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
147 /***************************************************************/
148 typedef union {
149 mlib_d64 d64;
150 struct {
151 mlib_s32 i0;
152 mlib_s32 i1;
153 } i32s;
154 } d64_2x32;
155
156 /***************************************************************/
157 #define GET_SRC_DST_PARAMETERS(type) \
158 hgt = mlib_ImageGetHeight(src); \
159 wid = mlib_ImageGetWidth(src); \
160 nchannel = mlib_ImageGetChannels(src); \
161 sll = mlib_ImageGetStride(src) / sizeof(type); \
162 dll = mlib_ImageGetStride(dst) / sizeof(type); \
163 adr_src = (type *)mlib_ImageGetData(src); \
164 adr_dst = (type *)mlib_ImageGetData(dst)
165
166 /***************************************************************/
167 #ifndef __sparc
168 #if IMG_TYPE == 1
169
170 /*
171 * Test for the presence of any "1" bit in bits
172 8 to 31 of val. If present, then val is either
173 negative or >255. If over/underflows of 8 bits
174 are uncommon, then this technique can be a win,
175 since only a single test, rather than two, is
176 necessary to determine if clamping is needed.
177 On the other hand, if over/underflows are common,
178 it adds an extra test.
179 */
180 #define CLAMP_STORE(dst, val) \
181 if (val & 0xffffff00) { \
182 if (val < MLIB_U8_MIN) \
183 dst = MLIB_U8_MIN; \
184 else \
185 dst = MLIB_U8_MAX; \
186 } else { \
187 dst = (mlib_u8)val; \
191
192 #define CLAMP_STORE(dst, val) \
193 if (val >= MLIB_S16_MAX) \
194 dst = MLIB_S16_MAX; \
195 else if (val <= MLIB_S16_MIN) \
196 dst = MLIB_S16_MIN; \
197 else \
198 dst = (mlib_s16)val
199
200 #elif IMG_TYPE == 3
201
202 #define CLAMP_STORE(dst, val) \
203 if (val >= MLIB_U16_MAX) \
204 dst = MLIB_U16_MAX; \
205 else if (val <= MLIB_U16_MIN) \
206 dst = MLIB_U16_MIN; \
207 else \
208 dst = (mlib_u16)val
209
210 #endif /* IMG_TYPE == 1 */
211 #endif /* __sparc */
212
213 /***************************************************************/
214 #define MAX_KER 7
215 #define MAX_N 15
216 #define BUFF_SIZE 1600
217 #define CACHE_SIZE (64*1024)
218
219 static mlib_status mlib_ImageConv1xN_ext(mlib_image *dst,
220 const mlib_image *src,
221 const mlib_d64 *k,
222 mlib_s32 n,
223 mlib_s32 dy_t,
224 mlib_s32 dy_b,
225 mlib_s32 cmask)
226 {
227 DTYPE *adr_src, *sl;
228 DTYPE *adr_dst, *dl, *dp;
229 FTYPE buff[BUFF_SIZE];
230 FTYPE *buffd;
231 FTYPE *pbuff = buff;
259
260 sbuff = pbuff;
261 buffd = sbuff + smax_hsize;
262
263 shgt -= (dy_t + dy_b);
264 k_off = 0;
265
266 for (l = 0; l < hgt; l += hsize) {
267 hsize = hgt - l;
268
269 if (hsize > max_hsize) hsize = max_hsize;
270
271 smax_hsize = hsize + (n - 1);
272
273 for (c = 0; c < nchannel; c++) {
274 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
275
276 sl = adr_src + c;
277 dl = adr_dst + c;
278
279 #ifdef __SUNPRO_C
280 #pragma pipeloop(0)
281 #endif /* __SUNPRO_C */
282 for (i = 0; i < hsize; i++) buffd[i] = 0.0;
283
284 for (j = 0; j < wid; j++) {
285 FTYPE *buff = sbuff;
286
287 for (i = k_off, ii = 0; (i < dy_t) && (ii < smax_hsize); i++, ii++) {
288 sbuff[i - k_off] = (FTYPE)sl[0];
289 }
290
291 #ifdef __SUNPRO_C
292 #pragma pipeloop(0)
293 #endif /* __SUNPRO_C */
294 for (; (i < shgt + dy_t) && (ii < smax_hsize); i++, ii++) {
295 sbuff[i - k_off] = (FTYPE)sl[(i - dy_t)*sll];
296 }
297
298 for (; (i < shgt + dy_t + dy_b) && (ii < smax_hsize); i++, ii++) {
299 sbuff[i - k_off] = (FTYPE)sl[(shgt - 1)*sll];
300 }
301
302 pk = k;
303
304 for (off = 0; off < (n - 4); off += 4) {
305
306 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
307 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
308
309 #ifdef __SUNPRO_C
310 #pragma pipeloop(0)
311 #endif /* __SUNPRO_C */
312 for (i = 0; i < hsize; i += 2) {
313 p0 = p2; p1 = p3; p2 = p4;
314
315 p3 = buff[i + 3]; p4 = buff[i + 4];
316
317 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
318 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
319 }
320
321 pk += 4;
322 buff += 4;
323 }
324
325 dp = dl;
326 kh = n - off;
327
328 if (kh == 4) {
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 - 2); i += 2) {
336 p0 = p2; p1 = p3; p2 = p4;
337
338 p3 = buff[i + 3]; p4 = buff[i + 4];
339
340 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
341 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
342
343 dp[0 ] = FROM_S32(d0);
344 dp[dll] = FROM_S32(d1);
345
346 buffd[i ] = 0.0;
347 buffd[i + 1] = 0.0;
348
349 dp += 2*dll;
350 }
351
352 if (i < hsize) {
353 p0 = p2; p1 = p3; p2 = p4;
354 p3 = buff[i + 3];
355 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i]);
356 dp[0] = FROM_S32(d0);
357 buffd[i] = 0.0;
358 }
359
360 } else if (kh == 3) {
361
362 p2 = buff[0]; p3 = buff[1];
363 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
364
365 #ifdef __SUNPRO_C
366 #pragma pipeloop(0)
367 #endif /* __SUNPRO_C */
368 for (i = 0; i <= (hsize - 2); i += 2) {
369 p0 = p2; p1 = p3;
370
371 p2 = buff[i + 2]; p3 = buff[i + 3];
372
373 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
374 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
375
376 dp[0 ] = FROM_S32(d0);
377 dp[dll] = FROM_S32(d1);
378
379 buffd[i ] = 0.0;
380 buffd[i + 1] = 0.0;
381
382 dp += 2*dll;
383 }
384
385 if (i < hsize) {
386 p0 = p2; p1 = p3;
387 p2 = buff[i + 2];
388 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i]);
389 dp[0] = FROM_S32(d0);
390
391 buffd[i] = 0.0;
392 }
393
394 } else if (kh == 2) {
395
396 p2 = buff[0];
397 k0 = pk[0]; k1 = pk[1];
398
399 #ifdef __SUNPRO_C
400 #pragma pipeloop(0)
401 #endif /* __SUNPRO_C */
402 for (i = 0; i <= (hsize - 2); i += 2) {
403 p0 = p2;
404
405 p1 = buff[i + 1]; p2 = buff[i + 2];
406
407 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
408 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
409
410 dp[0 ] = FROM_S32(d0);
411 dp[dll] = FROM_S32(d1);
412
413 buffd[i ] = 0.0;
414 buffd[i + 1] = 0.0;
415
416 dp += 2*dll;
417 }
418
419 if (i < hsize) {
420 p0 = p2;
421 p1 = buff[i + 1];
422 d0 = D2I(p0*k0 + p1*k1 + buffd[i]);
423 dp[0] = FROM_S32(d0);
424
425 buffd[i] = 0.0;
426 }
427
428 } else /* kh == 1 */{
429
430 k0 = pk[0];
431
432 #ifdef __SUNPRO_C
