cdeterman
gpuR installation problem Ubuntu 16.10
I was trying to install gpuR in my ubuntu 16.10. I did the following to install the package
devtools::install_github("cdeterman/RViennaCL") devtools::install_github("cdeterman/gpuR", ref = "develop")
But I am getting a error
/home/boby/R/x86_64-pc-linux-gnu-library/3.3/RViennaCL/include/viennacl/traits/size.hpp:164:44: error: ‘const class Eigen::Map<Eigen::Matrix<int, -1, -1>, 0,
/usr/lib/R/etc/Makeconf:141: recipe for target 'chol.o' failed make: *** [chol.o] Error 1 ERROR: compilation failed for package ‘gpuR’
It would be great if you can help me in this regard.
Boby Mathew
Thanks for reporting the problem. I will need to set up a 16.10 virtual machine to see if I can reproduce. I don't think anyone has tried this on 16.10 yet.
@bobymathew what OpenCL platform are you using as well? I would like to be as reproducible to your problem as possible.
@bobymathew How did you install the CUDA toolkit? Did you install the 16.04 version? Did you use the network or local files? It isn't going particularly smoothly for me at the moment so if you can explain how you did it we can try to solve the problem faster.
I am using 16.10. I installed them using the synaptic package manager with ubuntu. Many thanks.
@bobymathew are you able to call clinfo? I am having trouble getting the OpenCL to be found after installing:
sudo apt-get install nvidia-cuda-toolkit
I have a warning like this when I try to install it
/home/boby/R/x86_64-pc-linux-gnu-library/3.3/RViennaCL/include/viennacl/matrix.hpp:1186:38: note: mismatched types ‘Eigen::Stride<0, 0>’ and ‘Eigen::OuterStride<>’ assert( viennacl::traits::size2(cpu_matrix) == gpu_matrix.size2() && bool("Matrix dimensions mismatch: columns"));
May be this will help!
Boby
@bobymathew I don't know why you are receiving that error so I need to find a way to reproduce the problem first. The program like clinfo should work and at the very least return 1 platform. I suspect the NVIDIA toolkit install isn't working properly for 16.10. I am trying to get this reproduced on a virtual machine but it doesn't have a NVIDIA GPU installed so it is difficult. Until clinfo works properly I doubt gpuR will install regardless.
Another final test to make sure OpenCL is available and installed correctly is to use another one of my repositories:
git clone https://github.com/cdeterman/intel_opencl.git
cd intel_opencl.git
g++ clDeviceQuery.cpp -o clDeviceQuery -lOpenCL
./clDeviceQuery
Now I get this error when I run ./clDeviceQuery
clDeviceQuery Starting...
Error -1001 in clGetPlatformIDs Call!
Boby
@bobymathew exactly, that's my point. The NVIDIA OpenCL platform is not being found on your OS. I'm still trying to figure out why this is happening on 16.10.
@bobymathew okay, well I can at least get gpuR to compile although the package cannot load because the OpenCL driver is still not happy. Basically I set the environmental variable CUDA_HOME and ran the install:
Sys.setenv(CUDA_HOME = "/usr/local/cuda-8.0/targets/x86_64-linux/lib")
devtools::install_gitub('cdeterman/gpuR', ref = 'develop')
Many thanks.. May be I will wait until someone fix this issue of OpenCL with ubuntu 16.10. Boby
@bobymathew I actually opened a question with NVIDIA and it appears that the CUDA toolkit is just not supported yet on 16.10. Will need to wait and see when this support comes.
@bobymathew Checking in at the following: https://launchpad.net/ubuntu/yakkety/amd64/nvidia-cuda-toolkit https://devtalk.nvidia.com/default/topic/975944/cuda-setup-and-installation/ubuntu-16-10-cuda-toolkit-install-with-opencl/
It is likely that you have a broken driver installation. Speaking from past systems, the closed source AMD and Nvidia drivers can be very fragile.
