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 #ifndef __sparc
193 #if IMG_TYPE == 1
194
195 /*
196 * Test for the presence of any "1" bit in bits
197 8 to 31 of val. If present, then val is either
198 negative or >255. If over/underflows of 8 bits
199 are uncommon, then this technique can be a win,
200 since only a single test, rather than two, is
201 necessary to determine if clamping is needed.
202 On the other hand, if over/underflows are common,
203 it adds an extra test.
204 */
205 #define CLAMP_STORE(dst, val) \
206 if (val & 0xffffff00) { \
207 if (val < MLIB_U8_MIN) \
208 dst = MLIB_U8_MIN; \
209 else \
210 dst = MLIB_U8_MAX; \
211 } else { \
212 dst = (mlib_u8)val; \
213 }
214
215 #elif IMG_TYPE == 2
216
217 #define CLAMP_STORE(dst, val) \
218 if (val >= MLIB_S16_MAX) \
219 dst = MLIB_S16_MAX; \
220 else if (val <= MLIB_S16_MIN) \
221 dst = MLIB_S16_MIN; \
222 else \
223 dst = (mlib_s16)val
224
225 #elif IMG_TYPE == 3
226
227 #define CLAMP_STORE(dst, val) \
228 if (val >= MLIB_U16_MAX) \
229 dst = MLIB_U16_MAX; \
230 else if (val <= MLIB_U16_MIN) \
231 dst = MLIB_U16_MIN; \
232 else \
233 dst = (mlib_u16)val
234
235 #endif /* IMG_TYPE == 1 */
236 #endif /* __sparc */
237
238 /***************************************************************/
239 #define MAX_KER 7
240 #define MAX_N 15
241 #define BUFF_SIZE 1600
242 #define CACHE_SIZE (64*1024)
243
mlib_ImageConv1xN_ext(mlib_image * dst,const mlib_image * src,const mlib_d64 * k,mlib_s32 n,mlib_s32 dy_t,mlib_s32 dy_b,mlib_s32 cmask)244 static mlib_status mlib_ImageConv1xN_ext(mlib_image *dst,
245 const mlib_image *src,
246 const mlib_d64 *k,
247 mlib_s32 n,
248 mlib_s32 dy_t,
249 mlib_s32 dy_b,
250 mlib_s32 cmask)
251 {
252 DTYPE *adr_src, *sl;
253 DTYPE *adr_dst, *dl, *dp;
254 FTYPE buff[BUFF_SIZE];
255 FTYPE *buffd;
256 FTYPE *pbuff = buff;
257 const FTYPE *pk;
258 FTYPE k0, k1, k2, k3;
259 FTYPE p0, p1, p2, p3, p4;
260 FTYPE *sbuff;
261 mlib_s32 l, k_off, off, bsize;
262 mlib_s32 max_hsize, smax_hsize, shgt, hsize, kh;
263 mlib_s32 d0, d1, ii;
264 mlib_s32 wid, hgt, sll, dll;
265 mlib_s32 nchannel;
266 mlib_s32 i, j, c;
267 GET_SRC_DST_PARAMETERS(DTYPE);
268
269 max_hsize = ((CACHE_SIZE/sizeof(DTYPE))/sll) - (n - 1);
270
271 if (max_hsize < 1) max_hsize = 1;
272 if (max_hsize > hgt) max_hsize = hgt;
273
274 shgt = hgt + (n - 1);
275 smax_hsize = max_hsize + (n - 1);
276
277 bsize = 2 * (smax_hsize + 1);
278
279 if (bsize > BUFF_SIZE) {
280 pbuff = mlib_malloc(sizeof(FTYPE)*bsize);
281
282 if (pbuff == NULL) return MLIB_FAILURE;
283 }
284
285 sbuff = pbuff;
286 buffd = sbuff + smax_hsize;
287
288 shgt -= (dy_t + dy_b);
289 k_off = 0;
290
291 for (l = 0; l < hgt; l += hsize) {
292 hsize = hgt - l;
293
294 if (hsize > max_hsize) hsize = max_hsize;
295
296 smax_hsize = hsize + (n - 1);
297
298 for (c = 0; c < nchannel; c++) {
299 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
300
301 sl = adr_src + c;
302 dl = adr_dst + c;
303
304 #ifdef __SUNPRO_C
305 #pragma pipeloop(0)
306 #endif /* __SUNPRO_C */
307 for (i = 0; i < hsize; i++) buffd[i] = 0.0;
308
309 for (j = 0; j < wid; j++) {
310 FTYPE *buff = sbuff;
311
312 for (i = k_off, ii = 0; (i < dy_t) && (ii < smax_hsize); i++, ii++) {
313 sbuff[i - k_off] = (FTYPE)sl[0];
314 }
315
316 #ifdef __SUNPRO_C
317 #pragma pipeloop(0)
318 #endif /* __SUNPRO_C */
319 for (; (i < shgt + dy_t) && (ii < smax_hsize); i++, ii++) {
320 sbuff[i - k_off] = (FTYPE)sl[(i - dy_t)*sll];
321 }
322
323 for (; (i < shgt + dy_t + dy_b) && (ii < smax_hsize); i++, ii++) {
324 sbuff[i - k_off] = (FTYPE)sl[(shgt - 1)*sll];
325 }
326
327 pk = k;
328
329 for (off = 0; off < (n - 4); off += 4) {
330
331 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
332 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