433 #pragma pipeloop(0)
434 #endif /* __SUNPRO_C */
435 for (i = 0; i <= (hsize - 2); i += 2) {
436 p0 = buff[i]; p1 = buff[i + 1];
437
438 d0 = D2I(p0*k0 + buffd[i ]);
439 d1 = D2I(p1*k0 + buffd[i + 1]);
440
441 dp[0 ] = FROM_S32(d0);
442 dp[dll] = FROM_S32(d1);
443
444 buffd[i ] = 0.0;
445 buffd[i + 1] = 0.0;
446
447 dp += 2*dll;
448 }
449
450 if (i < hsize) {
451 p0 = buff[i];
452 d0 = D2I(p0*k0 + buffd[i]);
453 dp[0] = FROM_S32(d0);
454
537 buffi = (mlib_s32*)(buffd + swid);
538
539 chan1 = nchannel;
540 chan2 = chan1 + chan1;
541
542 swid -= (dx_l + dx_r);
543
544 for (c = 0; c < nchannel; c++) {
545 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
546
547 sl = adr_src + c;
548 dl = adr_dst + c;
549
550 for (l = 0; l < n; l++) {
551 FTYPE *buff = buffs[l];
552
553 for (i = 0; i < dx_l; i++) {
554 buff[i] = (FTYPE)sl[0];
555 }
556
557 #ifdef __SUNPRO_C
558 #pragma pipeloop(0)
559 #endif /* __SUNPRO_C */
560 for (i = 0; i < swid; i++) {
561 buff[i + dx_l] = (FTYPE)sl[i*chan1];
562 }
563
564 for (i = 0; i < dx_r; i++) {
565 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
566 }
567
568 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
569 }
570
571 buff_ind = 0;
572
573 #ifdef __SUNPRO_C
574 #pragma pipeloop(0)
575 #endif /* __SUNPRO_C */
576 for (i = 0; i < wid; i++) buffd[i] = 0.0;
577
578 for (j = 0; j < hgt; j++) {
579 FTYPE **buffc = buffs + buff_ind;
580 FTYPE *buffn = buffc[n];
581 FTYPE *pk = k;
582
583 for (l = 0; l < n; l++) {
584 FTYPE *buff_l = buffc[l];
585
586 for (off = 0; off < m;) {
587 FTYPE *buff = buff_l + off;
588
589 kw = m - off;
590
591 if (kw > 2*MAX_KER) kw = MAX_KER; else
592 if (kw > MAX_KER) kw = kw/2;
593 off += kw;
594
595 sp = sl;
596 dp = dl;
597
598 if (kw == 7) {
599
600 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
601 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
602
603 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
604 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
605
606 if (l < (n - 1) || off < m) {
607 #ifdef __SUNPRO_C
608 #pragma pipeloop(0)
609 #endif /* __SUNPRO_C */
610 for (i = 0; i <= (wid - 2); i += 2) {
611 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
612
613 p6 = buff[i + 6]; p7 = buff[i + 7];
614
615 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
616 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
617 }
618
619 } else {
620 #ifdef __SUNPRO_C
621 #pragma pipeloop(0)
622 #endif /* __SUNPRO_C */
623 for (i = 0; i <= (wid - 2); i += 2) {
624 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
625
626 p6 = buff[i + 6]; p7 = buff[i + 7];
627
628 LOAD_BUFF(buffi);
629
630 dd.d64 = *(FTYPE *)(buffi + i);
631 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
632 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
633
634 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
635 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
636
637 dp[0 ] = FROM_S32(d0);
638 dp[chan1] = FROM_S32(d1);
639
640 buffd[i ] = 0.0;
641 buffd[i + 1] = 0.0;
642
643 sp += chan2;
644 dp += chan2;
645 }
646 }
647
648 } else if (kw == 6) {
649
650 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
651 p5 = buff[3]; p6 = buff[4];
652
653 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
654 k4 = pk[4]; k5 = pk[5];
655
656 if (l < (n - 1) || off < m) {
657 #ifdef __SUNPRO_C
658 #pragma pipeloop(0)
659 #endif /* __SUNPRO_C */
660 for (i = 0; i <= (wid - 2); i += 2) {
661 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
662
663 p5 = buff[i + 5]; p6 = buff[i + 6];
664
665 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
666 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
667 }
668
669 } else {
670 #ifdef __SUNPRO_C
671 #pragma pipeloop(0)
672 #endif /* __SUNPRO_C */
673 for (i = 0; i <= (wid - 2); i += 2) {
674 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
675
676 p5 = buff[i + 5]; p6 = buff[i + 6];
677
678 LOAD_BUFF(buffi);
679
680 dd.d64 = *(FTYPE *)(buffi + i);
681 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
682 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
683
684 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
685 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
686
687 dp[0 ] = FROM_S32(d0);
688 dp[chan1] = FROM_S32(d1);
689
690 buffd[i ] = 0.0;
691 buffd[i + 1] = 0.0;
692
693 sp += chan2;
694 dp += chan2;
695 }
696 }
697
698 } else if (kw == 5) {
699
700 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
701 p5 = buff[3];
702
703 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
704 k4 = pk[4];
705
706 if (l < (n - 1) || off < m) {
707 #ifdef __SUNPRO_C
708 #pragma pipeloop(0)
709 #endif /* __SUNPRO_C */
710 for (i = 0; i <= (wid - 2); i += 2) {
711 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
712
713 p4 = buff[i + 4]; p5 = buff[i + 5];
714
715 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
716 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
717 }
718
719 } else {
720 #ifdef __SUNPRO_C
721 #pragma pipeloop(0)
722 #endif /* __SUNPRO_C */
723 for (i = 0; i <= (wid - 2); i += 2) {
724 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
725
726 p4 = buff[i + 4]; p5 = buff[i + 5];
727
728 LOAD_BUFF(buffi);
729
730 dd.d64 = *(FTYPE *)(buffi + i);
731 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
732 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
733
734 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
735 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
736
737 dp[0 ] = FROM_S32(d0);
738 dp[chan1] = FROM_S32(d1);
739
740 buffd[i ] = 0.0;
741 buffd[i + 1] = 0.0;
742
743 sp += chan2;
744 dp += chan2;
745 }
746 }
747
748 } else if (kw == 4) {
749
750 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
751
752 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
753
754 if (l < (n - 1) || off < m) {
755 #ifdef __SUNPRO_C
756 #pragma pipeloop(0)
757 #endif /* __SUNPRO_C */
758 for (i = 0; i <= (wid - 2); i += 2) {
759 p0 = p2; p1 = p3; p2 = p4;
760
761 p3 = buff[i + 3]; p4 = buff[i + 4];
762
763 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
