Closed dalexandrov closed 5 years ago
Regarding the missing JNvgraphJNI
directory: In addition to the other git clone
calls, you will need
git clone https://github.com/jcuda/jnvgraph.git
This was missing in the build instructions in the README, and I have just updated the README accordingly.
Regarding the missing NVRTC library:
This is similar to https://github.com/jcuda/jcuda/issues/3 , although this one referred to Linux. Some questions:
.dylib
file) involving "nvrtc" as part of its name in your CUDA install directory?As a quick workaround for you: If you find the NVRTC CUDA library, you can set the CMake CUDA_nvrtc_LIBRARY
variable in the cmake-gui to point to this library. Then it should be possible to build everything.
Of course, the library should be found automatically, but I may have to update the CMake find calls accordingly, as soon as I know the directory on MacOS where the NVRTC library is contained.
Hello, thank you for an update.
I have installed the latest CUDA toolkit 8.0. Everything is under path /Developer/NVIDIA/CUDA-8.0 Unfortunately I couldn't find any *nvrtc*.dylib on my mac
So I'm currently like this:
Thank you!
I have never used a Mac, so I'm afraid I can't really give a focussed advice. Quick websearches indicate that there have been some glitches in the CUDA installation at least for version 7.5: http://stackoverflow.com/questions/36105066/cuda-7-5-install-on-mac-missing-nvrtc - maybe this is still the same for version 8.0? If so, it seems like you have to install XCode first.
As a last resort, you could compile all libraries except for that of JNvrtc, but of course, a working solution here would be preferable...
I've managed to build it. Thank you!
Thanks for the info. I assume that it was solved by installing XCode (i.e. the installation instructions for Mac). If not, people might be interested in knowing how you eventually solved it.
XCode was already there. It was just use of the writing the right command
cmake ../jcuda-main -DCUDA_SDK_ROOT_DIR=/Developer/NVIDIA/CUDA-8.0 -DCUDA_nvrtc_LIBRARY=/Developer/NVIDIA/CUDA-8.0/lib/libnvrtc.dylib
Actually, it should find these directory- and library names automatically. So it should not be necessary to give them at the command line. I'll review the CMakeLists.txt, to see whether this can be fixed.
btw... may be you need these compiled for mac jars and native dylibs? For distribution?
They should be available (for 0.8.0) according to https://github.com/jcuda/jcuda-main/issues/14 I'll try to create a Maven release ASAP, but have some other tasks in the next few days.
ah, ok)
The latest release (10.1) contains binaries for MacOS as well. If there are problems during the build for future versions, they could be discussed in a new issue.
Any chance to build it on MacOS Sierra.
I've got this one:
Found CUDA: /usr/local/cuda/bin/nvcc CMake Error at CMakeLists.txt:45 (add_subdirectory): add_subdirectory given source "/Users/mitia/Documents/Java/JCuda/jcuda-main/../jnvgraph/JNvgraphJNI" which is not an existing directory.
CMake Error: The following variables are used in this project, but they are set to NOTFOUND. Please set them or make sure they are set and tested correctly in the CMake files: CUDA_nvrtc_LIBRARY linked by target "JNvrtc" in directory /Users/mitia/Documents/Java/JCuda/jcuda/JNvrtcJNI
Configuring incomplete, errors occurred! See also "/Users/mitia/Documents/Java/JCuda/bin/CMakeFiles/CMakeOutput.log".