1 /*
2 * Copyright (c) 2003, 2013, 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 U8/S16/U16 type and
30 * MLIB_EDGE_SRC_EXTEND mask
31 */
32
33 #include "mlib_image.h"
34 #include "mlib_ImageConv.h"
35 #include "mlib_c_ImageConv.h"
36
37 /*
38 * This define switches between functions of different data types
39 */
40
41 #define IMG_TYPE 3
42
43 /***************************************************************/
44 #if IMG_TYPE == 1
45
46 #define DTYPE mlib_u8
47 #define CONV_FUNC(KERN) mlib_c_conv##KERN##ext_u8(PARAM)
48 #define CONV_FUNC_MxN mlib_c_convMxNext_u8(PARAM_MxN)
49 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##ext_u8(PARAM)
50 #define CONV_FUNC_MxN_I mlib_i_convMxNext_u8(PARAM_MxN)
51 #define DSCALE (1 << 24)
52 #define FROM_S32(x) (((x) >> 24) ^ 128)
53 #define S64TOS32(x) (x)
54 #define SAT_OFF -(1u << 31)
55
56 #elif IMG_TYPE == 2
57
58 #define DTYPE mlib_s16
59 #define CONV_FUNC(KERN) mlib_conv##KERN##ext_s16(PARAM)
60 #define CONV_FUNC_MxN mlib_convMxNext_s16(PARAM_MxN)
61 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##ext_s16(PARAM)
62 #define CONV_FUNC_MxN_I mlib_i_convMxNext_s16(PARAM_MxN)
63 #define DSCALE 65536.0
64 #define FROM_S32(x) ((x) >> 16)
65 #define S64TOS32(x) ((x) & 0xffffffff)
66 #define SAT_OFF
67
68 #elif IMG_TYPE == 3
69
70 #define DTYPE mlib_u16
71 #define CONV_FUNC(KERN) mlib_conv##KERN##ext_u16(PARAM)
72 #define CONV_FUNC_MxN mlib_convMxNext_u16(PARAM_MxN)
73 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##ext_u16(PARAM)
74 #define CONV_FUNC_MxN_I mlib_i_convMxNext_u16(PARAM_MxN)
75 #define DSCALE 65536.0
76 #define FROM_S32(x) (((x) >> 16) ^ 0x8000)
77 #define S64TOS32(x) (x)
78 #define SAT_OFF -(1u << 31)
79
80 #endif /* IMG_TYPE == 1 */
81
82 /***************************************************************/
83 #define PARAM \
84 mlib_image *dst, \
85 const mlib_image *src, \
86 mlib_s32 dx_l, \
87 mlib_s32 dx_r, \
88 mlib_s32 dy_t, \
89 mlib_s32 dy_b, \
90 const mlib_s32 *kern, \
91 mlib_s32 scalef_expon, \
92 mlib_s32 cmask
93
94 /***************************************************************/
95 #define PARAM_MxN \
96 mlib_image *dst, \
97 const mlib_image *src, \
98 const mlib_s32 *kernel, \
99 mlib_s32 m, \
100 mlib_s32 n, \
101 mlib_s32 dx_l, \
102 mlib_s32 dx_r, \
103 mlib_s32 dy_t, \
104 mlib_s32 dy_b, \
105 mlib_s32 scale, \
106 mlib_s32 cmask
107
108 /***************************************************************/
109 #define FTYPE mlib_d64
110
111 #ifndef MLIB_USE_FTOI_CLAMPING
112
113 #define CLAMP_S32(x) \
114 (((x) <= MLIB_S32_MIN) ? MLIB_S32_MIN : (((x) >= MLIB_S32_MAX) ? MLIB_S32_MAX : (mlib_s32)(x)))
115
116 #else
117
118 #define CLAMP_S32(x) ((mlib_s32)(x))
119
120 #endif /* MLIB_USE_FTOI_CLAMPING */
121
122 /***************************************************************/
123 #define D2I(x) CLAMP_S32((x) SAT_OFF)
124
125 /***************************************************************/
126 #ifdef _LITTLE_ENDIAN
127
128 #define STORE2(res0, res1) \
129 dp[0 ] = res1; \
130 dp[chan1] = res0
131
132 #else
133
134 #define STORE2(res0, res1) \
135 dp[0 ] = res0; \
136 dp[chan1] = res1
137
138 #endif /* _LITTLE_ENDIAN */
139
140 /***************************************************************/
141 #ifdef _NO_LONGLONG
142
143 #define LOAD_BUFF(buff) \
144 buff[i ] = sp[0]; \
145 buff[i + 1] = sp[chan1]
146
147 #else /* _NO_LONGLONG */
148
149 #ifdef _LITTLE_ENDIAN
150
151 #define LOAD_BUFF(buff) \
152 *(mlib_s64*)(buff + i) = (((mlib_s64)sp[chan1]) << 32) | S64TOS32((mlib_s64)sp[0])
153
154 #else /* _LITTLE_ENDIAN */
155
156 #define LOAD_BUFF(buff) \
157 *(mlib_s64*)(buff + i) = (((mlib_s64)sp[0]) << 32) | S64TOS32((mlib_s64)sp[chan1])
158
159 #endif /* _LITTLE_ENDIAN */
160 #endif /* _NO_LONGLONG */
161
162 /***************************************************************/
163 typedef union {
164 mlib_d64 d64;
165 struct {
166 mlib_s32 i0;
167 mlib_s32 i1;
168 } i32s;
169 } d64_2x32;
170
171 /***************************************************************/
172 #define DEF_VARS(type) \
173 type *adr_src, *sl, *sp, *sl1; \
174 type *adr_dst, *dl, *dp; \
175 FTYPE *pbuff = buff; \
176 mlib_s32 *buffi, *buffo; \
177 mlib_s32 wid, hgt, sll, dll; \
178 mlib_s32 nchannel, chan1, chan2; \
179 mlib_s32 i, j, c, swid
180
181 /***************************************************************/
182 #define GET_SRC_DST_PARAMETERS(type) \
183 hgt = mlib_ImageGetHeight(src); \
184 wid = mlib_ImageGetWidth(src); \
185 nchannel = mlib_ImageGetChannels(src); \
186 sll = mlib_ImageGetStride(src) / sizeof(type); \
187 dll = mlib_ImageGetStride(dst) / sizeof(type); \
188 adr_src = (type *)mlib_ImageGetData(src); \
189 adr_dst = (type *)mlib_ImageGetData(dst)
190
191 /***************************************************************/
192 #if IMG_TYPE == 1
193
194 /*
195 * Test for the presence of any "1" bit in bits
196 8 to 31 of val. If present, then val is either
197 negative or >255. If over/underflows of 8 bits
198 are uncommon, then this technique can be a win,
199 since only a single test, rather than two, is
200 necessary to determine if clamping is needed.
