File size: 5,989 Bytes
78d2329
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
191
#include "../cuda_utils.h"
#include "ball_query_cuda_kernel.h"


namespace ball_query_utils{

template <typename DType>
__device__ void swap(DType *x, DType *y)
{
    DType tmp = *x;
    *x = *y;
    *y = tmp;
}

__device__ void reheap(float *dist, int *idx, int k)
{
    int root = 0;
    int child = root * 2 + 1;
    while (child < k)
    {
        if(child + 1 < k && dist[child+1] > dist[child])
            child++;
        if(dist[root] > dist[child])
            return;
        swap<float>(&dist[root], &dist[child]);
        swap<int>(&idx[root], &idx[child]);
        root = child;
        child = root * 2 + 1;
    }
}


__device__ void heap_sort(float *dist, int *idx, int k)
{
    int i;
    for (i = k - 1; i > 0; i--)
    {
        swap<float>(&dist[0], &dist[i]);
        swap<int>(&idx[0], &idx[i]);
        reheap(dist, idx, i);
    }
}

__device__ int get_bt_idx(int idx, const int *offset)
{
    int i = 0;
    while (1)
    {
        if (idx < offset[i])
            break;
        else
            i++;
    }
    return i;
}
}  // namespace ball_query_utils

__global__ void ball_query_cuda_kernel(int m, int nsample,
                                       float min_radius, float max_radius,
                                       const float *__restrict__ xyz, const float *__restrict__ new_xyz,
                                       const int *__restrict__ offset, const int *__restrict__ new_offset,
                                       int *__restrict__ idx, float *__restrict__ dist2) {
    // input: xyz (n, 3) new_xyz (m, 3)
    // output: idx (m, nsample) dist (m, nsample)
    int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (pt_idx >= m) return;

    new_xyz += pt_idx * 3;
    idx += pt_idx * nsample;
    dist2 += pt_idx * nsample;

    int bt_idx = ball_query_utils::get_bt_idx(pt_idx, new_offset);
    int start;
    if (bt_idx == 0)
        start = 0;
    else
        start = offset[bt_idx - 1];
    int end = offset[bt_idx];

    float max_radius2 = max_radius * max_radius;
    float min_radius2 = min_radius * min_radius;
    float new_x = new_xyz[0];
    float new_y = new_xyz[1];
    float new_z = new_xyz[2];

    float candi_dist[2048];
    int candi_idx[2048];
    int candi_num = 0;

    for(int i = start; i < end; i++){
        float x = xyz[i * 3 + 0];
        float y = xyz[i * 3 + 1];
        float z = xyz[i * 3 + 2];
        float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) + (new_z - z) * (new_z - z);

        if (d2 <= 1e-5 || (d2 >= min_radius2 && d2 < max_radius2)){
            // TODO: Check d2 <= 1e-5
            candi_dist[candi_num] = d2;
            candi_idx[candi_num] = i;
            candi_num += 1;
        }
    }
    ball_query_utils::heap_sort(candi_dist, candi_idx, candi_num);
    if(candi_num <= nsample){
        for(int i = 0; i < candi_num; i++){
            idx[i] = candi_idx[i];
            dist2[i] = candi_dist[i];
        }
        for(int i = candi_num; i < nsample; i++){
            idx[i] = -1;
            dist2[i] = 1e10;
        }
    }
    else{
        float sep = static_cast<float>(candi_num) / nsample;
        for(int i = 0; i < nsample; i++)
        {
            int index = static_cast<int>(sep * i);
            idx[i] = candi_idx[index];
            dist2[i] = candi_idx[index];
        }
    }
}

/* Random Sample Mode Ball Query */

// __global__ void ball_query_cuda_kernel(int m, int nsample,
//                                        float min_radius, float max_radius,
//                                        const float *__restrict__ xyz, const float *__restrict__ new_xyz,
//                                        const int *__restrict__ offset, const int *__restrict__ new_offset,
//                                        int *__restrict__ idx, float *__restrict__ dist2) {
//     // input: xyz (n, 3) new_xyz (m, 3)
//     // output: idx (m, nsample) dist (m, nsample)
//     int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
//     if (pt_idx >= m) return;
//
//     new_xyz += pt_idx * 3;
//     idx += pt_idx * nsample;
//     dist2 += pt_idx * nsample;
//
//     int bt_idx = ball_get_bt_idx(pt_idx, new_offset);
//     int start;
//     if (bt_idx == 0)
//         start = 0;
//     else
//         start = offset[bt_idx - 1];
//     int end = offset[bt_idx];
//
//     float max_radius2 = max_radius * max_radius;
//     float min_radius2 = min_radius * min_radius;
//     float new_x = new_xyz[0];
//     float new_y = new_xyz[1];
//     float new_z = new_xyz[2];
//
//     int cnt = 0;
//     for(int i = start; i < end; i++){
//         float x = xyz[i * 3 + 0];
//         float y = xyz[i * 3 + 1];
//         float z = xyz[i * 3 + 2];
//         float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) + (new_z - z) * (new_z - z);
//
//         if (d2 == 0 || (d2 >= min_radius2 && d2 < max_radius2)) {
//             if (cnt == 0) {
//                 for (int l = 0; l < nsample; ++l) {
//                     idx[l] = i;
//                     dist2[l] = d2;
//                 }
//             }
//             idx[cnt] = i;
//             ++cnt;
//             if (cnt >= nsample) break;
//         }
//     }
// }


void ball_query_cuda_launcher(int m, int nsample,
                              float min_radius, float max_radius,
                              const float *xyz, const float *new_xyz,
                              const int *offset, const int *new_offset,
                              int *idx, float *dist2) {
    // input: new_xyz: (m, 3), xyz: (n, 3), idx: (m, nsample)
    dim3 blocks(DIVUP(m, THREADS_PER_BLOCK));
    dim3 threads(THREADS_PER_BLOCK);
    ball_query_cuda_kernel<<<blocks, threads, 0>>>(m, nsample,
                                                  min_radius, max_radius,
                                                  xyz, new_xyz,
                                                  offset, new_offset,
                                                  idx, dist2);
}