-
-
Save bmmalone/1b5f9ff72754c7d4b313c0b044c42684 to your computer and use it in GitHub Desktop.
### Lines to add to .bashrc | |
# for the OpenBLAS library | |
export LD_LIBRARY_PATH=/path/to/OpenBLAS:$LD_LIBRARY_PATH | |
export BLAS=/path/to/libopenblas.a | |
export ATLAS=/path/to/libopenblas.a |
# download the OpenBLAS source | |
git clone https://github.com/xianyi/OpenBLAS | |
# compile the library | |
cd OpenBLAS && make FC=gfortran |
what is the "/path/to/" means and how should I define it. How "/path/to/" related to OpenBLAS directory?
@rahmahi: /path/to
just means the absolute path the respective libraries and folders. So, for example, something like: export BLAS=/home/bmmalone/local/lib/libopenblas.a
Hello,
I am using ubuntu 18.0.4, if I installed Openblas using sudo apt-get install libopenblas-dev
command. Does that mean it is not necessary to install again Openblas using git clone https://github.com/xianyi/OpenBLAS
? If not, what's the difference between those two packages.
Hey @nagarjunvinukonda: The versions are probably about the same; sometimes, the versions in the apt-get repositories are a little behind the version in the main repository.
The main difference is that using sudo
requires, well, sudo/root access, while installing via the script can be done without root access. Thus, if you are in a computing environment where you don't have root access, sudo isn't possible.
It's also a little easier if you need to install multiple versions of openblas for some reason (since the location of the versions is controlled explicitly, while the location of files from apt-get can be difficult to track).
The OpenBLAS library is used for efficiency. This library is installed by default on many versions of Unix. It is also available through many standard package managers. For example:
sudo apt-get install libopenblas-dev
sudo yum install openblas-devel
If the use of package managers are not an option (for example, because they require root access) or just not desired, then OpenBLAS can be installed locally. The following instructions have been tested extensively on Ubuntu, but they may require modification for other distributions.
For OSX, a custom installation of numpy linking against the OpenBLAS library is required. We do not officially support this, but example documentation is available elsewhere on the web.
N.B. In principle, any BLAS and ATLAS library, such as Intel's Math Kernel Library, can be used. This has not been tested and is not supported, though.