1 /*
   2  * Copyright (c) 2000, 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 S32 type and
  30  *   MLIB_EDGE_DST_NO_WRITE mask
  31  *
  32  */
  33 
  34 #include "mlib_image.h"
  35 #include "mlib_ImageConv.h"
  36 
  37 /***************************************************************/
  38 #define CACHE_SIZE (64*1024)
  39 
  40 /***************************************************************/
  41 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_s32
  42 
  43 /***************************************************************/
  44 #ifndef MLIB_USE_FTOI_CLAMPING
  45 
  46 #define CLAMP_S32(dst, src)                                       \
  47   if (src > (mlib_d64)MLIB_S32_MAX) src = (mlib_d64)MLIB_S32_MAX; \
  48   if (src < (mlib_d64)MLIB_S32_MIN) src = (mlib_d64)MLIB_S32_MIN; \
  49   dst = (mlib_s32)src
  50 
  51 #else
  52 
  53 #define CLAMP_S32(dst, src) dst = (mlib_s32)(src)
  54 
  55 #endif /* MLIB_USE_FTOI_CLAMPING */
  56 
  57 /***************************************************************/
  58 #define GET_SRC_DST_PARAMETERS(type)                            \
  59   mlib_s32 hgt = mlib_ImageGetHeight(src);                      \
  60   mlib_s32 wid = mlib_ImageGetWidth(src);                       \
  61   mlib_s32 sll = mlib_ImageGetStride(src) / sizeof(type);       \
  62   mlib_s32 dll = mlib_ImageGetStride(dst) / sizeof(type);       \
  63   type*    adr_src = mlib_ImageGetData(src);                    \
  64   type*    adr_dst = mlib_ImageGetData(dst);                    \
  65   mlib_s32 chan1 = mlib_ImageGetChannels(src)
  66 /*  mlib_s32 chan2 = chan1 + chan1 */
  67 
  68 /***************************************************************/
  69 #define DEF_VARS(type)                                          \
  70   GET_SRC_DST_PARAMETERS(type);                                 \
  71   type     *sl, *sp, *sl1, *dl, *dp;                            \
  72   mlib_d64 *pbuff = buff, *buff0, *buff1, *buff2, *buffT;       \
  73   mlib_s32 i, j, c;                                             \
  74   mlib_d64 scalef, d0, d1
  75 
  76 /***************************************************************/
  77 #define DEF_VARS_MxN(type)                                      \
  78   GET_SRC_DST_PARAMETERS(type);                                 \
  79   type     *sl, *sp = NULL, *dl, *dp = NULL;                    \
  80   mlib_d64 *pbuff = buff;                                       \
  81   mlib_s32 i, j, c
  82 
  83 /***************************************************************/
  84 #define FTYPE  mlib_d64
  85 #define DTYPE  mlib_s32
  86 
  87 #define BUFF_SIZE  1600
  88 
  89 static mlib_status mlib_ImageConv1xN(mlib_image       *dst,
  90                                      const mlib_image *src,
  91                                      const mlib_d64   *k,
  92                                      mlib_s32         n,
  93                                      mlib_s32         dn,
  94                                      mlib_s32         cmask)
  95 {
  96   FTYPE    buff[BUFF_SIZE];
  97   mlib_s32 off, kh;
  98   const FTYPE    *pk;
  99   FTYPE    k0, k1, k2, k3, d0, d1;
 100   FTYPE    p0, p1, p2, p3, p4;
 101   DTYPE    *sl_c, *dl_c, *sl0;
 102   mlib_s32 l, hsize, max_hsize;
 103   DEF_VARS_MxN(DTYPE);
 104 
 105   hgt -= (n - 1);
 106   adr_dst += dn*dll;
 107 
 108   max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll;
 109 
 110   if (!max_hsize) max_hsize = 1;
 111 
 112   if (max_hsize > BUFF_SIZE) {
 113     pbuff = mlib_malloc(sizeof(FTYPE)*max_hsize);
 114   }
 115 
