Satellite Application Facility for Numerical Weather Prediction › Forums › RTTOV › RTTOV v13 › RTTOV v13 General Discussion › question about RTTOV v13 Test
Tagged: apple silicon, arm64, hdf5, test_rttov13
- This topic has 11 replies, 4 voices, and was last updated 11 months, 1 week ago by James Hocking.
-
AuthorPosts
-
November 30, 2022 at 4:05 pm #48306yuhan jiangParticipant
helloļ¼
I’m having trouble with testing, when I finished compiling and then tested, all tests fail like this:
STOP 1
| 2022/11/30 23:57:05 fatal error in module ../../src/hdf/rttov_hdf_mod.F90:0106
|
| 2022/11/30 23:57:05 fatal error in module
| ../../src/coef_io/rttov_read_coefs.F90:0229
|
| 2022/11/30 23:57:05 fatal error in module ../../src/test/rttov_test.F90:1805
|
| 2022/11/30 23:57:05 fatal error in module ../../src/test/rttov_test.F90:0317
How can I solve this poblem?Wating for your reply.
Thanks
jiangNovember 30, 2022 at 6:29 pm #48313James HockingKeymasterDear Jiang,
This is occurring because the code is failing to open the HDF5 library. I haven’t seen this error before so I’m not sure what would cause it.
Which version of the HDF5 library are you using? It must be v1.8.8 or later.
Have you exported the HDF5 library to your LD_LIBRARY_PATH (or the equivalent variable on your system) before running the test?
$ export LD_LIBRARY_PATH=/full/path/to/your/hdf5_library/lib/:$LD_LIBRARY_PATH
Best wishes,
JamesMay 4, 2023 at 6:52 am #48569Andrew PrataParticipantDear James
Today I tried to install and compile RTTOV v13.2 and I ran into exactly the same error.
I should note that I’m trying to run RTTOV on my Mac which has the M2 chip (arm64 architecture).
I was able to get RTTOV to compile successfully in a conda environment (using mamba); however, as Jiang found, all of the tests fail with the HDF5 library issue.
Note that I did try:
$ export LD_LIBRARY_PATH=/Users/apra0062/mambaforge/envs/rttov/libAnd ran the following test:
$ ./test_rttov13.sh ARCH=gfortranAny advice would be much appreciated. Has anyone you know tried to compile and run on the new Apple Silicon machines?
Thanks
Andrew Prata | andrew.prata@monash.eduMay 4, 2023 at 7:46 am #48579James HockingKeymasterDear Andrew,
Could you confirm which version of the HDF5 library you are using?
When you update the LD_LIBRARY_PATH, are you pointing to the HDF5 lib/ directory, rather than your RTTOV lib/ directory:
$ export LD_LIBRARY_PATH=/full/path/to/your/hdf5_library/lib/:$LD_LIBRARY_PATH
I’m not sure if RTTOV has been successfully compiled on the M2 architecture, but I wouldn’t necessarily expect problems.
Best wishes,
JamesMay 5, 2023 at 5:24 am #48584Andrew PrataParticipantHi James
Thank you for the quick reply. I can see how my above comment might be confusing. I have created a conda environment called “rttov”. So all of the library files inside the environment get installed in my $CONDA_PREFIX, which in this case is: /Users/apra0062/mambaforge/envs/rttov and so I believe the lib directory for hdf5 is here: /Users/apra0062/mambaforge/envs/rttov/lib
If I had called the environment ‘my-env’, for example, then my library path would be: /Users/apra0062/mambaforge/envs/my-env/lib
For what it’s worth, my RTTOV lib directory is here: /Users/apra0062/packages/rttov132/lib
The version of my hdf5 library is 1.14.0, which I found by running:
(rttov)$ h5dump –version
h5dump: Version 1.14.0Presumably I can install a different version of hdf5 with conda – just let me know which version should work. I can also try and install with ‘brew’ but I would like to have rttov and its dependancies installed in a self-contained environment.
Thanks
AndrewMay 5, 2023 at 10:50 am #48585James HockingKeymasterHi Andrew,
I haven’t yet tested RTTOV with HDF5 v1.14.0, but it has been tested with HDF5 v1.12.0, v1.10.4, and v1.8.8+, so if you could try v1.12.0 that would be good to confirm or rule out a compatibility issue with the latest HDF5 release.
Thanks,
JamesMay 14, 2023 at 10:11 am #48603Andrew PrataParticipantHi James
Thanks for the suggestion. I ran with a conda environment using hdf5 v1.12.2, netcdf-fortran 4.6.0 and python 3.9.7. After a successful compile, all 11 tests passed. So it looks like it’s an hdf5 version issue (at least when trying to build with a conda environment).
Thanks for the help.
Cheers
AndyMay 15, 2023 at 6:44 am #48604James HockingKeymasterHi Andy,
Thanks for letting me know about this: I will look into the issue with RTTOV and HDF5 v1.14 for the next release.
Cheers,
JamesDecember 15, 2023 at 3:03 pm #49057Marco WurthParticipantHi James,
I am running into the same issue as Jiang and Andy and only hdf5 v1.12.3 is provided on my cluster (intel x86_64) and no other version unfortunately.
I find it suprising that Andy did not experience the issue with v1.12.2 but I do with the difference of only a minor revision of hdf5.Using RTTOV 13.2 I get slightly different error lines compared to what Jiang posted at the top but they are essentially the same:
2023/12/15 14:29:21 fatal error in module rttov_hdf_mod.F90:0106
2023/12/15 14:29:21 fatal error in module rttov_read_coefs.F90:0246
2023/12/15 14:29:21 fatal error in module rttov_test.F90:1814
2023/12/15 14:29:21 fatal error in module rttov_test.F90:0319Did you figure out a workaround for this issue already?
Cheers,
MarcoDecember 15, 2023 at 7:27 pm #49064James HockingKeymasterHi Marco,
This seems to be caused when RTTOV attempts to open the HDF5 library when it is already open. It doesn’t cause a problem with earlier versions of the HDF5 library.
In many user applications this “re-opening” would not occur, and indeed when running the example*.sh scripts (which run the src/test/example_*.F90 programs) there is no problem.
But the problem occurs with the RTTOV test suite (run using rttov_test.pl) due to the way this is configured.
I’ve made updates which resolve the issue with the test suite and should also prevent potential issues more generally in user code. The fix for RTTOV v13.2 is here:
https://nwp-saf.eumetsat.int/downloads/rttov13_files/rttov_read_coefs.F90Download this file to src/coef_io/ and recompile RTTOV. If you need a fix for RTTOV v13.1 or v13.0 let me know.
Best wishes,
JamesDecember 19, 2023 at 12:38 pm #49067Marco WurthParticipantHi James,
thank you so much for the bugfix, now the tests all pass!
All the best,
MarcoDecember 19, 2023 at 12:39 pm #49068James HockingKeymasterHi Marco,
Great, thanks for letting me know.
Best,
James -
AuthorPosts
- You must be logged in to reply to this topic.