764 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
765 }
766
767 } else {
768 #ifdef __SUNPRO_C
769 #pragma pipeloop(0)
770 #endif /* __SUNPRO_C */
771 for (i = 0; i <= (wid - 2); i += 2) {
772 p0 = p2; p1 = p3; p2 = p4;
773
774 p3 = buff[i + 3]; p4 = buff[i + 4];
775
776 LOAD_BUFF(buffi);
777
778 dd.d64 = *(FTYPE *)(buffi + i);
779 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
780 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
781
782 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
783 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
784
785 dp[0 ] = FROM_S32(d0);
786 dp[chan1] = FROM_S32(d1);
787
788 buffd[i ] = 0.0;
789 buffd[i + 1] = 0.0;
790
791 sp += chan2;
792 dp += chan2;
793 }
794 }
795
796 } else if (kw == 3) {
797
798 p2 = buff[0]; p3 = buff[1];
799 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
800
801 if (l < (n - 1) || off < m) {
802 #ifdef __SUNPRO_C
803 #pragma pipeloop(0)
804 #endif /* __SUNPRO_C */
805 for (i = 0; i <= (wid - 2); i += 2) {
806 p0 = p2; p1 = p3;
807
808 p2 = buff[i + 2]; p3 = buff[i + 3];
809
810 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
811 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
812 }
813
814 } else {
815 #ifdef __SUNPRO_C
816 #pragma pipeloop(0)
817 #endif /* __SUNPRO_C */
818 for (i = 0; i <= (wid - 2); i += 2) {
819 p0 = p2; p1 = p3;
820
821 p2 = buff[i + 2]; p3 = buff[i + 3];
822
823 LOAD_BUFF(buffi);
824
825 dd.d64 = *(FTYPE *)(buffi + i);
826 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
827 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
828
829 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
830 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
831
832 dp[0 ] = FROM_S32(d0);
833 dp[chan1] = FROM_S32(d1);
834
835 buffd[i ] = 0.0;
836 buffd[i + 1] = 0.0;
837
838 sp += chan2;
839 dp += chan2;
840 }
841 }
842
843 } else /* if (kw == 2) */ {
844
845 p2 = buff[0];
846 k0 = pk[0]; k1 = pk[1];
847
848 if (l < (n - 1) || off < m) {
849 #ifdef __SUNPRO_C
850 #pragma pipeloop(0)
851 #endif /* __SUNPRO_C */
852 for (i = 0; i <= (wid - 2); i += 2) {
853 p0 = p2;
854
855 p1 = buff[i + 1]; p2 = buff[i + 2];
856
857 buffd[i ] += p0*k0 + p1*k1;
858 buffd[i + 1] += p1*k0 + p2*k1;
859 }
860
861 } else {
862 #ifdef __SUNPRO_C
863 #pragma pipeloop(0)
864 #endif /* __SUNPRO_C */
865 for (i = 0; i <= (wid - 2); i += 2) {
866 p0 = p2;
867
868 p1 = buff[i + 1]; p2 = buff[i + 2];
869
870 LOAD_BUFF(buffi);
871
872 dd.d64 = *(FTYPE *)(buffi + i);
873 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
874 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
875
876 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
877 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
878
879 dp[0 ] = FROM_S32(d0);
880 dp[chan1] = FROM_S32(d1);
881
882 buffd[i ] = 0.0;
883 buffd[i + 1] = 0.0;
884
918 }
919
920 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
921 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
922
923 /* next line */
924
925 if (j < hgt - dy_b - 2) sl += sll;
926 dl += dll;
927
928 buff_ind++;
929
930 if (buff_ind >= n + 1) buff_ind = 0;
931 }
932 }
933
934 FREE_AND_RETURN_STATUS;
935 }
936
937 /***************************************************************/
938 #ifndef __sparc /* for x86, using integer multiplies is faster */
939
940 #define STORE_RES(res, x) \
941 x >>= shift2; \
942 CLAMP_STORE(res, x)
943
944 mlib_status CONV_FUNC_MxN_I
945 {
946 DTYPE *adr_src, *sl, *sp = NULL;
947 DTYPE *adr_dst, *dl, *dp = NULL;
948 mlib_s32 buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
949 mlib_s32 *pbuff = buff;
950 mlib_s32 **buffs = buffs_arr, *buffd;
951 mlib_s32 l, off, kw, bsize, buff_ind;
952 mlib_s32 d0, d1, shift1, shift2;
953 mlib_s32 k0, k1, k2, k3, k4, k5, k6;
954 mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
955 mlib_s32 wid, hgt, sll, dll;
956 mlib_s32 nchannel, chan1;
957 mlib_s32 i, j, c, swid;
958 mlib_s32 chan2;
995
996 for (i = 0; i < m*n; i++) {
997 k[i] = kernel[i] >> shift1;
998 }
999
1000 swid -= (dx_l + dx_r);
1001
1002 for (c = 0; c < nchannel; c++) {
1003 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
1004
1005 sl = adr_src + c;
1006 dl = adr_dst + c;
1007
1008 for (l = 0; l < n; l++) {
1009 mlib_s32 *buff = buffs[l];
1010
1011 for (i = 0; i < dx_l; i++) {
1012 buff[i] = (mlib_s32)sl[0];
1013 }
1014
1015 #ifdef __SUNPRO_C
1016 #pragma pipeloop(0)
1017 #endif /* __SUNPRO_C */
1018 for (i = 0; i < swid; i++) {
1019 buff[i + dx_l] = (mlib_s32)sl[i*chan1];
1020 }
1021
1022 for (i = 0; i < dx_r; i++) {
1023 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
1024 }
1025
1026 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
1027 }
1028
1029 buff_ind = 0;
1030
1031 #ifdef __SUNPRO_C
1032 #pragma pipeloop(0)
1033 #endif /* __SUNPRO_C */
1034 for (i = 0; i < wid; i++) buffd[i] = 0;
1035
1036 for (j = 0; j < hgt; j++) {
1037 mlib_s32 **buffc = buffs + buff_ind;
1038 mlib_s32 *buffn = buffc[n];
1039 mlib_s32 *pk = k;
1040
1041 for (l = 0; l < n; l++) {
1042 mlib_s32 *buff_l = buffc[l];
1043
1044 for (off = 0; off < m;) {
1045 mlib_s32 *buff = buff_l + off;
1046
1047 sp = sl;
1048 dp = dl;
1049
1050 kw = m - off;
1051
1052 if (kw > 2*MAX_KER) kw = MAX_KER; else
1053 if (kw > MAX_KER) kw = kw/2;
1054 off += kw;
1055
1056 if (kw == 7) {
1057
1058 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1059 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
1060
1061 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1062 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
1063
1064 if (l < (n - 1) || off < m) {
1065 #ifdef __SUNPRO_C
1066 #pragma pipeloop(0)
1067 #endif /* __SUNPRO_C */
1068 for (i = 0; i <= (wid - 2); i += 2) {
1069 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1070
1071 p6 = buff[i + 6]; p7 = buff[i + 7];
1072
1073 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
1074 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
1075 }
1076
1077 } else {
1078 #ifdef __SUNPRO_C
1079 #pragma pipeloop(0)
1080 #endif /* __SUNPRO_C */
1081 for (i = 0; i <= (wid - 2); i += 2) {
1082 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1083
