2 * Copyright (c) 2017-2018 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/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.h"
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/GLES_COMPUTE/GCHelpers.h"
28 #include "arm_compute/core/GLES_COMPUTE/GCKernelLibrary.h"
29 #include "arm_compute/core/GLES_COMPUTE/IGCTensor.h"
30 #include "arm_compute/core/GLES_COMPUTE/OpenGLES.h"
31 #include "arm_compute/core/Helpers.h"
32 #include "arm_compute/core/Types.h"
33 #include "arm_compute/core/Validate.h"
34 #include "arm_compute/core/Window.h"
36 using namespace arm_compute;
38 GCGEMMMatrixAdditionKernel::GCGEMMMatrixAdditionKernel()
39 : _input(nullptr), _output(nullptr)
43 void GCGEMMMatrixAdditionKernel::configure(const IGCTensor *input, IGCTensor *output, float beta)
45 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
46 ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
47 ARM_COMPUTE_ERROR_ON(input->info()->dimension(0) != output->info()->dimension(0));
48 ARM_COMPUTE_ERROR_ON(input->info()->dimension(1) != output->info()->dimension(1));
52 const unsigned int num_elems_processed_per_iteration = max_gc_vector_width / data_size_from_type(input->info()->data_type());
54 std::set<std::string> build_opts;
55 std::string dt_name = (input->info()->data_type() == DataType::F32) ? "DATA_TYPE_FP32" : "DATA_TYPE_FP16";
56 build_opts.emplace(("#define " + dt_name));
57 build_opts.emplace("#define LOCAL_SIZE_X " + support::cpp11::to_string(1));
58 build_opts.emplace("#define LOCAL_SIZE_Y " + support::cpp11::to_string(1));
59 build_opts.emplace("#define LOCAL_SIZE_Z " + support::cpp11::to_string(1));
60 build_opts.emplace("#define BETA " + float_to_string_with_full_precision(beta));
63 build_opts.emplace("#define GEMM_MATRIXADDITION");
64 std::string data_type_name = lower_string(string_from_data_type(input->info()->data_type()));
65 _kernel = GCKernelLibrary::get().create_kernel(("gemm_ma"), build_opts);
67 // Configure kernel window
68 Window win = calculate_max_window(*_input->info(), Steps(num_elems_processed_per_iteration));
70 AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration);
71 AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
73 update_window_and_padding(win, input_access, output_access);
75 output_access.set_valid_region(win, input->info()->valid_region());
77 IGCKernel::configure(win);
80 void GCGEMMMatrixAdditionKernel::run(const Window &window)
82 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
83 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IGCKernel::window(), window);
87 Window slice = window.first_slice_window_2D();
92 add_2D_tensor_argument(idx, _input, 1, slice);
93 add_2D_tensor_argument(idx, _output, 2, slice);
95 _kernel.update_shader_params();
97 enqueue(*this, slice);
99 while(window.slide_window_slice_2D(slice));