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
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
|
/*
* This work is part of the Core Imaging Library developed by
* Visual Analytics and Imaging System Group of the Science Technology
* Facilities Council, STFC and Diamond Light Source Ltd.
*
* Copyright 2017 Daniil Kazantsev
* Copyright 2017 Srikanth Nagella, Edoardo Pasca
* Copyright 2018 Diamond Light Source Ltd.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
* http://www.apache.org/licenses/LICENSE-2.0
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "PatchSelect_GPU_core.h"
#include "shared.h"
/* CUDA implementation of non-local weight pre-calculation for non-local priors
* Weights and associated indices are stored into pre-allocated arrays and passed
* to the regulariser
*
*
* Input Parameters:
* 1. 2D grayscale image (classical 3D version will not be supported but rather 2D + dim extension (TODO))
* 2. Searching window (half-size of the main bigger searching window, e.g. 11)
* 3. Similarity window (half-size of the patch window, e.g. 2)
* 4. The number of neighbours to take (the most prominent after sorting neighbours will be taken)
* 5. noise-related parameter to calculate non-local weights
*
* Output [2D]:
* 1. AR_i - indeces of i neighbours
* 2. AR_j - indeces of j neighbours
* 3. Weights_ij - associated weights
*/
#define BLKXSIZE 16
#define BLKYSIZE 16
#define idivup(a, b) ( ((a)%(b) != 0) ? (a)/(b)+1 : (a)/(b) )
#define M_PI 3.14159265358979323846
#define EPS 1.0e-8
#define CONSTVECSIZE5 121
#define CONSTVECSIZE7 225
#define CONSTVECSIZE9 361
#define CONSTVECSIZE11 529
#define CONSTVECSIZE13 729
__device__ void swap(float *xp, float *yp)
{
float temp = *xp;
*xp = *yp;
*yp = temp;
}
__device__ void swapUS(unsigned short *xp, unsigned short *yp)
{
unsigned short temp = *xp;
*xp = *yp;
*yp = temp;
}
/********************************************************************************/
__global__ void IndexSelect2D_5_kernel(float *Ad, unsigned short *H_i_d, unsigned short *H_j_d, float *Weights_d, float *Eucl_Vec_d, int N, int M, int SearchWindow, int SearchW_full, int SimilarWin, int NumNeighb, float h2)
{
long i1, j1, i_m, j_m, i_c, j_c, i2, j2, i3, j3, counter, x, y, counterG, index2;
float normsum;
float Weight_Vec[CONSTVECSIZE5];
unsigned short ind_i[CONSTVECSIZE5];
unsigned short ind_j[CONSTVECSIZE5];
int i = blockDim.x * blockIdx.x + threadIdx.x;
int j = blockDim.y * blockIdx.y + threadIdx.y;
long index = i*M+j;
counter = 0;
for(i_m=-SearchWindow; i_m<=SearchWindow; i_m++) {
for(j_m=-SearchWindow; j_m<=SearchWindow; j_m++) {
i1 = i+i_m;
j1 = j+j_m;
if (((i1 >= 0) && (i1 < N)) && ((j1 >= 0) && (j1 < M))) {
normsum = 0.0f; counterG = 0;
for(i_c=-SimilarWin; i_c<=SimilarWin; i_c++) {
for(j_c=-SimilarWin; j_c<=SimilarWin; j_c++) {
i2 = i1 + i_c;
j2 = j1 + j_c;
i3 = i + i_c;
j3 = j + j_c;
if (((i2 >= 0) && (i2 < N)) && ((j2 >= 0) && (j2 < M))) {
if (((i3 >= 0) && (i3 < N)) && ((j3 >= 0) && (j3 < M))) {