201 On the other hand, if over/underflows are common,
202 it adds an extra test.
203 */
204 #define CLAMP_STORE(dst, val) \
205 if (val & 0xffffff00) { \
206 if (val < MLIB_U8_MIN) \
207 dst = MLIB_U8_MIN; \
208 else \
209 dst = MLIB_U8_MAX; \
210 } else { \
211 dst = (mlib_u8)val; \
212 }
213
214 #elif IMG_TYPE == 2
215
216 #define CLAMP_STORE(dst, val) \
217 if (val >= MLIB_S16_MAX) \
218 dst = MLIB_S16_MAX; \
219 else if (val <= MLIB_S16_MIN) \
220 dst = MLIB_S16_MIN; \
221 else \
222 dst = (mlib_s16)val
223
224 #elif IMG_TYPE == 3
225
226 #define CLAMP_STORE(dst, val) \
227 if (val >= MLIB_U16_MAX) \
228 dst = MLIB_U16_MAX; \
229 else if (val <= MLIB_U16_MIN) \
230 dst = MLIB_U16_MIN; \
231 else \
232 dst = (mlib_u16)val
233
234 #endif /* IMG_TYPE == 1 */
235
236 /***************************************************************/
237 #define MAX_KER 7
238 #define MAX_N 15
239 #define BUFF_SIZE 1600
240 #define CACHE_SIZE (64*1024)
241
242 static mlib_status mlib_ImageConv1xN_ext(mlib_image *dst,
243 const mlib_image *src,
244 const mlib_d64 *k,
245 mlib_s32 n,
246 mlib_s32 dy_t,
247 mlib_s32 dy_b,
248 mlib_s32 cmask)
249 {
250 DTYPE *adr_src, *sl;
251 DTYPE *adr_dst, *dl, *dp;
252 FTYPE buff[BUFF_SIZE];
253 FTYPE *buffd;
254 FTYPE *pbuff = buff;
255 const FTYPE *pk;
256 FTYPE k0, k1, k2, k3;
257 FTYPE p0, p1, p2, p3, p4;
258 FTYPE *sbuff;
259 mlib_s32 l, k_off, off, bsize;
260 mlib_s32 max_hsize, smax_hsize, shgt, hsize, kh;
261 mlib_s32 d0, d1, ii;
262 mlib_s32 wid, hgt, sll, dll;
263 mlib_s32 nchannel;
264 mlib_s32 i, j, c;
265 GET_SRC_DST_PARAMETERS(DTYPE);
266
267 max_hsize = ((CACHE_SIZE/sizeof(DTYPE))/sll) - (n - 1);
268
269 if (max_hsize < 1) max_hsize = 1;
270 if (max_hsize > hgt) max_hsize = hgt;
271
272 shgt = hgt + (n - 1);
273 smax_hsize = max_hsize + (n - 1);
274
275 bsize = 2 * (smax_hsize + 1);
276
277 if (bsize > BUFF_SIZE) {
278 pbuff = mlib_malloc(sizeof(FTYPE)*bsize);
279
280 if (pbuff == NULL) return MLIB_FAILURE;
281 }
282
283 sbuff = pbuff;
284 buffd = sbuff + smax_hsize;
285
286 shgt -= (dy_t + dy_b);
287 k_off = 0;
288
289 for (l = 0; l < hgt; l += hsize) {
290 hsize = hgt - l;
291
292 if (hsize > max_hsize) hsize = max_hsize;
293
294 smax_hsize = hsize + (n - 1);
295
296 for (c = 0; c < nchannel; c++) {
297 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
298
299 sl = adr_src + c;
300 dl = adr_dst + c;
301
302 #ifdef __SUNPRO_C
303 #pragma pipeloop(0)
304 #endif /* __SUNPRO_C */
305 for (i = 0; i < hsize; i++) buffd[i] = 0.0;
306
307 for (j = 0; j < wid; j++) {
308 FTYPE *buff = sbuff;
309
310 for (i = k_off, ii = 0; (i < dy_t) && (ii < smax_hsize); i++, ii++) {
311 sbuff[i - k_off] = (FTYPE)sl[0];
312 }
313
314 #ifdef __SUNPRO_C
315 #pragma pipeloop(0)
316 #endif /* __SUNPRO_C */
317 for (; (i < shgt + dy_t) && (ii < smax_hsize); i++, ii++) {
318 sbuff[i - k_off] = (FTYPE)sl[(i - dy_t)*sll];
319 }
320
321 for (; (i < shgt + dy_t + dy_b) && (ii < smax_hsize); i++, ii++) {
322 sbuff[i - k_off] = (FTYPE)sl[(shgt - 1)*sll];
323 }
324
325 pk = k;
326
327 for (off = 0; off < (n - 4); off += 4) {
328
329 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
330 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
331
332 #ifdef __SUNPRO_C
333 #pragma pipeloop(0)
334 #endif /* __SUNPRO_C */
335 for (i = 0; i < hsize; i += 2) {
336 p0 = p2; p1 = p3; p2 = p4;
337
338 p3 = buff[i + 3]; p4 = buff[i + 4];
339
340 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
341 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
342 }
343
344 pk += 4;
345 buff += 4;
346 }
347
348 dp = dl;
349 kh = n - off;
350
351 if (kh == 4) {
352 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
353 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
354
355 #ifdef __SUNPRO_C
356 #pragma pipeloop(0)
357 #endif /* __SUNPRO_C */
358 for (i = 0; i <= (hsize - 2); i += 2) {
359 p0 = p2; p1 = p3; p2 = p4;
360
361 p3 = buff[i + 3]; p4 = buff[i + 4];
362
363 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
364 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
365
366 dp[0 ] = FROM_S32(d0);
367 dp[dll] = FROM_S32(d1);
368
369 buffd[i ] = 0.0;
370 buffd[i + 1] = 0.0;
371
372 dp += 2*dll;
373 }
374
