xref: /aosp_15_r20/external/ComputeLibrary/src/runtime/Utils.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
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 "src/runtime/Utils.h"
25 
26 #include "arm_compute/runtime/NEON/NEScheduler.h"
27 
28 #include <cmath>
29 #include <map>
30 #include <string>
31 
32 namespace arm_compute
33 {
34 namespace utils
35 {
36 #ifndef DOXYGEN_SKIP_THIS
37 static const std::string information =
38 #include "arm_compute_version.embed"
39     ;
40 #endif /* DOXYGEN_SKIP_THIS */
41 
string_from_scheduler_type(Scheduler::Type t)42 const std::string &string_from_scheduler_type(Scheduler::Type t)
43 {
44     static std::map<Scheduler::Type, const std::string> scheduler_type_map =
45     {
46         { Scheduler::Type::ST, "Single Thread" },
47         { Scheduler::Type::CPP, "C++11 Threads" },
48         { Scheduler::Type::OMP, "OpenMP Threads" },
49         { Scheduler::Type::CUSTOM, "Custom" }
50     };
51 
52     return scheduler_type_map[t];
53 }
54 
schedule_kernel_on_ctx(IRuntimeContext * ctx,ICPPKernel * kernel,const IScheduler::Hints & hints)55 void schedule_kernel_on_ctx(IRuntimeContext *ctx, ICPPKernel *kernel, const IScheduler::Hints &hints)
56 {
57     if(ctx)
58     {
59         ARM_COMPUTE_ERROR_ON(ctx->scheduler() == nullptr);
60         ctx->scheduler()->schedule(kernel, hints);
61     }
62     else
63     {
64         NEScheduler::get().schedule(kernel, hints);
65     }
66 }
67 
calculate_number_of_stages_only_x_axis(size_t input_x_dimension,unsigned int axis)68 unsigned int calculate_number_of_stages_only_x_axis(size_t input_x_dimension, unsigned int axis)
69 {
70     // We need only 1 stage for all axis except x-axis
71     if(axis != 0)
72     {
73         return 1;
74     }
75     // Calculate number of WGs. 16 elements per thread, 8 threads per WG
76     const auto num_of_wg = static_cast<unsigned int>(ceil(input_x_dimension / 128.f));
77 
78     // Calculate number of stages. First stage performs op and the rest reduction sum
79     // depending on the size of the input. Last stage should have only 1 WG.
80     const unsigned int num_of_stages = num_of_wg / 128 + 2;
81     return num_of_stages;
82 }
83 } // namespace utils
84 } // namespace arm_compute
85