Close #111
The Makefile now reads the .env as a default properly. The previous behavior where environment variables in the host shell were read by the Makefile, which was fortunately never documented in the README.md, was removed as it would cause too much confusion and the .env file could now perform the role of specifying the SERVICE and other Make variables.
This PR also updates the documentation on how to use the new features that have recently been implemented. It is still not enough but more will be coming.
Also, the simple service now uses Python and Numpy compiled with Intel(R) compilers for better compatibility with MKL. According to the Intel blog, Numpy built with GCC or other C compilers may be slower than Numpy built with Intel compilers. https://www.intel.com/content/www/us/en/developer/articles/technical/easy-intro-to-the-intel-distribution-for-python.html
The major downside is that versions of the Intel compiled libraries are always older than the latest version on the conda-forge channel. However, using the Intel version is just a default, and users can change this setting if they so desire.
Close #111 The Makefile now reads the
.envas a default properly. The previous behavior where environment variables in the host shell were read by the Makefile, which was fortunately never documented in theREADME.md, was removed as it would cause too much confusion and the.envfile could now perform the role of specifying theSERVICEand other Make variables. This PR also updates the documentation on how to use the new features that have recently been implemented. It is still not enough but more will be coming. Also, thesimpleservice now uses Python and Numpy compiled with Intel(R) compilers for better compatibility with MKL. According to the Intel blog, Numpy built with GCC or other C compilers may be slower than Numpy built with Intel compilers. https://www.intel.com/content/www/us/en/developer/articles/technical/easy-intro-to-the-intel-distribution-for-python.html The major downside is that versions of the Intel compiled libraries are always older than the latest version on theconda-forgechannel. However, using the Intel version is just a default, and users can change this setting if they so desire.