375 if (i < hsize) {
376 p0 = p2; p1 = p3; p2 = p4;
377 p3 = buff[i + 3];
378 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i]);
379 dp[0] = FROM_S32(d0);
380 buffd[i] = 0.0;
381 }
382
383 } else if (kh == 3) {
384
385 p2 = buff[0]; p3 = buff[1];
386 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
387
388 #ifdef __SUNPRO_C
389 #pragma pipeloop(0)
390 #endif /* __SUNPRO_C */
391 for (i = 0; i <= (hsize - 2); i += 2) {
392 p0 = p2; p1 = p3;
393
394 p2 = buff[i + 2]; p3 = buff[i + 3];
395
396 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
397 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
398
399 dp[0 ] = FROM_S32(d0);
400 dp[dll] = FROM_S32(d1);
401
402 buffd[i ] = 0.0;
403 buffd[i + 1] = 0.0;
404
405 dp += 2*dll;
406 }
407
408 if (i < hsize) {
409 p0 = p2; p1 = p3;
410 p2 = buff[i + 2];
411 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i]);
412 dp[0] = FROM_S32(d0);
413
414 buffd[i] = 0.0;
415 }
416
417 } else if (kh == 2) {
418
419 p2 = buff[0];
420 k0 = pk[0]; k1 = pk[1];
421
422 #ifdef __SUNPRO_C
423 #pragma pipeloop(0)
424 #endif /* __SUNPRO_C */
425 for (i = 0; i <= (hsize - 2); i += 2) {
426 p0 = p2;
427
428 p1 = buff[i + 1]; p2 = buff[i + 2];
429
430 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
431 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
432
433 dp[0 ] = FROM_S32(d0);
434 dp[dll] = FROM_S32(d1);
435
436 buffd[i ] = 0.0;
437 buffd[i + 1] = 0.0;
438
439 dp += 2*dll;
440 }
441
442 if (i < hsize) {
443 p0 = p2;
444 p1 = buff[i + 1];
445 d0 = D2I(p0*k0 + p1*k1 + buffd[i]);
446 dp[0] = FROM_S32(d0);
447
448 buffd[i] = 0.0;
449 }
450
451 } else /* kh == 1 */{
452
453 k0 = pk[0];
454
455 #ifdef __SUNPRO_C
456 #pragma pipeloop(0)
457 #endif /* __SUNPRO_C */
458 for (i = 0; i <= (hsize - 2); i += 2) {
459 p0 = buff[i]; p1 = buff[i + 1];
460
461 d0 = D2I(p0*k0 + buffd[i ]);
462 d1 = D2I(p1*k0 + buffd[i + 1]);
463
464 dp[0 ] = FROM_S32(d0);
465 dp[dll] = FROM_S32(d1);
466
467 buffd[i ] = 0.0;
468 buffd[i + 1] = 0.0;
469
470 dp += 2*dll;
471 }
472
473 if (i < hsize) {
474 p0 = buff[i];
475 d0 = D2I(p0*k0 + buffd[i]);
476 dp[0] = FROM_S32(d0);
477
478 buffd[i] = 0.0;
479 }
480 }
481
482 /* next line */
483 sl += nchannel;
484 dl += nchannel;
485 }
486 }
487
488 k_off += max_hsize;
489 adr_dst += max_hsize*dll;
490 }
491
492 if (pbuff != buff) mlib_free(pbuff);
493
494 return MLIB_SUCCESS;
495 }
496
497 /***************************************************************/
498 mlib_status CONV_FUNC_MxN
499 {
500 DTYPE *adr_src, *sl, *sp = NULL;
501 DTYPE *adr_dst, *dl, *dp = NULL;
502 FTYPE buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
503 FTYPE **buffs = buffs_arr, *buffd;
504 FTYPE akernel[256], *k = akernel, fscale = DSCALE;
505 FTYPE *pbuff = buff;
506 FTYPE k0, k1, k2, k3, k4, k5, k6;
507 FTYPE p0, p1, p2, p3, p4, p5, p6, p7;
508 mlib_s32 *buffi;
509 mlib_s32 mn, l, off, kw, bsize, buff_ind;
510 mlib_s32 d0, d1;
511 mlib_s32 wid, hgt, sll, dll;
512 mlib_s32 nchannel, chan1, chan2;
513 mlib_s32 i, j, c, swid;
514 d64_2x32 dd;
515 mlib_status status = MLIB_SUCCESS;
516
517 GET_SRC_DST_PARAMETERS(DTYPE);
518
519 if (scale > 30) {
520 fscale *= 1.0/(1 << 30);
521 scale -= 30;
522 }
523
524 fscale /= (1 << scale);
525
526 mn = m*n;
527
528 if (mn > 256) {
529 k = mlib_malloc(mn*sizeof(mlib_d64));
530
531 if (k == NULL) return MLIB_FAILURE;
532 }
533
534 for (i = 0; i < mn; i++) {
535 k[i] = kernel[i]*fscale;
536 }
537
538 if (m == 1) {
539 status = mlib_ImageConv1xN_ext(dst, src, k, n, dy_t, dy_b, cmask);
540 FREE_AND_RETURN_STATUS;
541 }
542
543 swid = wid + (m - 1);
544
545 bsize = (n + 3)*swid;
546
547 if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
548 pbuff = mlib_malloc(sizeof(FTYPE)*bsize + sizeof(FTYPE *)*2*(n + 1));
549
550 if (pbuff == NULL) {
551 status = MLIB_FAILURE;
552 FREE_AND_RETURN_STATUS;
553 }
554 buffs = (FTYPE **)(pbuff + bsize);
555 }
556
557 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*swid;
558 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
559 buffd = buffs[n] + swid;
560 buffi = (mlib_s32*)(buffd + swid);
561
562 chan1 = nchannel;
563 chan2 = chan1 + chan1;
564
565 swid -= (dx_l + dx_r);
566
