Georgios Pinitas | 6d9d6f4 | 2018-12-24 16:10:47 +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/NEStridedSliceKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/CPP/Validate.h" |
| 27 | #include "arm_compute/core/IAccessWindow.h" |
| 28 | #include "arm_compute/core/ITensor.h" |
| 29 | #include "arm_compute/core/TensorInfo.h" |
| 30 | #include "arm_compute/core/Window.h" |
| 31 | |
| 32 | #include "arm_compute/core/Types.h" |
| 33 | #include "arm_compute/core/utils/helpers/bit_ops.h" |
| 34 | #include "arm_compute/core/utils/helpers/tensor_transform.h" |
| 35 | #include "arm_compute/core/utils/misc/ShapeCalculator.h" |
| 36 | |
| 37 | namespace arm_compute |
| 38 | { |
| 39 | namespace |
| 40 | { |
| 41 | Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, |
| 42 | const Coordinates &starts, const Coordinates &ends, const BiStrides &strides, |
| 43 | int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask) |
| 44 | { |
| 45 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); |
Georgios Pinitas | 3384356 | 2019-12-10 13:33:18 +0000 | [diff] [blame] | 46 | ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN); |
Georgios Pinitas | 6d9d6f4 | 2018-12-24 16:10:47 +0000 | [diff] [blame] | 47 | |
| 48 | ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().num_dimensions() > 4); |
| 49 | ARM_COMPUTE_RETURN_ERROR_ON(starts.num_dimensions() > input->num_dimensions()); |
| 50 | ARM_COMPUTE_RETURN_ERROR_ON(ends.num_dimensions() > input->num_dimensions()); |
| 51 | ARM_COMPUTE_RETURN_ERROR_ON(strides.num_dimensions() > input->num_dimensions()); |
| 52 | ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(strides.cbegin(), strides.cbegin() + strides.num_dimensions(), [](int i) |
| 53 | { |
| 54 | return i == 0; |
| 55 | })); |
| 56 | |
| 57 | // Get expected output shape |
| 58 | const TensorShape exp_output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(*input, |
| 59 | starts, ends, strides, |
| 60 | begin_mask, end_mask, shrink_axis_mask); |
| 61 | ARM_COMPUTE_RETURN_ERROR_ON(exp_output_shape.total_size() == 0); |
| 62 | |
| 63 | // Checks output if configured |
| 64 | if(output->total_size() != 0) |
| 65 | { |
| 66 | const TensorInfo exp_output_info = output->clone()->set_tensor_shape(exp_output_shape); |
| 67 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &exp_output_info); |
| 68 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); |
| 69 | } |
| 70 | |
| 71 | return Status{}; |
| 72 | } |
| 73 | |
| 74 | std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, |
| 75 | const Coordinates &starts, const Coordinates &ends, const BiStrides &strides, |
| 76 | int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask) |
| 77 | { |
| 78 | // Output tensor auto initialization if not yet initialized |
| 79 | const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(*input, |
| 80 | starts, ends, strides, |
| 81 | begin_mask, end_mask, shrink_axis_mask); |
| 82 | auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape)); |
| 83 | |
| 84 | // Create window |
Michele Di Giorgio | 4646d2e | 2019-06-19 12:28:47 +0100 | [diff] [blame] | 85 | Window win = calculate_max_window(*output, Steps()); |
Georgios Pinitas | 6d9d6f4 | 2018-12-24 16:10:47 +0000 | [diff] [blame] | 86 | output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape())); |
| 87 | |
| 88 | return std::make_pair(Status{}, win); |
| 89 | } |
| 90 | |
| 91 | void strided_slice_generic(const ITensor *input, ITensor *output, |
| 92 | const Coordinates &starts, const BiStrides &strides, int32_t shrink_axis_mask, |
| 93 | const Window &window) |
| 94 | { |
| 95 | Iterator output_it(output, window); |
| 96 | const size_t width_size = input->info()->element_size(); |
| 97 | |
| 98 | const bool is_shrink_w = arm_compute::helpers::bit_ops::is_bit_set(shrink_axis_mask, 0); |
| 99 | const bool is_shrink_h = arm_compute::helpers::bit_ops::is_bit_set(shrink_axis_mask, 1); |
