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/CLGEMMMatrixAccumulateBiasesKernel.h"
26 #include "arm_compute/core/AccessWindowStatic.h"
27 #include "arm_compute/core/CL/CLHelpers.h"
28 #include "arm_compute/core/CL/CLKernelLibrary.h"
29 #include "arm_compute/core/CL/ICLTensor.h"
30 #include "arm_compute/core/CL/OpenCL.h"
31 #include "arm_compute/core/Error.h"
32 #include "arm_compute/core/Helpers.h"
33 #include "arm_compute/core/Types.h"
34 #include "arm_compute/core/Utils.h"
35 #include "arm_compute/core/Validate.h"
37 using namespace arm_compute;
39 CLGEMMMatrixAccumulateBiasesKernel::CLGEMMMatrixAccumulateBiasesKernel()
40 : _accum(nullptr), _biases(nullptr)
44 void CLGEMMMatrixAccumulateBiasesKernel::configure(ICLTensor *accum, const ICLTensor *biases)
46 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::F16, DataType::F32);
47 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::F16, DataType::F32);
48 ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
49 ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() != 1);
55 std::string data_type_name = lower_string(string_from_data_type(accum->info()->data_type()));
56 _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemm_accumulate_biases_" + data_type_name));
58 // Configure kernel window
59 const unsigned int processed_elements = max_cl_vector_width / data_size_from_type(accum->info()->data_type());
60 Window win = calculate_max_window(*_accum->info(), Steps(processed_elements));
61 AccessWindowStatic biases_access(biases->info(), 0, 0, biases->info()->dimension(0), biases->info()->dimension(1));
62 AccessWindowHorizontal accum_access(_accum->info(), 0, processed_elements);
63 update_window_and_padding(win, biases_access, accum_access);
64 ICLKernel::configure(win);
67 void CLGEMMMatrixAccumulateBiasesKernel::run(const Window &window, cl::CommandQueue &queue)
69 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
70 ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
72 Window accum_slice = window.first_slice_window_2D();
74 Window biases_slice(accum_slice);
75 biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
82 add_2D_tensor_argument(idx, _accum, accum_slice);
83 add_1D_tensor_argument(idx, _biases, biases_slice);
85 enqueue(queue, *this, accum_slice);
87 while(window.slide_window_slice_2D(accum_slice));