Compiling Error

[Daniel LaBuda]

Dear all,

I am a new user of hoomd. I am currently working inside of miniconda in a cluster. Initially I used conda-forge to install v3.9.0 of hoomd (I cannot use the newest version because my scripts aren't up to date yet). I was following steps on how to compile hoomd from someone who had previously done it before in my research group. I was able to make it to nearly compiling and then I get a huge error. I could use hoomd with the GPU enabled from conda-forge but it doesn't detect any GPUs when I run my programs.

I am using python v3.11.5. Cuda v12.2. Matplotlib v3.7.2. Cmake v3.27.4. Pybind11 v2.11.1. Cudatoolkit v11.8.0. Numpy v1.25.2. And finally eigen v3.4.0. Please tell me if any other information helps. Below is the error. 

What I am inputting before I receive the error message is this "cmake --build build/hoomd -j 40". The error I keep receiving while compiling is 

"[  1%] Built target copy_filter
[  2%] Built target copy_update
[  3%] Built target copy_write
[  4%] Built target copy_custom
[  5%] Built target copy_md-long_range
[  6%] Built target copy_hoomd
[  6%] Built target copy_md-methods
[  7%] Built target copy_tune
[  7%] Built target quickhull
[  8%] Built target copy_data
[  8%] Built target copy_minimize
[ 12%] Built target copy_md
[ 13%] Built target copy_md-pair
[ 13%] Built target copy_hoomd_pytest
[ 13%] Built target copy_md-external
[ 17%] Built target copy_hpmc-nec
[ 19%] Built target copy_hpmc-pair
[ 19%] Built target copy_mpcd
[ 19%] Built target copy_md-tune
[ 15%] Built target copy_hpmc_tune
[ 15%] Built target copy_updater_plugin_pytest
[ 23%] Built target copy_md_pytest
[ 23%] Built target copy_md-data
[ 24%] Built target copy_metal
[ 16%] Built target copy_hpmc-external
[ 24%] Built target copy_md-alchemy
[ 25%] Built target copy_hoomd_pytest_programs
[ 25%] Built target copy_hpmc
[ 25%] Built target copy_shape_plugin_pytest
[ 25%] Built target copy_hpmc_pytest
[ 24%] Built target copy_pair_plugin_pytest
[ 25%] Built target copy_shape_plugin
[ 25%] Built target copy_pair_plugin
[ 26%] Built target copy_md-mesh
[ 26%] Built target copy_updater_plugin
[ 26%] Building CUDA object hoomd/CMakeFiles/_hoomd.dir/ParticleGroup.cu.o
/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1237): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1242): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1247): error: "__uint128_t ()" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1252): error: "__uint128_t ()" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1272): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1277): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1282): error: "__int128_t ()" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1288): error: "__int128_t ()" contains a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1284): error: "__uint128_t" is a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1285): error: "__int128_t" is a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1290): error: "__uint128_t" is a 128-bit integer, which is not supported in device code

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1291): error: "__int128_t" is a 128-bit integer, which is not supported in device code

12 errors detected in the compilation of "/tmp/tmpxft_0001e022_00000000-8_ParticleGroup.compute_70.cpp1.ii".
gmake[2]: *** [hoomd/CMakeFiles/_hoomd.dir/ParticleGroup.cu.o] Error 1
gmake[1]: *** [hoomd/CMakeFiles/_hoomd.dir/all] Error 2
gmake: *** [all] Error 2" 

Thanks,
Daniel LaBuda

[Joshua Anderson]

Daniel,

1) Uninstall hoodlt and try again. The error is appearing in what appears to be an old version of cub embedded in hoodlt:

/home/djlabuda/miniconda3/envs/hoodlt/include/cub/util_type.cuh(1237): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code

HOOMD-blue expects to use the latest version of CUB packaged with the CUDA toolkit.

2) Print the return value of `hoomd.device.GPU.get_unavailable_device_reasons()` to see why your devices are unavailable.

2a) If your cluster has CUDA 11 capable drivers (not CUDA 12), then you can install the appropriate GPU enabled version following the instructions here: https://hoomd-blue.readthedocs.io/en/v4.1.0/installation.html (see the first note on how to install hoomd built against CUDA 11.2 instead of 12.0)

3) Updating from 3.9.0 to 4.1.0 requires changing a very small number of lines of code (0-5 in typical simulation scripts): https://hoomd-blue.readthedocs.io/en/v4.1.0/migrating.html . This is *far* less work than building HOOMD-blue from source. You should receive deprecation warnings on all the lines that need to be changed when running in 3.9.

------

Joshua A. Anderson, Ph.D.

Research Area Specialist, Chemical Engineering, University of Michigan

--
You received this message because you are subscribed to the Google Groups "hoomd-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hoomd-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hoomd-users/7caab81b-fed4-483b-ad45-145e6cad849fn%40googlegroups.com.