333
334 #ifdef __SUNPRO_C
335 #pragma pipeloop(0)
336 #endif /* __SUNPRO_C */
337 for (i = 0; i < hsize; i += 2) {
338 p0 = p2; p1 = p3; p2 = p4;
339
340 p3 = buff[i + 3]; p4 = buff[i + 4];
341
342 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
343 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
344 }
345
346 pk += 4;
347 buff += 4;
348 }
349
350 dp = dl;
351 kh = n - off;
352
353 if (kh == 4) {
354 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
355 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
356
357 #ifdef __SUNPRO_C
358 #pragma pipeloop(0)
359 #endif /* __SUNPRO_C */
360 for (i = 0; i <= (hsize - 2); i += 2) {
361 p0 = p2; p1 = p3; p2 = p4;
362
363 p3 = buff[i + 3]; p4 = buff[i + 4];
364
365 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
366 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
367
368 dp[0 ] = FROM_S32(d0);
369 dp[dll] = FROM_S32(d1);
370
371 buffd[i ] = 0.0;
372 buffd[i + 1] = 0.0;
373
374 dp += 2*dll;
375 }
376
377 if (i < hsize) {
378 p0 = p2; p1 = p3; p2 = p4;
379 p3 = buff[i + 3];
380 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i]);
381 dp[0] = FROM_S32(d0);
382 buffd[i] = 0.0;
383 }
384
385 } else if (kh == 3) {
386
387 p2 = buff[0]; p3 = buff[1];
388 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
389
390 #ifdef __SUNPRO_C
391 #pragma pipeloop(0)
392 #endif /* __SUNPRO_C */
393 for (i = 0; i <= (hsize - 2); i += 2) {
394 p0 = p2; p1 = p3;
395
396 p2 = buff[i + 2]; p3 = buff[i + 3];
397
398 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
399 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
400
401 dp[0 ] = FROM_S32(d0);
402 dp[dll] = FROM_S32(d1);
403
404 buffd[i ] = 0.0;
405 buffd[i + 1] = 0.0;
406
407 dp += 2*dll;
408 }
409
410 if (i < hsize) {
411 p0 = p2; p1 = p3;
412 p2 = buff[i + 2];
413 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i]);
414 dp[0] = FROM_S32(d0);
415
416 buffd[i] = 0.0;
417 }
418
419 } else if (kh == 2) {
420
421 p2 = buff[0];
422 k0 = pk[0]; k1 = pk[1];
423
424 #ifdef __SUNPRO_C
425 #pragma pipeloop(0)
426 #endif /* __SUNPRO_C */
427 for (i = 0; i <= (hsize - 2); i += 2) {
428 p0 = p2;
429
430 p1 = buff[i + 1]; p2 = buff[i + 2];
431
432 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
433 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
434
435 dp[0 ] = FROM_S32(d0);
436 dp[dll] = FROM_S32(d1);
437
438 buffd[i ] = 0.0;
439 buffd[i + 1] = 0.0;
440
441 dp += 2*dll;
442 }
443
444 if (i < hsize) {
445 p0 = p2;
446 p1 = buff[i + 1];
447 d0 = D2I(p0*k0 + p1*k1 + buffd[i]);
448 dp[0] = FROM_S32(d0);
449
450 buffd[i] = 0.0;
451 }
452
453 } else /* kh == 1 */{
454
455 k0 = pk[0];
456
457 #ifdef __SUNPRO_C
458 #pragma pipeloop(0)
459 #endif /* __SUNPRO_C */
460 for (i = 0; i <= (hsize - 2); i += 2) {
461 p0 = buff[i]; p1 = buff[i + 1];
462
463 d0 = D2I(p0*k0 + buffd[i ]);
464 d1 = D2I(p1*k0 + buffd[i + 1]);
465
466 dp[0 ] = FROM_S32(d0);
467 dp[dll] = FROM_S32(d1);
468
469 buffd[i ] = 0.0;
470 buffd[i + 1] = 0.0;
471
472 dp += 2*dll;
473 }
474
475 if (i < hsize) {
476 p0 = buff[i];
477 d0 = D2I(p0*k0 + buffd[i]);
478 dp[0] = FROM_S32(d0);
479
480 buffd[i] = 0.0;
481 }
482 }
483
484 /* next line */
485 sl += nchannel;
486 dl += nchannel;
487 }
488 }
489
490 k_off += max_hsize;
491 adr_dst += max_hsize*dll;
492 }
493
494 if (pbuff != buff) mlib_free(pbuff);
495
496 return MLIB_SUCCESS;
497 }
498
499 /***************************************************************/
500 mlib_status CONV_FUNC_MxN
501 {
502 DTYPE *adr_src, *sl, *sp = NULL;
503 DTYPE *adr_dst, *dl, *dp = NULL;
504 FTYPE buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
505 FTYPE **buffs = buffs_arr, *buffd;
506 FTYPE akernel[256], *k = akernel, fscale = DSCALE;
507 FTYPE *pbuff = buff;
