File size: 8,606 Bytes
e8ffc70
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
#include <math.h>
#include <stdio.h>
#include <stdlib.h>


#include "ball_query_gpu.h"
#include "cuda_utils.h"


__global__ void ball_query_kernel_fast(int b, int n, int m, float radius, int nsample,
    const float *__restrict__ new_xyz, const float *__restrict__ xyz, int *__restrict__ idx) {
    // new_xyz: (B, M, 3)
    // xyz: (B, N, 3)
    // output:
    //      idx: (B, M, nsample)
    int bs_idx = blockIdx.y;
    int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (bs_idx >= b || pt_idx >= m) return;

    new_xyz += bs_idx * m * 3 + pt_idx * 3;
    xyz += bs_idx * n * 3;
    idx += bs_idx * m * nsample + pt_idx * nsample;

    float radius2 = radius * radius;
    float new_x = new_xyz[0];
    float new_y = new_xyz[1];
    float new_z = new_xyz[2];

    int cnt = 0;
    for (int k = 0; k < n; ++k) {
        float x = xyz[k * 3 + 0];
        float y = xyz[k * 3 + 1];
        float z = xyz[k * 3 + 2];
        float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) + (new_z - z) * (new_z - z);
        if (d2 < radius2){
            if (cnt == 0){
                for (int l = 0; l < nsample; ++l) {
                    idx[l] = k;
                }
            }
            idx[cnt] = k;
            ++cnt;
            if (cnt >= nsample) break;
        }
    }
}


void ball_query_kernel_launcher_fast(int b, int n, int m, float radius, int nsample, \
    const float *new_xyz, const float *xyz, int *idx) {
    // new_xyz: (B, M, 3)
    // xyz: (B, N, 3)
    // output:
    //      idx: (B, M, nsample)

    cudaError_t err;

    dim3 blocks(DIVUP(m, THREADS_PER_BLOCK), b);  // blockIdx.x(col), blockIdx.y(row)
    dim3 threads(THREADS_PER_BLOCK);

    ball_query_kernel_fast<<<blocks, threads>>>(b, n, m, radius, nsample, new_xyz, xyz, idx);
    // cudaDeviceSynchronize();  // for using printf in kernel function
    err = cudaGetLastError();
    if (cudaSuccess != err) {
        fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
        exit(-1);
    }
}


__global__ void ball_center_query_kernel_fast(int b, int n, int m, float radius, \
    const float *__restrict__ point, const float *__restrict__ key_point, int *__restrict__ idx) {
    // key_point: (B, M, 3)
    // point: (B, N, 3)
    // output:
    //      idx: (B, N)
    int bs_idx = blockIdx.y;
    int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (bs_idx >= b || pt_idx >= n) return;

    point += bs_idx * n * 3 + pt_idx * 3;
    key_point += bs_idx * m * 3;
    idx += bs_idx * n + pt_idx;

    float radius2 = radius * radius;
    float point_x = point[0];
    float point_y = point[1];
    float point_z = point[2];

    float bestd = 1e8;
    for (int k = 0; k < m; ++k) {
        float x = key_point[k * 3 + 0];
        float y = key_point[k * 3 + 1];
        float z = key_point[k * 3 + 2];
        if (((x + 1) * (x + 1) + (y + 1) * (y + 1) + (z + 1) * (z + 1)) < 1e-4) break;
        float d2 = (point_x - x) * (point_x - x) + (point_y - y) * (point_y - y) + (point_z - z) * (point_z - z);
        if (d2 < radius2 && d2 < bestd){
            idx[0] = k;
            bestd = d2;
        }
    }
}


void ball_center_query_kernel_launcher_fast(int b, int n, int m, float radius, \
    const float *point, const float *key_point, int *idx) {
    // point: (B, n, 3)
    // key_point: (B, m, 3)
    // output:
    //      idx: (B, n)

    cudaError_t err;

    dim3 blocks(DIVUP(n, THREADS_PER_BLOCK), b);  // blockIdx.x(col), blockIdx.y(row)
    dim3 threads(THREADS_PER_BLOCK);

    ball_center_query_kernel_fast<<<blocks, threads>>>(b, n, m, radius, point, key_point, idx);
    // cudaDeviceSynchronize();  // for using printf in kernel function
    err = cudaGetLastError();
    if (cudaSuccess != err) {
        fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
        exit(-1);
    }
}





__global__ void knn_query_kernel_fast(int b, int n, int m, int nsample, const float *__restrict__ new_xyz,
    const float *__restrict__ xyz, float *__restrict__ dist2, int *__restrict__ idx) {

