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/CLStackLayerKernel.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 "arm_compute/core/utils/misc/ShapeCalculator.h"
33 #include "src/core/CL/CLValidate.h"
34 #include "src/core/helpers/AutoConfiguration.h"
35 #include "src/core/helpers/WindowHelpers.h"
36
37 #include "support/StringSupport.h"
38
39 using namespace arm_compute::misc::shape_calculator;
40
41 namespace arm_compute
42 {
43 namespace
44 {
validate_arguments(const ITensorInfo * input,unsigned int axis,unsigned int idx_input,unsigned int num_tensors,const ITensorInfo * output)45 Status validate_arguments(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
46 {
47 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
48 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
49 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
50 ARM_COMPUTE_RETURN_ERROR_ON(idx_input >= num_tensors);
51 ARM_COMPUTE_RETURN_ERROR_ON(axis > input->num_dimensions());
52 ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
53
54 if(output->total_size() != 0)
55 {
56 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_stack_shape(*input, axis, num_tensors));
57 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
58 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
59 }
60
61 return Status{};
62 }
63
validate_and_configure_window(ITensorInfo * input,unsigned int axis,unsigned int num_tensors,ITensorInfo * output)64 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, unsigned int axis, unsigned int num_tensors, ITensorInfo *output)
65 {
66 // Output auto inizialitation if not yet initialized
67 auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_stack_shape(*input, axis, num_tensors)));
68
69 // Configure kernel window
70 Window win = calculate_max_window(*input);
71
72 return std::make_pair(Status{}, win);
73 }
74 } // namespace
75
CLStackLayerKernel()76 CLStackLayerKernel::CLStackLayerKernel()
77 : _input(nullptr), _output(nullptr)
78 {
79 _type = CLKernelType::ELEMENTWISE;
80 }
81
configure(const ICLTensor * input,unsigned int axis,unsigned int idx_input,unsigned int num_tensors,ICLTensor * output)82 void CLStackLayerKernel::configure(const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
83 {
84 configure(CLKernelLibrary::get().get_compile_context(), input, axis, idx_input, num_tensors, output);
85 }
86
configure(const CLCompileContext & compile_context,const ICLTensor * input,unsigned int axis,unsigned int idx_input,unsigned int num_tensors,ICLTensor * output)87 void CLStackLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
88 {
89 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
90 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), axis, idx_input, num_tensors, output->info()));
91
92 _input = input;
93 _output = output;
94
95 // Configure kernel window
96 auto win_config = validate_and_configure_window(input->info(), axis, num_tensors, output->info());
97
98 // Add 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("-DAXIS=" + support::cpp11::to_string(axis));
102 build_opts.add_option("-DSRC_DIM2=" + support::cpp11::to_string(input->info()->dimension(2)));
103 build_opts.add_option("-DDST_DIM3=" + support::cpp11::to_string(output->info()->dimension(3)));
104
105 // Create kernel
106 _kernel = create_kernel(compile_context, "stack_layer", build_opts.options());
107
108 ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
109 ICLKernel::configure_internal(win_config.second);
110
111 const unsigned int idx = 2 * num_arguments_per_4D_tensor();
112 _kernel.setArg<cl_uint>(idx, idx_input);
113 }
114
validate(const ITensorInfo * input,unsigned int axis,unsigned int idx_input,unsigned int num_tensors,const ITensorInfo * output)115 Status CLStackLayerKernel::validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
116 {
117 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, axis, idx_input, num_tensors, output));
118 ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), axis, num_tensors, output->clone().get()).first);
119 return Status{};
120 }
121
run(const Window & window,cl::CommandQueue & queue)122 void CLStackLayerKernel::run(const Window &window, cl::CommandQueue &queue)
123 {
124 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
125 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
126
127 Window window_out;
128 window_out.use_tensor_dimensions(_output->info()->tensor_shape());
129
130 Window collapsed = window.collapse(ICLKernel::window(), Window::DimZ);
131
132 Window slice_in = collapsed.first_slice_window_4D();
133 Window slice_out = window_out.first_slice_window_4D();
134
135 unsigned int idx = 0;
136 add_4D_tensor_argument(idx, _input, slice_in);
137 add_4D_tensor_argument(idx, _output, slice_out);
138 enqueue(queue, *this, slice_in, lws_hint());
139 }
140 } // namespace arm_compute
141