1 /*
2 * Copyright (c) 2000, 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 S32 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 #define CACHE_SIZE (64*1024)
39
40 /***************************************************************/
41 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_s32
42
43 /***************************************************************/
44 #ifndef MLIB_USE_FTOI_CLAMPING
45
46 #define CLAMP_S32(dst, src) \
47 if (src > (mlib_d64)MLIB_S32_MAX) src = (mlib_d64)MLIB_S32_MAX; \
48 if (src < (mlib_d64)MLIB_S32_MIN) src = (mlib_d64)MLIB_S32_MIN; \
49 dst = (mlib_s32)src
50
51 #else
52
53 #define CLAMP_S32(dst, src) dst = (mlib_s32)(src)
54
55 #endif /* MLIB_USE_FTOI_CLAMPING */
56
57 /***************************************************************/
58 #define GET_SRC_DST_PARAMETERS(type) \
59 mlib_s32 hgt = mlib_ImageGetHeight(src); \
60 mlib_s32 wid = mlib_ImageGetWidth(src); \
61 mlib_s32 sll = mlib_ImageGetStride(src) / sizeof(type); \
62 mlib_s32 dll = mlib_ImageGetStride(dst) / sizeof(type); \
63 type* adr_src = mlib_ImageGetData(src); \
64 type* adr_dst = mlib_ImageGetData(dst); \
65 mlib_s32 chan1 = mlib_ImageGetChannels(src)
66 /* mlib_s32 chan2 = chan1 + chan1 */
67
68 /***************************************************************/
69 #define DEF_VARS(type) \
70 GET_SRC_DST_PARAMETERS(type); \
71 type *sl, *sp, *sl1, *dl, *dp; \
72 mlib_d64 *pbuff = buff, *buff0, *buff1, *buff2, *buffT; \
73 mlib_s32 i, j, c; \
74 mlib_d64 scalef, d0, d1
75
76 /***************************************************************/
77 #define DEF_VARS_MxN(type) \
78 GET_SRC_DST_PARAMETERS(type); \
79 type *sl, *sp = NULL, *dl, *dp = NULL; \
80 mlib_d64 *pbuff = buff; \
81 mlib_s32 i, j, c
82
83 /***************************************************************/
84 #define FTYPE mlib_d64
85 #define DTYPE mlib_s32
86
87 #define BUFF_SIZE 1600
88
89 static mlib_status mlib_ImageConv1xN(mlib_image *dst,
90 const mlib_image *src,
91 const mlib_d64 *k,
92 mlib_s32 n,
93 mlib_s32 dn,
94 mlib_s32 cmask)
95 {
96 FTYPE buff[BUFF_SIZE];
97 mlib_s32 off, kh;
98 const FTYPE *pk;
99 FTYPE k0, k1, k2, k3, d0, d1;
100 FTYPE p0, p1, p2, p3, p4;
101 DTYPE *sl_c, *dl_c, *sl0;
102 mlib_s32 l, hsize, max_hsize;
103 DEF_VARS_MxN(DTYPE);
104
105 hgt -= (n - 1);
106 adr_dst += dn*dll;
107
108 max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll;
109
110 if (!max_hsize) max_hsize = 1;
111
112 if (max_hsize > BUFF_SIZE) {
113 pbuff = mlib_malloc(sizeof(FTYPE)*max_hsize);
114 }
115
116 sl_c = adr_src;
117 dl_c = adr_dst;
118
119 for (l = 0; l < hgt; l += hsize) {
120 hsize = hgt - l;
121
122 if (hsize > max_hsize) hsize = max_hsize;
123
124 for (c = 0; c < chan1; c++) {
125 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
126
127 sl = sl_c + c;
128 dl = dl_c + c;
129
130 for (j = 0; j < hsize; j++) pbuff[j] = 0.0;
131
132 for (i = 0; i < wid; i++) {
133 sl0 = sl;
134
135 for (off = 0; off < (n - 4); off += 4) {
136 pk = k + off;
137 sp = sl0;
138
139 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
140 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
141 sp += 3*sll;
142
143 for (j = 0; j < hsize; j += 2) {
144 p0 = p2; p1 = p3; p2 = p4;
145 p3 = sp[0];
146 p4 = sp[sll];
147
148 pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
149 pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
150
151 sp += 2*sll;
152 }
153
154 sl0 += 4*sll;
155 }
156
157 pk = k + off;
158 sp = sl0;
159
160 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
