-
Notifications
You must be signed in to change notification settings - Fork 11
/
RotConstraint.cu
190 lines (169 loc) · 5.21 KB
/
RotConstraint.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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
#include "RotConstraint.h"
#include <helper_cuda.h>
#include <device_launch_parameters.h>
#include "InputData.h"
#include "GNSolver.h"
#include "Helpers/UtilsMath.h"
bool RotConstraint::init(GNSolver* gnSolver, float weight)
{
assert(gnSolver);
m_gnSolver = gnSolver;
m_inputData = gnSolver->m_inputData;
setWeight(weight);
return true;
}
bool RotConstraint::init()
{
return true;
}
__global__ void bKernelRot(float* dB,
int nodeNum,
float3* dVars,
float weight)
{
int idx = blockDim.x * blockIdx.x + threadIdx.x;
if (idx >= nodeNum)
{
return;
}
int node_seri;
node_seri = idx * 12;
int currentIdx = idx * 6;
float3* R = dVars + 4 * idx;
dB[currentIdx + 0] = weight * dot(R[0], R[1]);
dB[currentIdx + 1] = weight * dot(R[0], R[2]);
dB[currentIdx + 2] = weight * dot(R[1], R[2]);
dB[currentIdx + 3] = weight * (dot(R[0], R[0]) - 1);
dB[currentIdx + 4] = weight * (dot(R[1], R[1]) - 1);
dB[currentIdx + 5] = weight * (dot(R[2], R[2]) - 1);
}
void RotConstraint::b(float3* dVars)
{
int nodeNum = m_inputData->m_source.m_nodeNum;
m_dB.resize(nodeNum * 6);
int block = 512;
int grid = (block + nodeNum - 1) / block;
bKernelRot << <grid, block >> >(RAW_PTR(m_dB),
nodeNum,
dVars,
m_weight);
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaGetLastError());
}
__global__ void DirectiveJTJAndJTbKernelRot(float* dJTJ_a,
int* dJTJ_ia,
float* dJTb,
float* dB,
int* dNnzPre,
int nodeNum,
float* dVars,
float weight)
{
__shared__ float s_res[81];
__shared__ float s_x[12];
__shared__ float s_J[9 * 6];
__shared__ float s_JTb[12];
if (threadIdx.x < 81) {
s_res[threadIdx.x] = 0.0f;
}
if (threadIdx.x < 9 * 6) {
s_J[threadIdx.x] = 0.0f;
}
if (threadIdx.x < 12) {
s_x[threadIdx.x] = dVars[blockIdx.x * 12 + threadIdx.x];
s_JTb[threadIdx.x] = 0.0f;
}
__syncthreads();
if (threadIdx.x == 0) {
// c1c2
s_J[9 * 0 + 0] = s_x[3];
s_J[9 * 0 + 1] = s_x[4];
s_J[9 * 0 + 2] = s_x[5];
s_J[9 * 0 + 3] = s_x[0];
s_J[9 * 0 + 4] = s_x[1];
s_J[9 * 0 + 5] = s_x[2];
// c1c3
s_J[9 * 1 + 0] = s_x[6];
s_J[9 * 1 + 1] = s_x[7];
s_J[9 * 1 + 2] = s_x[8];
s_J[9 * 1 + 6] = s_x[0];
s_J[9 * 1 + 7] = s_x[1];
s_J[9 * 1 + 8] = s_x[2];
// c2c3
s_J[9 * 2 + 3] = s_x[6];
s_J[9 * 2 + 4] = s_x[7];
s_J[9 * 2 + 5] = s_x[8];
s_J[9 * 2 + 6] = s_x[3];
s_J[9 * 2 + 7] = s_x[4];
s_J[9 * 2 + 8] = s_x[5];
// c1c1
s_J[9 * 3 + 0] = 2 * s_x[0];
s_J[9 * 3 + 1] = 2 * s_x[1];
s_J[9 * 3 + 2] = 2 * s_x[2];
// c2c2
s_J[9 * 4 + 3] = 2 * s_x[3];
s_J[9 * 4 + 4] = 2 * s_x[4];
s_J[9 * 4 + 5] = 2 * s_x[5];
// c3c3
s_J[9 * 5 + 6] = 2 * s_x[6];
s_J[9 * 5 + 7] = 2 * s_x[7];
s_J[9 * 5 + 8] = 2 * s_x[8];
}
__syncthreads();
int row_res = threadIdx.x / 9;
int col_res = threadIdx.x % 9;
// reduction
if (threadIdx.x < 81) {
float squ_weight = weight * weight;
#pragma unroll
for (int iter = 0; iter < 6; iter++) {
s_res[threadIdx.x] += s_J[iter * 9 + row_res] * s_J[iter * 9 + col_res] * squ_weight;
}
}
__syncthreads();
// write back
int start_pos;
if (threadIdx.x < 81) {
start_pos = dNnzPre[blockIdx.x * nodeNum + blockIdx.x] * 12;
dJTJ_a[dJTJ_ia[blockIdx.x * 12 + row_res] + start_pos + col_res] += s_res[row_res * 9 + col_res];
}
// calculate JTb
__syncthreads();
if (threadIdx.x < 9) {
start_pos = blockIdx.x * 6;
#pragma unroll
for (int iter = 0; iter < 6; iter++) {
s_JTb[threadIdx.x] += s_J[iter * 9 + threadIdx.x] * dB[start_pos + iter] * weight;
}
}
__syncthreads();
// write back
if (threadIdx.x < 9) {
int save_start_pos = blockIdx.x * 12;
dJTb[save_start_pos + threadIdx.x] -= s_JTb[threadIdx.x];
}
}
void RotConstraint::getJTJAndJTb(float* dJTJ_a, int* dJTJ_ia, thrust::device_vector<float>& dJTb, float3* dVars)
{
b(dVars);
int nodeNum = m_inputData->m_source.m_nodeNum;
int block = 96;
int grid = nodeNum;
DirectiveJTJAndJTbKernelRot << <grid, block >> >(dJTJ_a,
dJTJ_ia,
RAW_PTR(dJTb),
RAW_PTR(m_dB),
RAW_PTR(m_inputData->m_Iij.m_dNnzPre),
nodeNum,
reinterpret_cast<float*>(dVars),
m_weight);
checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaGetLastError());
}
void RotConstraint::directiveJTJ(float* JTJ_a,
int* JTJ_ia)
{
}
void RotConstraint::directiveJTb(thrust::device_vector<float> &JTb)
{
}