xref: /aosp_15_r20/external/pthreadpool/README.md (revision b095b0533730c2930f947df924a4486d266faa1a)
1*b095b053SXin Li# pthreadpool
2*b095b053SXin Li
3*b095b053SXin Li[![BSD (2 clause) License](https://img.shields.io/badge/License-BSD%202--Clause%20%22Simplified%22%20License-blue.svg)](https://github.com/Maratyszcza/pthreadpool/blob/master/LICENSE)
4*b095b053SXin Li[![Build Status](https://img.shields.io/travis/Maratyszcza/pthreadpool.svg)](https://travis-ci.org/Maratyszcza/pthreadpool)
5*b095b053SXin Li
6*b095b053SXin Li**pthreadpool** is a portable and efficient thread pool implementation.
7*b095b053SXin LiIt provides similar functionality to `#pragma omp parallel for`, but with additional features.
8*b095b053SXin Li
9*b095b053SXin Li## Features:
10*b095b053SXin Li
11*b095b053SXin Li* C interface (C++-compatible).
12*b095b053SXin Li* 1D-6D loops with step parameters.
13*b095b053SXin Li* Run on user-specified or auto-detected number of threads.
14*b095b053SXin Li* Work-stealing scheduling for efficient work balancing.
15*b095b053SXin Li* Wait-free synchronization of work items.
16*b095b053SXin Li* Compatible with Linux (including Android), macOS, iOS, Windows, Emscripten environments.
17*b095b053SXin Li* 100% unit tests coverage.
18*b095b053SXin Li* Throughput and latency microbenchmarks.
19*b095b053SXin Li
20*b095b053SXin Li## Example
21*b095b053SXin Li
22*b095b053SXin Li  The following example demonstates using the thread pool for parallel addition of two arrays:
23*b095b053SXin Li
24*b095b053SXin Li```c
25*b095b053SXin Listatic void add_arrays(struct array_addition_context* context, size_t i) {
26*b095b053SXin Li  context->sum[i] = context->augend[i] + context->addend[i];
27*b095b053SXin Li}
28*b095b053SXin Li
29*b095b053SXin Li#define ARRAY_SIZE 4
30*b095b053SXin Li
31*b095b053SXin Liint main() {
32*b095b053SXin Li  double augend[ARRAY_SIZE] = { 1.0, 2.0, 4.0, -5.0 };
33*b095b053SXin Li  double addend[ARRAY_SIZE] = { 0.25, -1.75, 0.0, 0.5 };
34*b095b053SXin Li  double sum[ARRAY_SIZE];
35*b095b053SXin Li
36*b095b053SXin Li  pthreadpool_t threadpool = pthreadpool_create(0);
37*b095b053SXin Li  assert(threadpool != NULL);
38*b095b053SXin Li
39*b095b053SXin Li  const size_t threads_count = pthreadpool_get_threads_count(threadpool);
40*b095b053SXin Li  printf("Created thread pool with %zu threads\n", threads_count);
41*b095b053SXin Li
42*b095b053SXin Li  struct array_addition_context context = { augend, addend, sum };
43*b095b053SXin Li  pthreadpool_parallelize_1d(threadpool,
44*b095b053SXin Li    (pthreadpool_task_1d_t) add_arrays,
45*b095b053SXin Li    (void*) &context,
46*b095b053SXin Li    ARRAY_SIZE,
47*b095b053SXin Li    PTHREADPOOL_FLAG_DISABLE_DENORMALS /* flags */);
48*b095b053SXin Li
49*b095b053SXin Li  pthreadpool_destroy(threadpool);
50*b095b053SXin Li  threadpool = NULL;
51*b095b053SXin Li
52*b095b053SXin Li  printf("%8s\t%.2lf\t%.2lf\t%.2lf\t%.2lf\n", "Augend",
53*b095b053SXin Li    augend[0], augend[1], augend[2], augend[3]);
54*b095b053SXin Li  printf("%8s\t%.2lf\t%.2lf\t%.2lf\t%.2lf\n", "Addend",
55*b095b053SXin Li    addend[0], addend[1], addend[2], addend[3]);
56*b095b053SXin Li  printf("%8s\t%.2lf\t%.2lf\t%.2lf\t%.2lf\n", "Sum",
57*b095b053SXin Li    sum[0], sum[1], sum[2], sum[3]);
58*b095b053SXin Li
59*b095b053SXin Li  return 0;
60*b095b053SXin Li}
61*b095b053SXin Li```
62