161 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
162
163 dp = dl;
164 kh = n - off;
165
166 if (kh == 4) {
167 sp += 3*sll;
168
169 for (j = 0; j <= (hsize - 2); j += 2) {
170 p0 = p2; p1 = p3; p2 = p4;
171 p3 = sp[0];
172 p4 = sp[sll];
173
174 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
175 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1];
176 CLAMP_S32(dp[0 ], d0);
177 CLAMP_S32(dp[dll], d1);
178
179 pbuff[j] = 0;
180 pbuff[j + 1] = 0;
181
182 sp += 2*sll;
183 dp += 2*dll;
184 }
185
186 if (j < hsize) {
187 p0 = p2; p1 = p3; p2 = p4;
188 p3 = sp[0];
189
190 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
191 CLAMP_S32(dp[0], d0);
192
193 pbuff[j] = 0;
194 }
195
196 } else if (kh == 3) {
197 sp += 2*sll;
198
199 for (j = 0; j <= (hsize - 2); j += 2) {
200 p0 = p2; p1 = p3;
201 p2 = sp[0];
202 p3 = sp[sll];
203
204 d0 = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
205 d1 = p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1];
206 CLAMP_S32(dp[0 ], d0);
207 CLAMP_S32(dp[dll], d1);
208
209 pbuff[j] = 0;
210 pbuff[j + 1] = 0;
211
212 sp += 2*sll;
213 dp += 2*dll;
214 }
215
216 if (j < hsize) {
217 p0 = p2; p1 = p3;
218 p2 = sp[0];
219
220 d0 = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
221 CLAMP_S32(dp[0], d0);
222
223 pbuff[j] = 0;
224 }
225
226 } else if (kh == 2) {
227 sp += sll;
228
229 for (j = 0; j <= (hsize - 2); j += 2) {
230 p0 = p2;
231 p1 = sp[0];
232 p2 = sp[sll];
233
234 d0 = p0*k0 + p1*k1 + pbuff[j];
235 d1 = p1*k0 + p2*k1 + pbuff[j + 1];
236 CLAMP_S32(dp[0 ], d0);
237 CLAMP_S32(dp[dll], d1);
238
239 pbuff[j] = 0;
240 pbuff[j + 1] = 0;
241
242 sp += 2*sll;
243 dp += 2*dll;
244 }
245
246 if (j < hsize) {
247 p0 = p2;
248 p1 = sp[0];
249
250 d0 = p0*k0 + p1*k1 + pbuff[j];
251 CLAMP_S32(dp[0], d0);
252
253 pbuff[j] = 0;
254 }
255
256 } else /* if (kh == 1) */ {
257 for (j = 0; j < hsize; j++) {
258 p0 = sp[0];
259
260 d0 = p0*k0 + pbuff[j];
261 CLAMP_S32(dp[0], d0);
262
263 pbuff[j] = 0;
264
265 sp += sll;
266 dp += dll;
267 }
268 }
269
270 sl += chan1;
271 dl += chan1;
272 }
273 }
274
275 sl_c += max_hsize*sll;
276 dl_c += max_hsize*dll;
277 }
278
279 if (pbuff != buff) mlib_free(pbuff);
280
281 return MLIB_SUCCESS;
282 }
283
284 /***************************************************************/
285 #define MAX_KER 7
286
287 #define MAX_N 15
288
289 #undef BUFF_SIZE
290 #define BUFF_SIZE 1500
291
292 mlib_status CONV_FUNC(MxN)(mlib_image *dst,
293 const mlib_image *src,
294 const mlib_s32 *kernel,
295 mlib_s32 m,
296 mlib_s32 n,
297 mlib_s32 dm,
298 mlib_s32 dn,
299 mlib_s32 scale,
300 mlib_s32 cmask)
301 {
302 mlib_d64 buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
303 mlib_d64 **buffs = buffs_arr, *buffd;
304 mlib_d64 akernel[256], *k = akernel, fscale = 1.0;
305 mlib_s32 l, off, kw, bsize, buff_ind, mn;
306 mlib_d64 d0, d1;
307 mlib_d64 k0, k1, k2, k3, k4, k5, k6;
308 mlib_d64 p0, p1, p2, p3, p4, p5, p6, p7;
309 DEF_VARS_MxN(mlib_s32);
310 mlib_s32 chan2 = chan1 + chan1;
311
312 mlib_status status = MLIB_SUCCESS;
313
314 if (scale > 30) {
315 fscale *= 1.0/(1 << 30);
316 scale -= 30;
317 }
318
319 fscale /= (1 << scale);
320
321 mn = m*n;
322
323 if (mn > 256) {
324 k = mlib_malloc(mn*sizeof(mlib_d64));
325
326 if (k == NULL) return MLIB_FAILURE;
327 }
328
329 for (i = 0; i < mn; i++) {
330 k[i] = kernel[i]*fscale;
331 }
332
333 if (m == 1) {
334 status = mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
335 FREE_AND_RETURN_STATUS;
336 }
337
338 bsize = (n + 2)*wid;
339