508 FTYPE k0, k1, k2, k3, k4, k5, k6;
509 FTYPE p0, p1, p2, p3, p4, p5, p6, p7;
510 mlib_s32 *buffi;
511 mlib_s32 mn, l, off, kw, bsize, buff_ind;
512 mlib_s32 d0, d1;
513 mlib_s32 wid, hgt, sll, dll;
514 mlib_s32 nchannel, chan1, chan2;
515 mlib_s32 i, j, c, swid;
516 d64_2x32 dd;
517 mlib_status status = MLIB_SUCCESS;
518
519 GET_SRC_DST_PARAMETERS(DTYPE);
520
521 if (scale > 30) {
522 fscale *= 1.0/(1 << 30);
523 scale -= 30;
524 }
525
526 fscale /= (1 << scale);
527
528 mn = m*n;
529
530 if (mn > 256) {
531 k = mlib_malloc(mn*sizeof(mlib_d64));
532
533 if (k == NULL) return MLIB_FAILURE;
534 }
535
536 for (i = 0; i < mn; i++) {
537 k[i] = kernel[i]*fscale;
538 }
539
540 if (m == 1) {
541 status = mlib_ImageConv1xN_ext(dst, src, k, n, dy_t, dy_b, cmask);
542 FREE_AND_RETURN_STATUS;
543 }
544
545 swid = wid + (m - 1);
546
547 bsize = (n + 3)*swid;
548
549 if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
550 pbuff = mlib_malloc(sizeof(FTYPE)*bsize + sizeof(FTYPE *)*2*(n + 1));
551
552 if (pbuff == NULL) {
553 status = MLIB_FAILURE;
554 FREE_AND_RETURN_STATUS;
555 }
556 buffs = (FTYPE **)(pbuff + bsize);
557 }
558
559 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*swid;
560 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
561 buffd = buffs[n] + swid;
562 buffi = (mlib_s32*)(buffd + swid);
563
564 chan1 = nchannel;
565 chan2 = chan1 + chan1;
566
567 swid -= (dx_l + dx_r);
568
569 for (c = 0; c < nchannel; c++) {
570 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
571
572 sl = adr_src + c;
573 dl = adr_dst + c;
574
575 for (l = 0; l < n; l++) {
576 FTYPE *buff = buffs[l];
577
578 for (i = 0; i < dx_l; i++) {
579 buff[i] = (FTYPE)sl[0];
580 }
581
582 #ifdef __SUNPRO_C
583 #pragma pipeloop(0)
584 #endif /* __SUNPRO_C */
585 for (i = 0; i < swid; i++) {
586 buff[i + dx_l] = (FTYPE)sl[i*chan1];
587 }
588
589 for (i = 0; i < dx_r; i++) {
590 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
591 }
592
593 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
594 }
595
596 buff_ind = 0;
597
598 #ifdef __SUNPRO_C
599 #pragma pipeloop(0)
600 #endif /* __SUNPRO_C */
601 for (i = 0; i < wid; i++) buffd[i] = 0.0;
602
603 for (j = 0; j < hgt; j++) {
604 FTYPE **buffc = buffs + buff_ind;
605 FTYPE *buffn = buffc[n];
606 FTYPE *pk = k;
607
608 for (l = 0; l < n; l++) {
609 FTYPE *buff_l = buffc[l];
610
611 for (off = 0; off < m;) {
612 FTYPE *buff = buff_l + off;
613
614 kw = m - off;
615
616 if (kw > 2*MAX_KER) kw = MAX_KER; else
617 if (kw > MAX_KER) kw = kw/2;
618 off += kw;
619
620 sp = sl;
621 dp = dl;
622
623 if (kw == 7) {
624
625 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
626 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
627
628 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
629 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
630
631 if (l < (n - 1) || off < m) {
632 #ifdef __SUNPRO_C
633 #pragma pipeloop(0)
634 #endif /* __SUNPRO_C */
635 for (i = 0; i <= (wid - 2); i += 2) {
636 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
637
638 p6 = buff[i + 6]; p7 = buff[i + 7];
639
640 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
641 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
642 }
643
644 } else {
645 #ifdef __SUNPRO_C
646 #pragma pipeloop(0)
647 #endif /* __SUNPRO_C */
648 for (i = 0; i <= (wid - 2); i += 2) {
649 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
650
651 p6 = buff[i + 6]; p7 = buff[i + 7];
652
653 LOAD_BUFF(buffi);
654
655 dd.d64 = *(FTYPE *)(buffi + i);
656 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
657 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
658
659 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
660 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
