-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathemd_cuda.cu
348 lines (338 loc) · 12.7 KB
/
emd_cuda.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
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
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
#include <stdio.h>
#include "emd_cuda.h"
__global__ void approxmatch(int b,int n,int m,const float * __restrict__ xyz1,const float * __restrict__ xyz2,float * __restrict__ match,float * temp){
float * remainL=temp+blockIdx.x*(n+m)*2, * remainR=temp+blockIdx.x*(n+m)*2+n,*ratioL=temp+blockIdx.x*(n+m)*2+n+m,*ratioR=temp+blockIdx.x*(n+m)*2+n+m+n;
float multiL,multiR;
if (n>=m){
multiL=1;
multiR=n/m;
}else{
multiL=m/n;
multiR=1;
}
const int Block=1024;
__shared__ float buf[Block*7];
for (int i=blockIdx.x;i<b;i+=gridDim.x){
for (int j=threadIdx.x;j<n*m;j+=blockDim.x)
match[i*n*m+j]=0;
for (int j=threadIdx.x;j<n;j+=blockDim.x)
remainL[j]=multiL;
for (int j=threadIdx.x;j<m;j+=blockDim.x)
remainR[j]=multiR;
__syncthreads();
for (int j=7;j>=-2;j--){
float level=-powf(4.0f,j);
if (j==-2){
level=0;
}
for (int k0=0;k0<n;k0+=blockDim.x){
int k=k0+threadIdx.x;
float x1=0,y1=0,z1=0, xn1=0,yn1=0,zn1=0;
if (k<n){
x1=xyz1[i*n*6+k*6+0];
y1=xyz1[i*n*6+k*6+1];
z1=xyz1[i*n*6+k*6+2];
xn1=xyz1[i*n*6+k*6+3];
yn1=xyz1[i*n*6+k*6+4];
zn1=xyz1[i*n*6+k*6+5];
}
float suml=1e-9f;
for (int l0=0;l0<m;l0+=Block){
int lend=min(m,l0+Block)-l0;
for (int l=threadIdx.x;l<lend;l+=blockDim.x){
float x2=xyz2[i*m*6+l0*6+l*6+0];
float y2=xyz2[i*m*6+l0*6+l*6+1];
float z2=xyz2[i*m*6+l0*6+l*6+2];
float xn2=xyz2[i*m*6+l0*6+l*6+3];
float yn2=xyz2[i*m*6+l0*6+l*6+4];
float zn2=xyz2[i*m*6+l0*6+l*6+5];
buf[l*7+0]=x2;
buf[l*7+1]=y2;
buf[l*7+2]=z2;
buf[l*7+3]=xn2;
buf[l*7+4]=yn2;
buf[l*7+5]=zn2;
buf[l*7+6]=remainR[l0+l];
}
__syncthreads();
for (int l=0;l<lend;l++){
float x2=buf[l*7+0];
float y2=buf[l*7+1];
float z2=buf[l*7+2];
float xn2=buf[l*7+3];
float yn2=buf[l*7+4];
float zn2=buf[l*7+5];
float d=level*((x2-x1)*(x2-x1)+(y2-y1)*(y2-y1)+(z2-z1)*(z2-z1)+(xn1-xn2)*(xn1-xn2)+(yn1-yn2)*(yn1-yn2)+(zn1-zn2)*(zn1-zn2));
float w=__expf(d)*buf[l*7+6];
suml+=w;
}
__syncthreads();
}
if (k<n)
ratioL[k]=remainL[k]/suml;
}
__syncthreads();
for (int l0=0;l0<m;l0+=blockDim.x){
int l=l0+threadIdx.x;
float x2=0,y2=0,z2=0,xn2=0,yn2=0,zn2=0;
if (l<m){
x2=xyz2[i*m*6+l*6+0];
y2=xyz2[i*m*6+l*6+1];
z2=xyz2[i*m*6+l*6+2];
xn2=xyz2[i*m*6+l*6+3];
yn2=xyz2[i*m*6+l*6+4];
