Vikas/docker by davschneller · Pull Request #62 · SeisSol/Training

davschneller · 2026-05-21T19:35:35Z

PR for a PR's sake (i.e. so that we get images built).

Thomas-Ulrich

Shall we compile with Intel compilers (else the RS setups might be very slow)?

Thomas-Ulrich · 2026-06-02T11:46:46Z

+       --disable-dependency-tracking \
    && make -j$(nproc) && make install

+# 2. Build libxsmm (Compiled ONLY on Intel hosts, cleanly skipped on ARM)


"Compiled ONLY on amd64 hosts, "
(e.g. amd processors)

Thomas-Ulrich · 2026-06-02T11:54:56Z

-COPY --from=0 /home/tools tools
-### need to specify --user for gmsh installation, otherwise the tpv13 notebook can't execute !gmsh
-RUN conda install \
+# Cleanly removed the 'gmsh' token from the pip mapping layer


?
remove comment?

davschneller · 2026-06-02T12:00:07Z

-ENV PYTHONPATH="${PYTHONPATH}:/home/tools/lib"
+
 COPY entrypoint.sh /entrypoint.sh
-RUN cd /home/tools/bin/ \


One more thing; this line and the two below should be added again—but with dneon instead of dnoarch .

Otherwise, the ARM build will not really run.

(or you switch everything to a better name, e.g. with the new binary naming)

One more thing; this line and the two below should be added again—but with dneon instead of dnoarch .

Otherwise, the ARM build will not really run.

(or you switch everything to a better name, e.g. with the new binary naming)

Done. Please let me know if I missed naming somewhere. :)

vikaskurapati · 2026-06-02T12:06:28Z

Shall we compile with Intel compilers (else the RS setups might be very slow)?

I thought the jupyter lab container always used GCC and hence I did not make the switch. I can try to build an intel compilers version with icx, icpx on a separate branch if required.

Thomas-Ulrich · 2026-06-02T12:14:19Z

I thought we used ghcr.io/seisscoped/container-base instead of ubuntu to be able to use Intel compilers:

https://github.com/SeisSol/Training/blob/main/Dockerfile#L35-L36

vikaskurapati · 2026-06-02T12:21:33Z

I thought we used ghcr.io/seisscoped/container-base instead of ubuntu to be able to use Intel compilers:

https://github.com/SeisSol/Training/blob/main/Dockerfile#L35-L36

Yes, but the compilers and base Ubuntu of these are quite old. I am not sure if the seisscoped has the latest Intel compilers.

https://hub.docker.com/r/seisscoped/container-base shows that the latest one is 22.04.

Do you know where this is? I can check if we could update the base container to a higher Ubuntu and get the more recent Intel compilers there. But for these smaller test cases, for jupyter lab to be run locally, I think gcc should suffice.

vikaskurapati · 2026-06-02T12:25:23Z

I thought we used ghcr.io/seisscoped/container-base instead of ubuntu to be able to use Intel compilers:
https://github.com/SeisSol/Training/blob/main/Dockerfile#L35-L36

Yes, but the compilers and base Ubuntu of these are quite old. I am not sure if the seisscoped has the latest Intel compilers.

https://hub.docker.com/r/seisscoped/container-base shows that the latest one is 22.04.

Do you know where this is? I can check if we could update the base container to a higher Ubuntu and get the more recent Intel compilers there. But for these smaller test cases, for jupyter lab to be run locally, I think gcc should suffice.

Another consideration would be Intel compilers' support for ARM64. I am not sure if they have support for ARM.

davschneller · 2026-06-02T13:18:23Z

You can also use the oneAPI base compiler images if needed.
Or add the oneAPI repositories to the build (to apt).

And yeah—we might need to switch compilers etc. by arch. It's possible, but probably more involved.

Another question would be—clang? (no icx, but the latest clang)

vikaskurapati · 2026-06-02T13:19:51Z

You can also use the oneAPI base compiler images if needed. Or add the oneAPI repositories to the build (to apt).

And yeah—we might need to switch compilers etc. by arch. It's possible, but probably more involved.

Another question would be—clang? (no icx, but the latest clang)

I have nothing against it, but if @Thomas-Ulrich 's concern is performance with R+S, is the clang performance benchmarked with those simulations?

davschneller · 2026-06-02T13:21:27Z

Personally, I'd rather try sticking with gcc for now (ICX sometimes even seemed a bit slower to me by now)—or how hard would it be to benchmark all 4 scenarios now quickly?

vikaskurapati · 2026-06-02T13:25:35Z

Personally, I'd rather try sticking with gcc for now (ICX sometimes even seemed a bit slower to me by now)—or how hard would it be to benchmark all 4 scenarios now quickly?

