This is a recipe (actually a set of scripts) that will help you build Python 2.7.16 natively for the SX-Aurora TSUBASA Vector Engine (VE). It is based upon the work of Laurent Gatineau from NEC, slightly adjusted and made to download required packages from the internet.
The code is built in scalar mode with no vectorization at all. Therefore this is not to be used for any serious numerical simulations inside Python. It is rather good for programs which use Python as a driver which controls code from other highly optimized libraries.
A VH (vector host) machine with a normal installation including VE compilers (ncc, nc++, nfort), MPI for VE version 2.2.0 and the NLC 2.0.0 optimized library collection for the VE.
The machine should have a connection to the internet or the proxy
settings should be configured properly, by setting the environment
variables https_proxy, http_proxy
.
Clone this repository:
git clone https://github.com/sx-aurora/ve-python2
then change working directory to it:
cd ve-python2
Decide for the target directory of the installation. Edit the file
env.sh
and adjust the variable PREFIX
accordingly.
Build by doing