f4334277
 Hu Chunming
提交3rdparty
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
|
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#ifndef OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_CONCAT_HPP
#define OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_CONCAT_HPP
#include "../../op_cuda.hpp"
#include "../csl/stream.hpp"
#include "../csl/pointer.hpp"
#include "../kernels/fill_copy.hpp"
#include "../kernels/concat.hpp"
#include <opencv2/core.hpp>
#include <cstddef>
#include <vector>
#include <utility>
namespace cv { namespace dnn { namespace cuda4dnn {
template <class T>
class ConcatOp final : public CUDABackendNode {
public:
using wrapper_type = GetCUDABackendWrapperType<T>;
ConcatOp(csl::Stream stream_, std::size_t concat_axis, bool zero_padding)
: stream(std::move(stream_)), concat_axis{ concat_axis }, zero_padding{ zero_padding }
{
}
void forward(
const std::vector<cv::Ptr<BackendWrapper>>& inputs,
const std::vector<cv::Ptr<BackendWrapper>>& outputs,
csl::Workspace& workspace) override
{
CV_Assert(outputs.size() == 1);
auto output_wrapper = outputs[0].dynamicCast<wrapper_type>();
auto output = output_wrapper->getSpan();
if(zero_padding)
{
auto output_shape = output_wrapper->getShape();
kernels::fill<T>(stream, output, 0.0);
std::size_t output_concat_axis_offset = 0;
for (int i = 0; i < inputs.size(); i++)
{
auto input_wrapper = inputs[i].dynamicCast<wrapper_type>();
auto input = input_wrapper->getView();
auto input_shape = input_wrapper->getShape();
std::vector<std::size_t> offsets(input_shape.size());
for (int j = 0; j < offsets.size(); j++)
offsets[j] = (output_shape[j] - input_shape[j]) / 2;
offsets[concat_axis] = output_concat_axis_offset;
kernels::concat_with_offsets(stream, output, input, offsets);
output_concat_axis_offset += input.get_axis_size(concat_axis);
}
}
else
{
std::size_t output_axis_offset = 0;
for (int i = 0; i < inputs.size(); i++)
{
auto input_wrapper = inputs[i].dynamicCast<wrapper_type>();
auto input = input_wrapper->getView();
kernels::concat(stream, output, output_axis_offset, input, concat_axis);
output_axis_offset += input.get_axis_size(concat_axis);
}
}
}
private:
csl::Stream stream;
std::size_t concat_axis;
bool zero_padding;
};
}}} /* namespace cv::dnn::cuda4dnn */
#endif /* OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_CONCAT_HPP */
|
