Lots of Tasks (not Threads)
The advice in this chapter is this: you need lots of task-level parallelism and you should first consider using OpenMP, Fortran 2008 DO CONCURRENT, Intel® Threading Building Blocks (TBB), and Intel® Cilk™ Plus. Alternatives such as direct use of pthreads or use of OpenCL can deliver excellent performance results, but the limitations in terms of being portable and difficulties in maintaining can be substantial. All together, the numerous methods to achieve scaling on Intel® Xeon Phi™ coprocessors are plentiful enough to help you have enough task-level parallelism, and the resulting code will work on processors and coprocessors alike.
It is important to understand that these tasking/threading models dispatch ...
Get Intel Xeon Phi Coprocessor High Performance Programming now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
