CUDA

Repository with sample codes using the CUDA/C++ library.

Pre-requisite

1. Install and configure NVIDIA drivers
2. Install and configure the CUDA library

Installing NVIDIA driver on Fedora 27

To install the NVIDIA proprietary drivers on Fedora 27 you must first update your system:

$ sudo dnf update

Then install the lastest NVIDIA drivers from the DNF repository.

$ sudo dnf install xorg-x11-drv-nvidia akmod-nvidia
$ sudo dnf install xorg-x11-drv-nvidia-cuda

Installing CUDA libraries on Fedora 27

Download the last version of the CUDA Toolkit runfile. For me it was the Cuda Version v9.1 Linux-x86_64-Fedora-25-v9.1.run

CUDA Toolkits = https://developer.nvidia.com/cuda-downloads

The latest NVIDIA compiler (NVCC) only works with a version of GNU/GCC compiler that is previous than version 7. So you need to install GCC 6.3.0 for instance, which RPM package could be obtain here:

GNU/GCC Compiler v6.3.0 = https://drive.google.com/file/d/1t4WrgvpEP-6_NN3qMJhz9MS3CJhHrHKc/view

GNU/GCC Compiler v5.3.0 = https://drive.google.com/file/d/0B7S255p3kFXNbTBneHgwSzBodFE/view

GNU/GCC Compiler v4.9.0 = https://drive.google.com/file/d/1Pwq1ua80dGM72i7rpDNAIIdfcR1WK-hG/view

After downloading it make a symbolic link between this version of GCC and the version NVCC uses by using these commands:

$ sudo ln -s /usr/bin/gcc-6 /usr/local/cuda/bin/gcc 
$ sudo ln -s /usr/bin/g++-6 /usr/local/cuda/bin/g++

The GNU/GCC version for each CUDA release can be checked below:

• CUDA 4.1 release
  • gcc 4.5 is now supported. gcc 4.6 and 4.7 are unsupported.
• CUDA 5.0 release
  • gcc 4.6 is now supported. gcc 4.7 is unsupported.
• CUDA 6.0 release
  • gcc 4.7 is now supported.
• CUDA 7.0 release
  • gcc 4.8 is fully supported, with 4.9 support on Ubuntu 14.04 and Fedora 21.
• CUDA 7.5 release
  • gcc 4.8 is fully supported, with 4.9 support on Ubuntu 14.04 and Fedora 21.
• CUDA 8 release
  • gcc 5.3 is fully supported on Ubuntu 16.06 and Fedora 23.
• CUDA 9 release
  • gcc 6 is fully supported on Ubuntu 16.04, Ubuntu 17.04 and Fedora 25.

The next step is to solve some bugs with the floatn.h header file. There is an problem when we define large float types. To prevent this, we need to add the following lines to the header file (After line 37):

		+ #if CUDART_VERSION
		+ #undef __HAVE_FLOAT128 0
		+ #define __HAVE_FLOAT128 0
		+ #endif
		

After this we proceed normally installing the remaning CUDA-drivers.

$ sudo dnf install nvidia-driver akmod-nvidia kernel-devel nvidia-driver-libs.i686 vulkan.i686 cuda nvidia-driver-cuda cuda-devel nvidia-driver-NVML-devel

Finally execute the runfile that we downloaded, specifing a folder to store some temporary files during the setup.

$ sudo sh cuda_9.1.85_387.26_linux.run --tmpdir=/home/username/tmp/ --override

A minor issue that you might encounter is related to errors on the host_config.h file, which should be located on /usr/local/cuda-9.1/include/crt. The errors might be related to the version of GCC, so I just comment with a simple // the error line.

#if __GNUC__ > 6

//#error -- unsupported GNU version! gcc versions later than 6 are not supported!

#endif /* __GNUC__ > 6 */