1084 p6 = buff[i + 6]; p7 = buff[i + 7];
1085
1086 buffn[i + dx_l ] = (mlib_s32)sp[0];
1087 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1088
1089 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
1090 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
1091
1092 STORE_RES(dp[0 ], d0);
1093 STORE_RES(dp[chan1], d1);
1094
1095 buffd[i ] = 0;
1096 buffd[i + 1] = 0;
1097
1098 sp += chan2;
1099 dp += chan2;
1100 }
1101 }
1102
1103 } else if (kw == 6) {
1104
1105 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1106 p5 = buff[3]; p6 = buff[4];
1107
1108 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1109 k4 = pk[4]; k5 = pk[5];
1110
1111 if (l < (n - 1) || off < m) {
1112 #ifdef __SUNPRO_C
1113 #pragma pipeloop(0)
1114 #endif /* __SUNPRO_C */
1115 for (i = 0; i <= (wid - 2); i += 2) {
1116 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1117
1118 p5 = buff[i + 5]; p6 = buff[i + 6];
1119
1120 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
1121 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
1122 }
1123
1124 } else {
1125 #ifdef __SUNPRO_C
1126 #pragma pipeloop(0)
1127 #endif /* __SUNPRO_C */
1128 for (i = 0; i <= (wid - 2); i += 2) {
1129 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1130
1131 p5 = buff[i + 5]; p6 = buff[i + 6];
1132
1133 buffn[i + dx_l ] = (mlib_s32)sp[0];
1134 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1135
1136 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
1137 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
1138
1139 STORE_RES(dp[0 ], d0);
1140 STORE_RES(dp[chan1], d1);
1141
1142 buffd[i ] = 0;
1143 buffd[i + 1] = 0;
1144
1145 sp += chan2;
1146 dp += chan2;
1147 }
1148 }
1149
1150 } else if (kw == 5) {
1151
1152 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1153 p5 = buff[3];
1154
1155 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1156 k4 = pk[4];
1157
1158 if (l < (n - 1) || off < m) {
1159 #ifdef __SUNPRO_C
1160 #pragma pipeloop(0)
1161 #endif /* __SUNPRO_C */
1162 for (i = 0; i <= (wid - 2); i += 2) {
1163 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1164
1165 p4 = buff[i + 4]; p5 = buff[i + 5];
1166
1167 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
1168 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
1169 }
1170
1171 } else {
1172 #ifdef __SUNPRO_C
1173 #pragma pipeloop(0)
1174 #endif /* __SUNPRO_C */
1175 for (i = 0; i <= (wid - 2); i += 2) {
1176 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1177
1178 p4 = buff[i + 4]; p5 = buff[i + 5];
1179
1180 buffn[i + dx_l ] = (mlib_s32)sp[0];
1181 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1182
1183 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
1184 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
1185
1186 STORE_RES(dp[0 ], d0);
1187 STORE_RES(dp[chan1], d1);
1188
1189 buffd[i ] = 0;
1190 buffd[i + 1] = 0;
1191
1192 sp += chan2;
1193 dp += chan2;
1194 }
1195 }
1196
1197 } else if (kw == 4) {
1198
1199 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1200
1201 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1202
1203 if (l < (n - 1) || off < m) {
1204 #ifdef __SUNPRO_C
1205 #pragma pipeloop(0)
1206 #endif /* __SUNPRO_C */
1207 for (i = 0; i <= (wid - 2); i += 2) {
1208 p0 = p2; p1 = p3; p2 = p4;
1209
1210 p3 = buff[i + 3]; p4 = buff[i + 4];
1211
1212 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
1213 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
1214 }
1215
1216 } else {
1217 #ifdef __SUNPRO_C
1218 #pragma pipeloop(0)
1219 #endif /* __SUNPRO_C */
1220 for (i = 0; i <= (wid - 2); i += 2) {
1221 p0 = p2; p1 = p3; p2 = p4;
1222
1223 p3 = buff[i + 3]; p4 = buff[i + 4];
1224
1225 buffn[i + dx_l ] = (mlib_s32)sp[0];
1226 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1227
1228 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
1229 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
1230
1231 STORE_RES(dp[0 ], d0);
1232 STORE_RES(dp[chan1], d1);
1233
1234 buffd[i ] = 0;
1235 buffd[i + 1] = 0;
1236
1237 sp += chan2;
1238 dp += chan2;
1239 }
1240 }
1241
1242 } else if (kw == 3) {
1243
1244 p2 = buff[0]; p3 = buff[1];
1245 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
1246
1247 if (l < (n - 1) || off < m) {
1248 #ifdef __SUNPRO_C
1249 #pragma pipeloop(0)
1250 #endif /* __SUNPRO_C */
1251 for (i = 0; i <= (wid - 2); i += 2) {
1252 p0 = p2; p1 = p3;
1253
1254 p2 = buff[i + 2]; p3 = buff[i + 3];
1255
1256 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
1257 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
1258 }
1259
1260 } else {
1261 #ifdef __SUNPRO_C
1262 #pragma pipeloop(0)
1263 #endif /* __SUNPRO_C */
1264 for (i = 0; i <= (wid - 2); i += 2) {
1265 p0 = p2; p1 = p3;
1266
1267 p2 = buff[i + 2]; p3 = buff[i + 3];
1268
1269 buffn[i + dx_l ] = (mlib_s32)sp[0];
1270 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1271
1272 d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
1273 d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
1274
1275 STORE_RES(dp[0 ], d0);
1276 STORE_RES(dp[chan1], d1);
1277
1278 buffd[i ] = 0;
1279 buffd[i + 1] = 0;
1280
1281 sp += chan2;
1282 dp += chan2;
1283 }
1284 }
1285
1286 } else if (kw == 2) {
1287
1288 p2 = buff[0];
1289 k0 = pk[0]; k1 = pk[1];
1290
1291 if (l < (n - 1) || off < m) {
1292 #ifdef __SUNPRO_C
1293 #pragma pipeloop(0)
1294 #endif /* __SUNPRO_C */
1295 for (i = 0; i <= (wid - 2); i += 2) {
1296 p0 = p2;
1297
1298 p1 = buff[i + 1]; p2 = buff[i + 2];
1299
1300 buffd[i ] += p0*k0 + p1*k1;
1301 buffd[i + 1] += p1*k0 + p2*k1;
1302 }
1303
1304 } else {
1305 #ifdef __SUNPRO_C
1306 #pragma pipeloop(0)
1307 #endif /* __SUNPRO_C */
1308 for (i = 0; i <= (wid - 2); i += 2) {
1309 p0 = p2;
1310
1311 p1 = buff[i + 1]; p2 = buff[i + 2];
1312
1313 buffn[i + dx_l ] = (mlib_s32)sp[0];
1314 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1315
1316 d0 = (p0*k0 + p1*k1 + buffd[i ]);
1317 d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
1318
1319 STORE_RES(dp[0 ], d0);
1320 STORE_RES(dp[chan1], d1);
1321
1322 buffd[i ] = 0;
1323 buffd[i + 1] = 0;
1324
1325 sp += chan2;
