JOSS Review - Cannot compile #3256
Assignees
Labels
type: new
A new issue has been created and requires attention
Comments
berenger-eu
added
the
type: new
|
@berenger-eu thank you for your review! Could you double check the following? In addition, could you attach the files Thank you! |
|
That seems to exist: I did not find a CMakeError.log, but I have the file attached to this message. |
|
Thank you for providing the log file. Yes, this looks like an MPI installation issue and I would recommend checking with your HPC system admin: |
|
Hello @berenger-eu ,
|
|
Hi, not sure if it's good to add this here but the error is similar and I'm also a JOSS reviewer. Same error but on a different system (macOS arm with OpenMPI 5.0.3). Note: The error seemed to have only appeared after installing openblas through homebrew. I usually use openblas on macOS through framework.Accelerate. Before You find attached my host-config file and the CMakeOutput from myhost.cmake.txt (added .txt to be able to upload to github) I guess I would have to remove |
|
@timokoch maybe this is relevant to you: https://discourse.cmake.org/t/appleclang-15-new-linker-general-issues/9106 |
|
Hello @timokoch,
vs was this required for you?
I have to configure a travel laptop (MacBook) from scratch this afternoon. I will confirm the steps that work for that system. |
|
|
My issue seems to be fixed in a recent open-mpi version (using brew install open-mpi --HEAD which gives 5.1.0a1). Possibly open-mpi/ompi#12664 which fixes something related from 5.0.4. Stable homebrew version is 5.0.3 though which is unfortunate and the version that seems to be incompatible with apple clang 14. I'm pretty sure there is no issue with apple clang 15 as well from what I read. |
|
@timokoch I have openmpi 5.0.3 working on my primary laptop. I will do a vanilla install on a fresh system this afternoon and try to reproduce the errors. |
|
Hello @berenger-eu , @timokoch |
|
@berenger-eu what time zone are you in? Perhaps we can do a zoom call to get to a solution faster? |
|
So I was able to configure with CMake, but it seems that getting the dependencies does not work: The files seems to exist: |
|
@berenger-eu Could you run the following and try again? Note you need |
|
Right, lfs seems not to be present. |
|
@berenger-eu You actually have to install git-lfs (https://git-lfs.com/) which is a separate plugin to make this command work. |
|
Almost work, git lfs now works, but there are other issues. But for some dependencies it worked: |
|
@berenger-eu can you retry the download step? If it doesn't work, can you attach the error log? |
|
Yes it seems the archives are incomplete: git lfs seems to freeze. Where are those files hosted ? It could be that it is forbidden on the supercomputer I am using. |
|
They are hosted on github cloud on dedicated LFS servers. You can find direct links to each package here: https://github.com/GEOS-DEV/thirdPartyLibs/blob/master/tplMirror |
|
Hello @berenger-eu. The use of "git lfs pull" is documented, but perhaps more emphasis should be placed on it: |
|
@berenger-eu Also, just for informational purposes, what institutional system are you installing on? |
|
I am running on a small supercomputer at inria (https://www.plafrim.fr). |
|
Ok it seems thirdPartyLibs compiled correctly. Maybe I forgot something, but my command is really similar to the one in the doc |
|
Hello. There should be an install directory under the tpl directory. For instance, on my Mac, the install is at: |
|
Everything is installed correctly for third party: It is the build of GEOS that fails. |
|
@berenger-eu, can you try building GEOS with the following command? |
|
Seems not to work. |
|
Could you please verify the output of Also, could you please verify the lines 31-33 of if(EXISTS ${GEOS_TPL_DIR}/conduit)
set(CONDUIT_DIR ${GEOS_TPL_DIR}/conduit CACHE PATH "" FORCE)
endif() |
|
hi @berenger-eu is the file |
|
Good point, @timokoch, we will highlight that in the documentation. Thanks! @berenger-eu, could you please add the include above to your In addition:
|
|
@victorapm I think we need a single dedicated template hostconfig rather than having folks try to template off one of ours that are typically using some sort of file hierarchy. |
|
Agreed! |
|
@berenger-eu Can you open a pull request with your host-config included so that we can more accurately help? |
|
So I am progressing. And: |
|
@berenger-eu Could you try using |
|
@victorapm thanks, it is currently compiling. By the way, in the quickstart, there is this line commands) |
|
The compilation did not finish: |
|
@berenger-eu, thank you for the recommendation. We will implement the changes in the documentation. Regarding your compilation error, it seems you are using an old version of GEOS. This is how that call should look like: GEOS/src/coreComponents/linearAlgebra/interfaces/hypre/HypreMatrix.cpp Lines 1052 to 1059 in f3eb4dc Could you please do the following: |
|
Unfortunately, it does not work. It fails to buid I feel like it is an internal compiler error.... Maybe I should try another nvcc. What version do you use? |
|
@berenger-eu This is an internal compiler error. We have seen this in our RockyLinux CI builds which are currently not enabled. I think we can put this up on the priority list. @CusiniM can you confirm that this is the same error we saw? Alternatively, if you can try cuda 11.8, this error has not come up in our testing. |
yes, it's the same exact error we see on the rockylinux builds with |
|
Ok, I will recompile everything with NVCC 11.7, I hope the host compiler has nothing to do with that. |
|
Ok, I compiled it, and it worked. I was able to run some tutorials. So I will close this issue. As a last remark, it could be great if you were sharing an archive (like a .tar.gz) that would contain the snapshot of the two git repo and allow getting everything in one file, like a release (especially, users might not be interested in having all the history of the repo, etc.). I will end my review on JOSS. Thanks for the support. Good work. |
Thanks for your patience and persistence. We are discussing a way to link the two repos. In reality we are supposed to be using Spack with a single repository, but that hasn't gone so smoothly, so we are sort of in limbo with the TPL solution that nobody was ever supposed to see! We will come up with a strategy for ensuring that the main GEOS repo and the TPL's are in sync when a build is attempted. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hello there,
I would like to compile GEOS on an HPC platform.
I was able to checkout the two repo, and then I loaded the following modules
module load compiler/cuda/12.3 compiler/gcc/11.2.0 mpi/openmpi/5.0.1 build/cmake/3.27.0 linalg/mkl/2020_update4.I added my hostfile,
nano ../GEOS/host-configs/plafrim/a100.cmakeAnd then when I execute the python command
python scripts/config-build.py -hc ../GEOS/host-configs/plafrim/a100.cmake -bt Release, I get:Any help will be welcome.
The text was updated successfully, but these errors were encountered: