OpenCV453
block.hpp
[詳解]
1/*M///////////////////////////////////////////////////////////////////////////////////////
2//
3// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4//
5// By downloading, copying, installing or using the software you agree to this license.
6// If you do not agree to this license, do not download, install,
7// copy or use the software.
8//
9//
10// License Agreement
11// For Open Source Computer Vision Library
12//
13// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
14// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
15// Third party copyrights are property of their respective owners.
16//
17// Redistribution and use in source and binary forms, with or without modification,
18// are permitted provided that the following conditions are met:
19//
20// * Redistribution's of source code must retain the above copyright notice,
21// this list of conditions and the following disclaimer.
22//
23// * Redistribution's in binary form must reproduce the above copyright notice,
24// this list of conditions and the following disclaimer in the documentation
25// and/or other materials provided with the distribution.
26//
27// * The name of the copyright holders may not be used to endorse or promote products
28// derived from this software without specific prior written permission.
29//
30// This software is provided by the copyright holders and contributors "as is" and
31// any express or implied warranties, including, but not limited to, the implied
32// warranties of merchantability and fitness for a particular purpose are disclaimed.
33// In no event shall the Intel Corporation or contributors be liable for any direct,
34// indirect, incidental, special, exemplary, or consequential damages
35// (including, but not limited to, procurement of substitute goods or services;
36// loss of use, data, or profits; or business interruption) however caused
37// and on any theory of liability, whether in contract, strict liability,
38// or tort (including negligence or otherwise) arising in any way out of
39// the use of this software, even if advised of the possibility of such damage.
40//
41//M*/
42
43#ifndef OPENCV_CUDA_DEVICE_BLOCK_HPP
44#define OPENCV_CUDA_DEVICE_BLOCK_HPP
45
51
52namespace cv { namespace cuda { namespace device
53{
54 struct Block
55 {
56 static __device__ __forceinline__ unsigned int id()
57 {
58 return blockIdx.x;
59 }
60
61 static __device__ __forceinline__ unsigned int stride()
62 {
63 return blockDim.x * blockDim.y * blockDim.z;
64 }
65
66 static __device__ __forceinline__ void sync()
67 {
68 __syncthreads();
69 }
70
71 static __device__ __forceinline__ int flattenedThreadId()
72 {
73 return threadIdx.z * blockDim.x * blockDim.y + threadIdx.y * blockDim.x + threadIdx.x;
74 }
75
76 template<typename It, typename T>
77 static __device__ __forceinline__ void fill(It beg, It end, const T& value)
78 {
79 int STRIDE = stride();
80 It t = beg + flattenedThreadId();
81
82 for(; t < end; t += STRIDE)
83 *t = value;
84 }
85
86 template<typename OutIt, typename T>
87 static __device__ __forceinline__ void yota(OutIt beg, OutIt end, T value)
88 {
89 int STRIDE = stride();
90 int tid = flattenedThreadId();
91 value += tid;
92
93 for(OutIt t = beg + tid; t < end; t += STRIDE, value += STRIDE)
94 *t = value;
95 }
96
97 template<typename InIt, typename OutIt>
98 static __device__ __forceinline__ void copy(InIt beg, InIt end, OutIt out)
99 {
100 int STRIDE = stride();
101 InIt t = beg + flattenedThreadId();
102 OutIt o = out + (t - beg);
103
104 for(; t < end; t += STRIDE, o += STRIDE)
105 *o = *t;
106 }
107
108 template<typename InIt, typename OutIt, class UnOp>
109 static __device__ __forceinline__ void transform(InIt beg, InIt end, OutIt out, UnOp op)
110 {
111 int STRIDE = stride();
112 InIt t = beg + flattenedThreadId();
113 OutIt o = out + (t - beg);
114