1326 dp += chan2;
1327 }
1328 }
1329
1330 } else /* kw == 1 */{
1331
1332 k0 = pk[0];
1333
1334 if (l < (n - 1) || off < m) {
1335 #ifdef __SUNPRO_C
1336 #pragma pipeloop(0)
1337 #endif /* __SUNPRO_C */
1338 for (i = 0; i <= (wid - 2); i += 2) {
1339 p0 = buff[i]; p1 = buff[i + 1];
1340
1341 buffd[i ] += p0*k0;
1342 buffd[i + 1] += p1*k0;
1343 }
1344
1345 } else {
1346 #ifdef __SUNPRO_C
1347 #pragma pipeloop(0)
1348 #endif /* __SUNPRO_C */
1349 for (i = 0; i <= (wid - 2); i += 2) {
1350 p0 = buff[i]; p1 = buff[i + 1];
1351
1352 buffn[i + dx_l ] = (mlib_s32)sp[0];
1353 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1354
1355 d0 = (p0*k0 + buffd[i ]);
1356 d1 = (p1*k0 + buffd[i + 1]);
1357
1358 STORE_RES(dp[0 ], d0);
1359 STORE_RES(dp[chan1], d1);
1360
1361 buffd[i ] = 0;
1362 buffd[i + 1] = 0;
1363
1364 sp += chan2;
1365 dp += chan2;
1366 }
1367 }
1368 }
1396
1397 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
1398 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
1399
1400 /* next line */
1401
1402 if (j < hgt - dy_b - 2) sl += sll;
1403 dl += dll;
1404
1405 buff_ind++;
1406
1407 if (buff_ind >= n + 1) buff_ind = 0;
1408 }
1409 }
1410
1411 if (pbuff != buff) mlib_free(pbuff);
1412 if (k != k_locl) mlib_free(k);
1413
1414 return MLIB_SUCCESS;
1415 }
1416
1417 #endif /* __sparc ( for x86, using integer multiplies is faster ) */
1418
1419 /***************************************************************/
|
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
147 /***************************************************************/
148 typedef union {
149 mlib_d64 d64;
150 struct {
151 mlib_s32 i0;
152 mlib_s32 i1;
153 } i32s;
154 } d64_2x32;
155
156 /***************************************************************/
157 #define GET_SRC_DST_PARAMETERS(type) \
158 hgt = mlib_ImageGetHeight(src); \
159 wid = mlib_ImageGetWidth(src); \
160 nchannel = mlib_ImageGetChannels(src); \
161 sll = mlib_ImageGetStride(src) / sizeof(type); \
162 dll = mlib_ImageGetStride(dst) / sizeof(type); \
163 adr_src = (type *)mlib_ImageGetData(src); \
164 adr_dst = (type *)mlib_ImageGetData(dst)
165
166 /***************************************************************/
167 #if IMG_TYPE == 1
168
169 /*
170 * Test for the presence of any "1" bit in bits
171 8 to 31 of val. If present, then val is either
172 negative or >255. If over/underflows of 8 bits
173 are uncommon, then this technique can be a win,
174 since only a single test, rather than two, is
175 necessary to determine if clamping is needed.
176 On the other hand, if over/underflows are common,
177 it adds an extra test.
178 */
179 #define CLAMP_STORE(dst, val) \
180 if (val & 0xffffff00) { \
181 if (val < MLIB_U8_MIN) \
182 dst = MLIB_U8_MIN; \
183 else \
184 dst = MLIB_U8_MAX; \
185 } else { \
186 dst = (mlib_u8)val; \
190
191 #define CLAMP_STORE(dst, val) \
192 if (val >= MLIB_S16_MAX) \
193 dst = MLIB_S16_MAX; \
194 else if (val <= MLIB_S16_MIN) \
195 dst = MLIB_S16_MIN; \
196 else \
197 dst = (mlib_s16)val
198
199 #elif IMG_TYPE == 3
200
201 #define CLAMP_STORE(dst, val) \
202 if (val >= MLIB_U16_MAX) \
203 dst = MLIB_U16_MAX; \
204 else if (val <= MLIB_U16_MIN) \
205 dst = MLIB_U16_MIN; \
206 else \
207 dst = (mlib_u16)val
208
209 #endif /* IMG_TYPE == 1 */
210
211 /***************************************************************/
212 #define MAX_KER 7
213 #define MAX_N 15
214 #define BUFF_SIZE 1600
215 #define CACHE_SIZE (64*1024)
216
217 static mlib_status mlib_ImageConv1xN_ext(mlib_image *dst,
218 const mlib_image *src,
219 const mlib_d64 *k,
220 mlib_s32 n,
221 mlib_s32 dy_t,
222 mlib_s32 dy_b,
223 mlib_s32 cmask)
224 {
225 DTYPE *adr_src, *sl;
226 DTYPE *adr_dst, *dl, *dp;
227 FTYPE buff[BUFF_SIZE];
228 FTYPE *buffd;
229 FTYPE *pbuff = buff;
257
258 sbuff = pbuff;
259 buffd = sbuff + smax_hsize;
260
261 shgt -= (dy_t + dy_b);
262 k_off = 0;
263
264 for (l = 0; l < hgt; l += hsize) {
265 hsize = hgt - l;
266
267 if (hsize > max_hsize) hsize = max_hsize;
268
269 smax_hsize = hsize + (n - 1);
270
271 for (c = 0; c < nchannel; c++) {
272 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
273
274 sl = adr_src + c;
275 dl = adr_dst + c;
276
277 for (i = 0; i < hsize; i++) buffd[i] = 0.0;
278
279 for (j = 0; j < wid; j++) {
280 FTYPE *buff = sbuff;
281
282 for (i = k_off, ii = 0; (i < dy_t) && (ii < smax_hsize); i++, ii++) {
283 sbuff[i - k_off] = (FTYPE)sl[0];
284 }
285
286 for (; (i < shgt + dy_t) && (ii < smax_hsize); i++, ii++) {
287 sbuff[i - k_off] = (FTYPE)sl[(i - dy_t)*sll];
288 }
289
290 for (; (i < shgt + dy_t + dy_b) && (ii < smax_hsize); i++, ii++) {
291 sbuff[i - k_off] = (FTYPE)sl[(shgt - 1)*sll];
292 }
293
294 pk = k;
295
296 for (off = 0; off < (n - 4); off += 4) {
297
298 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
299 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
300
301 for (i = 0; i < hsize; i += 2) {
302 p0 = p2; p1 = p3; p2 = p4;
303
304 p3 = buff[i + 3]; p4 = buff[i + 4];
305
306 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
307 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
308 }
309
310 pk += 4;
311 buff += 4;
312 }
313
314 dp = dl;
315 kh = n - off;
316
317 if (kh == 4) {
318 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
319 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
320
321 for (i = 0; i <= (hsize - 2); i += 2) {
322 p0 = p2; p1 = p3; p2 = p4;
323
324 p3 = buff[i + 3]; p4 = buff[i + 4];
325
326 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
327 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
328
329 dp[0 ] = FROM_S32(d0);
330 dp[dll] = FROM_S32(d1);
331
332 buffd[i ] = 0.0;
333 buffd[i + 1] = 0.0;
334
335 dp += 2*dll;
336 }
337
338 if (i < hsize) {
339 p0 = p2; p1 = p3; p2 = p4;
340 p3 = buff[i + 3];
341 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i]);
342 dp[0] = FROM_S32(d0);
343 buffd[i] = 0.0;
344 }
345
346 } else if (kh == 3) {
347
348 p2 = buff[0]; p3 = buff[1];
349 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
350
351 for (i = 0; i <= (hsize - 2); i += 2) {
352 p0 = p2; p1 = p3;
353
354 p2 = buff[i + 2]; p3 = buff[i + 3];
355