567 for (c = 0; c < nchannel; c++) {
568 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
569
570 sl = adr_src + c;
571 dl = adr_dst + c;
572
573 for (l = 0; l < n; l++) {
574 FTYPE *buff = buffs[l];
575
576 for (i = 0; i < dx_l; i++) {
577 buff[i] = (FTYPE)sl[0];
578 }
579
580 #ifdef __SUNPRO_C
581 #pragma pipeloop(0)
582 #endif /* __SUNPRO_C */
583 for (i = 0; i < swid; i++) {
584 buff[i + dx_l] = (FTYPE)sl[i*chan1];
585 }
586
587 for (i = 0; i < dx_r; i++) {
588 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
589 }
590
591 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
592 }
593
594 buff_ind = 0;
595
596 #ifdef __SUNPRO_C
597 #pragma pipeloop(0)
598 #endif /* __SUNPRO_C */
599 for (i = 0; i < wid; i++) buffd[i] = 0.0;
600
601 for (j = 0; j < hgt; j++) {
602 FTYPE **buffc = buffs + buff_ind;
603 FTYPE *buffn = buffc[n];
604 FTYPE *pk = k;
605
606 for (l = 0; l < n; l++) {
607 FTYPE *buff_l = buffc[l];
608
609 for (off = 0; off < m;) {
610 FTYPE *buff = buff_l + off;
611
612 kw = m - off;
613
614 if (kw > 2*MAX_KER) kw = MAX_KER; else
615 if (kw > MAX_KER) kw = kw/2;
616 off += kw;
617
618 sp = sl;
619 dp = dl;
620
621 if (kw == 7) {
622
623 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
624 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
625
626 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
627 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
628
629 if (l < (n - 1) || off < m) {
630 #ifdef __SUNPRO_C
631 #pragma pipeloop(0)
632 #endif /* __SUNPRO_C */
633 for (i = 0; i <= (wid - 2); i += 2) {
634 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
635
636 p6 = buff[i + 6]; p7 = buff[i + 7];
637
638 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
639 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
640 }
641
642 } else {
643 #ifdef __SUNPRO_C
644 #pragma pipeloop(0)
645 #endif /* __SUNPRO_C */
646 for (i = 0; i <= (wid - 2); i += 2) {
647 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
648
649 p6 = buff[i + 6]; p7 = buff[i + 7];
650
651 LOAD_BUFF(buffi);
652
653 dd.d64 = *(FTYPE *)(buffi + i);
654 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
655 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
656
657 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
658 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
659
660 dp[0 ] = FROM_S32(d0);
661 dp[chan1] = FROM_S32(d1);
662
663 buffd[i ] = 0.0;
664 buffd[i + 1] = 0.0;
665
666 sp += chan2;
667 dp += chan2;
668 }
669 }
670
671 } else if (kw == 6) {
672
673 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
674 p5 = buff[3]; p6 = buff[4];
675
676 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
677 k4 = pk[4]; k5 = pk[5];
678
679 if (l < (n - 1) || off < m) {
680 #ifdef __SUNPRO_C
681 #pragma pipeloop(0)
682 #endif /* __SUNPRO_C */
683 for (i = 0; i <= (wid - 2); i += 2) {
684 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
685
686 p5 = buff[i + 5]; p6 = buff[i + 6];
687
688 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
689 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
690 }
691
692 } else {
693 #ifdef __SUNPRO_C
694 #pragma pipeloop(0)
695 #endif /* __SUNPRO_C */
696 for (i = 0; i <= (wid - 2); i += 2) {
697 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
698
699 p5 = buff[i + 5]; p6 = buff[i + 6];
700
701 LOAD_BUFF(buffi);
702
703 dd.d64 = *(FTYPE *)(buffi + i);
704 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
705 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
706
707 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
708 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
709
710 dp[0 ] = FROM_S32(d0);
711 dp[chan1] = FROM_S32(d1);
712
713 buffd[i ] = 0.0;
714 buffd[i + 1] = 0.0;
715
716 sp += chan2;
717 dp += chan2;
718 }
719 }
720
721 } else if (kw == 5) {
722
723 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
724 p5 = buff[3];
725
726 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
727 k4 = pk[4];
728
729 if (l < (n - 1) || off < m) {
730 #ifdef __SUNPRO_C
731 #pragma pipeloop(0)
732 #endif /* __SUNPRO_C */
733 for (i = 0; i <= (wid - 2); i += 2) {
734 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
735
