1 /*
2 * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation. Oracle designates this
8 * particular file as subject to the "Classpath" exception as provided
9 * by Oracle in the LICENSE file that accompanied this code.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 */
25
26
27 /*
28 * FUNCTION
29 * Internal functions for mlib_ImageConv* on D64/F32 type and
30 * MLIB_EDGE_DST_NO_WRITE mask
31 *
32 */
33
34 #include "mlib_image.h"
35 #include "mlib_ImageConv.h"
36
37 /***************************************************************/
38 /*
39 This define switches between functions of MLIB_DOUBLE and MLIB_FLOAT types:
40 Files mlib_ImageConv_D64nw.c and mlib_ImageConv_F32nw.c
41 */
42
43 #define TYPE_DOUBLE
44
45 /***************************************************************/
46 #ifdef TYPE_DOUBLE
47
48 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_d64
49
50 #define DTYPE mlib_d64
51
52 #else
53
54 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_f32
55
56 #define DTYPE mlib_f32
57
58 #endif /* TYPE_DOUBLE */
59
60 /***************************************************************/
61 #define GET_SRC_DST_PARAMETERS(type) \
62 mlib_s32 hgt = mlib_ImageGetHeight(src); \
63 mlib_s32 wid = mlib_ImageGetWidth(src); \
64 mlib_s32 sll = mlib_ImageGetStride(src) / sizeof(type); \
65 mlib_s32 dll = mlib_ImageGetStride(dst) / sizeof(type); \
66 type* adr_src = mlib_ImageGetData(src); \
67 type* adr_dst = mlib_ImageGetData(dst); \
68 mlib_s32 chan1 = mlib_ImageGetChannels(src)
69
70 /***************************************************************/
71 #define DEF_VARS(type) \
72 GET_SRC_DST_PARAMETERS(type); \
73 type *sl; \
74 type *dl, *dp = NULL; \
75 mlib_s32 i = 0, j, c
76
77 /***************************************************************/
78 #define BUFF_SIZE 1600
79
80 #define CACHE_SIZE (64*1024)
81
82 static mlib_status mlib_ImageConv1xN(mlib_image *dst,
83 const mlib_image *src,
84 const DTYPE *k,
85 mlib_s32 n,
86 mlib_s32 dn,
87 mlib_s32 cmask)
88 {
89 DTYPE buff[BUFF_SIZE], *pbuff = buff;
90 const DTYPE *pk;
91 DTYPE k0, k1, k2, k3;
92 DTYPE p0, p1, p2, p3, p4;
93 DTYPE *sp, *sl_c, *dl_c, *sl0;
94 DEF_VARS(DTYPE);
95 mlib_s32 off, kh;
96 mlib_s32 l, hsize, max_hsize;
97
98 hgt -= (n - 1);
99 adr_dst += dn*dll;
100
101 max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll;
102
103 if (!max_hsize) max_hsize = 1;
104
105 if (max_hsize > BUFF_SIZE) {
106 pbuff = mlib_malloc(sizeof(DTYPE)*max_hsize);
107 }
108
109 sl_c = adr_src;
110 dl_c = adr_dst;
111
112 for (l = 0; l < hgt; l += hsize) {
113 hsize = hgt - l;
114
115 if (hsize > max_hsize) hsize = max_hsize;
116
117 for (c = 0; c < chan1; c++) {
118 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
119
120 sl = sl_c + c;
121 dl = dl_c + c;
122
123 for (j = 0; j < hsize; j++) pbuff[j] = 0.0;
124
125 for (i = 0; i < wid; i++) {
126 sl0 = sl;
127
128 for (off = 0; off < (n - 4); off += 4) {
129 pk = k + off;
130 sp = sl0;
131
132 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
133 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
134 sp += 3*sll;
135
136 for (j = 0; j < hsize; j += 2) {
137 p0 = p2; p1 = p3; p2 = p4;
138 p3 = sp[0];
139 p4 = sp[sll];
140
141 pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
142 pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
143
144 sp += 2*sll;
145 }
146
147 sl0 += 4*sll;
148 }
149
150 pk = k + off;
151 sp = sl0;
152
153 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
154 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
155
156 dp = dl;
157 kh = n - off;
158
159 if (kh == 4) {
160 sp += 3*sll;
161
162 for (j = 0; j <= (hsize - 2); j += 2) {
163 p0 = p2; p1 = p3; p2 = p4;
164 p3 = sp[0];
165 p4 = sp[sll];
166
167 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
168 dp[dll] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1];
169