 116   sl_c = adr_src;
 117   dl_c = adr_dst;
 118 
 119   for (l = 0; l < hgt; l += hsize) {
 120     hsize = hgt - l;
 121 
 122     if (hsize > max_hsize) hsize = max_hsize;
 123 
 124     for (c = 0; c < chan1; c++) {
 125     if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
 126 
 127       sl = sl_c + c;
 128       dl = dl_c + c;
 129 
 130 #ifdef __SUNPRO_C
 131 #pragma pipeloop(0)
 132 #endif /* __SUNPRO_C */
 133       for (j = 0; j < hsize; j++) pbuff[j] = 0.0;
 134 
 135       for (i = 0; i < wid; i++) {
 136         sl0 = sl;
 137 
 138         for (off = 0; off < (n - 4); off += 4) {
 139           pk = k + off;
 140           sp = sl0;
 141 
 142           k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
 143           p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
 144           sp += 3*sll;
 145 
 146 #ifdef __SUNPRO_C
 147 #pragma pipeloop(0)
 148 #endif /* __SUNPRO_C */
 149           for (j = 0; j < hsize; j += 2) {
 150             p0 = p2; p1 = p3; p2 = p4;
 151             p3 = sp[0];
 152             p4 = sp[sll];
 153 
 154             pbuff[j    ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
 155             pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
 156 
 157             sp += 2*sll;
 158           }
 159 
 160           sl0 += 4*sll;
 161         }
 162 
 163         pk = k + off;
 164         sp = sl0;
 165 
 166         k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
 167         p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
 168 
 169         dp = dl;
 170         kh = n - off;
 171 
 172         if (kh == 4) {
 173           sp += 3*sll;
 174 
 175 #ifdef __SUNPRO_C
 176 #pragma pipeloop(0)
 177 #endif /* __SUNPRO_C */
 178           for (j = 0; j <= (hsize - 2); j += 2) {
 179             p0 = p2; p1 = p3; p2 = p4;
 180             p3 = sp[0];
 181             p4 = sp[sll];
 182 
 183             d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
 184             d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1];
 185             CLAMP_S32(dp[0  ], d0);
 186             CLAMP_S32(dp[dll], d1);
 187 
 188             pbuff[j] = 0;
 189             pbuff[j + 1] = 0;
 190 
 191             sp += 2*sll;
 192             dp += 2*dll;
 193           }
 194 
 195           if (j < hsize) {
 196             p0 = p2; p1 = p3; p2 = p4;
 197             p3 = sp[0];
 198 
 199             d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j];
 200             CLAMP_S32(dp[0], d0);
 201 
 202             pbuff[j] = 0;
 203           }
 204 
 205         } else if (kh == 3) {
 206           sp += 2*sll;
 207 
 208 #ifdef __SUNPRO_C
 209 #pragma pipeloop(0)
 210 #endif /* __SUNPRO_C */
 211           for (j = 0; j <= (hsize - 2); j += 2) {
 212             p0 = p2; p1 = p3;
 213             p2 = sp[0];
 214             p3 = sp[sll];
 215 
 216             d0 = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
 217             d1 = p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1];
 218             CLAMP_S32(dp[0  ], d0);
 219             CLAMP_S32(dp[dll], d1);
 220 
 221             pbuff[j] = 0;
 222             pbuff[j + 1] = 0;
 223 
 224             sp += 2*sll;
 225             dp += 2*dll;
 226           }
 227 
 228           if (j < hsize) {
 229             p0 = p2; p1 = p3;
 230             p2 = sp[0];
 231 
 232             d0 = p0*k0 + p1*k1 + p2*k2 + pbuff[j];
 233             CLAMP_S32(dp[0], d0);
 234 
 235             pbuff[j] = 0;
 236           }
 237 
 238         } else if (kh == 2) {
 239           sp += sll;
 240 
 241 #ifdef __SUNPRO_C
