1 /*
2 * Copyright (c) 2017-2020 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #include "arm_compute/runtime/OMP/OMPScheduler.h"
25
26 #include "arm_compute/core/CPP/ICPPKernel.h"
27 #include "arm_compute/core/Error.h"
28 #include "arm_compute/core/Helpers.h"
29 #include "arm_compute/core/Utils.h"
30 #include "src/runtime/CPUUtils.h"
31 #include <omp.h>
32
33 namespace arm_compute
34 {
OMPScheduler()35 OMPScheduler::OMPScheduler() // NOLINT
36 : _num_threads(omp_get_max_threads())
37 {
38 }
39
num_threads() const40 unsigned int OMPScheduler::num_threads() const
41 {
42 return _num_threads;
43 }
44
set_num_threads(unsigned int num_threads)45 void OMPScheduler::set_num_threads(unsigned int num_threads)
46 {
47 const unsigned int num_cores = omp_get_max_threads();
48 _num_threads = (num_threads == 0) ? num_cores : num_threads;
49 }
50
schedule(ICPPKernel * kernel,const Hints & hints)51 void OMPScheduler::schedule(ICPPKernel *kernel, const Hints &hints)
52 {
53 ITensorPack tensors;
54 schedule_common(kernel, hints, tensors);
55 }
56
schedule_op(ICPPKernel * kernel,const Hints & hints,ITensorPack & tensors)57 void OMPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, ITensorPack &tensors)
58 {
59 ARM_COMPUTE_ERROR_ON_MSG(!kernel, "The child class didn't set the kernel");
60 ARM_COMPUTE_ERROR_ON_MSG(hints.strategy() == StrategyHint::DYNAMIC,
61 "Dynamic scheduling is not supported in OMPScheduler");
62
63 const Window &max_window = kernel->window();
64 const unsigned int num_iterations = max_window.num_iterations(hints.split_dimension());
65 const unsigned int num_threads = std::min(num_iterations, _num_threads);
66
67 if(!kernel->is_parallelisable() || num_threads == 1)
68 {
69 ThreadInfo info;
70 info.cpu_info = &_cpu_info;
71 kernel->run_op(tensors, max_window, info);
72 }
73 else
74 {
75 const unsigned int num_windows = num_threads;
76 std::vector<IScheduler::Workload> workloads(num_windows);
77 for(unsigned int t = 0; t < num_windows; t++)
78 {
79 //Capture 't' by copy, all the other variables by reference:
80 workloads[t] = [t, &hints, &max_window, &num_windows, &kernel, &tensors](const ThreadInfo & info)
81 {
82 Window win = max_window.split_window(hints.split_dimension(), t, num_windows);
83 win.validate();
84 kernel->run_op(tensors, win, info);
85 };
86 }
87 run_workloads(workloads);
88 }
89 }
90 #ifndef DOXYGEN_SKIP_THIS
run_workloads(std::vector<arm_compute::IScheduler::Workload> & workloads)91 void OMPScheduler::run_workloads(std::vector<arm_compute::IScheduler::Workload> &workloads)
92 {
93 const unsigned int num_threads = std::min(_num_threads, static_cast<unsigned int>(workloads.size()));
94 if(num_threads < 1)
95 {
96 return;
97 }
98
99 ThreadInfo info;
100 info.cpu_info = &_cpu_info;
101 info.num_threads = num_threads;
102 #pragma omp parallel firstprivate(info) num_threads(num_threads)
103 {
104 const int tid = omp_get_thread_num();
105 info.thread_id = tid;
106 workloads[tid](info);
107 }
108 }
109 #endif /* DOXYGEN_SKIP_THIS */
110 } // namespace arm_compute
111