- CUDA and C:
- C_only:
- Added C-only branch. use install_github(“IshmaelBelghazi/gpuClassifieR”, ref=”C_only”)
- Benchmark against glmnet. In misc/test.md
- A CUDA enabled gpu with compute capabilities 2.0 or above
- CUDA toolkit 5.5 or above (although the package was only tested with 6.5)
- Be sure to have a CUDA_HOME environment variable defined. Typically its value would be CUDA_HOME=/usr/local/cuda-M-m/. Where M and m refer to the Major and minor version number respectively.
- Add CUDA_HOME/lib64 to your LD_LIBRARY_PATH. i.e: Typically export LD_LIBRARY_PATH=/usr/local/cuda-M-m/lib64 (or lib if on 32bit system)
- Use devtools::install_github(“IshmaelBelghazi/gpuClassifieR”)
- Build: the package builds successfully. Tested on Ubuntu 14.10 x64 and travis (Ubuntu 12.04 LTS x64)
- Installation: The package installs successfully using devtools::install_github() or R CMD BUILD followed by R CMD INSTALL.
- Check: The package fails R CMD check. This seems to a common problem with packages containing CUDA code and using nvcc. Even packages currently on CRAN (such as gputools and wideLM) do no pass the check. Still, the package does pass R CMD INSTALL –fake. If you know how to deal with this, please let me know.
This package was built on Ubuntu 14.10 x64 with CUDA toolkit 6.5.