Controlling OpenMP threads

[reguly]

I compiled with clang++ and -fopenmp, yet I am having trouble controlling the number of threads. It seems triSYCL launches twice or even three times as many threads as CPU cores, which I suspect is the main reason for the poor performance of my code. changing OMP_NUM_THREADS does not seem to have any effect. Any suggestions?

[keryell]

If you are not using hierarchical parallelism, by default SYCL assumes that you might have some barriers in the work-group, which means, to support this with OpenMP and a plain C++ compiler, we need 1 thread per work-item... :-(

If you know that you will not be in that case, you can play with

#define TRISYCL_NO_BARRIER

like in https://github.com/triSYCL/triSYCL/blob/master/tests/parallel_for/hierarchical.cpp

See also https://github.com/triSYCL/triSYCL/blob/master/doc/macros.rst

But it might be also a bug...