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 U8/S16/U16 types and 30 * MLIB_EDGE_DST_NO_WRITE mask 31 */ 32 33 #include "mlib_image.h" 34 #include "mlib_c_ImageConv.h" 35 36 /* 37 This define switches between functions of different data types 38 */ 39 #define IMG_TYPE 2 40 41 /***************************************************************/ 42 #if IMG_TYPE == 1 43 44 #define DTYPE mlib_u8 45 #define CONV_FUNC(KERN) mlib_c_conv##KERN##nw_u8 46 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u8 47 #define DSCALE (1 << 24) 48 #define FROM_S32(x) (((x) >> 24) ^ 128) 49 #define S64TOS32(x) (x) 50 #define SAT_OFF -(1u << 31) 51 52 #elif IMG_TYPE == 2 53 54 #define DTYPE mlib_s16 55 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_s16 56 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_s16 57 #define DSCALE 65536.0 58 #define FROM_S32(x) ((x) >> 16) 59 #define S64TOS32(x) ((x) & 0xffffffff) 60 #define SAT_OFF 61 62 #elif IMG_TYPE == 3 63 64 #define DTYPE mlib_u16 65 #define CONV_FUNC(KERN) mlib_conv##KERN##nw_u16 66 #define CONV_FUNC_I(KERN) mlib_i_conv##KERN##nw_u16 67 #define DSCALE 65536.0 68 #define FROM_S32(x) (((x) >> 16) ^ 0x8000) 69 #define S64TOS32(x) (x) 70 #define SAT_OFF -(1u << 31) 71 72 #endif /* IMG_TYPE == 1 */ 73 74 /***************************************************************/ 75 #define BUFF_SIZE 1600 76 77 #define CACHE_SIZE (64*1024) 78 79 /***************************************************************/ 80 #define FTYPE mlib_d64 81 82 #ifndef MLIB_USE_FTOI_CLAMPING 83 84 #define CLAMP_S32(x) \ 85 (((x) <= MLIB_S32_MIN) ? MLIB_S32_MIN : (((x) >= MLIB_S32_MAX) ? MLIB_S32_MAX : (mlib_s32)(x))) 86 87 #else 88 89 #define CLAMP_S32(x) ((mlib_s32)(x)) 90 91 #endif /* MLIB_USE_FTOI_CLAMPING */ 92 93 /***************************************************************/ 94 #define D2I(x) CLAMP_S32((x) SAT_OFF) 95 96 /***************************************************************/ 97 #ifdef _LITTLE_ENDIAN 98 99 #define STORE2(res0, res1) \ 100 dp[0 ] = res1; \ 101 dp[chan1] = res0 102 103 #else 104 105 #define STORE2(res0, res1) \ 106 dp[0 ] = res0; \ 107 dp[chan1] = res1 108 109 #endif /* _LITTLE_ENDIAN */ 110 111 /***************************************************************/ 112 #ifdef _NO_LONGLONG 113 114 #define LOAD_BUFF(buff) \ 115 buff[i ] = sp[0]; \ 116 buff[i + 1] = sp[chan1] 117 118 #else /* _NO_LONGLONG */ 119 120 #ifdef _LITTLE_ENDIAN 121 122 #define LOAD_BUFF(buff) \ 123 *(mlib_s64*)(buff + i) = (((mlib_s64)sp[chan1]) << 32) | S64TOS32((mlib_s64)sp[0]) 124 125 #else /* _LITTLE_ENDIAN */ 126 127 #define LOAD_BUFF(buff) \ 128 *(mlib_s64*)(buff + i) = (((mlib_s64)sp[0]) << 32) | S64TOS32((mlib_s64)sp[chan1]) 129 130 #endif /* _LITTLE_ENDIAN */ 131 #endif /* _NO_LONGLONG */ 132 133 /***************************************************************/ 134 typedef union { 135 mlib_d64 d64; 136 struct { 137 mlib_s32 i0; 138 mlib_s32 i1; 139 } i32s; 140 struct { 141 mlib_s32 f0; 142 mlib_s32 f1; 143 } f32s; 144 } d64_2x32; 145 146 /***************************************************************/ 147 #define DEF_VARS(type) \ 148 type *adr_src, *sl, *sp = NULL; \ 149 type *adr_dst, *dl, *dp = NULL; \ 150 FTYPE *pbuff = buff; \ 151 mlib_s32 wid, hgt, sll, dll; \ 152 mlib_s32 nchannel, chan1; \ 153 mlib_s32 i, j, c 154 155 /***************************************************************/ 