170 pbuff[j] = 0;
171 pbuff[j + 1] = 0;
172
173 sp += 2*sll;
174 dp += 2*dll;
175 }
176
177 if (j < hsize) {
178 p0 = p2; p1 = p3; p2 = p4;
179 p3 = sp[0];
180
181 dp[0] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
182
183 pbuff[j] = 0;
184 }
185
186 } else if (kh == 3) {
187 sp += 2*sll;
188
189 for (j = 0; j <= (hsize - 2); j += 2) {
190 p0 = p2; p1 = p3;
191 p2 = sp[0];
192 p3 = sp[sll];
193
194 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
195 dp[dll] = p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1];
196
197 pbuff[j] = 0;
198 pbuff[j + 1] = 0;
199
200 sp += 2*sll;
201 dp += 2*dll;
202 }
203
204 if (j < hsize) {
205 p0 = p2; p1 = p3;
206 p2 = sp[0];
207
208 dp[0] = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
209
210 pbuff[j] = 0;
211 }
212
213 } else if (kh == 2) {
214 sp += sll;
215
216 for (j = 0; j <= (hsize - 2); j += 2) {
217 p0 = p2;
218 p1 = sp[0];
219 p2 = sp[sll];
220
221 dp[0 ] = p0*k0 + p1*k1 + pbuff[j];
222 dp[dll] = p1*k0 + p2*k1 + pbuff[j + 1];
223
224 pbuff[j] = 0;
225 pbuff[j + 1] = 0;
226
227 sp += 2*sll;
228 dp += 2*dll;
229 }
230
231 if (j < hsize) {
232 p0 = p2;
233 p1 = sp[0];
234
235 dp[0] = p0*k0 + p1*k1 + pbuff[j];
236
237 pbuff[j] = 0;
238 }
239
240 } else /* if (kh == 1) */ {
241 for (j = 0; j < hsize; j++) {
242 p0 = sp[0];
243
244 dp[0] = p0*k0 + pbuff[j];
245
246 pbuff[j] = 0;
247
248 sp += sll;
249 dp += dll;
250 }
251 }
252
253 sl += chan1;
254 dl += chan1;
255 }
256 }
257
258 sl_c += max_hsize*sll;
259 dl_c += max_hsize*dll;
260 }
261
262 if (pbuff != buff) mlib_free(pbuff);
263
264 return MLIB_SUCCESS;
265 }
266
267 /***************************************************************/
268 #define MAX_KER 7
269 #define MAX_NM 81
270
271 mlib_status CONV_FUNC(MxN)(mlib_image *dst,
272 const mlib_image *src,
273 const mlib_d64 *ker,
274 mlib_s32 m,
275 mlib_s32 n,
276 mlib_s32 dm,
277 mlib_s32 dn,
278 mlib_s32 cmask)
279 {
280 DTYPE k0, k1, k2, k3, k4, k5, k6, *sp;
281 DTYPE p0, p1, p2, p3, p4, p5, p6, p7;
282 mlib_s32 l, off, kw;
283 DEF_VARS(DTYPE);
284 mlib_s32 chan2 = chan1 + chan1;
285 mlib_s32 chan3 = chan1 + chan2;
286
287 #ifdef TYPE_DOUBLE
288 const mlib_d64 *k = ker;
289 #else
290 mlib_f32 k_arr[MAX_NM], *k = k_arr;
291
292 if (n*m > MAX_NM) {
293 k = mlib_malloc(n*m*sizeof(mlib_f32));
294
295 if (k == NULL) return MLIB_FAILURE;
296 }
297
298 for (i = 0; i < n*m; i++) k[i] = (mlib_f32)ker[i];
299 #endif /* TYPE_DOUBLE */
300
301 if (m == 1) return mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
302
303 wid -= (m - 1);
304 hgt -= (n - 1);
305 adr_dst += dn*dll + dm*chan1;
306
307 for (c = 0; c < chan1; c++) {
308 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
309
310 sl = adr_src + c;
311 dl = adr_dst + c;
312
313 for (j = 0; j < hgt; j++) {
314 const DTYPE *pk = k;
315
316 for (l = 0; l < n; l++) {
317 DTYPE *sp0 = sl + l*sll;
318
319 for (off = 0; off < m; off += kw, pk += kw, sp0 += chan1) {
320 kw = m - off;
321
322 if (kw > 2*MAX_KER) kw = MAX_KER; else
323 if (kw > MAX_KER) kw = kw/2;
324
325 p2 = sp0[0]; p3 = sp0[chan1]; p4 = sp0[chan2];
326 sp0 += chan3;
327 p5 = sp0[0]; p6 = sp0[chan1]; p7 = sp0[chan2];
328
329 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
330 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
331
332 dp = dl;
333
334 if (kw == 7) {
335 sp = sp0 += chan3;
336
337 if (pk == k) {
338 for (i = 0; i <= (wid - 2); i += 2) {
339 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
340
341 p5 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
342
343 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
344 dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
345
346 sp += chan2;
347 dp += chan2;
348 }
349
350 } else {
351 for (i = 0; i <= (wid - 2); i += 2) {
352 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
353
354 p5 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