736 p4 = buff[i + 4]; p5 = buff[i + 5];
737
738 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
739 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
740 }
741
742 } else {
743 #ifdef __SUNPRO_C
744 #pragma pipeloop(0)
745 #endif /* __SUNPRO_C */
746 for (i = 0; i <= (wid - 2); i += 2) {
747 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
748
749 p4 = buff[i + 4]; p5 = buff[i + 5];
750
751 LOAD_BUFF(buffi);
752
753 dd.d64 = *(FTYPE *)(buffi + i);
754 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
755 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
756
757 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
758 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
759
760 dp[0 ] = FROM_S32(d0);
761 dp[chan1] = FROM_S32(d1);
762
763 buffd[i ] = 0.0;
764 buffd[i + 1] = 0.0;
765
766 sp += chan2;
767 dp += chan2;
768 }
769 }
770
771 } else if (kw == 4) {
772
773 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
774
775 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
776
777 if (l < (n - 1) || off < m) {
778 #ifdef __SUNPRO_C
779 #pragma pipeloop(0)
780 #endif /* __SUNPRO_C */
781 for (i = 0; i <= (wid - 2); i += 2) {
782 p0 = p2; p1 = p3; p2 = p4;
783
784 p3 = buff[i + 3]; p4 = buff[i + 4];
785
786 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
787 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
788 }
789
790 } else {
791 #ifdef __SUNPRO_C
792 #pragma pipeloop(0)
793 #endif /* __SUNPRO_C */
794 for (i = 0; i <= (wid - 2); i += 2) {
795 p0 = p2; p1 = p3; p2 = p4;
796
797 p3 = buff[i + 3]; p4 = buff[i + 4];
798
799 LOAD_BUFF(buffi);
800
801 dd.d64 = *(FTYPE *)(buffi + i);
802 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
803 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
804
805 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
806 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
807
808 dp[0 ] = FROM_S32(d0);
809 dp[chan1] = FROM_S32(d1);
810
811 buffd[i ] = 0.0;
812 buffd[i + 1] = 0.0;
813
814 sp += chan2;
815 dp += chan2;
816 }
817 }
818
819 } else if (kw == 3) {
820
821 p2 = buff[0]; p3 = buff[1];
822 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
823
824 if (l < (n - 1) || off < m) {
825 #ifdef __SUNPRO_C
826 #pragma pipeloop(0)
827 #endif /* __SUNPRO_C */
828 for (i = 0; i <= (wid - 2); i += 2) {
829 p0 = p2; p1 = p3;
830
831 p2 = buff[i + 2]; p3 = buff[i + 3];
832
833 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
834 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
835 }
836
837 } else {
838 #ifdef __SUNPRO_C
839 #pragma pipeloop(0)
840 #endif /* __SUNPRO_C */
841 for (i = 0; i <= (wid - 2); i += 2) {
842 p0 = p2; p1 = p3;
843
844 p2 = buff[i + 2]; p3 = buff[i + 3];
845
846 LOAD_BUFF(buffi);
847
848 dd.d64 = *(FTYPE *)(buffi + i);
849 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
850 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
851
852 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
853 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
854
855 dp[0 ] = FROM_S32(d0);
856 dp[chan1] = FROM_S32(d1);
857
858 buffd[i ] = 0.0;
859 buffd[i + 1] = 0.0;
860
861 sp += chan2;
862 dp += chan2;
863 }
864 }
865
866 } else /* if (kw == 2) */ {
867
868 p2 = buff[0];
869 k0 = pk[0]; k1 = pk[1];
870
871 if (l < (n - 1) || off < m) {
872 #ifdef __SUNPRO_C
873 #pragma pipeloop(0)
874 #endif /* __SUNPRO_C */
875 for (i = 0; i <= (wid - 2); i += 2) {
876 p0 = p2;
877
878 p1 = buff[i + 1]; p2 = buff[i + 2];
879
880 buffd[i ] += p0*k0 + p1*k1;
881 buffd[i + 1] += p1*k0 + p2*k1;
882 }
883
884 } else {
885 #ifdef __SUNPRO_C
886 #pragma pipeloop(0)
887 #endif /* __SUNPRO_C */
888 for (i = 0; i <= (wid - 2); i += 2) {
889 p0 = p2;
890
891 p1 = buff[i + 1]; p2 = buff[i + 2];
892
893 LOAD_BUFF(buffi);
894
895 dd.d64 = *(FTYPE *)(buffi + i);
896 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
897 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
898
899 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
900 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
901
902 dp[0 ] = FROM_S32(d0);
903 dp[chan1] = FROM_S32(d1);
904
905 buffd[i ] = 0.0;
906 buffd[i + 1] = 0.0;
907
908 sp += chan2;
909 dp += chan2;
