Michele Di Giorgio | 5b48ad7 | 2019-06-04 18:43:35 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2019 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 "arm_compute/core/CL/kernels/CLMeanStdDevNormalizationKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/CL/CLHelpers.h" |
| 27 | #include "arm_compute/core/CL/CLKernelLibrary.h" |
| 28 | #include "arm_compute/core/CL/CLValidate.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/TensorInfo.h" |
| 34 | #include "arm_compute/core/Types.h" |
| 35 | #include "arm_compute/core/Window.h" |
| 36 | |
| 37 | namespace arm_compute |
| 38 | { |
| 39 | namespace |
| 40 | { |
| 41 | Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, float epsilon) |
| 42 | { |
| 43 | ARM_COMPUTE_UNUSED(epsilon); |
| 44 | ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input); |
| 45 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input); |
| 46 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_dimensions() > 2, "Input tensor cannot have more than 2 dimensions"); |
| 47 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32); |
| 48 | |
| 49 | // Checks performed when output is configured |
| 50 | if((output != nullptr) && (output->total_size() != 0)) |
| 51 | { |
| 52 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); |
| 53 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); |
| 54 | } |
| 55 | return Status{}; |
| 56 | } |
| 57 | |
| 58 | std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output) |
| 59 | { |
| 60 | if(output != nullptr) |
| 61 | { |
| 62 | ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); |
| 63 | // Output auto inizialitation if not yet initialized |
| 64 | auto_init_if_empty(*output, *input); |
| 65 | } |
| 66 | |
| 67 | const unsigned int num_elems_processed_per_iteration = 16 / input->element_size(); |
| 68 | |
| 69 | // This kernel doesn't need padding |
| 70 | Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration)); |
| 71 | if(output != nullptr) |
| 72 | { |
| 73 | output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape())); |
| 74 | } |
| 75 | |
| 76 | return std::make_pair(Status{}, win); |
| 77 | } |
| 78 | } // namespace |
| 79 | |
| 80 | CLMeanStdDevNormalizationKernel::CLMeanStdDevNormalizationKernel() |
| 81 | : _input(nullptr), _output(nullptr), _run_in_place(false) |
| 82 | { |
| 83 | } |
| 84 | |
| 85 | void CLMeanStdDevNormalizationKernel::configure(ICLTensor *input, ICLTensor *output, float epsilon) |
| 86 | { |
| 87 | ARM_COMPUTE_ERROR_ON_NULLPTR(input); |
| 88 | |
| 89 | _run_in_place = (output == nullptr) || (output == input); |
| 90 | |
| 91 | ARM_COMPUTE_ERROR_THROW_ON(CLMeanStdDevNormalizationKernel::validate(input->info(), (output != nullptr) ? output->info() : nullptr, epsilon)); |
| 92 | |
| 93 | _input = input; |
| 94 | _output = output; |
| 95 | |
| 96 | const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size(); |
| 97 | |
| 98 | // Set build options |
| 99 | CLBuildOptions build_opts; |
| 100 | build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())); |
| 101 | build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration)); |
| 102 | build_opts.add_option("-DEPSILON=" + float_to_string_with_full_precision(epsilon)); |
| 103 | build_opts.add_option("-DWIDTH=" + support::cpp11::to_string(input->info()->dimension(0))); |
| 104 | build_opts.add_option_if(_run_in_place, "-DIN_PLACE"); |
| 105 | |
| 106 | // Create kernel |
| 107 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("mean_stddev_normalization", build_opts.options())); |
| 108 | |
| 109 | // Configure kernel window |
| 110 | auto win_config = validate_and_configure_window(input->info(), (_run_in_place) ? nullptr : output->info()); |
| 111 | ARM_COMPUTE_ERROR_THROW_ON(win_config.first); |
| 112 | ICLKernel::configure_internal(win_config.second); |
| 113 | |
| 114 | // Set config_id for enabling LWS tuning |
| 115 | _config_id = "mean_stddev_normalization_layer_"; |
| 116 | _config_id += lower_string(string_from_data_type(input->info()->data_type())); |
| 117 | _config_id += "_"; |
| 118 | _config_id += support::cpp11::to_string(input->info()->dimension(0)); |
| 119 | _config_id += "_"; |
| 120 | _config_id += support::cpp11::to_string(input->info()->dimension(1)); |
| 121 | } |
| 122 | |
| 123 | Status CLMeanStdDevNormalizationKernel::validate(const ITensorInfo *input, const ITensorInfo *output, float epsilon) |
| 124 | { |
| 125 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, epsilon)); |
| 126 | ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), (output != nullptr) ? output->clone().get() : nullptr).first); |
| 127 | return Status{}; |
| 128 | } |
| 129 | |
| 130 | void CLMeanStdDevNormalizationKernel::run(const Window &window, cl::CommandQueue &queue) |
| 131 | { |
| 132 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 133 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); |
| 134 | |
| 135 | Window slice = window.first_slice_window_2D(); |
| 136 | // Set slice step equal to width to force gws[0] to 1, as each thread normalizes across all rows |
| 137 | slice.set_dimension_step(Window::DimX, _input->info()->dimension(0)); |
| 138 | |
| 139 | do |
| 140 | { |
| 141 | unsigned int idx = 0; |
| 142 | add_2D_tensor_argument(idx, _input, slice); |
Michalis Spyrou | e1651a5 | 2019-07-11 15:00:49 +0100 | [diff] [blame] | 143 | add_2D_tensor_argument_if((!_run_in_place), idx, _output, slice); |
| 144 | |
Michele Di Giorgio | 5b48ad7 | 2019-06-04 18:43:35 +0100 | [diff] [blame] | 145 | enqueue(queue, *this, slice, lws_hint()); |
| 146 | } |
| 147 | while(window.slide_window_slice_2D(slice)); |
| 148 | } |
| 149 | } // namespace arm_compute |