1 /*
2 * Copyright (c) 2018-2021, 2023 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 "src/core/NEON/kernels/NEStridedSliceKernel.h"
25
26 #include "arm_compute/core/ITensor.h"
27 #include "arm_compute/core/TensorInfo.h"
28 #include "arm_compute/core/Types.h"
29 #include "arm_compute/core/Window.h"
30 #include "arm_compute/core/utils/helpers/tensor_transform.h"
31 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
32 #include "src/core/CPP/Validate.h"
33 #include "src/core/helpers/AutoConfiguration.h"
34 #include "src/core/helpers/WindowHelpers.h"
35 #include "src/core/utils/helpers/bit_ops.h"
36
37 namespace arm_compute
38 {
39 namespace
40 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,const Coordinates & starts,const Coordinates & ends,const BiStrides & strides,int32_t begin_mask,int32_t end_mask,int32_t shrink_axis_mask)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);
46 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
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
validate_and_configure_window(const ITensorInfo * input,ITensorInfo * output,const Coordinates & starts,const Coordinates & ends,const BiStrides & strides,int32_t begin_mask,int32_t end_mask,int32_t shrink_axis_mask)74 std::pair<Status, Window> validate_and_configure_window(const 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
85 Window win = calculate_max_window(*output, Steps());
86
87 return std::make_pair(Status{}, win);
88 }
89 } // namespace
90
NEStridedSliceKernel()91 NEStridedSliceKernel::NEStridedSliceKernel()
92 : _starts_abs(), _final_strides(), _shrink_mask()
93 {
94 }
95
configure(const ITensorInfo * input,ITensorInfo * output,const Coordinates & starts,const Coordinates & ends,const BiStrides & strides,int32_t begin_mask,int32_t end_mask,int32_t shrink_axis_mask)96 void NEStridedSliceKernel::configure(const ITensorInfo *input, ITensorInfo *output,
97 const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
98 int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
99 {
100 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
101 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
102 _shrink_mask = shrink_axis_mask;
103 const TensorShape &input_shape = input->tensor_shape();
104 Coordinates ends_abs;
105 std::tie(_starts_abs, ends_abs, _final_strides) = arm_compute::helpers::tensor_transform::calculate_strided_slice_coords(
106 input_shape,
107 starts, ends, strides,
108 begin_mask, end_mask, shrink_axis_mask);
109 // Configure kernel window
110 auto win_config = validate_and_configure_window(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
111 ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
112 INEKernel::configure(win_config.second);
113 }
114
validate(const ITensorInfo * input,const ITensorInfo * output,const Coordinates & starts,const Coordinates & ends,const BiStrides & strides,int32_t begin_mask,int32_t end_mask,int32_t shrink_axis_mask)115 Status NEStridedSliceKernel::validate(const ITensorInfo *input, const ITensorInfo *output,
116 const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
117 int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
118 {
119 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
120 ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(),
121 starts, ends, strides, begin_mask, end_mask, shrink_axis_mask)
122 .first);
123
124 return Status{};
125 }
126
run_op(ITensorPack & tensors,const Window & window,const ThreadInfo & info)127 void NEStridedSliceKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
128 {
129 ARM_COMPUTE_UNUSED(info);
130 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
131 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
132
133 const ITensor *input = tensors.get_const_tensor(TensorType::ACL_SRC_0);
134 const ITensor *output = tensors.get_tensor(TensorType::ACL_DST);
135
136 size_t width_size = input->info()->element_size();
137
138 const bool is_shrink_x = arm_compute::helpers::bit_ops::is_bit_set(_shrink_mask, 0);
139 const bool is_shrink_y = arm_compute::helpers::bit_ops::is_bit_set(_shrink_mask, 1);
140 const bool is_shrink_z = arm_compute::helpers::bit_ops::is_bit_set(_shrink_mask, 2);
141 const bool is_shrink_w = arm_compute::helpers::bit_ops::is_bit_set(_shrink_mask, 3);
142
143 unsigned int index = 0;
144 const int idx_x = is_shrink_x ? 0 : index++;
145 const int idx_y = is_shrink_y ? 0 : index++;
146 const int idx_z = is_shrink_z ? 0 : index++;
147 const int idx_w = is_shrink_w ? 0 : index;
148
149 BiStrides shrinked_strides;
150 shrinked_strides.set(0, is_shrink_x ? 0 : _final_strides[0]);
151 shrinked_strides.set(1, is_shrink_y ? 0 : _final_strides[1]);
152 shrinked_strides.set(2, is_shrink_z ? 0 : _final_strides[2]);
153 shrinked_strides.set(3, is_shrink_w ? 0 : _final_strides[3]);
154
155 Window win = window;
156
157 size_t length_x = win.shape()[0];
158
159 if(_final_strides[0] == 1 && !is_shrink_x)
160 {
161 win.set(Window::DimX, Window::Dimension(0, 1, 1));
162 width_size = width_size * length_x;
163 }
164
165 Iterator output_it(output, win);
166
167 const int start_0 = _starts_abs[0];
168 const int start_1 = _starts_abs[1];
169 const int start_2 = _starts_abs[2];
170 const int start_3 = _starts_abs[3];
171
172 const int shrinked_stride_0 = shrinked_strides[0];
173 const int shrinked_stride_1 = shrinked_strides[1];
174 const int shrinked_stride_2 = shrinked_strides[2];
175 const int shrinked_stride_3 = shrinked_strides[3];
176
177 const int byte_increment_0 = static_cast<int>(input->info()->strides_in_bytes()[0]);
178 const int byte_increment_1 = static_cast<int>(input->info()->strides_in_bytes()[1]);
179 const int byte_increment_2 = static_cast<int>(input->info()->strides_in_bytes()[2]);
180 const int byte_increment_3 = static_cast<int>(input->info()->strides_in_bytes()[3]);
181
182 uint8_t *input_base = input->ptr_to_element(Coordinates(0, 0, 0, 0));
183 uint8_t *cur_ptr;
184
185 execute_window_loop(
186 win, [&](const Coordinates & id)
187 {
188 cur_ptr = input_base;
189 cur_ptr += (start_0 + (id[idx_x] * shrinked_stride_0)) * byte_increment_0;
190 cur_ptr += (start_1 + (id[idx_y] * shrinked_stride_1)) * byte_increment_1;
191 cur_ptr += (start_2 + (id[idx_z] * shrinked_stride_2)) * byte_increment_2;
192 cur_ptr += (start_3 + (id[idx_w] * shrinked_stride_3)) * byte_increment_3;
193
194 std::copy_n(cur_ptr, width_size, output_it.ptr());
195 },
196 output_it);
197 }
198 } // namespace arm_compute
199