910 }
911 }
912 }
913
914 pk += kw;
915 }
916 }
917
918 /* last pixels */
919 for (; i < wid; i++) {
920 FTYPE *pk = k, s = 0;
921 mlib_s32 x, d0;
922
923 for (l = 0; l < n; l++) {
924 FTYPE *buff = buffc[l] + i;
925
926 for (x = 0; x < m; x++) s += buff[x] * (*pk++);
927 }
928
929 d0 = D2I(s);
930 dp[0] = FROM_S32(d0);
931
932 buffn[i + dx_l] = (FTYPE)sp[0];
933
934 sp += chan1;
935 dp += chan1;
936 }
937
938 for (; i < swid; i++) {
939 buffn[i + dx_l] = (FTYPE)sp[0];
940 sp += chan1;
941 }
942
943 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
944 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
945
946 /* next line */
947
948 if (j < hgt - dy_b - 2) sl += sll;
949 dl += dll;
950
951 buff_ind++;
952
953 if (buff_ind >= n + 1) buff_ind = 0;
954 }
955 }
956
957 FREE_AND_RETURN_STATUS;
958 }
959
960 /***************************************************************/
961 #define STORE_RES(res, x) \
962 x >>= shift2; \
963 CLAMP_STORE(res, x)
964
965 mlib_status CONV_FUNC_MxN_I
966 {
967 DTYPE *adr_src, *sl, *sp = NULL;
968 DTYPE *adr_dst, *dl, *dp = NULL;
969 mlib_s32 buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
970 mlib_s32 *pbuff = buff;
971 mlib_s32 **buffs = buffs_arr, *buffd;
972 mlib_s32 l, off, kw, bsize, buff_ind;
973 mlib_s32 d0, d1, shift1, shift2;
974 mlib_s32 k0, k1, k2, k3, k4, k5, k6;
975 mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
976 mlib_s32 wid, hgt, sll, dll;
977 mlib_s32 nchannel, chan1;
978 mlib_s32 i, j, c, swid;
979 mlib_s32 chan2;
980 mlib_s32 k_locl[MAX_N*MAX_N], *k = k_locl;
981 GET_SRC_DST_PARAMETERS(DTYPE);
982
983 #if IMG_TYPE != 1
984 shift1 = 16;
985 #else
986 shift1 = 8;
987 #endif /* IMG_TYPE != 1 */
988 shift2 = scale - shift1;
989
990 chan1 = nchannel;
991 chan2 = chan1 + chan1;
992
993 swid = wid + (m - 1);
994
995 bsize = (n + 2)*swid;
996
997 if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
998 pbuff = mlib_malloc(sizeof(mlib_s32)*bsize + sizeof(mlib_s32 *)*2*(n + 1));
999
1000 if (pbuff == NULL) return MLIB_FAILURE;
1001 buffs = (mlib_s32 **)(pbuff + bsize);
1002 }
1003
1004 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*swid;
1005 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
1006 buffd = buffs[n] + swid;
1007
1008 if (m*n > MAX_N*MAX_N) {
1009 k = mlib_malloc(sizeof(mlib_s32)*(m*n));
1010
1011 if (k == NULL) {
1012 if (pbuff != buff) mlib_free(pbuff);
1013 return MLIB_FAILURE;
1014 }
1015 }
1016
1017 for (i = 0; i < m*n; i++) {
1018 k[i] = kernel[i] >> shift1;
1019 }
1020
1021 swid -= (dx_l + dx_r);
1022
1023 for (c = 0; c < nchannel; c++) {
1024 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
1025
1026 sl = adr_src + c;
1027 dl = adr_dst + c;
1028
1029 for (l = 0; l < n; l++) {
1030 mlib_s32 *buff = buffs[l];
1031
1032 for (i = 0; i < dx_l; i++) {
1033 buff[i] = (mlib_s32)sl[0];
1034 }
1035
1036 #ifdef __SUNPRO_C
1037 #pragma pipeloop(0)
1038 #endif /* __SUNPRO_C */
1039 for (i = 0; i < swid; i++) {
1040 buff[i + dx_l] = (mlib_s32)sl[i*chan1];
1041 }
1042
1043 for (i = 0; i < dx_r; i++) {
1044 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
1045 }
1046
1047 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
1048 }
1049
1050 buff_ind = 0;
1051
1052 #ifdef __SUNPRO_C
1053 #pragma pipeloop(0)
1054 #endif /* __SUNPRO_C */
1055 for (i = 0; i < wid; i++) buffd[i] = 0;
1056
1057 for (j = 0; j < hgt; j++) {
1058 mlib_s32 **buffc = buffs + buff_ind;
1059 mlib_s32 *buffn = buffc[n];
1060 mlib_s32 *pk = k;
1061
1062 for (l = 0; l < n; l++) {
1063 mlib_s32 *buff_l = buffc[l];
1064
1065 for (off = 0; off < m;) {
1066 mlib_s32 *buff = buff_l + off;
1067
1068 sp = sl;
1069 dp = dl;
1070
1071 kw = m - off;
1072
1073 if (kw > 2*MAX_KER) kw = MAX_KER; else
1074 if (kw > MAX_KER) kw = kw/2;
1075 off += kw;
1076
1077 if (kw == 7) {
1078
1079 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1080 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
1081
1082 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1083 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
1084
1085 if (l < (n - 1) || off < m) {
1086 #ifdef __SUNPRO_C
1087 #pragma pipeloop(0)
1088 #endif /* __SUNPRO_C */
1089 for (i = 0; i <= (wid - 2); i += 2) {