661
662 dp[0 ] = FROM_S32(d0);
663 dp[chan1] = FROM_S32(d1);
664
665 buffd[i ] = 0.0;
666 buffd[i + 1] = 0.0;
667
668 sp += chan2;
669 dp += chan2;
670 }
671 }
672
673 } else if (kw == 6) {
674
675 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
676 p5 = buff[3]; p6 = buff[4];
677
678 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
679 k4 = pk[4]; k5 = pk[5];
680
681 if (l < (n - 1) || off < m) {
682 #ifdef __SUNPRO_C
683 #pragma pipeloop(0)
684 #endif /* __SUNPRO_C */
685 for (i = 0; i <= (wid - 2); i += 2) {
686 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
687
688 p5 = buff[i + 5]; p6 = buff[i + 6];
689
690 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
691 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
692 }
693
694 } else {
695 #ifdef __SUNPRO_C
696 #pragma pipeloop(0)
697 #endif /* __SUNPRO_C */
698 for (i = 0; i <= (wid - 2); i += 2) {
699 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
700
701 p5 = buff[i + 5]; p6 = buff[i + 6];
702
703 LOAD_BUFF(buffi);
704
705 dd.d64 = *(FTYPE *)(buffi + i);
706 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
707 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
708
709 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
710 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
711
712 dp[0 ] = FROM_S32(d0);
713 dp[chan1] = FROM_S32(d1);
714
715 buffd[i ] = 0.0;
716 buffd[i + 1] = 0.0;
717
718 sp += chan2;
719 dp += chan2;
720 }
721 }
722
723 } else if (kw == 5) {
724
725 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
726 p5 = buff[3];
727
728 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
729 k4 = pk[4];
730
731 if (l < (n - 1) || off < m) {
732 #ifdef __SUNPRO_C
733 #pragma pipeloop(0)
734 #endif /* __SUNPRO_C */
735 for (i = 0; i <= (wid - 2); i += 2) {
736 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
737
738 p4 = buff[i + 4]; p5 = buff[i + 5];
739
740 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
741 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
742 }
743
744 } else {
745 #ifdef __SUNPRO_C
746 #pragma pipeloop(0)
747 #endif /* __SUNPRO_C */
748 for (i = 0; i <= (wid - 2); i += 2) {
749 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
750
751 p4 = buff[i + 4]; p5 = buff[i + 5];
752
753 LOAD_BUFF(buffi);
754
755 dd.d64 = *(FTYPE *)(buffi + i);
756 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
757 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
758
759 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
760 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
761
762 dp[0 ] = FROM_S32(d0);
763 dp[chan1] = FROM_S32(d1);
764
765 buffd[i ] = 0.0;
766 buffd[i + 1] = 0.0;
767
768 sp += chan2;
769 dp += chan2;
770 }
771 }
772
773 } else if (kw == 4) {
774
775 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
776
777 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
778
779 if (l < (n - 1) || off < m) {
780 #ifdef __SUNPRO_C
781 #pragma pipeloop(0)
782 #endif /* __SUNPRO_C */
783 for (i = 0; i <= (wid - 2); i += 2) {
784 p0 = p2; p1 = p3; p2 = p4;
785
786 p3 = buff[i + 3]; p4 = buff[i + 4];
787
788 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
789 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
790 }
791
792 } else {
793 #ifdef __SUNPRO_C
794 #pragma pipeloop(0)
795 #endif /* __SUNPRO_C */
796 for (i = 0; i <= (wid - 2); i += 2) {
797 p0 = p2; p1 = p3; p2 = p4;
798
799 p3 = buff[i + 3]; p4 = buff[i + 4];
800
801 LOAD_BUFF(buffi);
802
803 dd.d64 = *(FTYPE *)(buffi + i);
804 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
805 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
806
807 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
808 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
809
