2 * Copyright (c) 2020 Samsung Electronics Co., Ltd. All Rights Reserved
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * Copyright (c) 2017-2019 ARM Limited.
20 * SPDX-License-Identifier: MIT
22 * Permission is hereby granted, free of charge, to any person obtaining a copy
23 * of this software and associated documentation files (the "Software"), to
24 * deal in the Software without restriction, including without limitation the
25 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
26 * sell copies of the Software, and to permit persons to whom the Software is
27 * furnished to do so, subject to the following conditions:
29 * The above copyright notice and this permission notice shall be included in all
30 * copies or substantial portions of the Software.
32 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
33 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
34 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
35 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
36 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
37 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
41 #include "arm_compute/core/CL/kernels/CLMultiplyScaleFactorKernel.h"
43 #include "arm_compute/core/AccessWindowStatic.h"
44 #include "arm_compute/core/CL/CLHelpers.h"
45 #include "arm_compute/core/CL/CLKernelLibraryEx.h"
46 #include "arm_compute/core/CL/CLValidate.h"
47 #include "arm_compute/core/CL/ICLTensor.h"
48 #include "arm_compute/core/TensorInfo.h"
49 #include "arm_compute/core/Utils.h"
50 #include "arm_compute/core/Validate.h"
51 #include "arm_compute/core/Window.h"
52 #include "support/StringSupport.h"
54 using namespace arm_compute;
58 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *scale_factor,
59 const ITensorInfo *output)
61 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
62 ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 2);
63 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
64 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(scale_factor, 1, DataType::F16,
66 ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->tensor_shape().total_size() == 0);
67 ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->num_dimensions() > 1);
68 ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->dimension(0) != input->dimension(1));
69 ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape().total_size() == 0);
70 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
71 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
73 // Checks performed when output is configured
74 if ((output->total_size() != 0))
76 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
82 std::tuple<Status, Window> validate_and_configure_window(const ITensorInfo *input,
85 // Configure kernel window
86 Window win = calculate_max_window(*input, Steps());
88 // Output tensor auto initialization if not yet initialized
89 auto_init_if_empty(*output, input->tensor_shape(), 1, DataType::F32);
91 // CLMultiplyScaleFactorKernel doesn't need padding so update_window_and_padding() can be
94 coord.set_num_dimensions(output->num_dimensions());
95 output->set_valid_region(ValidRegion(coord, output->tensor_shape()));
97 return std::make_tuple(Status{}, win);
101 CLMultiplyScaleFactorKernel::CLMultiplyScaleFactorKernel()
102 : _input(nullptr), _scale_factor(nullptr), _output(nullptr), _multiplier(1.f)
106 void CLMultiplyScaleFactorKernel::configure(const ICLTensor *input, const ICLTensor *scale_factor,
107 ICLTensor *output, float multiplier)
109 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
110 ARM_COMPUTE_ERROR_THROW_ON(
111 validate_arguments(input->info(), scale_factor->info(), output->info()));
114 _scale_factor = scale_factor;
116 _multiplier = multiplier;
118 const int vec_size_x = 16 / output->info()->element_size();
119 const int output_width_x = output->info()->tensor_shape().x();
120 const bool multi_access_x = (output_width_x / vec_size_x > 0);
122 // Create and update the window (if needed)
123 Window win = calculate_max_window(*output->info());
126 win.set(Window::DimX,
127 Window::Dimension(win.x().start(), ceil_to_multiple(win.x().end(), vec_size_x),
130 ICLKernel::configure_internal(win);
133 CLBuildOptions build_opts;
134 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
135 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
136 build_opts.add_option_if(
137 multi_access_x, "-DLAST_ACCESSED_X=" +
138 support::cpp11::to_string(std::max<int>(output_width_x - vec_size_x, 0)));
140 _kernel = static_cast<cl::Kernel>(
141 CLKernelLibraryEx::get().create_kernel("multiply_scale_factor", build_opts.options()));
144 Status CLMultiplyScaleFactorKernel::validate(const ITensorInfo *input,
145 const ITensorInfo *scale_factor,
146 const ITensorInfo *output)
148 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, scale_factor, output));
149 ARM_COMPUTE_RETURN_ON_ERROR(
150 std::get<0>(validate_and_configure_window(input->clone().get(), output->clone().get())));
154 void CLMultiplyScaleFactorKernel::run(const Window &window, cl::CommandQueue &queue)
156 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
157 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
159 Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
160 Window slice = window_collapsed.first_slice_window_2D();
162 // Set scale_factor window
163 Window win_scale = calculate_max_window(*_scale_factor->info(), Steps());
167 unsigned int idx = 0;
168 add_2D_tensor_argument(idx, _input, slice);
169 add_1D_tensor_argument(idx, _scale_factor, win_scale);
170 add_2D_tensor_argument(idx, _output, slice);
171 _kernel.setArg<float>(idx++, _multiplier);
172 enqueue(queue, *this, slice, lws_hint());
173 } while (window_collapsed.slide_window_slice_2D(slice));