zn2=xyz2[i*m*6+l*6+5];
}
float sumr=0;
for (int k0=0;k0<n;k0+=Block){
int kend=min(n,k0+Block)-k0;
for (int k=threadIdx.x;k<kend;k+=blockDim.x){
buf[k*7+0]=xyz1[i*n*6+k0*6+k*6+0];
buf[k*7+1]=xyz1[i*n*6+k0*6+k*6+1];
buf[k*7+2]=xyz1[i*n*6+k0*6+k*6+2];
buf[k*7+3]=xyz1[i*n*6+k0*6+k*6+3];
buf[k*7+4]=xyz1[i*n*6+k0*6+k*6+4];
buf[k*7+5]=xyz1[i*n*6+k0*6+k*6+5];
buf[k*7+6]=ratioL[k0+k];
}
__syncthreads();
for (int k=0;k<kend;k++){
float x1=buf[k*7+0];
float y1=buf[k*7+1];
float z1=buf[k*7+2];
float xn1=buf[k*7+3];
float yn1=buf[k*7+4];
float zn1=buf[k*7+5];
float w=__expf(level*((x2-x1)*(x2-x1)+(y2-y1)*(y2-y1)+(z2-z1)*(z2-z1)+(xn1-xn2)*(xn1-xn2)+(yn1-yn2)*(yn1-yn2)+(zn1-zn2)*(zn1-zn2)))*buf[k*7+6];
sumr+=w;
}
__syncthreads();
}
if (l<m){
sumr*=remainR[l];
float consumption=fminf(remainR[l]/(sumr+1e-9f),1.0f);
ratioR[l]=consumption*remainR[l];
remainR[l]=fmaxf(0.0f,remainR[l]-sumr);
}
}
__syncthreads();
for (int k0=0;k0<n;k0+=blockDim.x){
int k=k0+threadIdx.x;
float x1=0,y1=0,z1=0,xn1=0,yn1=0,zn1=0;
if (k<n){
x1=xyz1[i*n*6+k*6+0];
y1=xyz1[i*n*6+k*6+1];
z1=xyz1[i*n*6+k*6+2];
xn1=xyz1[i*n*6+k*6+3];
yn1=xyz1[i*n*6+k*6+4];
zn1=xyz1[i*n*6+k*6+5];
}
float suml=0;
for (int l0=0;l0<m;l0+=Block){
int lend=min(m,l0+Block)-l0;
for (int l=threadIdx.x;l<lend;l+=blockDim.x){
buf[l*7+0]=xyz2[i*m*6+l0*6+l*6+0];
buf[l*7+1]=xyz2[i*m*6+l0*6+l*6+1];
buf[l*7+2]=xyz2[i*m*6+l0*6+l*6+2];
buf[l*7+3]=xyz2[i*m*6+l0*6+l*6+3];
buf[l*7+4]=xyz2[i*m*6+l0*6+l*6+4];
buf[l*7+5]=xyz2[i*m*6+l0*6+l*6+5];
buf[l*7+6]=ratioR[l0+l];
}
__syncthreads();
float rl=ratioL[k];
if (k<n){
for (int l=0;l<lend;l++){
float x2=buf[l*7+0];
float y2=buf[l*7+1];
float z2=buf[l*7+2];
float xn2=buf[l*7+3];
float yn2=buf[l*7+4];
float zn2=buf[l*7+5];
float w=__expf(level*((x2-x1)*(x2-x1)+(y2-y1)*(y2-y1)+(z2-z1)*(z2-z1)+(xn1-xn2)*(xn1-xn2)+(yn1-yn2)*(yn1-yn2)+(zn1-zn2)*(zn1-zn2)))*rl*buf[l*7+6];
match[i*n*m+(l0+l)*n+k]+=w;
suml+=w;
}
}
__syncthreads();
}
if (k<n)
remainL[k]=fmaxf(0.0f,remainL[k]-suml);
}
__syncthreads();
}
}
}
int approxmatchLauncher(int b,int n,int m,const float * xyz1,const float * xyz2,float * match, float * temp){
approxmatch<<<32,512>>>(b,n,m,xyz1,xyz2,match,temp);
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("error in matching: %s\n", cudaGetErrorString(err));
//THError("aborting");
return 0;
}
return 1;
}