normsum += Eucl_Vec_d[counterG]*powf(Ad[i3*M + j3] - Ad[i2*M + j2], 2);
counterG++;
}}
}}
/* writing temporarily into vectors */
if (normsum > EPS) {
Weight_Vec[counter] = __expf(-normsum/h2);
ind_i[counter] = i1;
ind_j[counter] = j1;
counter++;
}
}
}}
/* do sorting to choose the most prominent weights [HIGH to LOW] */
/* and re-arrange indeces accordingly */
for (x = 0; x < counter-1; x++) {
for (y = 0; y < counter-x-1; y++) {
if (Weight_Vec[y] < Weight_Vec[y+1]) {
swap(&Weight_Vec[y], &Weight_Vec[y+1]);
swapUS(&ind_i[y], &ind_i[y+1]);
swapUS(&ind_j[y], &ind_j[y+1]);
}
}
}
/*sorting loop finished*/
/*now select the NumNeighb more prominent weights and store into arrays */
for(x=0; x < NumNeighb; x++) {
index2 = (N*M*x) + index;
H_i_d[index2] = ind_i[x];
H_j_d[index2] = ind_j[x];
Weights_d[index2] = Weight_Vec[x];
}
}
/********************************************************************************/
__global__ void IndexSelect2D_7_kernel(float *Ad, unsigned short *H_i_d, unsigned short *H_j_d, float *Weights_d, float *Eucl_Vec_d, int N, int M, int SearchWindow, int SearchW_full, int SimilarWin, int NumNeighb, float h2)
{
long i1, j1, i_m, j_m, i_c, j_c, i2, j2, i3, j3, counter, x, y, counterG, index2;
float normsum;
float Weight_Vec[CONSTVECSIZE7];
unsigned short ind_i[CONSTVECSIZE7];
unsigned short ind_j[CONSTVECSIZE7];
int i = blockDim.x * blockIdx.x + threadIdx.x;
int j = blockDim.y * blockIdx.y + threadIdx.y;
long index = i*M+j;
counter = 0;
for(i_m=-SearchWindow; i_m<=SearchWindow; i_m++) {
for(j_m=-SearchWindow; j_m<=SearchWindow; j_m++) {
i1 = i+i_m;
j1 = j+j_m;
if (((i1 >= 0) && (i1 < N)) && ((j1 >= 0) && (j1 < M))) {
normsum = 0.0f; counterG = 0;
for(i_c=-SimilarWin; i_c<=SimilarWin; i_c++) {
for(j_c=-SimilarWin; j_c<=SimilarWin; j_c++) {
i2 = i1 + i_c;
j2 = j1 + j_c;
i3 = i + i_c;
j3 = j + j_c;
if (((i2 >= 0) && (i2 < N)) && ((j2 >= 0) && (j2 < M))) {
if (((i3 >= 0) && (i3 < N)) && ((j3 >= 0) && (j3 < M))) {
normsum += Eucl_Vec_d[counterG]*powf(Ad[i3*M + j3] - Ad[i2*M + j2], 2);
counterG++;
}}
}}
/* writing temporarily into vectors */
if (normsum > EPS) {
Weight_Vec[counter] = __expf(-normsum/h2);
ind_i[counter] = i1;
ind_j[counter] = j1;
counter++;
}
}
}}
/* do sorting to choose the most prominent weights [HIGH to LOW] */
/* and re-arrange indeces accordingly */
for (x = 0; x < counter-1; x++) {
for (y = 0; y < counter-x-1; y++) {
if (Weight_Vec[y] < Weight_Vec[y+1]) {
swap(&Weight_Vec[y], &Weight_Vec[y+1]);
swapUS(&ind_i[y], &ind_i[y+1]);
swapUS(&ind_j[y], &ind_j[y+1]);
}
}
}
/*sorting loop finished*/
/*now select the NumNeighb more prominent weights and store into arrays */
for(x=0; x < NumNeighb; x++) {
index2 = (N*M*x) + index;
H_i_d[index2] = ind_i[x];
H_j_d[index2] = ind_j[x];
Weights_d[index2] = Weight_Vec[x];
}
}
__global__ void IndexSelect2D_9_kernel(float *Ad, unsigned short *H_i_d, unsigned short *H_j_d, float *Weights_d, float *Eucl_Vec_d, int N, int M, int SearchWindow, int SearchW_full, int SimilarWin, int NumNeighb, float h2)
{
