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/NEON/kernels/NEMaxUnpoolingLayerKernel.h"
25
26 #include "arm_compute/core/TensorInfo.h"
27 #include "arm_compute/core/Validate.h"
28 #include "arm_compute/core/Window.h"
29 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
30 #include "src/core/CPP/Validate.h"
31 #include "src/core/helpers/AutoConfiguration.h"
32 #include "src/core/helpers/WindowHelpers.h"
33
34 #include "support/ToolchainSupport.h"
35
36 namespace arm_compute
37 {
38 using namespace misc::shape_calculator;
39
40 namespace
41 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,const PoolingLayerInfo & pool_info,const ITensorInfo * indices)42 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
43 {
44 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, indices);
45 ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
46 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
47 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32);
48 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, indices);
49
50 int pool_stride_x = 0;
51 int pool_stride_y = 0;
52 PoolingType pool_type = pool_info.pool_type;
53 const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
54 std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
55 const int pool_size_x = pool_info.pool_size.width;
56 const int pool_size_y = pool_info.pool_size.height;
57 const Size2D pool_size(pool_size_x, pool_size_y);
58
59 ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX, "Pooling indices only supported for MAX pooling method");
60 ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_size != Size2D(2, 2)), "Pooling indices only supported for pool size 2x2");
61 if(output->total_size() != 0)
62 {
63 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
64 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
65 }
66
67 return Status{};
68 }
69 } // namespace
70
NEMaxUnpoolingLayerKernel()71 NEMaxUnpoolingLayerKernel::NEMaxUnpoolingLayerKernel()
72 : _func(nullptr), _input(nullptr), _output(nullptr), _indices(nullptr)
73 {
74 }
75
configure(const ITensor * input,const ITensor * indices,ITensor * output,const PoolingLayerInfo & pool_info)76 void NEMaxUnpoolingLayerKernel::configure(const ITensor *input, const ITensor *indices, ITensor *output, const PoolingLayerInfo &pool_info)
77 {
78 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
79 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), pool_info, indices->info()));
80
81 _input = input;
82 _output = output;
83 _indices = indices;
84
85 switch(input->info()->data_type())
86 {
87 case DataType::F32:
88 _func = &NEMaxUnpoolingLayerKernel::unpooling2<float>;
89 break;
90 case DataType::QASYMM8:
91 _func = &NEMaxUnpoolingLayerKernel::unpooling2<uint8_t>;
92 break;
93 case DataType::QASYMM8_SIGNED:
94 _func = &NEMaxUnpoolingLayerKernel::unpooling2<int8_t>;
95 break;
96 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
97 case DataType::F16:
98 _func = &NEMaxUnpoolingLayerKernel::unpooling2<float16_t>;
99 break;
100 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
101 default:
102 break;
103 }
104 const TensorShape output_shape = compute_unpool_shape(*input->info(), pool_info);
105 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
106
107 auto window = calculate_max_window(*input->info(), Steps());
108 INEKernel::configure(window);
109 }
110 template <typename T>
unpooling2(const Window & window)111 void NEMaxUnpoolingLayerKernel::unpooling2(const Window &window)
112 {
113 Iterator input(_input, window);
114 Iterator indices(_indices, window);
115 auto out_ptr = reinterpret_cast<T *>(_output->buffer());
116 const int out_stride_w = static_cast<int>(_output->info()->strides_in_bytes()[3]);
117 execute_window_loop(window, [&](const Coordinates & id)
118 {
119 auto vindices = reinterpret_cast<uint32_t *>(indices.ptr());
120 auto vinput = reinterpret_cast<T *>(input.ptr());
121 out_ptr[id[3] * out_stride_w / sizeof(T) + *vindices] = *vinput;
122 },
123 input, indices);
124 }
125
validate(const ITensorInfo * input,const ITensorInfo * indices,const ITensorInfo * output,const PoolingLayerInfo & pool_info)126 Status NEMaxUnpoolingLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info)
127 {
128 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
129 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, pool_info, indices));
130 return Status{};
131 }
132
run(const Window & window,const ThreadInfo & info)133 void NEMaxUnpoolingLayerKernel::run(const Window &window, const ThreadInfo &info)
134 {
135 ARM_COMPUTE_UNUSED(info);
136 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
137 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
138 ARM_COMPUTE_ERROR_ON(_func == nullptr);
139 // Run function
140 (this->*_func)(window);
141 }
142 } // namespace arm_compute
143