810 dp[0 ] = FROM_S32(d0);
811 dp[chan1] = FROM_S32(d1);
812
813 buffd[i ] = 0.0;
814 buffd[i + 1] = 0.0;
815
816 sp += chan2;
817 dp += chan2;
818 }
819 }
820
821 } else if (kw == 3) {
822
823 p2 = buff[0]; p3 = buff[1];
824 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
825
826 if (l < (n - 1) || off < m) {
827 #ifdef __SUNPRO_C
828 #pragma pipeloop(0)
829 #endif /* __SUNPRO_C */
830 for (i = 0; i <= (wid - 2); i += 2) {
831 p0 = p2; p1 = p3;
832
833 p2 = buff[i + 2]; p3 = buff[i + 3];
834
835 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
836 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
837 }
838
839 } else {
840 #ifdef __SUNPRO_C
841 #pragma pipeloop(0)
842 #endif /* __SUNPRO_C */
843 for (i = 0; i <= (wid - 2); i += 2) {
844 p0 = p2; p1 = p3;
845
846 p2 = buff[i + 2]; p3 = buff[i + 3];
847
848 LOAD_BUFF(buffi);
849
850 dd.d64 = *(FTYPE *)(buffi + i);
851 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
852 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
853
854 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
855 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
856
857 dp[0 ] = FROM_S32(d0);
858 dp[chan1] = FROM_S32(d1);
859
860 buffd[i ] = 0.0;
861 buffd[i + 1] = 0.0;
862
863 sp += chan2;
864 dp += chan2;
865 }
866 }
867
868 } else /* if (kw == 2) */ {
869
870 p2 = buff[0];
871 k0 = pk[0]; k1 = pk[1];
872
873 if (l < (n - 1) || off < m) {
874 #ifdef __SUNPRO_C
875 #pragma pipeloop(0)
876 #endif /* __SUNPRO_C */
877 for (i = 0; i <= (wid - 2); i += 2) {
878 p0 = p2;
879
880 p1 = buff[i + 1]; p2 = buff[i + 2];
881
882 buffd[i ] += p0*k0 + p1*k1;
883 buffd[i + 1] += p1*k0 + p2*k1;
884 }
885
886 } else {
887 #ifdef __SUNPRO_C
888 #pragma pipeloop(0)
889 #endif /* __SUNPRO_C */
890 for (i = 0; i <= (wid - 2); i += 2) {
891 p0 = p2;
892
893 p1 = buff[i + 1]; p2 = buff[i + 2];
894
895 LOAD_BUFF(buffi);
896
897 dd.d64 = *(FTYPE *)(buffi + i);
898 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
899 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
900
901 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
902 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
903
904 dp[0 ] = FROM_S32(d0);
905 dp[chan1] = FROM_S32(d1);
906
907 buffd[i ] = 0.0;
908 buffd[i + 1] = 0.0;
909
910 sp += chan2;
911 dp += chan2;
912 }
913 }
914 }
915
916 pk += kw;
917 }
918 }
919
920 /* last pixels */
921 for (; i < wid; i++) {
922 FTYPE *pk = k, s = 0;
923 mlib_s32 x, d0;
924
925 for (l = 0; l < n; l++) {
926 FTYPE *buff = buffc[l] + i;
927
928 for (x = 0; x < m; x++) s += buff[x] * (*pk++);
929 }
930
931 d0 = D2I(s);
932 dp[0] = FROM_S32(d0);
933
934 buffn[i + dx_l] = (FTYPE)sp[0];
935
936 sp += chan1;
937 dp += chan1;
938 }
939
940 for (; i < swid; i++) {
941 buffn[i + dx_l] = (FTYPE)sp[0];
942 sp += chan1;
943 }
944
945 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
946 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
947
948 /* next line */
949
950 if (j < hgt - dy_b - 2) sl += sll;
951 dl += dll;
952
953 buff_ind++;
954
955 if (buff_ind >= n + 1) buff_ind = 0;
956 }
957 }
958
959 FREE_AND_RETURN_STATUS;
960 }
961
962 /***************************************************************/
963 #ifndef __sparc /* for x86, using integer multiplies is faster */
964
965 #define STORE_RES(res, x) \
966 x >>= shift2; \
967 CLAMP_STORE(res, x)
968
969 mlib_status CONV_FUNC_MxN_I
970 {
971 DTYPE *adr_src, *sl, *sp = NULL;
972 DTYPE *adr_dst, *dl, *dp = NULL;
973 mlib_s32 buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
974 mlib_s32 *pbuff = buff;
975 mlib_s32 **buffs = buffs_arr, *buffd;
976 mlib_s32 l, off, kw, bsize, buff_ind;
977 mlib_s32 d0, d1, shift1, shift2;
978 mlib_s32 k0, k1, k2, k3, k4, k5, k6;
979 mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
