vspline_threadpool Namespace Reference

Classes
class	thread_pool

Variables
const int	ncores = std::thread::hardware_concurrency()
	number of CPU cores in the system More...
const int	default_njobs = 2 * ncores
	when multithreading, use this number of jobs per default. This looks like overkill and unnecessary signalling overhead, but it improves performance over just having as many threads as there are physical cores. Why is this so? There are several possibilities I've considered: More...

Variable Documentation

default_njobs

const int vspline_threadpool::default_njobs = 2 * ncores

when multithreading, use this number of jobs per default. This looks like overkill and unnecessary signalling overhead, but it improves performance over just having as many threads as there are physical cores. Why is this so? There are several possibilities I've considered:

• while one task waits e.g. for memory, another task can perform computations
• the scheduler might assign time slices to each thread, so having more threads yields more time slices

Definition at line 99 of file thread_pool.h.

ncores

const int vspline_threadpool::ncores = std::thread::hardware_concurrency()

number of CPU cores in the system

Definition at line 87 of file thread_pool.h.