__global__ void matchcost(int b,int n,int m,const float * __restrict__ xyz1,const float * __restrict__ xyz2,const float * __restrict__ match,float * __restrict__ out){
__shared__ float allsum[512];
const int Block=1024;
__shared__ float buf[Block*6];
for (int i=blockIdx.x;i<b;i+=gridDim.x){
float subsum=0;
for (int k0=0;k0<n;k0+=blockDim.x){
int k=k0+threadIdx.x;
float x1=0,y1=0,z1=0, xn1=0,yn1=0,zn1=0;
if (k<n){
x1=xyz1[i*n*6+k*6+0];
y1=xyz1[i*n*6+k*6+1];
z1=xyz1[i*n*6+k*6+2];
xn1=xyz1[i*n*6+k*6+3];
yn1=xyz1[i*n*6+k*6+4];
zn1=xyz1[i*n*6+k*6+5];
}
for (int l0=0;l0<m;l0+=Block){
int lend=min(m,l0+Block)-l0;
for (int l=threadIdx.x;l<lend*6;l+=blockDim.x)
buf[l]=xyz2[i*m*6+l0*6+l];
__syncthreads();
if (k<n){
for (int l=0;l<lend;l++){
float x2=buf[l*6+0];
float y2=buf[l*6+1];
float z2=buf[l*6+2];
float xn2=buf[l*6+3];
float yn2=buf[l*6+4];
float zn2=buf[l*6+5];
float d=sqrtf((x2-x1)*(x2-x1)+(y2-y1)*(y2-y1)+(z2-z1)*(z2-z1)+(xn1-xn2)*(xn1-xn2)+(yn1-yn2)*(yn1-yn2)+(zn1-zn2)*(zn1-zn2));
subsum+=d*match[i*n*m+(l0+l)*n+k];
}
}
__syncthreads();
}
}
allsum[threadIdx.x]=subsum;
for (int j=1;j<blockDim.x;j<<=1){
__syncthreads();
if ((threadIdx.x&j)==0 && threadIdx.x+j<blockDim.x){
allsum[threadIdx.x]+=allsum[threadIdx.x+j];
}
}
if (threadIdx.x==0)
out[i]=allsum[0];
__syncthreads();
}
}
int matchcostLauncher(int b,int n,int m,const float * xyz1,const float * xyz2,const float * match,float * out){
matchcost<<<32,512>>>(b,n,m,xyz1,xyz2,match,out);
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("error in emd updateOutput: %s\n", cudaGetErrorString(err));
//THError("aborting");
return 0;
}
return 1;
}
__global__ void matchcostgrad2(int b,int n,int m,const float * __restrict__ xyz1,const float * __restrict__ xyz2,const float * __restrict__ match,float * __restrict__ grad2){
__shared__ float sum_grad[256*6];
for (int i=blockIdx.x;i<b;i+=gridDim.x){
int kbeg=m*blockIdx.y/gridDim.y;
int kend=m*(blockIdx.y+1)/gridDim.y;
for (int k=kbeg;k<kend;k++){
float x2=xyz2[(i*m+k)*6+0];
float y2=xyz2[(i*m+k)*6+1];
float z2=xyz2[(i*m+k)*6+2];
float xn2=xyz2[(i*m+k)*6+3];
float yn2=xyz2[(i*m+k)*6+4];
float zn2=xyz2[(i*m+k)*6+5];
float subsumx=0,subsumy=0,subsumz=0,subsumxn=0,subsumyn=0,subsumzn=0;
for (int j=threadIdx.x;j<n;j+=blockDim.x){
float x1=x2-xyz1[(i*n+j)*6+0];
float y1=y2-xyz1[(i*n+j)*6+1];
float z1=z2-xyz1[(i*n+j)*6+2];
float xn1=xn2-xyz1[(i*n+j)*6+3];
float yn1=yn2-xyz1[(i*n+j)*6+4];
float zn1=zn2-xyz1[(i*n+j)*6+5];
float d=match[i*n*m+k*n+j]*rsqrtf(fmaxf(x1*x1+y1*y1+z1*z1+xn1*xn1+yn1*yn1+zn1*zn1,1e-20f));