 242 #pragma pipeloop(0)
 243 #endif /* __SUNPRO_C */
 244           for (j = 0; j <= (hsize - 2); j += 2) {
 245             p0 = p2;
 246             p1 = sp[0];
 247             p2 = sp[sll];
 248 
 249             d0 = p0*k0 + p1*k1 + pbuff[j];
 250             d1 = p1*k0 + p2*k1 + pbuff[j + 1];
 251             CLAMP_S32(dp[0  ], d0);
 252             CLAMP_S32(dp[dll], d1);
 253 
 254             pbuff[j] = 0;
 255             pbuff[j + 1] = 0;
 256 
 257             sp += 2*sll;
 258             dp += 2*dll;
 259           }
 260 
 261           if (j < hsize) {
 262             p0 = p2;
 263             p1 = sp[0];
 264 
 265             d0 = p0*k0 + p1*k1 + pbuff[j];
 266             CLAMP_S32(dp[0], d0);
 267 
 268             pbuff[j] = 0;
 269           }
 270 
 271         } else /* if (kh == 1) */ {
 272 #ifdef __SUNPRO_C
 273 #pragma pipeloop(0)
 274 #endif /* __SUNPRO_C */
 275           for (j = 0; j < hsize; j++) {
 276             p0 = sp[0];
 277 
 278             d0 = p0*k0 + pbuff[j];
 279             CLAMP_S32(dp[0], d0);
 280 
 281             pbuff[j] = 0;
 282 
 283             sp += sll;
 284             dp += dll;
 285           }
 286         }
 287 
 288         sl += chan1;
 289         dl += chan1;
 290       }
 291     }
 292 
 293     sl_c += max_hsize*sll;
 294     dl_c += max_hsize*dll;
 295   }
 296 
 297   if (pbuff != buff) mlib_free(pbuff);
 298 
 299   return MLIB_SUCCESS;
 300 }
 301 
 302 /***************************************************************/
 303 #define MAX_KER 7
 304 
 305 #define MAX_N     15
 306 
 307 #undef  BUFF_SIZE
 308 #define BUFF_SIZE 1500
 309 
 310 mlib_status CONV_FUNC(MxN)(mlib_image       *dst,
 311                            const mlib_image *src,
 312                            const mlib_s32   *kernel,
 313                            mlib_s32         m,
 314                            mlib_s32         n,
 315                            mlib_s32         dm,
 316                            mlib_s32         dn,
 317                            mlib_s32         scale,
 318                            mlib_s32         cmask)
 319 {
 320   mlib_d64  buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
 321   mlib_d64  **buffs = buffs_arr, *buffd;
 322   mlib_d64  akernel[256], *k = akernel, fscale = 1.0;
 323   mlib_s32  l, off, kw, bsize, buff_ind, mn;
 324   mlib_d64  d0, d1;
 325   mlib_d64  k0, k1, k2, k3, k4, k5, k6;
 326   mlib_d64  p0, p1, p2, p3, p4, p5, p6, p7;
 327   DEF_VARS_MxN(mlib_s32);
 328   mlib_s32 chan2 = chan1 + chan1;
 329 
 330   mlib_status status = MLIB_SUCCESS;
 331 
 332   if (scale > 30) {
 333     fscale *= 1.0/(1 << 30);
 334     scale -= 30;
 335   }
 336 
 337   fscale /= (1 << scale);
 338 
 339   mn = m*n;
 340 
 341   if (mn > 256) {
 342     k = mlib_malloc(mn*sizeof(mlib_d64));
 343 
 344     if (k == NULL) return MLIB_FAILURE;
 345   }
 346 
 347   for (i = 0; i < mn; i++) {
 348     k[i] = kernel[i]*fscale;
 349   }
 350 
 351   if (m == 1) {
 352     status = mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
 353     FREE_AND_RETURN_STATUS;
 354   }
 355 
 356   bsize = (n + 2)*wid;
 357 
 358   if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
 359     pbuff = mlib_malloc(sizeof(mlib_d64)*bsize + sizeof(mlib_d64*)*2*(n + 1));
 360 
 361     if (pbuff == NULL) {
 362       status = MLIB_FAILURE;
 363       FREE_AND_RETURN_STATUS;
 364     }
 365     buffs = (mlib_d64**)(pbuff + bsize);
 366   }
 367 
 368   for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*wid;
 369   for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
