OpenACC Resource Site

I hope eventually to develop this site as a resource for OpenACC users. At this point, however, it will primarily serve as a site for OpenACC learners.

It is assumed that the reader has background in OpenMP and GPUs. If not, he/she may acquire this background by reading my open source book on parallel computation, or my published book, Parallel Computation for Data Science (Chapman and Hall, 2015).

Here are some resources for learning OpenACC. Suggestions for others are highly welcome!

• There is first the issue of what compiler to use. PGI is expensive and difficult to install, and GCC is not quite OpenACC-ready. I suggest using the Omni compiler, which installs "right out of the box" if you have the components, e.g. MPI.

  On the system I am using Omni, I started with

  configure --prefix=/home/matloff/Pub/omni --enable-openacc --enable-shared --with-cuda=/usr/local/cuda
  

  and then ran the usual make; make install.

  Also, I set LD_LIBRARY_PATH to /home/matloff/Pub/omni/lib

The command line is of the form

% ompcc -acc yourfile.c

• For a 2-minute introduction, see the code for calculating π in this site. Its similarity to OpenMP should make it easy to follow. The line

  #pragma acc parallel vector_length(vl) 
  

  is similar to the parallel pragma in OpenMP, while

  #pragma acc loop reduction(+:pi)
  

  is like reduction on OpenMP.

• I actually had the students in my undergrad class write an introduction to OpenACC as their term project. Some of their writeups were especially clear and insightful, so I am posting them here with their permission.