    // new_xyz: (B, M, 3)
    // xyz: (B, N, 3)
    // output:
    //      dist2: (B, M, nsample)
    //      idx: (B, M, nsample)

    int bs_idx = blockIdx.y;
    int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (bs_idx >= b || pt_idx >= m) return;

    new_xyz += bs_idx * m * 3 + pt_idx * 3;
    xyz += bs_idx * n * 3;
    dist2 += bs_idx * m * nsample + pt_idx * nsample;
    idx += bs_idx * m * nsample + pt_idx * nsample;

    float nx = new_xyz[0];
    float ny = new_xyz[1];
    float nz = new_xyz[2];

    for (int i = 0; i < n; ++i) {
        float x = xyz[i * 3 + 0];
        float y = xyz[i * 3 + 1];
        float z = xyz[i * 3 + 2];
        float d2 = (nx - x) * (nx - x) + (ny - y) * (ny - y) + (nz - z) * (nz - z);
        if (d2 < dist2[nsample - 1]) {
            dist2[nsample - 1] = d2;
            idx[nsample - 1] = i;
            for (int j = nsample - 2; j >= 0; j--) {
                if (d2 < dist2[j]){
                    dist2[j + 1] = dist2[j];
                    dist2[j] = d2;
                    idx[j + 1] = idx[j];
                    idx[j] = i;
                }
            }
        }
    }
}


void knn_query_kernel_launcher_fast(int b, int n, int m, int nsample, \
    const float *new_xyz, const float *xyz, float *dist2, int *idx) {
    cudaError_t err;

    dim3 blocks(DIVUP(m, THREADS_PER_BLOCK), b);  // blockIdx.x(col), blockIdx.y(row)
    dim3 threads(THREADS_PER_BLOCK);

    knn_query_kernel_fast<<<blocks, threads>>>(b, n, m, nsample, new_xyz, xyz, dist2, idx);
    // cudaDeviceSynchronize();  // for using printf in kernel function
    err = cudaGetLastError();
    if (cudaSuccess != err) {
        fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
        exit(-1);
    }
}








__global__ void ball_query_kernel_stack(int B, int M, float radius, int nsample, \
    const float *new_xyz, const int *new_xyz_batch_cnt, const float *xyz, const int *xyz_batch_cnt, int *idx) {
    // :param xyz: (N1 + N2 ..., 3) xyz coordinates of the features
    // :param xyz_batch_cnt: (batch_size), [N1, N2, ...]
    // :param new_xyz: (M1 + M2 ..., 3) centers of the ball query
    // :param new_xyz_batch_cnt: (batch_size), [M1, M2, ...]
    // output:
    //      idx: (M, nsample)
    int pt_idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (pt_idx >= M) return;

    int bs_idx = 0, pt_cnt = new_xyz_batch_cnt[0];
    for (int k = 1; k < B; k++){
        if (pt_idx < pt_cnt) break;
        pt_cnt += new_xyz_batch_cnt[k];
        bs_idx = k;
    }

    int xyz_batch_start_idx = 0;
    for (int k = 0; k < bs_idx; k++) xyz_batch_start_idx += xyz_batch_cnt[k];
    // for (int k = 0; k < bs_idx; k++) new_xyz_batch_start_idx += new_xyz_batch_cnt[k];

    new_xyz += pt_idx * 3;
    xyz += xyz_batch_start_idx * 3;
    idx += pt_idx * nsample;

    float radius2 = radius * radius;
    float new_x = new_xyz[0];
    float new_y = new_xyz[1];
    float new_z = new_xyz[2];
    int n = xyz_batch_cnt[bs_idx];

    int cnt = 0;
    for (int k = 0; k < n; ++k) {
        float x = xyz[k * 3 + 0];
        float y = xyz[k * 3 + 1];
        float z = xyz[k * 3 + 2];
        float d2 = (new_x - x) * (new_x - x) + (new_y - y) * (new_y - y) + (new_z - z) * (new_z - z);
        if (d2 < radius2){
            if (cnt == 0){
                for (int l = 0; l < nsample; ++l) {
                    idx[l] = k;
                }
            }
            idx[cnt] = k;
            ++cnt;
            if (cnt >= nsample) break;
        }
    }
    if (cnt == 0) idx[0] = -1;
}


void ball_query_kernel_launcher_stack(int B, int M, float radius, int nsample,
    const float *new_xyz, const int *new_xyz_batch_cnt, const float *xyz, const int *xyz_batch_cnt, int *idx){
    // :param xyz: (N1 + N2 ..., 3) xyz coordinates of the features
    // :param xyz_batch_cnt: (batch_size), [N1, N2, ...]
    // :param new_xyz: (M1 + M2 ..., 3) centers of the ball query
    // :param new_xyz_batch_cnt: (batch_size), [M1, M2, ...]
    // output:
    //      idx: (M, nsample)

    cudaError_t err;

    dim3 blocks(DIVUP(M, THREADS_PER_BLOCK));  // blockIdx.x(col), blockIdx.y(row)
    dim3 threads(THREADS_PER_BLOCK);

    ball_query_kernel_stack<<<blocks, threads>>>(B, M, radius, nsample, new_xyz, new_xyz_batch_cnt, xyz, xyz_batch_cnt, idx);
    // cudaDeviceSynchronize();  // for using printf in kernel function
    err = cudaGetLastError();
    if (cudaSuccess != err) {
        fprintf(stderr, "CUDA kernel failed : %s\n", cudaGetErrorString(err));
        exit(-1);
    }
}