1090 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1091
1092 p6 = buff[i + 6]; p7 = buff[i + 7];
1093
1094 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
1095 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
1096 }
1097
1098 } else {
1099 #ifdef __SUNPRO_C
1100 #pragma pipeloop(0)
1101 #endif /* __SUNPRO_C */
1102 for (i = 0; i <= (wid - 2); i += 2) {
1103 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1104
1105 p6 = buff[i + 6]; p7 = buff[i + 7];
1106
1107 buffn[i + dx_l ] = (mlib_s32)sp[0];
1108 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1109
1110 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
1111 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
1112
1113 STORE_RES(dp[0 ], d0);
1114 STORE_RES(dp[chan1], d1);
1115
1116 buffd[i ] = 0;
1117 buffd[i + 1] = 0;
1118
1119 sp += chan2;
1120 dp += chan2;
1121 }
1122 }
1123
1124 } else if (kw == 6) {
1125
1126 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1127 p5 = buff[3]; p6 = buff[4];
1128
1129 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1130 k4 = pk[4]; k5 = pk[5];
1131
1132 if (l < (n - 1) || off < m) {
1133 #ifdef __SUNPRO_C
1134 #pragma pipeloop(0)
1135 #endif /* __SUNPRO_C */
1136 for (i = 0; i <= (wid - 2); i += 2) {
1137 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1138
1139 p5 = buff[i + 5]; p6 = buff[i + 6];
1140
1141 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
1142 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
1143 }
1144
1145 } else {
1146 #ifdef __SUNPRO_C
1147 #pragma pipeloop(0)
1148 #endif /* __SUNPRO_C */
1149 for (i = 0; i <= (wid - 2); i += 2) {
1150 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1151
1152 p5 = buff[i + 5]; p6 = buff[i + 6];
1153
1154 buffn[i + dx_l ] = (mlib_s32)sp[0];
1155 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1156
1157 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
1158 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
1159
1160 STORE_RES(dp[0 ], d0);
1161 STORE_RES(dp[chan1], d1);
1162
1163 buffd[i ] = 0;
1164 buffd[i + 1] = 0;
1165
1166 sp += chan2;
1167 dp += chan2;
1168 }
1169 }
1170
1171 } else if (kw == 5) {
1172
1173 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1174 p5 = buff[3];
1175
1176 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1177 k4 = pk[4];
1178
1179 if (l < (n - 1) || off < m) {
1180 #ifdef __SUNPRO_C
1181 #pragma pipeloop(0)
1182 #endif /* __SUNPRO_C */
1183 for (i = 0; i <= (wid - 2); i += 2) {
1184 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1185
1186 p4 = buff[i + 4]; p5 = buff[i + 5];
1187
1188 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
1189 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
1190 }
1191
1192 } else {
1193 #ifdef __SUNPRO_C
1194 #pragma pipeloop(0)
1195 #endif /* __SUNPRO_C */
1196 for (i = 0; i <= (wid - 2); i += 2) {
1197 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1198
1199 p4 = buff[i + 4]; p5 = buff[i + 5];
1200
1201 buffn[i + dx_l ] = (mlib_s32)sp[0];
1202 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1203
1204 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
1205 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
1206
1207 STORE_RES(dp[0 ], d0);
1208 STORE_RES(dp[chan1], d1);
1209
1210 buffd[i ] = 0;
1211 buffd[i + 1] = 0;
1212
1213 sp += chan2;
1214 dp += chan2;
1215 }
1216 }
1217
1218 } else if (kw == 4) {
1219
1220 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1221
1222 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1223
1224 if (l < (n - 1) || off < m) {
1225 #ifdef __SUNPRO_C
1226 #pragma pipeloop(0)
1227 #endif /* __SUNPRO_C */
1228 for (i = 0; i <= (wid - 2); i += 2) {
1229 p0 = p2; p1 = p3; p2 = p4;
1230
1231 p3 = buff[i + 3]; p4 = buff[i + 4];
1232
1233 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
1234 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
1235 }
1236
1237 } else {
1238 #ifdef __SUNPRO_C
1239 #pragma pipeloop(0)
1240 #endif /* __SUNPRO_C */
1241 for (i = 0; i <= (wid - 2); i += 2) {
1242 p0 = p2; p1 = p3; p2 = p4;
1243
1244 p3 = buff[i + 3]; p4 = buff[i + 4];
1245
1246 buffn[i + dx_l ] = (mlib_s32)sp[0];