| 100 | const bool is_shrink_c = arm_compute::helpers::bit_ops::is_bit_set(shrink_axis_mask, 2); |
| 101 | const bool is_shrink_n = arm_compute::helpers::bit_ops::is_bit_set(shrink_axis_mask, 3); |
| 102 | |
| 103 | unsigned int index = 0; |
| 104 | const int idx_w = is_shrink_w ? 0 : index++; |
| 105 | const int idx_h = is_shrink_h ? 0 : index++; |
| 106 | const int idx_c = is_shrink_c ? 0 : index++; |
| 107 | const int idx_n = is_shrink_n ? 0 : index; |
| 108 | |
| 109 | BiStrides shrinked_strides; |
| 110 | shrinked_strides.set(0, is_shrink_w ? 0 : strides[0]); |
| 111 | shrinked_strides.set(1, is_shrink_h ? 0 : strides[1]); |
| 112 | shrinked_strides.set(2, is_shrink_c ? 0 : strides[2]); |
| 113 | shrinked_strides.set(3, is_shrink_n ? 0 : strides[3]); |
| 114 | |
| 115 | execute_window_loop(window, [&](const Coordinates & id) |
| 116 | { |
| 117 | const int w_coord = starts[0] + (id[idx_w] * shrinked_strides[0]); |
| 118 | const int h_coord = starts[1] + (id[idx_h] * shrinked_strides[1]); |
| 119 | const int c_coord = starts[2] + (id[idx_c] * shrinked_strides[2]); |
| 120 | const int n_coord = starts[3] + (id[idx_n] * shrinked_strides[3]); |
| 121 | |
| 122 | Coordinates in_coords(w_coord, h_coord, c_coord, n_coord); |
| 123 | std::copy_n(input->ptr_to_element(in_coords), width_size, output_it.ptr()); |
| 124 | }, |
| 125 | output_it); |
| 126 | } |
| 127 | } // namespace |
| 128 | |
| 129 | NEStridedSliceKernel::NEStridedSliceKernel() |
| 130 | : _input(nullptr), _output(nullptr), _starts_abs(), _final_strides(), _shrink_mask() |
| 131 | { |
| 132 | } |
| 133 | |
| 134 | void NEStridedSliceKernel::configure(const ITensor *input, ITensor *output, |
| 135 | const Coordinates &starts, const Coordinates &ends, const BiStrides &strides, |
| 136 | int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask) |
| 137 | { |
| 138 | ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); |
| 139 | ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), starts, ends, strides, begin_mask, end_mask, shrink_axis_mask)); |
| 140 | |
| 141 | _input = input; |
| 142 | _output = output; |
| 143 | _shrink_mask = shrink_axis_mask; |
| 144 | |
| 145 | const TensorShape &input_shape = input->info()->tensor_shape(); |
| 146 | |
| 147 | Coordinates ends_abs; |
| 148 | std::tie(_starts_abs, ends_abs, _final_strides) = arm_compute::helpers::tensor_transform::calculate_strided_slice_coords( |
| 149 | input_shape, |
| 150 | starts, ends, strides, |
| 151 | begin_mask, end_mask, shrink_axis_mask); |
| 152 | |
| 153 | // Configure kernel window |
| 154 | auto win_config = validate_and_configure_window(input->info(), output->info(), starts, ends, strides, begin_mask, end_mask, shrink_axis_mask); |
| 155 | ARM_COMPUTE_ERROR_THROW_ON(win_config.first); |
| 156 | INEKernel::configure(win_config.second); |
| 157 | } |
| 158 | |
| 159 | Status NEStridedSliceKernel::validate(const ITensorInfo *input, const ITensorInfo *output, |
| 160 | const Coordinates &starts, const Coordinates &ends, const BiStrides &strides, |
| 161 | int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask) |
| 162 | { |
| 163 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask)); |
| 164 | ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), |
| 165 | starts, ends, strides, begin_mask, end_mask, shrink_axis_mask) |
| 166 | .first); |
| 167 | |
| 168 | return Status{}; |
| 169 | } |
| 170 | |
| 171 | void NEStridedSliceKernel::run(const Window &window, const ThreadInfo &info) |
| 172 | { |
| 173 | ARM_COMPUTE_UNUSED(info); |
| 174 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 175 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); |
| 176 | |
| 177 | // Dispatch kernel |
| 178 | strided_slice_generic(_input, _output, _starts_abs, _final_strides, _shrink_mask, window); |
| 179 | } |
| 180 | } // namespace arm_compute |