356 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
357 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
358
359 dp[0 ] = FROM_S32(d0);
360 dp[dll] = FROM_S32(d1);
361
362 buffd[i ] = 0.0;
363 buffd[i + 1] = 0.0;
364
365 dp += 2*dll;
366 }
367
368 if (i < hsize) {
369 p0 = p2; p1 = p3;
370 p2 = buff[i + 2];
371 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i]);
372 dp[0] = FROM_S32(d0);
373
374 buffd[i] = 0.0;
375 }
376
377 } else if (kh == 2) {
378
379 p2 = buff[0];
380 k0 = pk[0]; k1 = pk[1];
381
382 for (i = 0; i <= (hsize - 2); i += 2) {
383 p0 = p2;
384
385 p1 = buff[i + 1]; p2 = buff[i + 2];
386
387 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
388 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
389
390 dp[0 ] = FROM_S32(d0);
391 dp[dll] = FROM_S32(d1);
392
393 buffd[i ] = 0.0;
394 buffd[i + 1] = 0.0;
395
396 dp += 2*dll;
397 }
398
399 if (i < hsize) {
400 p0 = p2;
401 p1 = buff[i + 1];
402 d0 = D2I(p0*k0 + p1*k1 + buffd[i]);
403 dp[0] = FROM_S32(d0);
404
405 buffd[i] = 0.0;
406 }
407
408 } else /* kh == 1 */{
409
410 k0 = pk[0];
411
412 for (i = 0; i <= (hsize - 2); i += 2) {
413 p0 = buff[i]; p1 = buff[i + 1];
414
415 d0 = D2I(p0*k0 + buffd[i ]);
416 d1 = D2I(p1*k0 + buffd[i + 1]);
417
418 dp[0 ] = FROM_S32(d0);
419 dp[dll] = FROM_S32(d1);
420
421 buffd[i ] = 0.0;
422 buffd[i + 1] = 0.0;
423
424 dp += 2*dll;
425 }
426
427 if (i < hsize) {
428 p0 = buff[i];
429 d0 = D2I(p0*k0 + buffd[i]);
430 dp[0] = FROM_S32(d0);
431
514 buffi = (mlib_s32*)(buffd + swid);
515
516 chan1 = nchannel;
517 chan2 = chan1 + chan1;
518
519 swid -= (dx_l + dx_r);
520
521 for (c = 0; c < nchannel; c++) {
522 if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
523
524 sl = adr_src + c;
525 dl = adr_dst + c;
526
527 for (l = 0; l < n; l++) {
528 FTYPE *buff = buffs[l];
529
530 for (i = 0; i < dx_l; i++) {
531 buff[i] = (FTYPE)sl[0];
532 }
533
534 for (i = 0; i < swid; i++) {
535 buff[i + dx_l] = (FTYPE)sl[i*chan1];
536 }
537
538 for (i = 0; i < dx_r; i++) {
539 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
540 }
541
542 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
543 }
544
545 buff_ind = 0;
546
547 for (i = 0; i < wid; i++) buffd[i] = 0.0;
548
549 for (j = 0; j < hgt; j++) {
550 FTYPE **buffc = buffs + buff_ind;
551 FTYPE *buffn = buffc[n];
552 FTYPE *pk = k;
553
554 for (l = 0; l < n; l++) {
555 FTYPE *buff_l = buffc[l];
556
557 for (off = 0; off < m;) {
558 FTYPE *buff = buff_l + off;
559
560 kw = m - off;
561
562 if (kw > 2*MAX_KER) kw = MAX_KER; else
563 if (kw > MAX_KER) kw = kw/2;
564 off += kw;
565
566 sp = sl;
567 dp = dl;
568
569 if (kw == 7) {
570
571 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
572 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
573
574 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
575 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
576
577 if (l < (n - 1) || off < m) {
578 for (i = 0; i <= (wid - 2); i += 2) {
579 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
580
581 p6 = buff[i + 6]; p7 = buff[i + 7];
582
583 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
584 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
585 }
586
587 } else {
588 for (i = 0; i <= (wid - 2); i += 2) {
589 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
590
591 p6 = buff[i + 6]; p7 = buff[i + 7];
592
593 LOAD_BUFF(buffi);
594
595 dd.d64 = *(FTYPE *)(buffi + i);
596 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
597 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
598
599 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
600 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
601
602 dp[0 ] = FROM_S32(d0);
603 dp[chan1] = FROM_S32(d1);
604
605 buffd[i ] = 0.0;
606 buffd[i + 1] = 0.0;
607
608 sp += chan2;
609 dp += chan2;
610 }
611 }
612
613 } else if (kw == 6) {
614
615 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
616 p5 = buff[3]; p6 = buff[4];
617
618 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
619 k4 = pk[4]; k5 = pk[5];
620
621 if (l < (n - 1) || off < m) {
622 for (i = 0; i <= (wid - 2); i += 2) {
623 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
624
625 p5 = buff[i + 5]; p6 = buff[i + 6];
626
627 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
628 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
629 }
630
631 } else {
632 for (i = 0; i <= (wid - 2); i += 2) {
633 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
634
635 p5 = buff[i + 5]; p6 = buff[i + 6];
636
637 LOAD_BUFF(buffi);
638
639 dd.d64 = *(FTYPE *)(buffi + i);
640 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
641 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
642
643 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
644 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
645
646 dp[0 ] = FROM_S32(d0);
647 dp[chan1] = FROM_S32(d1);
648
649 buffd[i ] = 0.0;
650 buffd[i + 1] = 0.0;
651
652 sp += chan2;
653 dp += chan2;
654 }
655 }
656
657 } else if (kw == 5) {
658
659 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
660 p5 = buff[3];
661
662 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
663 k4 = pk[4];
664
665 if (l < (n - 1) || off < m) {
666 for (i = 0; i <= (wid - 2); i += 2) {
667 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
668
669 p4 = buff[i + 4]; p5 = buff[i + 5];
670
671 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
672 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
673 }
674
675 } else {
676 for (i = 0; i <= (wid - 2); i += 2) {
677 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
678
679 p4 = buff[i + 4]; p5 = buff[i + 5];
680
681 LOAD_BUFF(buffi);
682
683 dd.d64 = *(FTYPE *)(buffi + i);
684 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
685 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
686
687 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
688 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
689
690 dp[0 ] = FROM_S32(d0);
691 dp[chan1] = FROM_S32(d1);
692
693 buffd[i ] = 0.0;
694 buffd[i + 1] = 0.0;
695
696 sp += chan2;
697 dp += chan2;
698 }
699 }
700
701 } else if (kw == 4) {
