| JF Bastien | e637637 | 2018-12-19 17:45:32 | [diff] [blame] | 1 | # Parallel STL |
| 2 | |
| 3 | Parallel STL is an implementation of the C++ standard library algorithms with support for execution policies, |
| 4 | as specified in ISO/IEC 14882:2017 standard, commonly called C++17. The implementation also supports the unsequenced |
| 5 | execution policy specified in Parallelism TS version 2 and proposed for the next version of the C++ standard in the |
| Louis Dionne | 9bb9690 | 2019-01-07 16:54:02 | [diff] [blame] | 6 | C++ working group paper [P1001](https://wg21.link/p1001). |
| JF Bastien | e637637 | 2018-12-19 17:45:32 | [diff] [blame] | 7 | Parallel STL offers efficient support for both parallel and vectorized execution of algorithms. For sequential |
| 8 | execution, it relies on an available implementation of the C++ standard library. |
| 9 | |
| 10 | ## Prerequisites |
| Louis Dionne | 9bb9690 | 2019-01-07 16:54:02 | [diff] [blame] | 11 | |
| JF Bastien | e637637 | 2018-12-19 17:45:32 | [diff] [blame] | 12 | To use Parallel STL, you must have the following software installed: |
| 13 | * C++ compiler with: |
| 14 | * Support for C++11 |
| 15 | * Support for OpenMP* 4.0 SIMD constructs |
| Louis Dionne | 9bb9690 | 2019-01-07 16:54:02 | [diff] [blame] | 16 | * Threading Building Blocks (TBB) which is available for download at https://github.com/01org/tbb/ |
| JF Bastien | e637637 | 2018-12-19 17:45:32 | [diff] [blame] | 17 | |
| Louis Dionne | 9bb9690 | 2019-01-07 16:54:02 | [diff] [blame] | 18 | ## Known issues and limitations |
| 19 | |
| 20 | * `unseq` and `par_unseq` policies only have effect with compilers that support `#pragma omp simd` or `#pragma simd`. |
| 21 | * Parallel and vector execution is only supported for the algorithms if random access iterators are provided, |
| 22 | while for other iterator types the execution will remain serial. |
| 23 | * The following algorithms do not allow efficient SIMD execution: `includes`, `inplace_merge`, `merge`, `nth_element`, |
| 24 | `partial_sort`, `partial_sort_copy`, `set_difference`, `set_intersection`, `set_symmetric_difference`, `set_union`, |
| 25 | `sort`, `stable_partition`, `stable_sort`, `unique`. |
| 26 | * The initial value type for `exclusive_scan`, `inclusive_scan`, `transform_exclusive_scan`, `transform_inclusive_scan` |
| 27 | shall be DefaultConstructible. A default constructed-instance of the initial value type shall be the identity element |
| 28 | for the specified binary operation. |
| 29 | * For `max_element`, `min_element`, `minmax_element`, `partial_sort`, `partial_sort_copy`, `sort`, `stable_sort` |
| 30 | the dereferenced value type of the provided iterators shall be DefaultConstructible. |
| 31 | * For `remove`, `remove_if`, `unique` the dereferenced value type of the provided iterators shall be MoveConstructible. |
| 32 | * The following algorithms require additional O(n) memory space for parallel execution: `copy_if`, `inplace_merge`, |
| 33 | `partial_sort`, `partial_sort_copy`, `partition_copy`, `remove`, `remove_if`, `rotate`, `sort`, `stable_sort`, |
| 34 | `unique`, `unique_copy`. |
| Mikhail Dvorskiy | 209adba | 2019-06-03 08:23:30 | [diff] [blame] | 35 | |
