-
Notifications
You must be signed in to change notification settings - Fork 11
/
SparseMatrix.cu
136 lines (114 loc) · 4.36 KB
/
SparseMatrix.cu
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
#include "SparseMatrix.h"
#include <cusparse_v2.h>
#include <helper_cuda.h>
#include <thrust/sort.h>
#include <thrust/execution_policy.h>
#include <device_launch_parameters.h>
#include <thrust/device_ptr.h>
__global__ void AddVecVecKernel(float* vec1, float* vec2, int length)
{
int idx = blockDim.x * blockIdx.x + threadIdx.x;
if (idx < length)
{
vec1[idx] += vec2[idx];
}
}
void AddVecVec(float* vec1, float* vec2, int length)
{
int block = 512, grid = (block + length - 1) / block;
AddVecVecKernel << <grid, block >> >(vec1, vec2, length);
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaGetLastError());
}
__global__ void AddVecVecSE3Kernel(float* vec1, float* vec2, int length, bool doOrtho)
{
int idx = blockDim.x * blockIdx.x + threadIdx.x;
if (idx >= length)
{
return;
}
double rx, ry, rz, tx, ty, tz, theta;
rx = vec2[6 * idx];
ry = vec2[6 * idx + 1];
rz = vec2[6 * idx + 2];
tx = vec2[6 * idx + 3];
ty = vec2[6 * idx + 4];
tz = vec2[6 * idx + 5];
theta = sqrt(rx * rx + ry * ry + rz * rz);
float* vars_ptr = vec1 + 12 * idx;
//printf("theta: %f %f %f %f, ", theta, rx, ry, rz);
if (theta >= DBL_EPSILON)
{
const double I[] = {1, 0, 0, 0, 1, 0, 0, 0, 1};
float vars_tmp[9];
vars_tmp[0] = vars_ptr[0];
vars_tmp[1] = vars_ptr[1];
vars_tmp[2] = vars_ptr[2];
vars_tmp[3] = vars_ptr[3];
vars_tmp[4] = vars_ptr[4];
vars_tmp[5] = vars_ptr[5];
vars_tmp[6] = vars_ptr[6];
vars_tmp[7] = vars_ptr[7];
vars_tmp[8] = vars_ptr[8];
double c = cos(theta);
double s = sin(theta);
double c1 = 1. - c;
double itheta = theta ? 1. / theta : 0.;
rx *= itheta;
ry *= itheta;
rz *= itheta;
double rrt[] = {rx * rx, rx * ry, rx * rz, rx * ry, ry * ry, ry * rz, rx * rz, ry * rz, rz * rz};
double _r_x_[] = {0, -rz, ry, rz, 0, -rx, -ry, rx, 0};
double R[9];
for (int k = 0; k < 9; k++)
{
R[k] = c * I[k] + c1 * rrt[k] + s * _r_x_[k];
}
vars_ptr[0] = R[0] * vars_tmp[0] + R[1] * vars_tmp[3] + R[2] * vars_tmp[6];
vars_ptr[1] = R[0] * vars_tmp[1] + R[1] * vars_tmp[4] + R[2] * vars_tmp[7];
vars_ptr[2] = R[0] * vars_tmp[2] + R[1] * vars_tmp[5] + R[2] * vars_tmp[8];
vars_ptr[3] = R[3] * vars_tmp[0] + R[4] * vars_tmp[3] + R[5] * vars_tmp[6];
vars_ptr[4] = R[3] * vars_tmp[1] + R[4] * vars_tmp[4] + R[5] * vars_tmp[7];
vars_ptr[5] = R[3] * vars_tmp[2] + R[4] * vars_tmp[5] + R[5] * vars_tmp[8];
vars_ptr[6] = R[6] * vars_tmp[0] + R[7] * vars_tmp[3] + R[8] * vars_tmp[6];
vars_ptr[7] = R[6] * vars_tmp[1] + R[7] * vars_tmp[4] + R[8] * vars_tmp[7];
vars_ptr[8] = R[6] * vars_tmp[2] + R[7] * vars_tmp[5] + R[8] * vars_tmp[8];
if (doOrtho)
{
double factor1, factor2, norm1, norm2;
norm1 = vars_ptr[0] * vars_ptr[0] + vars_ptr[1] * vars_ptr[1] + vars_ptr[2] * vars_ptr[2];
factor1 = (vars_ptr[3] * vars_ptr[0] + vars_ptr[4] * vars_ptr[1] + vars_ptr[5] * vars_ptr[2]) / norm1;
vars_ptr[3] = vars_ptr[3] - factor1 * vars_ptr[0];
vars_ptr[4] = vars_ptr[4] - factor1 * vars_ptr[1];
vars_ptr[5] = vars_ptr[5] - factor1 * vars_ptr[2];
norm2 = vars_ptr[3] * vars_ptr[3] + vars_ptr[4] * vars_ptr[4] + vars_ptr[5] * vars_ptr[5];
factor1 = (vars_ptr[6] * vars_ptr[0] + vars_ptr[7] * vars_ptr[1] + vars_ptr[8] * vars_ptr[2]) / norm1;
factor2 = (vars_ptr[6] * vars_ptr[3] + vars_ptr[7] * vars_ptr[4] + vars_ptr[8] * vars_ptr[5]) / norm2;
vars_ptr[6] = vars_ptr[6] - factor1 * vars_ptr[0] - factor2 * vars_ptr[3];
vars_ptr[7] = vars_ptr[7] - factor1 * vars_ptr[1] - factor2 * vars_ptr[4];;
vars_ptr[8] = vars_ptr[8] - factor1 * vars_ptr[2] - factor2 * vars_ptr[5];;
norm1 = sqrt(vars_ptr[0] * vars_ptr[0] + vars_ptr[1] * vars_ptr[1] + vars_ptr[2] * vars_ptr[2]);
vars_ptr[0] /= norm1;
vars_ptr[1] /= norm1;
vars_ptr[2] /= norm1;
norm1 = sqrt(vars_ptr[3] * vars_ptr[3] + vars_ptr[4] * vars_ptr[4] + vars_ptr[5] * vars_ptr[5]);
vars_ptr[3] /= norm1;
vars_ptr[4] /= norm1;
vars_ptr[5] /= norm1;
norm1 = sqrt(vars_ptr[6] * vars_ptr[6] + vars_ptr[7] * vars_ptr[7] + vars_ptr[8] * vars_ptr[8]);
vars_ptr[6] /= norm1;
vars_ptr[7] /= norm1;
vars_ptr[8] /= norm1;
}
}
vars_ptr[9] += tx;
vars_ptr[10] += ty;
vars_ptr[11] += tz;
}
void AddVecVecSE3(float* dVec1, float* dVec2, int length)
{
int block = 256, grid = (block + length - 1) / block;
AddVecVecSE3Kernel << <grid, block >> >(dVec1, dVec2, length, false);
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaGetLastError());
}