long i1, j1, i_m, j_m, i_c, j_c, i2, j2, i3, j3, counter, x, y, counterG, index2;
float normsum;
float Weight_Vec[CONSTVECSIZE9];
unsigned short ind_i[CONSTVECSIZE9];
unsigned short ind_j[CONSTVECSIZE9];
int i = blockDim.x * blockIdx.x + threadIdx.x;
int j = blockDim.y * blockIdx.y + threadIdx.y;
long index = i*M+j;
counter = 0;
for(i_m=-SearchWindow; i_m<=SearchWindow; i_m++) {
for(j_m=-SearchWindow; j_m<=SearchWindow; j_m++) {
i1 = i+i_m;
j1 = j+j_m;
if (((i1 >= 0) && (i1 < N)) && ((j1 >= 0) && (j1 < M))) {
normsum = 0.0f; counterG = 0;
for(i_c=-SimilarWin; i_c<=SimilarWin; i_c++) {
for(j_c=-SimilarWin; j_c<=SimilarWin; j_c++) {
i2 = i1 + i_c;
j2 = j1 + j_c;
i3 = i + i_c;
j3 = j + j_c;
if (((i2 >= 0) && (i2 < N)) && ((j2 >= 0) && (j2 < M))) {
if (((i3 >= 0) && (i3 < N)) && ((j3 >= 0) && (j3 < M))) {
normsum += Eucl_Vec_d[counterG]*powf(Ad[i3*M + j3] - Ad[i2*M + j2], 2);
counterG++;
}}
}}
/* writing temporarily into vectors */
if (normsum > EPS) {
Weight_Vec[counter] = expf(-normsum/h2);
ind_i[counter] = i1;
ind_j[counter] = j1;
counter++;
}
}
}}
/* do sorting to choose the most prominent weights [HIGH to LOW] */
/* and re-arrange indeces accordingly */
for (x = 0; x < counter-1; x++) {
for (y = 0; y < counter-x-1; y++) {
if (Weight_Vec[y] < Weight_Vec[y+1]) {
swap(&Weight_Vec[y], &Weight_Vec[y+1]);
swapUS(&ind_i[y], &ind_i[y+1]);
swapUS(&ind_j[y], &ind_j[y+1]);
}
}
}
/*sorting loop finished*/
/*now select the NumNeighb more prominent weights and store into arrays */
for(x=0; x < NumNeighb; x++) {
index2 = (N*M*x) + index;
H_i_d[index2] = ind_i[x];
H_j_d[index2] = ind_j[x];
Weights_d[index2] = Weight_Vec[x];
}
}
__global__ void IndexSelect2D_11_kernel(float *Ad, unsigned short *H_i_d, unsigned short *H_j_d, float *Weights_d, float *Eucl_Vec_d, int N, int M, int SearchWindow, int SearchW_full, int SimilarWin, int NumNeighb, float h2)
{
long i1, j1, i_m, j_m, i_c, j_c, i2, j2, i3, j3, counter, x, y, counterG, index2;
float normsum;
float Weight_Vec[CONSTVECSIZE11];
unsigned short ind_i[CONSTVECSIZE11];
unsigned short ind_j[CONSTVECSIZE11];
int i = blockDim.x * blockIdx.x + threadIdx.x;
int j = blockDim.y * blockIdx.y + threadIdx.y;
long index = i*M+j;
counter = 0;
for(i_m=-SearchWindow; i_m<=SearchWindow; i_m++) {
for(j_m=-SearchWindow; j_m<=SearchWindow; j_m++) {
i1 = i+i_m;
j1 = j+j_m;
if (((i1 >= 0) && (i1 < N)) && ((j1 >= 0) && (j1 < M))) {
normsum = 0.0f; counterG = 0;
for(i_c=-SimilarWin; i_c<=SimilarWin; i_c++) {
for(j_c=-SimilarWin; j_c<=SimilarWin; j_c++) {
i2 = i1 + i_c;
j2 = j1 + j_c;
i3 = i + i_c;
j3 = j + j_c;
if (((i2 >= 0) && (i2 < N)) && ((j2 >= 0) && (j2 < M))) {
if (((i3 >= 0) && (i3 < N)) && ((j3 >= 0) && (j3 < M))) {
normsum += Eucl_Vec_d[counterG]*powf(Ad[i3*M + j3] - Ad[i2*M + j2], 2);
counterG++;
}}
}}
/* writing temporarily into vectors */
if (normsum > EPS) {
Weight_Vec[counter] = __expf(-normsum/h2);
ind_i[counter] = i1;
ind_j[counter] = j1;
counter++;
}
}
}}