702
703 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
704
705 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
706
707 if (l < (n - 1) || off < m) {
708 for (i = 0; i <= (wid - 2); i += 2) {
709 p0 = p2; p1 = p3; p2 = p4;
710
711 p3 = buff[i + 3]; p4 = buff[i + 4];
712
713 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
714 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
715 }
716
717 } else {
718 for (i = 0; i <= (wid - 2); i += 2) {
719 p0 = p2; p1 = p3; p2 = p4;
720
721 p3 = buff[i + 3]; p4 = buff[i + 4];
722
723 LOAD_BUFF(buffi);
724
725 dd.d64 = *(FTYPE *)(buffi + i);
726 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
727 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
728
729 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
730 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
731
732 dp[0 ] = FROM_S32(d0);
733 dp[chan1] = FROM_S32(d1);
734
735 buffd[i ] = 0.0;
736 buffd[i + 1] = 0.0;
737
738 sp += chan2;
739 dp += chan2;
740 }
741 }
742
743 } else if (kw == 3) {
744
745 p2 = buff[0]; p3 = buff[1];
746 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
747
748 if (l < (n - 1) || off < m) {
749 for (i = 0; i <= (wid - 2); i += 2) {
750 p0 = p2; p1 = p3;
751
752 p2 = buff[i + 2]; p3 = buff[i + 3];
753
754 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
755 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
756 }
757
758 } else {
759 for (i = 0; i <= (wid - 2); i += 2) {
760 p0 = p2; p1 = p3;
761
762 p2 = buff[i + 2]; p3 = buff[i + 3];
763
764 LOAD_BUFF(buffi);
765
766 dd.d64 = *(FTYPE *)(buffi + i);
767 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
768 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
769
770 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
771 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
772
773 dp[0 ] = FROM_S32(d0);
774 dp[chan1] = FROM_S32(d1);
775
776 buffd[i ] = 0.0;
777 buffd[i + 1] = 0.0;
778
779 sp += chan2;
780 dp += chan2;
781 }
782 }
783
784 } else /* if (kw == 2) */ {
785
786 p2 = buff[0];
787 k0 = pk[0]; k1 = pk[1];
788
789 if (l < (n - 1) || off < m) {
790 for (i = 0; i <= (wid - 2); i += 2) {
791 p0 = p2;
792
793 p1 = buff[i + 1]; p2 = buff[i + 2];
794
795 buffd[i ] += p0*k0 + p1*k1;
796 buffd[i + 1] += p1*k0 + p2*k1;
797 }
798
799 } else {
800 for (i = 0; i <= (wid - 2); i += 2) {
801 p0 = p2;
802
803 p1 = buff[i + 1]; p2 = buff[i + 2];
804
805 LOAD_BUFF(buffi);
806
807 dd.d64 = *(FTYPE *)(buffi + i);
808 buffn[i + dx_l ] = (FTYPE)dd.i32s.i0;
809 buffn[i + dx_l + 1] = (FTYPE)dd.i32s.i1;
810
811 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
812 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
813
814 dp[0 ] = FROM_S32(d0);
815 dp[chan1] = FROM_S32(d1);
816
817 buffd[i ] = 0.0;
818 buffd[i + 1] = 0.0;
819
853 }
854
855 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
856 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
857
858 /* next line */
859
860 if (j < hgt - dy_b - 2) sl += sll;
861 dl += dll;
862
863 buff_ind++;
864
865 if (buff_ind >= n + 1) buff_ind = 0;
866 }
867 }
868
869 FREE_AND_RETURN_STATUS;
870 }
871
872 /***************************************************************/
873 /* for x86, using integer multiplies is faster */
874
875 #define STORE_RES(res, x) \
876 x >>= shift2; \
877 CLAMP_STORE(res, x)
878
879 mlib_status CONV_FUNC_MxN_I
880 {
881 DTYPE *adr_src, *sl, *sp = NULL;
882 DTYPE *adr_dst, *dl, *dp = NULL;
883 mlib_s32 buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
884 mlib_s32 *pbuff = buff;
885 mlib_s32 **buffs = buffs_arr, *buffd;
886 mlib_s32 l, off, kw, bsize, buff_ind;
887 mlib_s32 d0, d1, shift1, shift2;
888 mlib_s32 k0, k1, k2, k3, k4, k5, k6;
889 mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
890 mlib_s32 wid, hgt, sll, dll;
891 mlib_s32 nchannel, chan1;
892 mlib_s32 i, j, c, swid;
893 mlib_s32 chan2;
930
931 for (i = 0; i < m*n; i++) {
932 k[i] = kernel[i] >> shift1;
933 }
934
935 swid -= (dx_l + dx_r);
936
937 for (c = 0; c < nchannel; c++) {
938 if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
939
940 sl = adr_src + c;
941 dl = adr_dst + c;
942
943 for (l = 0; l < n; l++) {
944 mlib_s32 *buff = buffs[l];
945
946 for (i = 0; i < dx_l; i++) {
947 buff[i] = (mlib_s32)sl[0];
948 }
949
950 for (i = 0; i < swid; i++) {
951 buff[i + dx_l] = (mlib_s32)sl[i*chan1];
952 }
953
954 for (i = 0; i < dx_r; i++) {
955 buff[swid + dx_l + i] = buff[swid + dx_l - 1];
956 }
957
958 if ((l >= dy_t) && (l < hgt + n - dy_b - 2)) sl += sll;
959 }
960
961 buff_ind = 0;
962
963 for (i = 0; i < wid; i++) buffd[i] = 0;
964
965 for (j = 0; j < hgt; j++) {
966 mlib_s32 **buffc = buffs + buff_ind;
967 mlib_s32 *buffn = buffc[n];
968 mlib_s32 *pk = k;
969
970 for (l = 0; l < n; l++) {
971 mlib_s32 *buff_l = buffc[l];
972
973 for (off = 0; off < m;) {
974 mlib_s32 *buff = buff_l + off;
975
976 sp = sl;
977 dp = dl;
978
979 kw = m - off;
980
981 if (kw > 2*MAX_KER) kw = MAX_KER; else
982 if (kw > MAX_KER) kw = kw/2;
983 off += kw;
984
985 if (kw == 7) {
986
987 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
988 p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
989
990 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
991 k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
992
993 if (l < (n - 1) || off < m) {
994 for (i = 0; i <= (wid - 2); i += 2) {
995 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
996
997 p6 = buff[i + 6]; p7 = buff[i + 7];
998
999 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
1000 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
1001 }
1002
1003 } else {
1004 for (i = 0; i <= (wid - 2); i += 2) {
1005 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
1006
1007 p6 = buff[i + 6]; p7 = buff[i + 7];
1008
1009 buffn[i + dx_l ] = (mlib_s32)sp[0];
1010 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1011
1012 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
1013 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
1014
1015 STORE_RES(dp[0 ], d0);
1016 STORE_RES(dp[chan1], d1);
1017
1018 buffd[i ] = 0;
1019 buffd[i + 1] = 0;
1020
