openmp

OpenMP and Python

Due to GIL there is no point to use threads for CPU intensive tasks in CPython. You need either multiprocessing (example) or use C extensions that release GIL during computations e.g., some of numpy functions, example. You could easily write C extensions that use multiple threads in Cython, example.

Using OpenMP with clang

Update Building the latest trunk of LLVM/Clang (clang-3.8), installing libiomp5, and specifying the location of the gomp omp header files worked. Note that the Ubuntu package for libiomp5 isn’t quite correct, so you will need to add a symlink in /usr/lib from /usr/lib/libiomp5.so to /usr/lib/libiomp5.so.5. I’m using g++-5.1 and clang++-3.6 on Linux Mint 17.2 (essentially … Read more

OpenMP set_num_threads() is not working

Besides calling omp_get_num_threads() outside of the parallel region in your case, calling omp_set_num_threads() still doesn’t guarantee that the OpenMP runtime will use exactly the specified number of threads. omp_set_num_threads() is used to override the value of the environment variable OMP_NUM_THREADS and they both control the upper limit of the size of the thread team that … Read more

tech