 370   buffd = buffs[n] + wid;
 371 
 372   wid -= (m - 1);
 373   hgt -= (n - 1);
 374   adr_dst += dn*dll + dm*chan1;
 375 
 376   for (c = 0; c < chan1; c++) {
 377     if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
 378 
 379     sl = adr_src + c;
 380     dl = adr_dst + c;
 381 
 382     for (l = 0; l < n; l++) {
 383       mlib_d64 *buff = buffs[l];
 384 
 385 #ifdef __SUNPRO_C
 386 #pragma pipeloop(0)
 387 #endif /* __SUNPRO_C */
 388       for (i = 0; i < wid + (m - 1); i++) {
 389         buff[i] = (mlib_d64)sl[i*chan1];
 390       }
 391 
 392       sl += sll;
 393     }
 394 
 395     buff_ind = 0;
 396 
 397 #ifdef __SUNPRO_C
 398 #pragma pipeloop(0)
 399 #endif /* __SUNPRO_C */
 400     for (i = 0; i < wid; i++) buffd[i] = 0.0;
 401 
 402     for (j = 0; j < hgt; j++) {
 403       mlib_d64 **buffc = buffs + buff_ind;
 404       mlib_d64 *buffn = buffc[n];
 405       mlib_d64 *pk = k;
 406 
 407       for (l = 0; l < n; l++) {
 408         mlib_d64 *buff_l = buffc[l];
 409 
 410         for (off = 0; off < m;) {
 411           mlib_d64 *buff = buff_l + off;
 412 
 413           kw = m - off;
 414 
 415           if (kw > 2*MAX_KER) kw = MAX_KER; else
 416             if (kw > MAX_KER) kw = kw/2;
 417           off += kw;
 418 
 419           sp = sl;
 420           dp = dl;
 421 
 422           p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
 423           p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
 424 
 425           k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
 426           k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
 427           pk += kw;
 428 
 429           if (kw == 7) {
 430 
 431             if (l < (n - 1) || off < m) {
 432 #ifdef __SUNPRO_C
 433 #pragma pipeloop(0)
 434 #endif /* __SUNPRO_C */
 435               for (i = 0; i <= (wid - 2); i += 2) {
 436                 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
 437 
 438                 p6 = buff[i + 6]; p7 = buff[i + 7];
 439 
 440                 buffd[i    ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
 441                 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
 442               }
 443 
 444             } else {
 445 #ifdef __SUNPRO_C
 446 #pragma pipeloop(0)
 447 #endif /* __SUNPRO_C */
 448               for (i = 0; i <= (wid - 2); i += 2) {
 449                 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
 450 
 451                 p6 = buff[i + 6]; p7 = buff[i + 7];
 452 
 453                 buffn[i    ] = (mlib_d64)sp[0];
 454                 buffn[i + 1] = (mlib_d64)sp[chan1];
 455 
 456                 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i    ];
 457                 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1];
 458 
 459                 CLAMP_S32(dp[0],     d0);
 460                 CLAMP_S32(dp[chan1], d1);
 461 
 462                 buffd[i    ] = 0.0;
 463                 buffd[i + 1] = 0.0;
 464 
 465                 sp += chan2;
 466                 dp += chan2;
 467               }
 468             }
 469 
 470           } else if (kw == 6) {
 471 
 472             if (l < (n - 1) || off < m) {
 473 #ifdef __SUNPRO_C
 474 #pragma pipeloop(0)
 475 #endif /* __SUNPRO_C */
 476               for (i = 0; i <= (wid - 2); i += 2) {
 477                 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
 478 
 479                 p5 = buff[i + 5]; p6 = buff[i + 6];
 480 
 481                 buffd[i    ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
 482                 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