/* do sorting to choose the most prominent weights [HIGH to LOW] */
/* and re-arrange indeces accordingly */
for (x = 0; x < counter-1; x++) {
for (y = 0; y < counter-x-1; y++) {
if (Weight_Vec[y] < Weight_Vec[y+1]) {
swap(&Weight_Vec[y], &Weight_Vec[y+1]);
swapUS(&ind_i[y], &ind_i[y+1]);
swapUS(&ind_j[y], &ind_j[y+1]);
}
}
}
/*sorting loop finished*/
/*now select the NumNeighb more prominent weights and store into arrays */
for(x=0; x < NumNeighb; x++) {
index2 = (N*M*x) + index;
H_i_d[index2] = ind_i[x];
H_j_d[index2] = ind_j[x];
Weights_d[index2] = Weight_Vec[x];
}
}
__global__ void IndexSelect2D_13_kernel(float *Ad, unsigned short *H_i_d, unsigned short *H_j_d, float *Weights_d, float *Eucl_Vec_d, int N, int M, int SearchWindow, int SearchW_full, int SimilarWin, int NumNeighb, float h2)
{
long i1, j1, i_m, j_m, i_c, j_c, i2, j2, i3, j3, counter, x, y, counterG, index2;
float normsum;
float Weight_Vec[CONSTVECSIZE13];
unsigned short ind_i[CONSTVECSIZE13];
unsigned short ind_j[CONSTVECSIZE13];
int i = blockDim.x * blockIdx.x + threadIdx.x;
int j = blockDim.y * blockIdx.y + threadIdx.y;
long index = i*M+j;
counter = 0;
for(i_m=-SearchWindow; i_m<=SearchWindow; i_m++) {
for(j_m=-SearchWindow; j_m<=SearchWindow; j_m++) {
i1 = i+i_m;
j1 = j+j_m;
if (((i1 >= 0) && (i1 < N)) && ((j1 >= 0) && (j1 < M))) {
normsum = 0.0f; counterG = 0;
for(i_c=-SimilarWin; i_c<=SimilarWin; i_c++) {
for(j_c=-SimilarWin; j_c<=SimilarWin; j_c++) {
i2 = i1 + i_c;
j2 = j1 + j_c;
i3 = i + i_c;
j3 = j + j_c;
if (((i2 >= 0) && (i2 < N)) && ((j2 >= 0) && (j2 < M))) {
if (((i3 >= 0) && (i3 < N)) && ((j3 >= 0) && (j3 < M))) {
normsum += Eucl_Vec_d[counterG]*powf(Ad[i3*M + j3] - Ad[i2*M + j2], 2);
counterG++;
}}
}}
/* writing temporarily into vectors */
if (normsum > EPS) {
Weight_Vec[counter] = __expf(-normsum/h2);
ind_i[counter] = i1;
ind_j[counter] = j1;
counter++;
}
}
}}
/* do sorting to choose the most prominent weights [HIGH to LOW] */
/* and re-arrange indeces accordingly */
for (x = 0; x < counter-1; x++) {
for (y = 0; y < counter-x-1; y++) {
if (Weight_Vec[y] < Weight_Vec[y+1]) {
swap(&Weight_Vec[y], &Weight_Vec[y+1]);
swapUS(&ind_i[y], &ind_i[y+1]);
swapUS(&ind_j[y], &ind_j[y+1]);
}
}
}
/*sorting loop finished*/
/*now select the NumNeighb more prominent weights and store into arrays */
for(x=0; x < NumNeighb; x++) {
index2 = (N*M*x) + index;
H_i_d[index2] = ind_i[x];
H_j_d[index2] = ind_j[x];
Weights_d[index2] = Weight_Vec[x];
}
}
/*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*/
/********************* MAIN HOST FUNCTION ******************/
/*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*/
extern "C" int PatchSelect_GPU_main(float *A, unsigned short *H_i, unsigned short *H_j, float *Weights, int N, int M, int SearchWindow, int SimilarWin, int NumNeighb, float h)
{
int deviceCount = -1; // number of devices
cudaGetDeviceCount(&deviceCount);
if (deviceCount == 0) {
fprintf(stderr, "No CUDA devices found\n");
return -1;
}
int SearchW_full, SimilW_full, counterG, i, j;
