| 1 | // -*- C++ -*- |
| 2 | //===-- parallel_impl.h ---------------------------------------------------===// |
| 3 | // |
| 4 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 5 | // See https://llvm.org/LICENSE.txt for license information. |
| 6 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 7 | // |
| 8 | //===----------------------------------------------------------------------===// |
| 9 | |
| 10 | #ifndef _PSTL_PARALLEL_IMPL_H |
| 11 | #define _PSTL_PARALLEL_IMPL_H |
| 12 | |
| 13 | #include <atomic> |
| 14 | // This header defines the minimum set of parallel routines required to support Parallel STL, |
| 15 | // implemented on top of Intel(R) Threading Building Blocks (Intel(R) TBB) library |
| 16 | |
| 17 | namespace __pstl |
| 18 | { |
| 19 | namespace __internal |
| 20 | { |
| 21 | |
| 22 | //------------------------------------------------------------------------ |
| 23 | // parallel_find |
| 24 | //----------------------------------------------------------------------- |
| 25 | /** Return extremum value returned by brick f[i,j) for subranges [i,j) of [first,last) |
| 26 | Each f[i,j) must return a value in [i,j). */ |
| 27 | template <class _ExecutionPolicy, class _Index, class _Brick, class _Compare> |
| 28 | _Index |
| 29 | __parallel_find(_ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f, _Compare __comp, bool __b_first) |
| 30 | { |
| 31 | typedef typename std::iterator_traits<_Index>::difference_type _DifferenceType; |
| 32 | const _DifferenceType __n = __last - __first; |
| 33 | _DifferenceType __initial_dist = __b_first ? __n : -1; |
| 34 | std::atomic<_DifferenceType> __extremum(__initial_dist); |
| 35 | // TODO: find out what is better here: parallel_for or parallel_reduce |
| 36 | __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last, |
| 37 | [__comp, __f, __first, &__extremum](_Index __i, _Index __j) { |
| 38 | // See "Reducing Contention Through Priority Updates", PPoPP '13, for discussion of |
| 39 | // why using a shared variable scales fairly well in this situation. |
| 40 | if (__comp(__i - __first, __extremum)) |
| 41 | { |
| 42 | _Index __res = __f(__i, __j); |
| 43 | // If not '__last' returned then we found what we want so put this to extremum |
| 44 | if (__res != __j) |
| 45 | { |
| 46 | const _DifferenceType __k = __res - __first; |
| 47 | for (_DifferenceType __old = __extremum; __comp(__k, __old); |
| 48 | __old = __extremum) |
| 49 | { |
| 50 | __extremum.compare_exchange_weak(__old, __k); |
| 51 | } |
| 52 | } |
| 53 | } |
| 54 | }); |
| 55 | return __extremum != __initial_dist ? __first + __extremum : __last; |
| 56 | } |
| 57 | |
| 58 | //------------------------------------------------------------------------ |
| 59 | // parallel_or |
| 60 | //------------------------------------------------------------------------ |
| 61 | //! Return true if brick f[i,j) returns true for some subrange [i,j) of [first,last) |
| 62 | template <class _ExecutionPolicy, class _Index, class _Brick> |
| 63 | bool |
| 64 | __parallel_or(_ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f) |
| 65 | { |
| 66 | std::atomic<bool> __found(false); |
| 67 | __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last, |
| 68 | [__f, &__found](_Index __i, _Index __j) { |
| 69 | if (!__found.load(m: std::memory_order_relaxed) && __f(__i, __j)) |
| 70 | { |
| 71 | __found.store(i: true, m: std::memory_order_relaxed); |
| 72 | __par_backend::__cancel_execution(); |
| 73 | } |
| 74 | }); |
| 75 | return __found; |
| 76 | } |
| 77 | |
| 78 | } // namespace __internal |
| 79 | } // namespace __pstl |
| 80 | |
| 81 | #endif /* _PSTL_PARALLEL_IMPL_H */ |
| 82 | |