1247 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1248
1249 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
1250 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
1251
1252 STORE_RES(dp[0 ], d0);
1253 STORE_RES(dp[chan1], d1);
1254
1255 buffd[i ] = 0;
1256 buffd[i + 1] = 0;
1257
1258 sp += chan2;
1259 dp += chan2;
1260 }
1261 }
1262
1263 } else if (kw == 3) {
1264
1265 p2 = buff[0]; p3 = buff[1];
1266 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
1267
1268 if (l < (n - 1) || off < m) {
1269 #ifdef __SUNPRO_C
1270 #pragma pipeloop(0)
1271 #endif /* __SUNPRO_C */
1272 for (i = 0; i <= (wid - 2); i += 2) {
1273 p0 = p2; p1 = p3;
1274
1275 p2 = buff[i + 2]; p3 = buff[i + 3];
1276
1277 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
1278 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
1279 }
1280
1281 } else {
1282 #ifdef __SUNPRO_C
1283 #pragma pipeloop(0)
1284 #endif /* __SUNPRO_C */
1285 for (i = 0; i <= (wid - 2); i += 2) {
1286 p0 = p2; p1 = p3;
1287
1288 p2 = buff[i + 2]; p3 = buff[i + 3];
1289
1290 buffn[i + dx_l ] = (mlib_s32)sp[0];
1291 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1292
1293 d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
1294 d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
1295
1296 STORE_RES(dp[0 ], d0);
1297 STORE_RES(dp[chan1], d1);
1298
1299 buffd[i ] = 0;
1300 buffd[i + 1] = 0;
1301
1302 sp += chan2;
1303 dp += chan2;
1304 }
1305 }
1306
1307 } else if (kw == 2) {
1308
1309 p2 = buff[0];
1310 k0 = pk[0]; k1 = pk[1];
1311
1312 if (l < (n - 1) || off < m) {
1313 #ifdef __SUNPRO_C
1314 #pragma pipeloop(0)
1315 #endif /* __SUNPRO_C */
1316 for (i = 0; i <= (wid - 2); i += 2) {
1317 p0 = p2;
1318
1319 p1 = buff[i + 1]; p2 = buff[i + 2];
1320
1321 buffd[i ] += p0*k0 + p1*k1;
1322 buffd[i + 1] += p1*k0 + p2*k1;
1323 }
1324
1325 } else {
1326 #ifdef __SUNPRO_C
1327 #pragma pipeloop(0)
1328 #endif /* __SUNPRO_C */
1329 for (i = 0; i <= (wid - 2); i += 2) {
1330 p0 = p2;
1331
1332 p1 = buff[i + 1]; p2 = buff[i + 2];
1333
1334 buffn[i + dx_l ] = (mlib_s32)sp[0];
1335 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1336
1337 d0 = (p0*k0 + p1*k1 + buffd[i ]);
1338 d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
1339
1340 STORE_RES(dp[0 ], d0);
1341 STORE_RES(dp[chan1], d1);
1342
1343 buffd[i ] = 0;
1344 buffd[i + 1] = 0;
1345
1346 sp += chan2;
1347 dp += chan2;
1348 }
1349 }
1350
1351 } else /* kw == 1 */{
1352
1353 k0 = pk[0];
1354
1355 if (l < (n - 1) || off < m) {
1356 #ifdef __SUNPRO_C
1357 #pragma pipeloop(0)
1358 #endif /* __SUNPRO_C */
1359 for (i = 0; i <= (wid - 2); i += 2) {
1360 p0 = buff[i]; p1 = buff[i + 1];
1361
1362 buffd[i ] += p0*k0;
1363 buffd[i + 1] += p1*k0;
1364 }
1365
1366 } else {
1367 #ifdef __SUNPRO_C
1368 #pragma pipeloop(0)
1369 #endif /* __SUNPRO_C */
1370 for (i = 0; i <= (wid - 2); i += 2) {
1371 p0 = buff[i]; p1 = buff[i + 1];
1372
1373 buffn[i + dx_l ] = (mlib_s32)sp[0];
1374 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1375
1376 d0 = (p0*k0 + buffd[i ]);
1377 d1 = (p1*k0 + buffd[i + 1]);
1378
1379 STORE_RES(dp[0 ], d0);
1380 STORE_RES(dp[chan1], d1);
1381
1382 buffd[i ] = 0;
1383 buffd[i + 1] = 0;
1384
1385 sp += chan2;
1386 dp += chan2;
1387 }
1388 }
1389 }
1390
1391 pk += kw;
1392 }
1393 }
1394
1395 /* last pixels */
1396 for (; i < wid; i++) {
1397 mlib_s32 *pk = k, x, s = 0;
1398
1399 for (l = 0; l < n; l++) {
1400 mlib_s32 *buff = buffc[l] + i;
1401
1402 for (x = 0; x < m; x++) s += buff[x] * (*pk++);
1403 }
1404
1405 STORE_RES(dp[0], s);
1406
1407 buffn[i + dx_l] = (mlib_s32)sp[0];
1408
1409 sp += chan1;
1410 dp += chan1;
1411 }
1412
1413 for (; i < swid; i++) {
1414 buffn[i + dx_l] = (mlib_s32)sp[0];
1415 sp += chan1;
1416 }
1417
1418 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
1419 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
1420
1421 /* next line */
1422
1423 if (j < hgt - dy_b - 2) sl += sll;
1424 dl += dll;
1425
1426 buff_ind++;
1427
1428 if (buff_ind >= n + 1) buff_ind = 0;
1429 }
1430 }
1431
1432 if (pbuff != buff) mlib_free(pbuff);
1433 if (k != k_locl) mlib_free(k);
1434
1435 return MLIB_SUCCESS;
1436 }
1437
1438 /***************************************************************/