980 mlib_s32 wid, hgt, sll, dll;
981 mlib_s32 nchannel, chan1;
982 mlib_s32 i, j, c, swid;
983 mlib_s32 chan2;
984 mlib_s32 k_locl[MAX_N*MAX_N], *k = k_locl;
985 GET_SRC_DST_PARAMETERS(DTYPE);
986
987 #if IMG_TYPE != 1
988 shift1 = 16;
989 #else
990 shift1 = 8;
991 #endif /* IMG_TYPE != 1 */
992 shift2 = scale - shift1;
993
994 chan1 = nchannel;
995 chan2 = chan1 + chan1;
996
997 swid = wid + (m - 1);
998
999 bsize = (n + 2)*swid;
1000
1001 if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
1002 pbuff = mlib_malloc(sizeof(mlib_s32)*bsize + sizeof(mlib_s32 *)*2*(n + 1));
1003
1004 if (pbuff == NULL) return MLIB_FAILURE;
1005 buffs = (mlib_s32 **)(pbuff + bsize);
1006 }
1007
1008 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*swid;
1009 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
1010 buffd = buffs[n] + swid;
1011
1012 if (m*n > MAX_N*MAX_N) {
1013 k = mlib_malloc(sizeof(mlib_s32)*(m*n));
1014
1015 if (k == NULL) {
1016 if (pbuff != buff) mlib_free(pbuff);
1017 return MLIB_FAILURE;
1018 }
1019 }
1020
1021 for (i = 0; i < m*n; i++) {
1022 k[i] = kernel[i] >> shift1;
1023 }
1024
1025 swid -= (dx_l + dx_r);
1026
1027 for (c = 0; c < nchannel; c++) {
1028 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
1029
1030 sl = adr_src + c;
1031 dl = adr_dst + c;
1032
1033 for (l = 0; l < n; l++) {
1034 mlib_s32 *buff = buffs[l];
1035
1036 for (i = 0; i < dx_l; i++) {
1037 buff[i] = (mlib_s32)sl[0];
1038 }
1039
1040 #ifdef __SUNPRO_C
1041 #pragma pipeloop(0)
1042 #endif /* __SUNPRO_C */
1043 for (i = 0; i < swid; i++) {
1044 buff[i + dx_l] = (mlib_s32)sl[i*chan1];
1045 }
1046
1047 for (i = 0; i < dx_r; i++) {
1048 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
1049 }
1050
1051 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
1052 }
1053
1054 buff_ind = 0;
1055
1056 #ifdef __SUNPRO_C
1057 #pragma pipeloop(0)
1058 #endif /* __SUNPRO_C */
1059 for (i = 0; i < wid; i++) buffd[i] = 0;
1060
1061 for (j = 0; j < hgt; j++) {
1062 mlib_s32 **buffc = buffs + buff_ind;
1063 mlib_s32 *buffn = buffc[n];
1064 mlib_s32 *pk = k;
1065
1066 for (l = 0; l < n; l++) {
1067 mlib_s32 *buff_l = buffc[l];
1068
1069 for (off = 0; off < m;) {
1070 mlib_s32 *buff = buff_l + off;
1071
1072 sp = sl;
1073 dp = dl;
1074
1075 kw = m - off;
1076
1077 if (kw > 2*MAX_KER) kw = MAX_KER; else
1078 if (kw > MAX_KER) kw = kw/2;
1079 off += kw;
1080
1081 if (kw == 7) {
1082
1083 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1084 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
1085
1086 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1087 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
1088
1089 if (l < (n - 1) || off < m) {
1090 #ifdef __SUNPRO_C
1091 #pragma pipeloop(0)
1092 #endif /* __SUNPRO_C */
1093 for (i = 0; i <= (wid - 2); i += 2) {
1094 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1095
1096 p6 = buff[i + 6]; p7 = buff[i + 7];
1097
1098 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
1099 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
1100 }
1101
1102 } else {
1103 #ifdef __SUNPRO_C
1104 #pragma pipeloop(0)
1105 #endif /* __SUNPRO_C */
1106 for (i = 0; i <= (wid - 2); i += 2) {
1107 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1108
1109 p6 = buff[i + 6]; p7 = buff[i + 7];
1110
1111 buffn[i + dx_l ] = (mlib_s32)sp[0];
1112 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1113
1114 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
1115 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
1116
1117 STORE_RES(dp[0 ], d0);
1118 STORE_RES(dp[chan1], d1);
1119
1120 buffd[i ] = 0;
1121 buffd[i + 1] = 0;
1122
1123 sp += chan2;
1124 dp += chan2;
1125 }
1126 }
1127
1128 } else if (kw == 6) {
1129
