2 * Copyright (c) 2017 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
24 #include "arm_compute/core/CL/kernels/CLDepthwiseVectorToTensorKernel.h"
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/CL/OpenCL.h"
30 #include "arm_compute/core/Error.h"
31 #include "arm_compute/core/Helpers.h"
32 #include "arm_compute/core/Types.h"
33 #include "support/ToolchainSupport.h"
35 using namespace arm_compute;
37 CLDepthwiseVectorToTensorKernel::CLDepthwiseVectorToTensorKernel()
38 : _input(nullptr), _output(nullptr)
42 void CLDepthwiseVectorToTensorKernel::configure(const ICLTensor *input, ICLTensor *output, size_t conv_w, size_t conv_h)
44 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
45 ARM_COMPUTE_ERROR_ON_NULLPTR(output);
47 TensorShape output_shape = input->info()->tensor_shape();
48 output_shape.set(0, conv_w);
49 output_shape.set(1, conv_h);
50 output_shape.set(2, input->info()->tensor_shape()[0] / (conv_w * conv_h));
52 // Output auto inizialitation if not yet initialized
53 auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->fixed_point_position());
55 ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
56 ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
57 ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
63 std::set<std::string> build_opts;
64 build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
65 build_opts.emplace("-DCONV_WIDTH=" + support::cpp11::to_string(conv_w));
66 build_opts.emplace("-DCONV_HEIGHT=" + support::cpp11::to_string(conv_h));
68 _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("depthwise_vector_to_tensor", build_opts));
70 // Configure kernel window
71 Window win = calculate_max_window(*input->info(), Steps());
72 // The CLDepthwisevectorToTensorKernel doesn't need padding so update_window_and_padding() can be skipped
73 output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
75 ICLKernel::configure(win);
78 void CLDepthwiseVectorToTensorKernel::run(const Window &window, cl::CommandQueue &queue)
80 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
81 ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
83 Window slice = window.first_slice_window_1D();
84 Window slice_out = window.first_slice_window_3D();
87 slice.set(Window::DimX, Window::Dimension(0, _input->info()->dimension(0), 1));
90 // The first three dimensions of the output are increased by the inner loops
91 slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
92 slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
93 slice_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
98 add_1D_tensor_argument(idx, _input, slice);
99 add_3D_tensor_argument(idx, _output, slice_out);
100 enqueue(queue, *this, slice);
102 while(window.slide_window_slice_1D(slice) && window.slide_window_slice_3D(slice_out));