340 if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
341 pbuff = mlib_malloc(sizeof(mlib_d64)*bsize + sizeof(mlib_d64*)*2*(n + 1));
342
343 if (pbuff == NULL) {
344 status = MLIB_FAILURE;
345 FREE_AND_RETURN_STATUS;
346 }
347 buffs = (mlib_d64**)(pbuff + bsize);
348 }
349
350 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*wid;
351 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
352 buffd = buffs[n] + wid;
353
354 wid -= (m - 1);
355 hgt -= (n - 1);
356 adr_dst += dn*dll + dm*chan1;
357
358 for (c = 0; c < chan1; c++) {
359 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
360
361 sl = adr_src + c;
362 dl = adr_dst + c;
363
364 for (l = 0; l < n; l++) {
365 mlib_d64 *buff = buffs[l];
366
367 for (i = 0; i < wid + (m - 1); i++) {
368 buff[i] = (mlib_d64)sl[i*chan1];
369 }
370
371 sl += sll;
372 }
373
374 buff_ind = 0;
375
376 for (i = 0; i < wid; i++) buffd[i] = 0.0;
377
378 for (j = 0; j < hgt; j++) {
379 mlib_d64 **buffc = buffs + buff_ind;
380 mlib_d64 *buffn = buffc[n];
381 mlib_d64 *pk = k;
382
383 for (l = 0; l < n; l++) {
384 mlib_d64 *buff_l = buffc[l];
385
386 for (off = 0; off < m;) {
387 mlib_d64 *buff = buff_l + off;
388
389 kw = m - off;
390
391 if (kw > 2*MAX_KER) kw = MAX_KER; else
392 if (kw > MAX_KER) kw = kw/2;
393 off += kw;
394
395 sp = sl;
396 dp = dl;
397
398 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
399 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
400
401 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
402 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
403 pk += kw;
404
405 if (kw == 7) {
406
407 if (l < (n - 1) || off < m) {
408 for (i = 0; i <= (wid - 2); i += 2) {
409 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
410
411 p6 = buff[i + 6]; p7 = buff[i + 7];
412
413 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
414 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
415 }
416
417 } else {
418 for (i = 0; i <= (wid - 2); i += 2) {
419 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
420
421 p6 = buff[i + 6]; p7 = buff[i + 7];
422
423 buffn[i ] = (mlib_d64)sp[0];
424 buffn[i + 1] = (mlib_d64)sp[chan1];
425
426 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ];
427 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1];
428
429 CLAMP_S32(dp[0], d0);
430 CLAMP_S32(dp[chan1], d1);
431
432 buffd[i ] = 0.0;
433 buffd[i + 1] = 0.0;
434
435 sp += chan2;
436 dp += chan2;
437 }
438 }
439
440 } else if (kw == 6) {
441
442 if (l < (n - 1) || off < m) {
443 for (i = 0; i <= (wid - 2); i += 2) {
444 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
445
446 p5 = buff[i + 5]; p6 = buff[i + 6];
447
448 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
449 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
450 }
451
452 } else {
453 for (i = 0; i <= (wid - 2); i += 2) {
454 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
455
456 p5 = buff[i + 5]; p6 = buff[i + 6];
457
458 buffn[i ] = (mlib_d64)sp[0];
459 buffn[i + 1] = (mlib_d64)sp[chan1];
460
461 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ];
462 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1];
463
464 CLAMP_S32(dp[0], d0);
465 CLAMP_S32(dp[chan1], d1);
466
467 buffd[i ] = 0.0;
468 buffd[i + 1] = 0.0;
469
470 sp += chan2;
471 dp += chan2;
472 }
473 }
474
475 } else if (kw == 5) {
476
477 if (l < (n - 1) || off < m) {
478 for (i = 0; i <= (wid - 2); i += 2) {
479 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
480
481 p4 = buff[i + 4]; p5 = buff[i + 5];
482
483 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