1130 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1131 p5 = buff[3]; p6 = buff[4];
1132
1133 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1134 k4 = pk[4]; k5 = pk[5];
1135
1136 if (l < (n - 1) || off < m) {
1137 #ifdef __SUNPRO_C
1138 #pragma pipeloop(0)
1139 #endif /* __SUNPRO_C */
1140 for (i = 0; i <= (wid - 2); i += 2) {
1141 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1142
1143 p5 = buff[i + 5]; p6 = buff[i + 6];
1144
1145 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
1146 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
1147 }
1148
1149 } else {
1150 #ifdef __SUNPRO_C
1151 #pragma pipeloop(0)
1152 #endif /* __SUNPRO_C */
1153 for (i = 0; i <= (wid - 2); i += 2) {
1154 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1155
1156 p5 = buff[i + 5]; p6 = buff[i + 6];
1157
1158 buffn[i + dx_l ] = (mlib_s32)sp[0];
1159 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1160
1161 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
1162 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
1163
1164 STORE_RES(dp[0 ], d0);
1165 STORE_RES(dp[chan1], d1);
1166
1167 buffd[i ] = 0;
1168 buffd[i + 1] = 0;
1169
1170 sp += chan2;
1171 dp += chan2;
1172 }
1173 }
1174
1175 } else if (kw == 5) {
1176
1177 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1178 p5 = buff[3];
1179
1180 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1181 k4 = pk[4];
1182
1183 if (l < (n - 1) || off < m) {
1184 #ifdef __SUNPRO_C
1185 #pragma pipeloop(0)
1186 #endif /* __SUNPRO_C */
1187 for (i = 0; i <= (wid - 2); i += 2) {
1188 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1189
1190 p4 = buff[i + 4]; p5 = buff[i + 5];
1191
1192 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
1193 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
1194 }
1195
1196 } else {
1197 #ifdef __SUNPRO_C
1198 #pragma pipeloop(0)
1199 #endif /* __SUNPRO_C */
1200 for (i = 0; i <= (wid - 2); i += 2) {
1201 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1202
1203 p4 = buff[i + 4]; p5 = buff[i + 5];
1204
1205 buffn[i + dx_l ] = (mlib_s32)sp[0];
1206 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1207
1208 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
1209 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
1210
1211 STORE_RES(dp[0 ], d0);
1212 STORE_RES(dp[chan1], d1);
1213
1214 buffd[i ] = 0;
1215 buffd[i + 1] = 0;
1216
1217 sp += chan2;
1218 dp += chan2;
1219 }
1220 }
1221
1222 } else if (kw == 4) {
1223
1224 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1225
1226 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1227
1228 if (l < (n - 1) || off < m) {
1229 #ifdef __SUNPRO_C
1230 #pragma pipeloop(0)
1231 #endif /* __SUNPRO_C */
1232 for (i = 0; i <= (wid - 2); i += 2) {
1233 p0 = p2; p1 = p3; p2 = p4;
1234
1235 p3 = buff[i + 3]; p4 = buff[i + 4];
1236
1237 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
1238 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
1239 }
1240
1241 } else {
1242 #ifdef __SUNPRO_C
1243 #pragma pipeloop(0)
1244 #endif /* __SUNPRO_C */
1245 for (i = 0; i <= (wid - 2); i += 2) {
1246 p0 = p2; p1 = p3; p2 = p4;
1247
1248 p3 = buff[i + 3]; p4 = buff[i + 4];
1249
1250 buffn[i + dx_l ] = (mlib_s32)sp[0];
1251 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1252
1253 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
1254 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
1255
1256 STORE_RES(dp[0 ], d0);
1257 STORE_RES(dp[chan1], d1);
1258
1259 buffd[i ] = 0;
1260 buffd[i + 1] = 0;
1261
1262 sp += chan2;
1263 dp += chan2;
1264 }
1265 }
1266
1267 } else if (kw == 3) {
1268
1269 p2 = buff[0]; p3 = buff[1];
1270 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
1271
1272 if (l < (n - 1) || off < m) {
1273 #ifdef __SUNPRO_C
