The AMD Core Math Library (ACML) is a set of numerical routines tuned specifically for AMD64 platform processors. It is now end of life.
Managed Linux workstations up to Ubuntu 16.04 (Xenial)
There are different ACMLs for different compilers. Pick the one that matches the compiler you want to use (module display acml shows the options) and then module load that one.
The 'up' versions are for running on a single cpu. You link those with -lacml. The 'mp' versions use OpenMP parallelization. You need to add an extra flag to tell your compiler to use this, usually -mp. You link those with -lacml_mp. With some compilers you need other flags too. If you do 'module help' on your ACML module you should get a list of possible link lines.
This is proprietary software. See /usr/local/shared/acml/version/ACML-EULA.txt for details.
There are info pages, and HTML and PDF documentation in /usr/local/shared/acml. The file /usr/local/shared/acml/version/Doc/acml.txt is usually a good place to get started.
The installer just untars the library and fixes up a few paths in the examples, so most versions of this can safely go in the generic directory on the NFS server. The modules use a 'magic' modulefile to work out what to do based on the name they're called under. There are versions installed with a 64-bit integer type that I haven't provided modules for. There are also version with support for FMA4 instructions. We don't currently own any cpus with these instruction sets so again no modules for them.
To test an ACML installation you load the acml module and the matching compiler and do:
compiler /usr/local/shared/acml/4.1/gfortran64/examples/acmlinfo.f \ -o acmlinfo -lacml
and then run the acmlinfo binary. You need to link with -lacml_mp for the mp versions of the library.