484 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
485 }
486
487 } else {
488 for (i = 0; i <= (wid - 2); i += 2) {
489 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
490
491 p4 = buff[i + 4]; p5 = buff[i + 5];
492
493 buffn[i ] = (mlib_d64)sp[0];
494 buffn[i + 1] = (mlib_d64)sp[chan1];
495
496 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ];
497 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1];
498
499 CLAMP_S32(dp[0], d0);
500 CLAMP_S32(dp[chan1], d1);
501
502 buffd[i ] = 0.0;
503 buffd[i + 1] = 0.0;
504
505 sp += chan2;
506 dp += chan2;
507 }
508 }
509
510 } else if (kw == 4) {
511
512 if (l < (n - 1) || off < m) {
513 for (i = 0; i <= (wid - 2); i += 2) {
514 p0 = p2; p1 = p3; p2 = p4;
515
516 p3 = buff[i + 3]; p4 = buff[i + 4];
517
518 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
519 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
520 }
521
522 } else {
523 for (i = 0; i <= (wid - 2); i += 2) {
524 p0 = p2; p1 = p3; p2 = p4;
525
526 p3 = buff[i + 3]; p4 = buff[i + 4];
527
528 buffn[i ] = (mlib_d64)sp[0];
529 buffn[i + 1] = (mlib_d64)sp[chan1];
530
531 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ];
532 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1];
533
534 CLAMP_S32(dp[0], d0);
535 CLAMP_S32(dp[chan1], d1);
536
537 buffd[i ] = 0.0;
538 buffd[i + 1] = 0.0;
539
540 sp += chan2;
541 dp += chan2;
542 }
543 }
544
545 } else if (kw == 3) {
546
547 if (l < (n - 1) || off < m) {
548 for (i = 0; i <= (wid - 2); i += 2) {
549 p0 = p2; p1 = p3;
550
551 p2 = buff[i + 2]; p3 = buff[i + 3];
552
553 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
554 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
555 }
556
557 } else {
558 for (i = 0; i <= (wid - 2); i += 2) {
559 p0 = p2; p1 = p3;
560
561 p2 = buff[i + 2]; p3 = buff[i + 3];
562
563 buffn[i ] = (mlib_d64)sp[0];
564 buffn[i + 1] = (mlib_d64)sp[chan1];
565
566 d0 = p0*k0 + p1*k1 + p2*k2 + buffd[i ];
567 d1 = p1*k0 + p2*k1 + p3*k2 + buffd[i + 1];
568
569 CLAMP_S32(dp[0], d0);
570 CLAMP_S32(dp[chan1], d1);
571
572 buffd[i ] = 0.0;
573 buffd[i + 1] = 0.0;
574
575 sp += chan2;
576 dp += chan2;
577 }
578 }
579
580 } else { /* kw == 2 */
581
582 if (l < (n - 1) || off < m) {
583 for (i = 0; i <= (wid - 2); i += 2) {
584 p0 = p2;
585
586 p1 = buff[i + 1]; p2 = buff[i + 2];
587
588 buffd[i ] += p0*k0 + p1*k1;
589 buffd[i + 1] += p1*k0 + p2*k1;
590 }
591
592 } else {
593 for (i = 0; i <= (wid - 2); i += 2) {
594 p0 = p2;
595
596 p1 = buff[i + 1]; p2 = buff[i + 2];
597
598 buffn[i ] = (mlib_d64)sp[0];
599 buffn[i + 1] = (mlib_d64)sp[chan1];
600
601 d0 = p0*k0 + p1*k1 + buffd[i ];
602 d1 = p1*k0 + p2*k1 + buffd[i + 1];
603
604 CLAMP_S32(dp[0], d0);
605 CLAMP_S32(dp[chan1], d1);
606
607 buffd[i ] = 0.0;
608 buffd[i + 1] = 0.0;
609
610 sp += chan2;
611 dp += chan2;
612 }
613 }
614 }
615 }
616 }
617
618 /* last pixels */
619 for (; i < wid; i++) {
620 mlib_d64 *pk = k, s = 0;
621 mlib_s32 x;
622
623 for (l = 0; l < n; l++) {
624 mlib_d64 *buff = buffc[l] + i;
625
626 for (x = 0; x < m; x++) s += buff[x] * (*pk++);
627 }
628
629 CLAMP_S32(dp[0], s);
630
631 buffn[i] = (mlib_d64)sp[0];
632
633 sp += chan1;
634 dp += chan1;
635 }
636
637 for (l = 0; l < (m - 1); l++) buffn[wid + l] = sp[l*chan1];
638
639 /* next line */
640 sl += sll;
641 dl += dll;
642
643 buff_ind++;
644
645 if (buff_ind >= n + 1) buff_ind = 0;
646 }
647 }
648
649 FREE_AND_RETURN_STATUS;
650 }
651
652 /***************************************************************/