1274 #pragma pipeloop(0)
1275 #endif /* __SUNPRO_C */
1276 for (i = 0; i <= (wid - 2); i += 2) {
1277 p0 = p2; p1 = p3;
1278
1279 p2 = buff[i + 2]; p3 = buff[i + 3];
1280
1281 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
1282 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
1283 }
1284
1285 } else {
1286 #ifdef __SUNPRO_C
1287 #pragma pipeloop(0)
1288 #endif /* __SUNPRO_C */
1289 for (i = 0; i <= (wid - 2); i += 2) {
1290 p0 = p2; p1 = p3;
1291
1292 p2 = buff[i + 2]; p3 = buff[i + 3];
1293
1294 buffn[i + dx_l ] = (mlib_s32)sp[0];
1295 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1296
1297 d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
1298 d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
1299
1300 STORE_RES(dp[0 ], d0);
1301 STORE_RES(dp[chan1], d1);
1302
1303 buffd[i ] = 0;
1304 buffd[i + 1] = 0;
1305
1306 sp += chan2;
1307 dp += chan2;
1308 }
1309 }
1310
1311 } else if (kw == 2) {
1312
1313 p2 = buff[0];
1314 k0 = pk[0]; k1 = pk[1];
1315
1316 if (l < (n - 1) || off < m) {
1317 #ifdef __SUNPRO_C
1318 #pragma pipeloop(0)
1319 #endif /* __SUNPRO_C */
1320 for (i = 0; i <= (wid - 2); i += 2) {
1321 p0 = p2;
1322
1323 p1 = buff[i + 1]; p2 = buff[i + 2];
1324
1325 buffd[i ] += p0*k0 + p1*k1;
1326 buffd[i + 1] += p1*k0 + p2*k1;
1327 }
1328
1329 } else {
1330 #ifdef __SUNPRO_C
1331 #pragma pipeloop(0)
1332 #endif /* __SUNPRO_C */
1333 for (i = 0; i <= (wid - 2); i += 2) {
1334 p0 = p2;
1335
1336 p1 = buff[i + 1]; p2 = buff[i + 2];
1337
1338 buffn[i + dx_l ] = (mlib_s32)sp[0];
1339 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1340
1341 d0 = (p0*k0 + p1*k1 + buffd[i ]);
1342 d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
1343
1344 STORE_RES(dp[0 ], d0);
1345 STORE_RES(dp[chan1], d1);
1346
1347 buffd[i ] = 0;
1348 buffd[i + 1] = 0;
1349
1350 sp += chan2;
1351 dp += chan2;
1352 }
1353 }
1354
1355 } else /* kw == 1 */{
1356
1357 k0 = pk[0];
1358
1359 if (l < (n - 1) || off < m) {
1360 #ifdef __SUNPRO_C
1361 #pragma pipeloop(0)
1362 #endif /* __SUNPRO_C */
1363 for (i = 0; i <= (wid - 2); i += 2) {
1364 p0 = buff[i]; p1 = buff[i + 1];
1365
1366 buffd[i ] += p0*k0;
1367 buffd[i + 1] += p1*k0;
1368 }
1369
1370 } else {
1371 #ifdef __SUNPRO_C
1372 #pragma pipeloop(0)
1373 #endif /* __SUNPRO_C */
1374 for (i = 0; i <= (wid - 2); i += 2) {
1375 p0 = buff[i]; p1 = buff[i + 1];
1376
1377 buffn[i + dx_l ] = (mlib_s32)sp[0];
1378 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1379
1380 d0 = (p0*k0 + buffd[i ]);
1381 d1 = (p1*k0 + buffd[i + 1]);
1382
1383 STORE_RES(dp[0 ], d0);
1384 STORE_RES(dp[chan1], d1);
1385
1386 buffd[i ] = 0;
1387 buffd[i + 1] = 0;
1388
1389 sp += chan2;
1390 dp += chan2;
1391 }
1392 }
1393 }
1394
1395 pk += kw;
1396 }
1397 }
1398
1399 /* last pixels */
1400 for (; i < wid; i++) {
1401 mlib_s32 *pk = k, x, s = 0;
1402
1403 for (l = 0; l < n; l++) {
1404 mlib_s32 *buff = buffc[l] + i;
1405
1406 for (x = 0; x < m; x++) s += buff[x] * (*pk++);
1407 }
1408
1409 STORE_RES(dp[0], s);
1410
1411 buffn[i + dx_l] = (mlib_s32)sp[0];
1412
1413 sp += chan1;
1414 dp += chan1;
1415 }
1416
1417 for (; i < swid; i++) {
1418 buffn[i + dx_l] = (mlib_s32)sp[0];
1419 sp += chan1;
1420 }
1421
1422 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
1423 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
1424
1425 /* next line */
1426
1427 if (j < hgt - dy_b - 2) sl += sll;
1428 dl += dll;
1429
1430 buff_ind++;
1431
1432 if (buff_ind >= n + 1) buff_ind = 0;
1433 }
1434 }
1435
1436 if (pbuff != buff) mlib_free(pbuff);
1437 if (k != k_locl) mlib_free(k);
1438
1439 return MLIB_SUCCESS;
1440 }
1441
1442 #endif /* __sparc ( for x86, using integer multiplies is faster ) */
1443
1444 /***************************************************************/
1445