internal::VectorImplementation Namespace Reference

Variables
unsigned int	minimum_parallel_grain_size = 4096

Variable Documentation

minimum_parallel_grain_size

unsigned int internal::VectorImplementation::minimum_parallel_grain_size = 4096

If we do computations on vectors in parallel (say, we add two vectors to get a third, and we do the loop over all elements in parallel), then this variable determines the minimum number of elements for which it is profitable to split a range of elements any further to distribute to different threads.

This variable is available as a global writable variable in order to allow the testsuite to also test the parallel case. By default, it is set to several thousand elements, which is a case that the testsuite would not normally encounter. As a consequence, in the testsuite we set it to one – a value that's hugely unprofitable but definitely tests parallel operations.

Definition at line 33 of file parallel.cc.