subsumx+=x1*d;
subsumy+=y1*d;
subsumz+=z1*d;
subsumxn+=xn1*d;
subsumyn+=yn1*d;
subsumzn+=zn1*d;
}
sum_grad[threadIdx.x*6+0]=subsumx;
sum_grad[threadIdx.x*6+1]=subsumy;
sum_grad[threadIdx.x*6+2]=subsumz;
sum_grad[threadIdx.x*6+3]=subsumxn;
sum_grad[threadIdx.x*6+4]=subsumyn;
sum_grad[threadIdx.x*6+5]=subsumzn;
for (int j=1;j<blockDim.x;j<<=1){
__syncthreads();
int j1=threadIdx.x;
int j2=threadIdx.x+j;
if ((j1&j)==0 && j2<blockDim.x){
sum_grad[j1*6+0]+=sum_grad[j2*6+0];
sum_grad[j1*6+1]+=sum_grad[j2*6+1];
sum_grad[j1*6+2]+=sum_grad[j2*6+2];
sum_grad[j1*6+3]+=sum_grad[j2*6+3];
sum_grad[j1*6+4]+=sum_grad[j2*6+4];
sum_grad[j1*6+5]+=sum_grad[j2*6+5];
}
}
if (threadIdx.x==0){
grad2[(i*m+k)*6+0]=sum_grad[0];
grad2[(i*m+k)*6+1]=sum_grad[1];
grad2[(i*m+k)*6+2]=sum_grad[2];
grad2[(i*m+k)*6+3]=sum_grad[3];
grad2[(i*m+k)*6+4]=sum_grad[4];
grad2[(i*m+k)*6+5]=sum_grad[5];
}
__syncthreads();
}
}
}
__global__ void matchcostgrad1(int b,int n,int m,const float * __restrict__ xyz1,const float * __restrict__ xyz2,const float * __restrict__ match,float * __restrict__ grad1){
for (int i=blockIdx.x;i<b;i+=gridDim.x){
for (int l=threadIdx.x;l<n;l+=blockDim.x){
float x1=xyz1[i*n*6+l*6+0];
float y1=xyz1[i*n*6+l*6+1];
float z1=xyz1[i*n*6+l*6+2];
float xn1=xyz1[i*n*6+l*6+3];
float yn1=xyz1[i*n*6+l*6+4];
float zn1=xyz1[i*n*6+l*6+5];
float dx=0,dy=0,dz=0,dxn=0,dyn=0,dzn=0;
for (int k=0;k<m;k++){
float x2=xyz2[i*m*6+k*6+0];
float y2=xyz2[i*m*6+k*6+1];
float z2=xyz2[i*m*6+k*6+2];
float xn2=xyz2[i*m*6+k*6+3];
float yn2=xyz2[i*m*6+k*6+4];
float zn2=xyz2[i*m*6+k*6+5];
float d=match[i*n*m+k*n+l]*rsqrtf(fmaxf((x1-x2)*(x1-x2)+(y1-y2)*(y1-y2)+(z1-z2)*(z1-z2)+(xn1-xn2)*(xn1-xn2)+(yn1-yn2)*(yn1-yn2)+(zn1-zn2)*(zn1-zn2),1e-20f));
dx+=(x1-x2)*d;
dy+=(y1-y2)*d;
dz+=(z1-z2)*d;
dxn+=(xn1-xn2)*d;
dyn+=(yn1-yn2)*d;
dzn+=(zn1-zn2)*d;
}
grad1[i*n*6+l*6+0]=dx;
grad1[i*n*6+l*6+1]=dy;
grad1[i*n*6+l*6+2]=dz;
grad1[i*n*6+l*6+3]=dxn;
grad1[i*n*6+l*6+4]=dyn;
grad1[i*n*6+l*6+5]=dzn;
}
}
}
int matchcostgradLauncher(int b,int n,int m,const float * xyz1,const float * xyz2,const float * match,float * grad1,float * grad2){
matchcostgrad1<<<32,512>>>(b,n,m,xyz1,xyz2,match,grad1);
matchcostgrad2<<<dim3(32,32),256>>>(b,n,m,xyz1,xyz2,match,grad2);
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("error in emd backward: %s\n", cudaGetErrorString(err));
//THError("aborting");
return 0;
}
return 1;
}