-
Notifications
You must be signed in to change notification settings - Fork 11
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Cannot compile via spack #8
Comments
Hi @svenevs, could you try building now with the latest vfd-gds code and using either HDF5 1.13.2 or the develop branch? I just pushed a commit that should update the VFD to work with those branches. If everything goes smoothly I can create a new release. |
@jhendersonHDF I was able to test this at two different sites, with hdf5 1.13.2 and Thanks! |
The VFD comes with a small test program that just does a basic read/write with the VFD, but I don't think it's integrated with spack yet, so you might need to build and run it by hand after setting
If you have issues with that, I think it's still reasonable to just tag a new version anyway. |
@jhendersonHDF AFAICT it worked and produced the Adding the Thanks for your help! cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
project(hdf5-test LANGUAGES C)
# NOTE: cmake configs are not valid
# 1. produced cmake targets forcibly include cuda toolkit directories
# (stop using find_package(CUDA) and include directly, use
# find_package(CUDAToolkit) and just link against CUDA::cudart.
# 2. produced cmake target depends specifically on hdf5-shared, not
# clear if an issue but the consumer may desire to link against
# hdf5-static?
find_package(hdf5_vfd_gds CONFIG REQUIRED)
find_package(CUDAToolkit REQUIRED)
# NOTE:
# 3. Seems like a bug that hdf5 isn't bringing MPI... (bug with hdf5 cmake)
find_package(MPI REQUIRED)
add_executable(h5gds_simple_dset_write simple_dset_write.c)
target_link_libraries(h5gds_simple_dset_write
PUBLIC
hdf5_vfd_gds
CUDA::cudart
MPI::MPI_C)
# NOTE:
# 4. hdf5-vfd-gds cmake target does not include its own headers install directory
# usually something like
# target_include_directories(tgt
# PUBLIC
# $<BUILD_INTERFACE:...${CMAKE_CURRENT_SOURCE_DIR}/include or something...>
# $<INSTALL_INTERFACE:include>) # partially depends on install() logic
# e.g., CMAKE_INSTALL_INCLUDEDIR could be used rather than hard-coded include
target_include_directories(h5gds_simple_dset_write
PUBLIC
/global/u1/s/svenevs/spack_hdf5_cuda/opt/spack/cray-sles15-zen3/gcc-11.2.0/hdf5-vfd-gds-master-bp627tb3l64qewewpkshmswo3n4e3zf7/include) |
Thanks for testing @svenevs! I'm currently awaiting a resolution for #5 and then we should be good to go for a new release. Would you mind creating a separate issue for the CMake issues you noted so we don't lose track of it and can close this current issue? I think I remember the issue with HDF5 not bring in MPI as a dependency, but I'll have to check with others on that one. |
@jhendersonHDF sure thing! Redirects:
OK, a quick look personally I do not endorse the changes there (the If no resolution is to be had by the end of the week I would encourage tagging a new release before finishing it. That will guarantee we have time to integrate into the ECP SDK before the E4S freeze -- if not this package simply gets yanked (next slot not until February). If resolution is to be had please @ me and I will re-test compilation from within the SDK! |
Awesome! It's alive here: spack/spack#33300 Very exciting to get this one in there, GDS is such cool stuff 😎 |
Hello, we are working on the hdf5 cuda integrations for the ECP datavis SDK, I have not been able to find a way to get
hdf5-vfd-gds
to compile. I've tried a handful of combinations between hdf5 1.13 versions and development branch as well as GCC compilers and cuda versions but haven't found anything successful.I'm attaching the full build output and will highlight the errors, these errors reproduce whether I'm trying to install the SDK or just install
hdf5-vfd-gds
(unrelated to the SDK, that is).spack-build-out.txt
error 1
error 2:
error 3:
error 4:
I took a cursory look but it is not obvious to me what the underlying problem here is. Any thoughts?
The text was updated successfully, but these errors were encountered: