Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2018-2019 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/NEON/kernels/NERangeKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/Error.h" |
| 27 | #include "arm_compute/core/Helpers.h" |
| 28 | #include "arm_compute/core/IAccessWindow.h" |
| 29 | #include "arm_compute/core/ITensor.h" |
| 30 | #include "arm_compute/core/NEON/NEAsymm.h" |
| 31 | #include "arm_compute/core/NEON/wrapper/wrapper.h" |
| 32 | #include "arm_compute/core/TensorInfo.h" |
| 33 | #include "arm_compute/core/Validate.h" |
| 34 | |
Michalis Spyrou | f63885b | 2019-01-16 14:18:09 +0000 | [diff] [blame] | 35 | #include "arm_compute/core/Utils.h" |
Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 36 | |
| 37 | namespace arm_compute |
| 38 | { |
| 39 | namespace |
| 40 | { |
| 41 | template <typename T> |
| 42 | void range_function(ITensor *output, float start, float step, const Window &window) |
| 43 | { |
| 44 | const unsigned int num_elems_processed_per_iteration = 16 / sizeof(T); |
| 45 | /** NEON vector tag type. */ |
| 46 | using ExactTagType = typename wrapper::traits::neon_bitvector<T, wrapper::traits::BitWidth::W128>::tag_type; |
| 47 | |
| 48 | const auto step_vec = wrapper::vdup_n(static_cast<T>(step), ExactTagType{}); |
| 49 | const auto start_vec = wrapper::vdup_n(static_cast<T>(start), ExactTagType{}); |
| 50 | auto id_vec = wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{}); |
| 51 | |
| 52 | Iterator output_it(output, window); |
| 53 | execute_window_loop(window, [&](const Coordinates & id) |
| 54 | { |
| 55 | for(unsigned int count = 0; count < num_elems_processed_per_iteration; ++count) |
| 56 | { |
| 57 | id_vec = wrapper::vsetlane(static_cast<T>(id.x() + count), id_vec, count); |
| 58 | } |
| 59 | // start + step * id |
| 60 | const auto res_vec = wrapper::vmla(start_vec, id_vec, step_vec); |
| 61 | const auto out_ptr = reinterpret_cast<T *>(output_it.ptr()); |
| 62 | wrapper::vstore(out_ptr, res_vec); |
| 63 | }, |
| 64 | output_it); |
| 65 | } |
| 66 | |
| 67 | Status validate_arguments(const ITensorInfo &output, const float start, const float end, const float step) |
| 68 | { |
| 69 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output, |
| 70 | 1, |
| 71 | DataType::U8, DataType::S8, |
| 72 | DataType::U16, DataType::S16, |
| 73 | DataType::U32, DataType::S32, |
| 74 | DataType::F16, DataType::F32); |
| 75 | |
| 76 | ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end"); |
| 77 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start < end) && (step <= 0)), "step must be greater than 0 when start < end"); |
| 78 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start > end) && (step >= 0)), "step must be less than 0 when start > end"); |
| 79 | |
Michalis Spyrou | f63885b | 2019-01-16 14:18:09 +0000 | [diff] [blame] | 80 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(start, output.data_type(), output.quantization_info()), "start value is outside the range of the data type"); |
| 81 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(end, output.data_type(), output.quantization_info()), "end value is outside the range of the data type"); |
| 82 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(step, output.data_type(), output.quantization_info()), "step value is outside the range of the data type"); |
Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 83 | |
| 84 | ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end"); |
| 85 | |
| 86 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.num_dimensions() != 1, "Output has to be a 1-D tensor"); |
Michalis Spyrou | f63885b | 2019-01-16 14:18:09 +0000 | [diff] [blame] | 87 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.tensor_shape().total_size() < num_of_elements_in_range(start, end, step), "Output tensor size is incorrect"); |
Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 88 | |
| 89 | return Status{}; |
| 90 | } |
| 91 | |
| 92 | std::pair<Status, Window> validate_and_configure_window(ITensorInfo &output, const float start, const float end, const float step) |
| 93 | { |
| 94 | const unsigned int num_elems_processed_per_iteration = 16 / output.element_size(); |
| 95 | |
| 96 | // Auto initialize output if not initialized |
Michalis Spyrou | f63885b | 2019-01-16 14:18:09 +0000 | [diff] [blame] | 97 | auto_init_if_empty(output, TensorShape(num_of_elements_in_range(start, end, step)), 1, output.data_type(), output.quantization_info()); |
Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 98 | |
| 99 | // Configure kernel window |
| 100 | Window win = calculate_max_window(output, Steps(num_elems_processed_per_iteration)); |
| 101 | AccessWindowHorizontal output_access(&output, 0, num_elems_processed_per_iteration); |
| 102 | bool window_changed = update_window_and_padding(win, output_access); |
Michalis Spyrou | f63885b | 2019-01-16 14:18:09 +0000 | [diff] [blame] | 103 | output_access.set_valid_region(win, ValidRegion(Coordinates(), TensorShape(num_of_elements_in_range(start, end, step)))); |
Manuel Bottini | 053e751 | 2018-12-28 15:05:20 +0000 | [diff] [blame] | 104 | Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; |
| 105 | return std::make_pair(err, win); |
| 106 | } |
| 107 | } // namespace |
| 108 | |
| 109 | NERangeKernel::NERangeKernel() |
| 110 | : _func(nullptr), _start(0), _end(1), _step(1), _output(nullptr) |
| 111 | { |
| 112 | } |
| 113 | |
| 114 | void NERangeKernel::configure(ITensor *output, float start, float end, float step) |
| 115 | { |
| 116 | ARM_COMPUTE_ERROR_ON_NULLPTR(output); |
| 117 | |
| 118 | ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*(output->info()), start, end, step)); |
| 119 | |
| 120 | // Configure kernel window |
| 121 | auto win_config = validate_and_configure_window(*(output->info()), start, end, step); |
| 122 | ARM_COMPUTE_ERROR_THROW_ON(win_config.first); |
| 123 | |
| 124 | _start = start; |
| 125 | _end = end; |
| 126 | _step = step; |
| 127 | _output = output; |
| 128 | switch(_output->info()->data_type()) |
| 129 | { |
| 130 | case DataType::U8: |
| 131 | _func = &range_function<uint8_t>; |
| 132 | break; |
| 133 | case DataType::U16: |
| 134 | _func = &range_function<uint16_t>; |
| 135 | break; |
| 136 | case DataType::U32: |
| 137 | _func = &range_function<uint32_t>; |
| 138 | break; |
| 139 | case DataType::S8: |
| 140 | _func = &range_function<int8_t>; |
| 141 | break; |
| 142 | case DataType::S16: |
| 143 | _func = &range_function<int16_t>; |
| 144 | break; |
| 145 | case DataType::S32: |
| 146 | _func = &range_function<int32_t>; |
| 147 | break; |
| 148 | case DataType::F32: |
| 149 | _func = &range_function<float>; |
| 150 | break; |
| 151 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 152 | case DataType::F16: |
| 153 | _func = &range_function<float16_t>; |
| 154 | break; |
| 155 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 156 | default: |
| 157 | ARM_COMPUTE_ERROR("Unsupported data type."); |
| 158 | break; |
| 159 | } |
| 160 | |
| 161 | INEKernel::configure(win_config.second); |
| 162 | } |
| 163 | |
| 164 | Status NERangeKernel::validate(const ITensorInfo *output, float start, float end, float step) |
| 165 | { |
| 166 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output); |
| 167 | |
| 168 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*output, start, end, step)); |
| 169 | ARM_COMPUTE_RETURN_ON_ERROR((validate_and_configure_window(*(output->clone()), start, end, step)).first); |
| 170 | |
| 171 | return Status{}; |
| 172 | } |
| 173 | |
| 174 | void NERangeKernel::run(const Window &window, const ThreadInfo &info) |
| 175 | { |
| 176 | ARM_COMPUTE_UNUSED(info); |
| 177 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 178 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); |
| 179 | ARM_COMPUTE_ERROR_ON(_func == nullptr); |
| 180 | |
| 181 | (*_func)(_output, _start, _step, window); |
| 182 | } |
| 183 | } // namespace arm_compute |