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
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
|
#include "test_precomp.hpp"
using namespace cv;
using namespace cv::cudev;
using namespace cvtest;
// BlockScanInt
template <int THREADS_NUM>
__global__ void int_kernel(int* data)
{
uint tid = Block::threadLineId();
#if CV_CUDEV_ARCH >= 300
const int n_warps = (THREADS_NUM - 1) / WARP_SIZE + 1;
__shared__ int smem[n_warps];
#else
__shared__ int smem[THREADS_NUM];
#endif
data[tid] = blockScanInclusive<THREADS_NUM>(data[tid], smem, tid);
}
#define BLOCK_SCAN_INT_TEST(block_size) \
TEST(BlockScanInt, BlockSize##block_size) \
{ \
Mat src = randomMat(Size(block_size, 1), CV_32SC1, 0, 1024); \
\
GpuMat d_src; \
d_src.upload(src); \
\
for (int col = 1; col < block_size; col++) \
src.at<int>(0, col) += src.at<int>(0, col - 1); \
\
int_kernel<block_size><<<1, block_size>>>((int*)d_src.data); \
\
CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize()); \
\
EXPECT_MAT_NEAR(d_src, src, 0); \
}
BLOCK_SCAN_INT_TEST(29)
BLOCK_SCAN_INT_TEST(30)
BLOCK_SCAN_INT_TEST(32)
BLOCK_SCAN_INT_TEST(40)
BLOCK_SCAN_INT_TEST(41)
BLOCK_SCAN_INT_TEST(59)
BLOCK_SCAN_INT_TEST(60)
BLOCK_SCAN_INT_TEST(64)
BLOCK_SCAN_INT_TEST(70)
BLOCK_SCAN_INT_TEST(71)
BLOCK_SCAN_INT_TEST(109)
BLOCK_SCAN_INT_TEST(110)
BLOCK_SCAN_INT_TEST(128)
BLOCK_SCAN_INT_TEST(130)
BLOCK_SCAN_INT_TEST(131)
BLOCK_SCAN_INT_TEST(189)
BLOCK_SCAN_INT_TEST(200)
BLOCK_SCAN_INT_TEST(256)
BLOCK_SCAN_INT_TEST(300)
BLOCK_SCAN_INT_TEST(311)
BLOCK_SCAN_INT_TEST(489)
BLOCK_SCAN_INT_TEST(500)
BLOCK_SCAN_INT_TEST(512)
BLOCK_SCAN_INT_TEST(600)
BLOCK_SCAN_INT_TEST(611)
BLOCK_SCAN_INT_TEST(1024)
// BlockScanDouble
template <int THREADS_NUM>
__global__ void double_kernel(double* data)
{
uint tid = Block::threadLineId();
#if CV_CUDEV_ARCH >= 300
const int n_warps = (THREADS_NUM - 1) / WARP_SIZE + 1;
__shared__ double smem[n_warps];
#else
__shared__ double smem[THREADS_NUM];
#endif
data[tid] = blockScanInclusive<THREADS_NUM>(data[tid], smem, tid);
}
#define BLOCK_SCAN_DOUBLE_TEST(block_size) \
TEST(BlockScanDouble, BlockSize##block_size) \
{ \
Mat src = randomMat(Size(block_size, 1), CV_64FC1, 0.0, 1.0); \
\
GpuMat d_src; \
d_src.upload(src); \
\
for (int col = 1; col < block_size; col++) \
src.at<double>(0, col) += src.at<double>(0, col - 1); \
\
double_kernel<block_size><<<1, block_size>>>((double*)d_src.data); \
\
CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize()); \
\
EXPECT_MAT_NEAR(d_src, src, 1e-10); \
}
BLOCK_SCAN_DOUBLE_TEST(29)
BLOCK_SCAN_DOUBLE_TEST(30)
BLOCK_SCAN_DOUBLE_TEST(32)
BLOCK_SCAN_DOUBLE_TEST(40)
BLOCK_SCAN_DOUBLE_TEST(41)
BLOCK_SCAN_DOUBLE_TEST(59)
BLOCK_SCAN_DOUBLE_TEST(60)
BLOCK_SCAN_DOUBLE_TEST(64)
BLOCK_SCAN_DOUBLE_TEST(70)
BLOCK_SCAN_DOUBLE_TEST(71)
BLOCK_SCAN_DOUBLE_TEST(109)
BLOCK_SCAN_DOUBLE_TEST(110)
BLOCK_SCAN_DOUBLE_TEST(128)
BLOCK_SCAN_DOUBLE_TEST(130)
BLOCK_SCAN_DOUBLE_TEST(131)
BLOCK_SCAN_DOUBLE_TEST(189)
BLOCK_SCAN_DOUBLE_TEST(200)
BLOCK_SCAN_DOUBLE_TEST(256)
BLOCK_SCAN_DOUBLE_TEST(300)
BLOCK_SCAN_DOUBLE_TEST(311)
BLOCK_SCAN_DOUBLE_TEST(489)
BLOCK_SCAN_DOUBLE_TEST(500)
BLOCK_SCAN_DOUBLE_TEST(512)
BLOCK_SCAN_DOUBLE_TEST(600)
BLOCK_SCAN_DOUBLE_TEST(611)
BLOCK_SCAN_DOUBLE_TEST(1024)
|