 483               }
 484 
 485             } else {
 486 #ifdef __SUNPRO_C
 487 #pragma pipeloop(0)
 488 #endif /* __SUNPRO_C */
 489               for (i = 0; i <= (wid - 2); i += 2) {
 490                 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
 491 
 492                 p5 = buff[i + 5]; p6 = buff[i + 6];
 493 
 494                 buffn[i    ] = (mlib_d64)sp[0];
 495                 buffn[i + 1] = (mlib_d64)sp[chan1];
 496 
 497                 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i    ];
 498                 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1];
 499 
 500                 CLAMP_S32(dp[0],     d0);
 501                 CLAMP_S32(dp[chan1], d1);
 502 
 503                 buffd[i    ] = 0.0;
 504                 buffd[i + 1] = 0.0;
 505 
 506                 sp += chan2;
 507                 dp += chan2;
 508               }
 509             }
 510 
 511           } else if (kw == 5) {
 512 
 513             if (l < (n - 1) || off < m) {
 514 #ifdef __SUNPRO_C
 515 #pragma pipeloop(0)
 516 #endif /* __SUNPRO_C */
 517               for (i = 0; i <= (wid - 2); i += 2) {
 518                 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
 519 
 520                 p4 = buff[i + 4]; p5 = buff[i + 5];
 521 
 522                 buffd[i    ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
 523                 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
 524               }
 525 
 526             } else {
 527 #ifdef __SUNPRO_C
 528 #pragma pipeloop(0)
 529 #endif /* __SUNPRO_C */
 530               for (i = 0; i <= (wid - 2); i += 2) {
 531                 p0 = p2; p1 = p3; p2 = p4; p3 = p5;
 532 
 533                 p4 = buff[i + 4]; p5 = buff[i + 5];
 534 
 535                 buffn[i    ] = (mlib_d64)sp[0];
 536                 buffn[i + 1] = (mlib_d64)sp[chan1];
 537 
 538                 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i    ];
 539                 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1];
 540 
 541                 CLAMP_S32(dp[0],     d0);
 542                 CLAMP_S32(dp[chan1], d1);
 543 
 544                 buffd[i    ] = 0.0;
 545                 buffd[i + 1] = 0.0;
 546 
 547                 sp += chan2;
 548                 dp += chan2;
 549               }
 550             }
 551 
 552           } else if (kw == 4) {
 553 
 554             if (l < (n - 1) || off < m) {
 555 #ifdef __SUNPRO_C
 556 #pragma pipeloop(0)
 557 #endif /* __SUNPRO_C */
 558               for (i = 0; i <= (wid - 2); i += 2) {
 559                 p0 = p2; p1 = p3; p2 = p4;
 560 
 561                 p3 = buff[i + 3]; p4 = buff[i + 4];
 562 
 563                 buffd[i    ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
 564                 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
 565               }
 566 
 567             } else {
 568 #ifdef __SUNPRO_C
 569 #pragma pipeloop(0)
 570 #endif /* __SUNPRO_C */
 571               for (i = 0; i <= (wid - 2); i += 2) {
 572                 p0 = p2; p1 = p3; p2 = p4;
 573 
 574                 p3 = buff[i + 3]; p4 = buff[i + 4];
 575 
 576                 buffn[i    ] = (mlib_d64)sp[0];
 577                 buffn[i + 1] = (mlib_d64)sp[chan1];
 578 
 579                 d0 = p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i    ];
 580                 d1 = p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1];
 581 
 582                 CLAMP_S32(dp[0],     d0);
 583                 CLAMP_S32(dp[chan1], d1);
 584 
 585                 buffd[i    ] = 0.0;