115 for(; t < end; t += STRIDE, o += STRIDE)
116 *o = op(*t);
117 }
118
119 template<typename InIt1, typename InIt2, typename OutIt, class BinOp>
120 static __device__ __forceinline__ void transform(InIt1 beg1, InIt1 end1, InIt2 beg2, OutIt out, BinOp op)
121 {
122 int STRIDE = stride();
123 InIt1 t1 = beg1 + flattenedThreadId();
124 InIt2 t2 = beg2 + flattenedThreadId();
125 OutIt o = out + (t1 - beg1);
126
127 for(; t1 < end1; t1 += STRIDE, t2 += STRIDE, o += STRIDE)
128 *o = op(*t1, *t2);
129 }
130
131 template<int CTA_SIZE, typename T, class BinOp>
132 static __device__ __forceinline__ void reduce(volatile T* buffer, BinOp op)
133 {
134 int tid = flattenedThreadId();
135 T val = buffer[tid];
136
137 if (CTA_SIZE >= 1024) { if (tid < 512) buffer[tid] = val = op(val, buffer[tid + 512]); __syncthreads(); }
138 if (CTA_SIZE >= 512) { if (tid < 256) buffer[tid] = val = op(val, buffer[tid + 256]); __syncthreads(); }
139 if (CTA_SIZE >= 256) { if (tid < 128) buffer[tid] = val = op(val, buffer[tid + 128]); __syncthreads(); }
140 if (CTA_SIZE >= 128) { if (tid < 64) buffer[tid] = val = op(val, buffer[tid + 64]); __syncthreads(); }
141
142 if (tid < 32)
143 {
144 if (CTA_SIZE >= 64) { buffer[tid] = val = op(val, buffer[tid + 32]); }
145 if (CTA_SIZE >= 32) { buffer[tid] = val = op(val, buffer[tid + 16]); }
146 if (CTA_SIZE >= 16) { buffer[tid] = val = op(val, buffer[tid + 8]); }
147 if (CTA_SIZE >= 8) { buffer[tid] = val = op(val, buffer[tid + 4]); }
148 if (CTA_SIZE >= 4) { buffer[tid] = val = op(val, buffer[tid + 2]); }
149 if (CTA_SIZE >= 2) { buffer[tid] = val = op(val, buffer[tid + 1]); }
150 }
151 }
152
153 template<int CTA_SIZE, typename T, class BinOp>
154 static __device__ __forceinline__ T reduce(volatile T* buffer, T init, BinOp op)
155 {
156 int tid = flattenedThreadId();
157 T val = buffer[tid] = init;
158 __syncthreads();
159
160 if (CTA_SIZE >= 1024) { if (tid < 512) buffer[tid] = val = op(val, buffer[tid + 512]); __syncthreads(); }
161 if (CTA_SIZE >= 512) { if (tid < 256) buffer[tid] = val = op(val, buffer[tid + 256]); __syncthreads(); }
162 if (CTA_SIZE >= 256) { if (tid < 128) buffer[tid] = val = op(val, buffer[tid + 128]); __syncthreads(); }
163 if (CTA_SIZE >= 128) { if (tid < 64) buffer[tid] = val = op(val, buffer[tid + 64]); __syncthreads(); }
164
165 if (tid < 32)
166 {
167 if (CTA_SIZE >= 64) { buffer[tid] = val = op(val, buffer[tid + 32]); }
168 if (CTA_SIZE >= 32) { buffer[tid] = val = op(val, buffer[tid + 16]); }
169 if (CTA_SIZE >= 16) { buffer[tid] = val = op(val, buffer[tid + 8]); }
170 if (CTA_SIZE >= 8) { buffer[tid] = val = op(val, buffer[tid + 4]); }
171 if (CTA_SIZE >= 4) { buffer[tid] = val = op(val, buffer[tid + 2]); }
172 if (CTA_SIZE >= 2) { buffer[tid] = val = op(val, buffer[tid + 1]); }
173 }
174 __syncthreads();
175 return buffer[0];
176 }
177
178 template <typename T, class BinOp>
179 static __device__ __forceinline__ void reduce_n(T* data, unsigned int n, BinOp op)
180 {
181 int ftid = flattenedThreadId();
182 int sft = stride();
183
184 if (sft < n)
185 {
186 for (unsigned int i = sft + ftid; i < n; i += sft)
187 data[ftid] = op(data[ftid], data[i]);
188
189 __syncthreads();
190
191 n = sft;
192 }
193
194 while (n > 1)
195 {
196 unsigned int half = n/2;
197
198 if (ftid < half)
199 data[ftid] = op(data[ftid], data[n - ftid - 1]);
200
201 __syncthreads();
202
203 n = n - half;
204 }
205 }
206 };
207}}}
208
210
211#endif /* OPENCV_CUDA_DEVICE_BLOCK_HPP */
cv::reduce
CV_EXPORTS_W void reduce(InputArray src, OutputArray dst, int dim, int rtype, int dtype=-1)
Reduces a matrix to a vector.
cv::transform
CV_EXPORTS_W void transform(InputArray src, OutputArray dst, InputArray m)
Performs the matrix transformation of every array element.
cv
"black box" representation of the file storage associated with a file on disk.
Definition: aruco.hpp:75
構築: doxygen 1.9.2