float *Ad, *Weights_d, h2, *Eucl_Vec, *Eucl_Vec_d;
unsigned short *H_i_d, *H_j_d;
h2 = h*h;
dim3 dimBlock(BLKXSIZE,BLKYSIZE);
dim3 dimGrid(idivup(N,BLKXSIZE), idivup(M,BLKYSIZE));
SearchW_full = (2*SearchWindow + 1)*(2*SearchWindow + 1); /* the full searching window size */
SimilW_full = (2*SimilarWin + 1)*(2*SimilarWin + 1); /* the full similarity window size */
/* generate a 2D Gaussian kernel for NLM procedure */
Eucl_Vec = (float*) calloc (SimilW_full,sizeof(float));
counterG = 0;
for(i=-SimilarWin; i<=SimilarWin; i++) {
for(j=-SimilarWin; j<=SimilarWin; j++) {
Eucl_Vec[counterG] = (float)exp(-(pow(((float) i), 2) + pow(((float) j), 2))/(2.0*SimilarWin*SimilarWin));
counterG++;
}} /*main neighb loop */
/*allocate space on the device*/
checkCudaErrors( cudaMalloc((void**)&Ad, N*M*sizeof(float)) );
checkCudaErrors( cudaMalloc((void**)&H_i_d, N*M*NumNeighb*sizeof(unsigned short)) );
checkCudaErrors( cudaMalloc((void**)&H_j_d, N*M*NumNeighb*sizeof(unsigned short)) );
checkCudaErrors( cudaMalloc((void**)&Weights_d, N*M*NumNeighb*sizeof(float)) );
checkCudaErrors( cudaMalloc((void**)&Eucl_Vec_d, SimilW_full*sizeof(float)) );
/* copy data from the host to the device */
checkCudaErrors( cudaMemcpy(Ad,A,N*M*sizeof(float),cudaMemcpyHostToDevice) );
checkCudaErrors( cudaMemcpy(Eucl_Vec_d,Eucl_Vec,SimilW_full*sizeof(float),cudaMemcpyHostToDevice) );
/********************** Run CUDA kernel here ********************/
if (SearchWindow == 5) IndexSelect2D_5_kernel<<<dimGrid,dimBlock>>>(Ad, H_i_d, H_j_d, Weights_d, Eucl_Vec_d, N, M, SearchWindow, SearchW_full, SimilarWin, NumNeighb, h2);
else if (SearchWindow == 7) IndexSelect2D_7_kernel<<<dimGrid,dimBlock>>>(Ad, H_i_d, H_j_d, Weights_d, Eucl_Vec_d, N, M, SearchWindow, SearchW_full, SimilarWin, NumNeighb, h2);
else if (SearchWindow == 9) IndexSelect2D_9_kernel<<<dimGrid,dimBlock>>>(Ad, H_i_d, H_j_d, Weights_d, Eucl_Vec_d, N, M, SearchWindow, SearchW_full, SimilarWin, NumNeighb, h2);
else if (SearchWindow == 11) IndexSelect2D_11_kernel<<<dimGrid,dimBlock>>>(Ad, H_i_d, H_j_d, Weights_d, Eucl_Vec_d, N, M, SearchWindow, SearchW_full, SimilarWin, NumNeighb, h2);
else if (SearchWindow == 13) IndexSelect2D_13_kernel<<<dimGrid,dimBlock>>>(Ad, H_i_d, H_j_d, Weights_d, Eucl_Vec_d, N, M, SearchWindow, SearchW_full, SimilarWin, NumNeighb, h2);
else {
fprintf(stderr, "Select the searching window size from 5, 7, 9, 11 or 13\n");
return -1;}
checkCudaErrors(cudaPeekAtLastError() );
checkCudaErrors(cudaDeviceSynchronize());
/***************************************************************/
checkCudaErrors(cudaMemcpy(H_i, H_i_d, N*M*NumNeighb*sizeof(unsigned short),cudaMemcpyDeviceToHost) );
checkCudaErrors(cudaMemcpy(H_j, H_j_d, N*M*NumNeighb*sizeof(unsigned short),cudaMemcpyDeviceToHost) );
checkCudaErrors(cudaMemcpy(Weights, Weights_d, N*M*NumNeighb*sizeof(float),cudaMemcpyDeviceToHost) );
cudaFree(Ad);
cudaFree(H_i_d);
cudaFree(H_j_d);
cudaFree(Weights_d);
cudaFree(Eucl_Vec_d);
cudaDeviceReset();
return 0;
}
|
