Benchmarks of the function queues

---

In this single-thread tests all the function queues of density are compared to a std::vector and a std::queue of std::functions. In every test a function queue is created, filled by many trivial functions, and then all the functions are consumed on-by-one. Only in the case of std::vector the queue is cleared in one time, rather than removing the functions one by one while they are invoked. This is because there is no constant-time function to remove the first element of a std::vector.

In the first test the function queues of density are instantiated with the default template parameters. In the second test they exploit all the available optimizations, that is:

• function_manual_clear, so that every element is actually a single pointer (plus an internal overhead pointer).
• concurrency_single, when it's applicable
• consistency_relaxed, when it's applicable

Summary of the results

All the tests are compiled with Visual Studio 2017 (version 15.1), and executed on Windows 10 with an int Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz, 2808 Mhz, 4 cores, 8 logical processors. The benchmarks of the whole library are shuffled and interleaved at the granularity of a single invocation with a given cardinality, and every invocation occurs 8 times.

Even unnecessarily paying the cost of multithreading, concurrent function queues are in general better than std::queue of std::function. lf_function_queue is much faster in the case of single consumer and single producers, as it can reduce considerably the number of atomic memory operations.

Results of the first test

Results of the second test