355
356 dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
357 dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
358
359 sp += chan2;
360 dp += chan2;
361 }
362 }
363
364 } else if (kw == 6) {
365 sp = sp0 += chan2;
366
367 if (pk == k) {
368 for (i = 0; i <= (wid - 2); i += 2) {
369 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
370
371 p5 = sp[0]; p6 = sp[chan1];
372
373 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
374 dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
375
376 sp += chan2;
377 dp += chan2;
378 }
379
380 } else {
381 for (i = 0; i <= (wid - 2); i += 2) {
382 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
383
384 p5 = sp[0]; p6 = sp[chan1];
385
386 dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
387 dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
388
389 sp += chan2;
390 dp += chan2;
391 }
392 }
393
394 } else if (kw == 5) {
395 sp = sp0 += chan1;
396
397 if (pk == k) {
398 for (i = 0; i <= (wid - 2); i += 2) {
399 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
400
401 p4 = sp[0]; p5 = sp[chan1];
402
403 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
404 dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
405
406 sp += chan2;
407 dp += chan2;
408 }
409
410 } else {
411 for (i = 0; i <= (wid - 2); i += 2) {
412 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
413
414 p4 = sp[0]; p5 = sp[chan1];
415
416 dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
417 dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
418
419 sp += chan2;
420 dp += chan2;
421 }
422 }
423
424 } else if (kw == 4) {
425
426 sp = sp0;
427
428 if (pk == k) {
429 for (i = 0; i <= (wid - 2); i += 2) {
430 p0 = p2; p1 = p3; p2 = p4;
431
432 p3 = sp[0]; p4 = sp[chan1];
433
434 dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3;
435 dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3;
436
437 sp += chan2;
438 dp += chan2;
439 }
440
441 } else {
442 for (i = 0; i <= (wid - 2); i += 2) {
443 p0 = p2; p1 = p3; p2 = p4;
444
445 p3 = sp[0]; p4 = sp[chan1];
446
447 dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
448 dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
449
450 sp += chan2;
451 dp += chan2;
452 }
453 }
454
455 } else if (kw == 3) {
456 sp = sp0 -= chan1;
457
458 if (pk == k) {
459 for (i = 0; i <= (wid - 2); i += 2) {
460 p0 = p2; p1 = p3;
461
462 p2 = sp[0]; p3 = sp[chan1];
463
464 dp[0 ] = p0*k0 + p1*k1 + p2*k2;
465 dp[chan1] = p1*k0 + p2*k1 + p3*k2;
466
467 sp += chan2;
468 dp += chan2;
469 }
470
471 } else {
472 for (i = 0; i <= (wid - 2); i += 2) {
473 p0 = p2; p1 = p3;
474
475 p2 = sp[0]; p3 = sp[chan1];
476
477 dp[0 ] += p0*k0 + p1*k1 + p2*k2;
478 dp[chan1] += p1*k0 + p2*k1 + p3*k2;
479
480 sp += chan2;
481 dp += chan2;
482 }
483 }
484
485 } else { /* kw == 2 */
486 sp = sp0 -= chan2;
487
488 if (pk == k) {
489 for (i = 0; i <= (wid - 2); i += 2) {
490 p0 = p2;
491
492 p1 = sp[0]; p2 = sp[chan1];
493
494 dp[0 ] = p0*k0 + p1*k1;
495 dp[chan1] = p1*k0 + p2*k1;
496
497 sp += chan2;
498 dp += chan2;
499 }
500
501 } else {
502 for (i = 0; i <= (wid - 2); i += 2) {
503 p0 = p2;
504
505 p1 = sp[0]; p2 = sp[chan1];
506
507 dp[0 ] += p0*k0 + p1*k1;
508 dp[chan1] += p1*k0 + p2*k1;
509
510 sp += chan2;
511 dp += chan2;
512 }
513 }
514 }
515 }
516 }
517
518 /* last pixels */
519
520 if (wid & 1) {
521 DTYPE *sp0 = sl + i*chan1, s = 0;
522 const DTYPE *pk = k;
523 mlib_s32 x;
524
525 for (l = 0; l < n; l++) {
526 DTYPE *sp = sp0 + l*sll;
527
528 for (x = 0; x < m; x++) s += sp[x*chan1] * (*pk++);
529 }
530
531 dp[0] = s;
532 }
533
534 /* next line */
535 sl += sll;
536 dl += dll;
537 }
538 }
539
540 #ifndef TYPE_DOUBLE
541
542 if (k != k_arr) mlib_free(k);
543 #endif /* TYPE_DOUBLE */
544
545 return MLIB_SUCCESS;
546 }
547
548 /***************************************************************/