README.md

What is test function?

First, the execution time of function is same (no matter what thread pool you use).
So, the function should be short, and what we need to do is test

how fast can a thread pool construct and destruct (test iteration times)
how fast can a thread pool complete N jobs, while N equals to the size of thread pool (test iteration times)
	-test by wait_single (if has)
	-test by wait_all (if has)
how fast can a thread pool complete iteration jobs

The version I use

progschj/ThreadPool commit: 9a42ec1329f259a5f4881a291db1dcb8f2ad9040
nbsdx/ThreadPool commit: 08905cfc36d8bae308488c56c505e9aefa2bcb78
philipphenkel/threadpool commit: bb6da2e40eacb3358e5c648f2242faf307a54268
tghosgor/threadpool11 commit: 8bfcf6afbdb60b2dc7976762f45fd424de06097a
Tyler-Hardin/thread_pool commit: 0b1eff9817d89266232489cd250d24841e147dc3
mtrebi/thread-pool commit: f54808b60544555bd8e1b1b748009d55a657e8c1

Environment

Motherboard: ASUS TUF X299 MARK 1
CPU: Intel(R) Core(TM) i7-7820X
Memory: 16 GiB DDR4 2400 MHz * 2
OS: Linux 4.14.15-1-ARCH #1 SMP PREEMPT
Compiler: gcc version 7.2.1 20180116
boost version: 1.66.0-1

Result

First, I will not compare nbsdx, philipphenkel, tghosgor and mtrebi. Because their works has potential bug.
You can get the following statistic in their directory. (out_100000)
Construct and destruct time:

progschj                 with 547785 ns
Tyler-Hardin             with 620148 ns
Fdhvdu's CThreadPool_Ret with 956302 ns
Fdhvdu's CThreadPool     with 959417 ns

Wait until specific job completes:

Fdhvdu's CThreadPool_Ret with 39995 ns
Fdhvdu's CThreadPool     with 40560 ns
progschj                 with 51185 ns
Tyler-Hardin             with 72200 ns

Launch a large number of jobs and wait until all jobs complete:

Fdhvdu's CThreadPool     with 2550 ns
progschj                 with 3107 ns
Tyler-Hardin             with 4638 ns

Use Fdhvdu's CThreadPool in most cases.
Use progschj only if you want to construct a thread pool multiple times.