Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 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 "CL/CLAccessor.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 25 | #include "Globals.h" |
| 26 | #include "TensorLibrary.h" |
| 27 | #include "benchmark/Datasets.h" |
| 28 | #include "benchmark/Profiler.h" |
| 29 | #include "benchmark/WallClockTimer.h" |
| 30 | |
| 31 | #include "arm_compute/core/Helpers.h" |
| 32 | #include "arm_compute/core/Types.h" |
| 33 | #include "arm_compute/runtime/CL/CLScheduler.h" |
| 34 | #include "arm_compute/runtime/CL/CLTensor.h" |
| 35 | #include "arm_compute/runtime/CL/CLTensorAllocator.h" |
| 36 | #include "arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h" |
| 37 | |
| 38 | #include "benchmark/benchmark_api.h" |
| 39 | |
| 40 | #include <memory> |
| 41 | #include <string> |
| 42 | |
| 43 | using namespace arm_compute; |
| 44 | using namespace arm_compute::test; |
| 45 | using namespace arm_compute::test::benchmark; |
| 46 | using namespace arm_compute::test::cl; |
| 47 | |
| 48 | #include "benchmark/common/FullyConnectedLayer.h" |
| 49 | |
| 50 | namespace |
| 51 | { |
| 52 | using FullyConnectedLayerAlexNet = FullyConnectedLayer<AlexNetFullyConnectedLayerDataset, CLTensor, CLAccessor, CLFullyConnectedLayer>; |
| 53 | using FullyConnectedLayerLeNet5 = FullyConnectedLayer<LeNet5FullyConnectedLayerDataset, CLTensor, CLAccessor, CLFullyConnectedLayer>; |
| 54 | using FullyConnectedLayerGoogLeNet = FullyConnectedLayer<GoogLeNetFullyConnectedLayerDataset, CLTensor, CLAccessor, CLFullyConnectedLayer>; |
| 55 | } // namespace |
| 56 | |
| 57 | BENCHMARK_DEFINE_F(FullyConnectedLayerAlexNet, cl_alexnet) |
| 58 | (::benchmark::State &state) |
| 59 | { |
| 60 | while(state.KeepRunning()) |
| 61 | { |
| 62 | // Run function |
| 63 | profiler.start(); |
| 64 | fc_layer->run(); |
| 65 | CLScheduler::get().sync(); |
| 66 | profiler.stop(); |
| 67 | } |
| 68 | } |
| 69 | |
| 70 | BENCHMARK_REGISTER_F(FullyConnectedLayerAlexNet, cl_alexnet) |
| 71 | ->Threads(1) |
| 72 | ->Apply(DataSetArgBatched<AlexNetFullyConnectedLayerDataset, 0, 1, 4, 8>); |
| 73 | BENCHMARK_REGISTER_F(FullyConnectedLayerAlexNet, cl_alexnet) |
| 74 | ->Threads(1) |
| 75 | ->Apply(DataSetArgBatched<AlexNetFullyConnectedLayerDataset, 1, 1, 4, 8>); |
| 76 | BENCHMARK_REGISTER_F(FullyConnectedLayerAlexNet, cl_alexnet) |
| 77 | ->Threads(1) |
| 78 | ->Apply(DataSetArgBatched<AlexNetFullyConnectedLayerDataset, 2, 1, 4, 8>); |
| 79 | |
| 80 | BENCHMARK_DEFINE_F(FullyConnectedLayerLeNet5, cl_lenet5) |
| 81 | (::benchmark::State &state) |
| 82 | { |
| 83 | while(state.KeepRunning()) |
| 84 | { |
| 85 | // Run function |
| 86 | profiler.start(); |
| 87 | fc_layer->run(); |
| 88 | CLScheduler::get().sync(); |
| 89 | profiler.stop(); |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | BENCHMARK_REGISTER_F(FullyConnectedLayerLeNet5, cl_lenet5) |
| 94 | ->Threads(1) |
| 95 | ->Apply(DataSetArgBatched<LeNet5FullyConnectedLayerDataset, 0, 1, 4, 8>); |
| 96 | BENCHMARK_REGISTER_F(FullyConnectedLayerLeNet5, cl_lenet5) |
| 97 | ->Threads(1) |
| 98 | ->Apply(DataSetArgBatched<LeNet5FullyConnectedLayerDataset, 1, 1, 4, 8>); |
| 99 | |
| 100 | BENCHMARK_DEFINE_F(FullyConnectedLayerGoogLeNet, cl_googlenet) |
| 101 | (::benchmark::State &state) |
| 102 | { |
| 103 | while(state.KeepRunning()) |
| 104 | { |
| 105 | // Run function |
| 106 | profiler.start(); |
| 107 | fc_layer->run(); |
| 108 | CLScheduler::get().sync(); |
| 109 | profiler.stop(); |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | BENCHMARK_REGISTER_F(FullyConnectedLayerGoogLeNet, cl_googlenet) |
| 114 | ->Threads(1) |
| 115 | ->Apply(DataSetArgBatched<GoogLeNetFullyConnectedLayerDataset, 0, 1, 4, 8>); |