1021 sp += chan2;
1022 dp += chan2;
1023 }
1024 }
1025
1026 } else if (kw == 6) {
1027
1028 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1029 p5 = buff[3]; p6 = buff[4];
1030
1031 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1032 k4 = pk[4]; k5 = pk[5];
1033
1034 if (l < (n - 1) || off < m) {
1035 for (i = 0; i <= (wid - 2); i += 2) {
1036 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1037
1038 p5 = buff[i + 5]; p6 = buff[i + 6];
1039
1040 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
1041 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
1042 }
1043
1044 } else {
1045 for (i = 0; i <= (wid - 2); i += 2) {
1046 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
1047
1048 p5 = buff[i + 5]; p6 = buff[i + 6];
1049
1050 buffn[i + dx_l ] = (mlib_s32)sp[0];
1051 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1052
1053 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
1054 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
1055
1056 STORE_RES(dp[0 ], d0);
1057 STORE_RES(dp[chan1], d1);
1058
1059 buffd[i ] = 0;
1060 buffd[i + 1] = 0;
1061
1062 sp += chan2;
1063 dp += chan2;
1064 }
1065 }
1066
1067 } else if (kw == 5) {
1068
1069 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1070 p5 = buff[3];
1071
1072 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1073 k4 = pk[4];
1074
1075 if (l < (n - 1) || off < m) {
1076 for (i = 0; i <= (wid - 2); i += 2) {
1077 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1078
1079 p4 = buff[i + 4]; p5 = buff[i + 5];
1080
1081 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
1082 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
1083 }
1084
1085 } else {
1086 for (i = 0; i <= (wid - 2); i += 2) {
1087 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
1088
1089 p4 = buff[i + 4]; p5 = buff[i + 5];
1090
1091 buffn[i + dx_l ] = (mlib_s32)sp[0];
1092 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1093
1094 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
1095 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
1096
1097 STORE_RES(dp[0 ], d0);
1098 STORE_RES(dp[chan1], d1);
1099
1100 buffd[i ] = 0;
1101 buffd[i + 1] = 0;
1102
1103 sp += chan2;
1104 dp += chan2;
1105 }
1106 }
1107
1108 } else if (kw == 4) {
1109
1110 p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
1111
1112 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
1113
1114 if (l < (n - 1) || off < m) {
1115 for (i = 0; i <= (wid - 2); i += 2) {
1116 p0 = p2; p1 = p3; p2 = p4;
1117
1118 p3 = buff[i + 3]; p4 = buff[i + 4];
1119
1120 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
1121 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
1122 }
1123
1124 } else {
1125 for (i = 0; i <= (wid - 2); i += 2) {
1126 p0 = p2; p1 = p3; p2 = p4;
1127
1128 p3 = buff[i + 3]; p4 = buff[i + 4];
1129
1130 buffn[i + dx_l ] = (mlib_s32)sp[0];
1131 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1132
1133 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
1134 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
1135
1136 STORE_RES(dp[0 ], d0);
1137 STORE_RES(dp[chan1], d1);
1138
1139 buffd[i ] = 0;
1140 buffd[i + 1] = 0;
1141
1142 sp += chan2;
1143 dp += chan2;
1144 }
1145 }
1146
1147 } else if (kw == 3) {
1148
1149 p2 = buff[0]; p3 = buff[1];
1150 k0 = pk[0]; k1 = pk[1]; k2 = pk[2];
1151
1152 if (l < (n - 1) || off < m) {
1153 for (i = 0; i <= (wid - 2); i += 2) {
1154 p0 = p2; p1 = p3;
1155
1156 p2 = buff[i + 2]; p3 = buff[i + 3];
1157
1158 buffd[i ] += p0*k0 + p1*k1 + p2*k2;
1159 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
1160 }
1161
1162 } else {
1163 for (i = 0; i <= (wid - 2); i += 2) {
1164 p0 = p2; p1 = p3;
1165
1166 p2 = buff[i + 2]; p3 = buff[i + 3];
1167
1168 buffn[i + dx_l ] = (mlib_s32)sp[0];
1169 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1170
1171 d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
1172 d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
1173
1174 STORE_RES(dp[0 ], d0);
1175 STORE_RES(dp[chan1], d1);
1176
1177 buffd[i ] = 0;
1178 buffd[i + 1] = 0;
1179
1180 sp += chan2;
1181 dp += chan2;
1182 }
1183 }
1184
1185 } else if (kw == 2) {
1186
1187 p2 = buff[0];
1188 k0 = pk[0]; k1 = pk[1];
1189
1190 if (l < (n - 1) || off < m) {
1191 for (i = 0; i <= (wid - 2); i += 2) {
1192 p0 = p2;
1193
1194 p1 = buff[i + 1]; p2 = buff[i + 2];
1195
1196 buffd[i ] += p0*k0 + p1*k1;
1197 buffd[i + 1] += p1*k0 + p2*k1;
1198 }
1199
1200 } else {
1201 for (i = 0; i <= (wid - 2); i += 2) {
1202 p0 = p2;
1203
1204 p1 = buff[i + 1]; p2 = buff[i + 2];
1205
1206 buffn[i + dx_l ] = (mlib_s32)sp[0];
1207 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1208
1209 d0 = (p0*k0 + p1*k1 + buffd[i ]);
1210 d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
1211
1212 STORE_RES(dp[0 ], d0);
1213 STORE_RES(dp[chan1], d1);
1214
1215 buffd[i ] = 0;
1216 buffd[i + 1] = 0;
1217
1218 sp += chan2;
1219 dp += chan2;
1220 }
1221 }
1222
1223 } else /* kw == 1 */{
1224
1225 k0 = pk[0];
1226
1227 if (l < (n - 1) || off < m) {
1228 for (i = 0; i <= (wid - 2); i += 2) {
1229 p0 = buff[i]; p1 = buff[i + 1];
1230
1231 buffd[i ] += p0*k0;
1232 buffd[i + 1] += p1*k0;
1233 }
1234
1235 } else {
1236 for (i = 0; i <= (wid - 2); i += 2) {
1237 p0 = buff[i]; p1 = buff[i + 1];
1238
1239 buffn[i + dx_l ] = (mlib_s32)sp[0];
1240 buffn[i + dx_l + 1] = (mlib_s32)sp[chan1];
1241
1242 d0 = (p0*k0 + buffd[i ]);
1243 d1 = (p1*k0 + buffd[i + 1]);
1244
1245 STORE_RES(dp[0 ], d0);
1246 STORE_RES(dp[chan1], d1);
1247
1248 buffd[i ] = 0;
1249 buffd[i + 1] = 0;
1250
1251 sp += chan2;
1252 dp += chan2;
1253 }
1254 }
1255 }
1283
1284 for (i = 0; i < dx_l; i++) buffn[i] = buffn[dx_l];
1285 for (i = 0; i < dx_r; i++) buffn[swid + dx_l + i] = buffn[swid + dx_l - 1];
1286
1287 /* next line */
1288
1289 if (j < hgt - dy_b - 2) sl += sll;
1290 dl += dll;
1291
1292 buff_ind++;
1293
1294 if (buff_ind >= n + 1) buff_ind = 0;
1295 }
1296 }
1297
1298 if (pbuff != buff) mlib_free(pbuff);
1299 if (k != k_locl) mlib_free(k);
1300
1301 return MLIB_SUCCESS;
1302 }
1303
1304 /***************************************************************/
|