1 /*
2 * Copyright (c) 2020 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/CLMaxUnpoolingLayerKernel.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/CL/OpenCL.h"
30 #include "arm_compute/core/Helpers.h"
31 #include "arm_compute/core/TensorInfo.h"
32 #include "arm_compute/core/Utils.h"
33 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
34 #include "src/core/AccessWindowStatic.h"
35 #include "src/core/CL/CLValidate.h"
36 #include "src/core/helpers/AutoConfiguration.h"
37 #include "src/core/helpers/WindowHelpers.h"
38 #include "support/StringSupport.h"
39
40 namespace arm_compute
41 {
42 using namespace misc::shape_calculator;
43
44 namespace
45 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,const PoolingLayerInfo & pool_info,const ITensorInfo * indices)46 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
47 {
48 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, indices);
49 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
50 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
51 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32);
52 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, indices);
53
54 int pool_stride_x = 0;
55 int pool_stride_y = 0;
56 PoolingType pool_type = pool_info.pool_type;
57 const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
58 std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
59 const int pool_size_x = pool_info.pool_size.width;
60 const int pool_size_y = pool_info.pool_size.height;
61 const Size2D pool_size(pool_size_x, pool_size_y);
62
63 ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX, "Pooling indices only supported for MAX pooling method");
64 ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_size != Size2D(2, 2)), "Pooling indices only supported for pool size 2x2");
65 if(output->total_size() != 0)
66 {
67 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
68 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
69 }
70
71 return Status{};
72 }
73 } // namespace
74
CLMaxUnpoolingLayerKernel()75 CLMaxUnpoolingLayerKernel::CLMaxUnpoolingLayerKernel()
76 : _input(nullptr), _output(nullptr), _indices(nullptr)
77 {
78 }
79
configure(const CLCompileContext & compile_context,const ICLTensor * input,const ICLTensor * indices,ICLTensor * output,const PoolingLayerInfo & pool_info)80 void CLMaxUnpoolingLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *indices, ICLTensor *output, const PoolingLayerInfo &pool_info)
81 {
82 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
83 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), pool_info, indices->info()));
84
85 _input = input;
86 _output = output;
87 _indices = indices;
88
89 // Create build options
90 CLBuildOptions build_opts;
91 build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->info()->element_size()));
92 build_opts.add_option("-DWIDTH_DST=" + support::cpp11::to_string(output->info()->dimension(0)));
93 build_opts.add_option("-DHEIGHT_DST=" + support::cpp11::to_string(output->info()->dimension(1)));
94 build_opts.add_option("-DDEPTH_DST=" + support::cpp11::to_string(output->info()->dimension(2)));
95
96 const std::string kernel_name("max_unpooling_layer_2");
97
98 // Create kernel
99 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
100
101 const TensorShape output_shape = compute_unpool_shape(*input->info(), pool_info);
102 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
103
104 auto window = calculate_max_window(*input->info(), Steps());
105 ICLKernel::configure_internal(window);
106
107 // Set config_id for enabling LWS tuning
108 _config_id = kernel_name;
109 _config_id += lower_string(string_from_data_type(input->info()->data_type()));
110 _config_id += "_";
111 _config_id += support::cpp11::to_string(output->info()->dimension(0));
112 _config_id += "_";
113 _config_id += support::cpp11::to_string(output->info()->dimension(1));
114 _config_id += "_";
115 _config_id += support::cpp11::to_string(output->info()->dimension(2));
116 _config_id += "_";
117 _config_id += support::cpp11::to_string(output->info()->dimension(3));
118 }
119
validate(const ITensorInfo * input,const ITensorInfo * indices,const ITensorInfo * output,const PoolingLayerInfo & pool_info)120 Status CLMaxUnpoolingLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info)
121 {
122 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
123 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, pool_info, indices));
124 return Status{};
125 }
126
run(const Window & window,cl::CommandQueue & queue)127 void CLMaxUnpoolingLayerKernel::run(const Window &window, cl::CommandQueue &queue)
128 {
129 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
130 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
131
132 Window slice = window.first_slice_window_3D();
133
134 do
135 {
136 unsigned int idx = 0;
137 add_3D_tensor_argument(idx, _input, slice);
138 add_3D_tensor_argument(idx, _output, slice);
139 add_3D_tensor_argument(idx, _indices, slice);
140 enqueue(queue, *this, slice, lws_hint());
141 }
142 while(window.slide_window_slice_3D(slice));
143 }
144 } // namespace arm_compute
145