro , could you point me to all the functions that DBCSR is using but is deprecated now? Is there an error report from CI that I could refer to?

from dbcsr.

in principle CMake should be adding header files from external packages with -isystem, on the other hand it will then just start failing hard when things are finally being removed

from dbcsr.

moving our docker image to use a tagged upstream and then use something like the pre-commit bot to scan the images periodically might be a better method?

from dbcsr.

in principle CMake should be adding header files from external packages with -isystem, on the other hand it will then just start failing hard when things are finally being removed

moving our docker image to use a tagged upstream and then use something like the pre-commit bot to scan the images periodically might be a better method?

Are your comments meant for this issue?

from dbcsr.