I would like to stick with gcc for now too, tbh. But I can try clang/icx/other compiler on a different branch sometime.

Do you mean all 4 scenarios with gcc? I think it is a matter of someone putting the effort to run them with different compilers enough number of times to compare them.

davschneller · 2026-06-02T13:33:02Z

Yeah, run all four with the current container (maybe just recording the ETA); if that's good enough we might be able to skip all the other compiler testing even.

Thomas-Ulrich · 2026-06-02T15:46:42Z


 The simplest way to access the input files for the Gateway is by running ``git clone https://github.com/SeisSol/Training.git`` from your terminal locally. Make sure to navigate into the folder _quakeworx_ when uploading the input files to the SeisSol App from the Gateway.
+
+SeisSol binaries in the training container follow NEW_BINARY_NAMING (for example, `seissol-cpu-elastic-p4-f64`).


NEW_BINARY_NAMING?
follow the new binary naming scheme?

Thomas-Ulrich · 2026-06-02T15:48:33Z

 - rconv (tool to describe point and finite source models in SeisSol's NetCDF Rupture Format, https://seissol.readthedocs.io/en/latest/standard-rupture-format.html#how-to-use-rconv)
 - SeisSol O4 (pre-compiled SeisSol with 4th order space-time accuracy for elastic and viscoelastic materials, https://seissol.readthedocs.io). We use the SeisSol `v1.0.1`. See the first lines of the SeisSol output to get the exact commit hash.

+The SeisSol binaries in the container use NEW_BINARY_NAMING (e.g., `seissol-cpu-elastic-p4-f64` and `seissol-cpu-viscoelastic-3-p4-f64`).


change label address comments

vikaskurapati · 2026-06-03T07:41:32Z

@Thomas-Ulrich @davschneller, any other comments? I think we are ready to merge unless you have any other issues you want to address as part of this PR. I think fixing the simulation files should be a separate PR.

Thomas-Ulrich · 2026-06-03T08:18:44Z

you can merge

davschneller · 2026-06-03T10:27:33Z

LGTM from my side as well

vikaskurapati added 5 commits May 20, 2026 17:07

Edited dokerfile

8282421

fix rconv builds

fda1dc0

moved modifications

a0cce61

docker file with arm and x86

fd44530

Working status on hsw

a85f4aa

vikaskurapati mentioned this pull request May 22, 2026

Update CI actions #63

Merged

Thomas-Ulrich approved these changes Jun 2, 2026

View reviewed changes

address comments

fb8fe56

davschneller commented Jun 2, 2026

View reviewed changes

modify binary naming

ae9946c

Thomas-Ulrich approved these changes Jun 2, 2026

View reviewed changes

add note in readme about memory

7701e3b

change label address comments

vikaskurapati force-pushed the vikas/docker branch from da90d75 to 7701e3b Compare June 2, 2026 16:19

vikaskurapati marked this pull request as ready for review June 3, 2026 07:33

Merge branch 'main' into vikas/docker

5a88cb5

vikaskurapati approved these changes Jun 3, 2026

View reviewed changes

davschneller merged commit 8e3ccb8 into main Jun 3, 2026
1 check passed

davschneller deleted the vikas/docker branch June 3, 2026 10:28


		The simplest way to access the input files for the Gateway is by running ``git clone https://github.com/SeisSol/Training.git`` from your terminal locally. Make sure to navigate into the folder _quakeworx_ when uploading the input files to the SeisSol App from the Gateway.

		SeisSol binaries in the training container follow NEW_BINARY_NAMING (for example, `seissol-cpu-elastic-p4-f64`).

Conversation

davschneller commented May 21, 2026

Uh oh!

Thomas-Ulrich left a comment

Choose a reason for hiding this comment

Uh oh!

Thomas-Ulrich Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

Thomas-Ulrich Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

davschneller Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vikaskurapati Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

vikaskurapati commented Jun 2, 2026

Uh oh!

Thomas-Ulrich commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vikaskurapati commented Jun 2, 2026

Uh oh!

vikaskurapati commented Jun 2, 2026

Uh oh!

davschneller commented Jun 2, 2026

Uh oh!

vikaskurapati commented Jun 2, 2026

Uh oh!

davschneller commented Jun 2, 2026

Uh oh!

vikaskurapati commented Jun 2, 2026

Uh oh!

davschneller commented Jun 2, 2026

Uh oh!

Thomas-Ulrich Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

Thomas-Ulrich Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

vikaskurapati commented Jun 3, 2026

Uh oh!

Thomas-Ulrich commented Jun 3, 2026

Uh oh!

davschneller commented Jun 3, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

davschneller Jun 2, 2026 •

edited

Loading

Thomas-Ulrich commented Jun 2, 2026 •

edited

Loading