1 /*
2 * Copyright (c) 2018-2021 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/CL/kernels/CLSelectKernel.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/TensorInfo.h"
31 #include "arm_compute/core/Utils.h"
32 #include "src/core/CL/CLValidate.h"
33 #include "src/core/helpers/AutoConfiguration.h"
34 #include "src/core/helpers/WindowHelpers.h"
35
36 #include "support/StringSupport.h"
37
38 namespace arm_compute
39 {
40 namespace
41 {
validate_arguments(const ITensorInfo * c,const ITensorInfo * x,const ITensorInfo * y,const ITensorInfo * output)42 Status validate_arguments(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
43 {
44 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(c, x, y, output);
45 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(x);
46 ARM_COMPUTE_RETURN_ERROR_ON(x->data_type() == DataType::UNKNOWN);
47 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(x, y);
48 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(x, y);
49 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(c, 1, DataType::U8);
50
51 const bool is_same_rank = (c->tensor_shape().num_dimensions() == x->tensor_shape().num_dimensions());
52 ARM_COMPUTE_RETURN_ERROR_ON(is_same_rank && (x->tensor_shape() != c->tensor_shape()));
53 ARM_COMPUTE_RETURN_ERROR_ON(!is_same_rank && ((c->tensor_shape().num_dimensions() > 1) || (c->tensor_shape().x() != x->tensor_shape()[x->tensor_shape().num_dimensions() - 1])));
54
55 if(output->total_size() != 0)
56 {
57 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(x, output);
58 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(x, output);
59 }
60
61 return Status{};
62 }
63 } // namespace
64
CLSelectKernel()65 CLSelectKernel::CLSelectKernel()
66 : _c(nullptr), _x(nullptr), _y(nullptr), _output(nullptr), _has_same_rank(false)
67 {
68 _type = CLKernelType::ELEMENTWISE;
69 }
70
configure(const CLCompileContext & compile_context,const ICLTensor * c,const ICLTensor * x,const ICLTensor * y,ICLTensor * output)71 void CLSelectKernel::configure(const CLCompileContext &compile_context, const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output)
72 {
73 ARM_COMPUTE_ERROR_ON_NULLPTR(c, x, y, output);
74 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(c->info(), x->info(), y->info(), output->info()));
75
76 _c = c;
77 _x = x;
78 _y = y;
79 _output = output;
80 _has_same_rank = (c->info()->tensor_shape().num_dimensions() == x->info()->tensor_shape().num_dimensions());
81
82 auto padding_info = get_padding_info({ c, x, y, output });
83 const unsigned int vec_size_x = adjust_vec_size(16 / x->info()->element_size(), x->info()->dimension(0));
84 const int vec_size_x_leftovers = output->info()->dimension(0) % vec_size_x;
85
86 // Set build options
87 CLBuildOptions build_opts;
88 build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(x->info()->element_size()));
89 build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
90 build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftovers));
91
92 // Create kernel
93 std::string kernel_name = "select";
94 if(_has_same_rank)
95 {
96 kernel_name += "_same_rank";
97 }
98 else
99 {
100 const bool is_input_rank_greater_than_two = x->info()->tensor_shape().num_dimensions() > 2;
101 if(is_input_rank_greater_than_two)
102 {
103 const size_t width = x->info()->tensor_shape().x();
104 const size_t height = x->info()->tensor_shape().y();
105 const size_t outer_size = x->info()->tensor_shape()[x->info()->tensor_shape().num_dimensions() - 1];
106 const size_t depth_size = x->info()->tensor_shape().total_size() / (width * height * outer_size);
107 build_opts.add_option("-DDEPTH_SIZE=" + support::cpp11::to_string(depth_size));
108 }
109 kernel_name += "_different_rank";
110 kernel_name += is_input_rank_greater_than_two ? "_n" : "_2";
111 }
112 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
113
114 // Configure kernel window
115 auto_init_if_empty(*output->info(), *x->info()->clone());
116 Window win = calculate_max_window(*x->info(), Steps(vec_size_x));
117 ICLKernel::configure_internal(win);
118
119 _config_id = "select_";
120 _config_id += string_from_data_type(x->info()->data_type());
121 _config_id += "_";
122 _config_id += support::cpp11::to_string(x->info()->dimension(0));
123 _config_id += "_";
124 _config_id += support::cpp11::to_string(x->info()->dimension(1));
125 _config_id += "_";
126 _config_id += support::cpp11::to_string(x->info()->dimension(2));
127 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
128 }
129
validate(const ITensorInfo * c,const ITensorInfo * x,const ITensorInfo * y,const ITensorInfo * output)130 Status CLSelectKernel::validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
131 {
132 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(c, x, y, output));
133 return Status{};
134 }
135
run(const arm_compute::Window & window,cl::CommandQueue & queue)136 void CLSelectKernel::run(const arm_compute::Window &window, cl::CommandQueue &queue)
137 {
138 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
139 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
140
141 Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
142 Window slice = collapsed.first_slice_window_3D();
143
144 if(!_has_same_rank)
145 {
146 Window vector_slice = window.first_slice_window_1D();
147 vector_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
148 unsigned int idx = 0;
149 add_1D_tensor_argument(idx, _c, vector_slice);
150 }
151
152 do
153 {
154 unsigned int idx = _has_same_rank ? 0 : num_arguments_per_1D_tensor();
155 if(_has_same_rank)
156 {
157 add_3D_tensor_argument(idx, _c, slice);
158 }
159 add_3D_tensor_argument(idx, _x, slice);
160 add_3D_tensor_argument(idx, _y, slice);
161 add_3D_tensor_argument(idx, _output, slice);
162
163 enqueue(queue, *this, slice, lws_hint());
164 }
165 while(collapsed.slide_window_slice_3D(slice));
166 }
167 } // namespace arm_compute
168