 586                 buffd[i + 1] = 0.0;
 587 
 588                 sp += chan2;
 589                 dp += chan2;
 590               }
 591             }
 592 
 593           } else if (kw == 3) {
 594 
 595             if (l < (n - 1) || off < m) {
 596 #ifdef __SUNPRO_C
 597 #pragma pipeloop(0)
 598 #endif /* __SUNPRO_C */
 599               for (i = 0; i <= (wid - 2); i += 2) {
 600                 p0 = p2; p1 = p3;
 601 
 602                 p2 = buff[i + 2]; p3 = buff[i + 3];
 603 
 604                 buffd[i    ] += p0*k0 + p1*k1 + p2*k2;
 605                 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
 606               }
 607 
 608             } else {
 609 #ifdef __SUNPRO_C
 610 #pragma pipeloop(0)
 611 #endif /* __SUNPRO_C */
 612               for (i = 0; i <= (wid - 2); i += 2) {
 613                 p0 = p2; p1 = p3;
 614 
 615                 p2 = buff[i + 2]; p3 = buff[i + 3];
 616 
 617                 buffn[i    ] = (mlib_d64)sp[0];
 618                 buffn[i + 1] = (mlib_d64)sp[chan1];
 619 
 620                 d0 = p0*k0 + p1*k1 + p2*k2 + buffd[i    ];
 621                 d1 = p1*k0 + p2*k1 + p3*k2 + buffd[i + 1];
 622 
 623                 CLAMP_S32(dp[0],     d0);
 624                 CLAMP_S32(dp[chan1], d1);
 625 
 626                 buffd[i    ] = 0.0;
 627                 buffd[i + 1] = 0.0;
 628 
 629                 sp += chan2;
 630                 dp += chan2;
 631               }
 632             }
 633 
 634           } else { /* kw == 2 */
 635 
 636             if (l < (n - 1) || off < m) {
 637 #ifdef __SUNPRO_C
 638 #pragma pipeloop(0)
 639 #endif /* __SUNPRO_C */
 640               for (i = 0; i <= (wid - 2); i += 2) {
 641                 p0 = p2;
 642 
 643                 p1 = buff[i + 1]; p2 = buff[i + 2];
 644 
 645                 buffd[i    ] += p0*k0 + p1*k1;
 646                 buffd[i + 1] += p1*k0 + p2*k1;
 647               }
 648 
 649             } else {
 650 #ifdef __SUNPRO_C
 651 #pragma pipeloop(0)
 652 #endif /* __SUNPRO_C */
 653               for (i = 0; i <= (wid - 2); i += 2) {
 654                 p0 = p2;
 655 
 656                 p1 = buff[i + 1]; p2 = buff[i + 2];
 657 
 658                 buffn[i    ] = (mlib_d64)sp[0];
 659                 buffn[i + 1] = (mlib_d64)sp[chan1];
 660 
 661                 d0 = p0*k0 + p1*k1 + buffd[i    ];
 662                 d1 = p1*k0 + p2*k1 + buffd[i + 1];
 663 
 664                 CLAMP_S32(dp[0],     d0);
 665                 CLAMP_S32(dp[chan1], d1);
 666 
 667                 buffd[i    ] = 0.0;
 668                 buffd[i + 1] = 0.0;
 669 
 670                 sp += chan2;
 671                 dp += chan2;
 672               }
 673             }
 674           }
 675         }
 676       }
 677 
 678       /* last pixels */
 679       for (; i < wid; i++) {
 680         mlib_d64 *pk = k, s = 0;
 681         mlib_s32 x;
 682 
 683         for (l = 0; l < n; l++) {
 684           mlib_d64 *buff = buffc[l] + i;
 685 
 686           for (x = 0; x < m; x++) s += buff[x] * (*pk++);
 687         }
 688 
 689         CLAMP_S32(dp[0], s);
 690 
 691         buffn[i] = (mlib_d64)sp[0];
 692 
 693         sp += chan1;
 694         dp += chan1;
 695       }
 696 
 697       for (l = 0; l < (m - 1); l++) buffn[wid + l] = sp[l*chan1];
 698 
 699       /* next line */
 700       sl += sll;
 701       dl += dll;
 702 
 703       buff_ind++;
 704 
 705       if (buff_ind >= n + 1) buff_ind = 0;
 706     }
 707   }
 708 
 709   FREE_AND_RETURN_STATUS;
 710 }
 711 
 712 /***************************************************************/