156 #define GET_SRC_DST_PARAMETERS(type) \ 157 hgt = mlib_ImageGetHeight(src); \ 158 wid = mlib_ImageGetWidth(src); \ 159 nchannel = mlib_ImageGetChannels(src); \ 160 sll = mlib_ImageGetStride(src) / sizeof(type); \ 161 dll = mlib_ImageGetStride(dst) / sizeof(type); \ 162 adr_src = (type *)mlib_ImageGetData(src); \ 163 adr_dst = (type *)mlib_ImageGetData(dst) 164 165 /***************************************************************/ 166 #if IMG_TYPE == 1 167 168 /* Test for the presence of any "1" bit in bits 169 8 to 31 of val. If present, then val is either 170 negative or >255. If over/underflows of 8 bits 171 are uncommon, then this technique can be a win, 172 since only a single test, rather than two, is 173 necessary to determine if clamping is needed. 174 On the other hand, if over/underflows are common, 175 it adds an extra test. 176 */ 177 #define CLAMP_STORE(dst, val) \ 178 if (val & 0xffffff00) { \ 179 if (val < MLIB_U8_MIN) \ 180 dst = MLIB_U8_MIN; \ 181 else \ 182 dst = MLIB_U8_MAX; \ 183 } else { \ 184 dst = (mlib_u8)val; \ 185 } 186 187 #elif IMG_TYPE == 2 188 189 #define CLAMP_STORE(dst, val) \ 190 if (val >= MLIB_S16_MAX) \ 191 dst = MLIB_S16_MAX; \ 192 else if (val <= MLIB_S16_MIN) \ 193 dst = MLIB_S16_MIN; \ 194 else \ 195 dst = (mlib_s16)val 196 197 #elif IMG_TYPE == 3 198 199 #define CLAMP_STORE(dst, val) \ 200 if (val >= MLIB_U16_MAX) \ 201 dst = MLIB_U16_MAX; \ 202 else if (val <= MLIB_U16_MIN) \ 203 dst = MLIB_U16_MIN; \ 204 else \ 205 dst = (mlib_u16)val 206 207 #endif /* IMG_TYPE == 1 */ 208 209 /***************************************************************/ 210 #define MAX_KER 7 211 #define MAX_N 15 212 213 static mlib_status mlib_ImageConv1xN(mlib_image *dst, 214 const mlib_image *src, 215 const mlib_d64 *k, 216 mlib_s32 n, 217 mlib_s32 dn, 218 mlib_s32 cmask) 219 { 220 FTYPE buff[BUFF_SIZE]; 221 mlib_s32 off, kh; 222 mlib_s32 d0, d1; 223 const FTYPE *pk; 224 FTYPE k0, k1, k2, k3; 225 FTYPE p0, p1, p2, p3, p4; 226 DEF_VARS(DTYPE); 227 DTYPE *sl_c, *dl_c, *sl0; 228 mlib_s32 l, hsize, max_hsize; 229 GET_SRC_DST_PARAMETERS(DTYPE); 230 231 hgt -= (n - 1); 232 adr_dst += dn*dll; 233 234 max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll; 235 236 if (!max_hsize) max_hsize = 1; 237 238 if (max_hsize > BUFF_SIZE) { 239 pbuff = mlib_malloc(sizeof(FTYPE)*max_hsize); 240 } 241 242 chan1 = nchannel; 243 244 sl_c = adr_src; 245 dl_c = adr_dst; 246 247 for (l = 0; l < hgt; l += hsize) { 248 hsize = hgt - l; 249 250 if (hsize > max_hsize) hsize = max_hsize; 251 252 for (c = 0; c < nchannel; c++) { 253 if (!(cmask & (1 << (chan1 - 1 - c)))) continue; 254 255 sl = sl_c + c; 256 dl = dl_c + c; 257 258 #ifdef __SUNPRO_C 259 #pragma pipeloop(0) 260 #endif /* __SUNPRO_C */ 261 for (j = 0; j < hsize; j++) pbuff[j] = 0.0; 262 263 for (i = 0; i < wid; i++) { 264 sl0 = sl; 265 266 for (off = 0; off < (n - 4); off += 4) { 267 pk = k + off; 268 sp = sl0; 269 270 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; 271 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll]; 272 sp += 3*sll; 273 274 #ifdef __SUNPRO_C 275 #pragma pipeloop(0) 276 #endif /* __SUNPRO_C */ 277 for (j = 0; j < hsize; j += 2) { 278 p0 = p2; p1 = p3; p2 = p4; 279 p3 = sp[0]; 280 p4 = sp[sll]; 281 282 pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3; 283 pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3; 284 285 sp += 2*sll; 286 } 287 288 sl0 += 4*sll; 289 } 290 291 pk = k + off; 292 sp = sl0; 293 294 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; 295 p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll]; 296 297 dp = dl; 298 kh = n - off; 299 300 if (kh == 4) { 301 sp += 3*sll; 302 303 #ifdef __SUNPRO_C 304 #pragma pipeloop(0) 305 #endif /* __SUNPRO_C */ 306 for (j = 0; j <= (hsize - 2); j += 2) { 307 p0 = p2; p1 = p3; p2 = p4; 308 p3 = sp[0]; 309 p4 = sp[sll]; 310 311 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]); 312 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1]); 313 314 dp[0 ] = FROM_S32(d0); 315 dp[dll] = FROM_S32(d1); 316 317 pbuff[j] = 0; 318 pbuff[j + 1] = 0; 319 320 sp += 2*sll; 321 dp += 2*dll; 322 } 323 324 if (j < hsize) { 325 p0 = p2; p1 = p3; p2 = p4; 326 p3 = sp[0]; 327 328 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]); 329 330 pbuff[j] = 0; 331 332 dp[0] = FROM_S32(d0); 333 } 334 335 } else if (kh == 3) { 336 sp += 2*sll; 337 338 #ifdef __SUNPRO_C 339 #pragma pipeloop(0) 340 #endif /* __SUNPRO_C */ 341 for (j = 0; j <= (hsize - 2); j += 2) { 342 p0 = p2; p1 = p3; 343 p2 = sp[0]; 344 p3 = sp[sll]; 345 346 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]); 347 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1]); 348 349 dp[0 ] = FROM_S32(d0); 350 dp[dll] = FROM_S32(d1); 351 352 pbuff[j] = 0; 353 pbuff[j + 1] = 0; 354 355 sp += 2*sll; 356 dp += 2*dll; 357 } 358 359 if (j < hsize) { 360 p0 = p2; p1 = p3; 361 p2 = sp[0]; 362 363 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]); 364 365 pbuff[j] = 0; 366 367 dp[0] = FROM_S32(d0); 368 } 369 370 } else if (kh == 2) { 371 sp += sll; 372 373 #ifdef __SUNPRO_C 374 #pragma pipeloop(0) 375 #endif /* __SUNPRO_C */ 376 for (j = 0; j <= (hsize - 2); j += 2) { 377 p0 = p2; 378 p1 = sp[0]; 379 p2 = sp[sll]; 380 381 d0 = D2I(p0*k0 + p1*k1 + pbuff[j]); 382 d1 = D2I(p1*k0 + p2*k1 + pbuff[j + 1]); 383 384 dp[0 ] = FROM_S32(d0); 385 dp[dll] = FROM_S32(d1); 386 387 pbuff[j] = 0; 388 pbuff[j + 1] = 0; 389 390 sp += 2*sll; 391 dp += 2*dll; 392 } 393 394 if (j < hsize) { 395 p0 = p2; 396 p1 = sp[0]; 397 398 d0 = D2I(p0*k0 + p1*k1 + pbuff[j]); 399 400 pbuff[j] = 0; 401 402 dp[0] = FROM_S32(d0); 403 } 404 405 } else /* if (kh == 1) */ { 406 #ifdef __SUNPRO_C 407 #pragma pipeloop(0) 408 #endif /* __SUNPRO_C */ 409 for (j = 0; j < hsize; j++) { 410 p0 = sp[0]; 411 412 d0 = D2I(p0*k0 + pbuff[j]); 413 414 dp[0] = FROM_S32(d0); 415 416 pbuff[j] = 0; 417 418 sp += sll; 419 dp += dll; 420 } 421 } 422 423 sl += chan1; 424 dl += chan1; 425 } 426 } 427 428 sl_c += max_hsize*sll; 429 dl_c += max_hsize*dll; 430 } 431 432 if (pbuff != buff) mlib_free(pbuff); 433 434 return MLIB_SUCCESS; 435 } 436 437 /***************************************************************/ 438 mlib_status CONV_FUNC(MxN)(mlib_image *dst, 439 const mlib_image *src, 440 const mlib_s32 *kernel, 441 mlib_s32 m, 442 mlib_s32 n, 443 mlib_s32 dm, 444 mlib_s32 dn, 445 mlib_s32 scale, 446 mlib_s32 cmask) 447 { 448 FTYPE buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)]; 449 FTYPE **buffs = buffs_arr, *buffd; 450 FTYPE akernel[256], *k = akernel, fscale = DSCALE; 451 mlib_s32 mn, l, off, kw, bsize, buff_ind; 452 mlib_s32 d0, d1; 453 FTYPE k0, k1, k2, k3, k4, k5, k6; 454 FTYPE p0, p1, p2, p3, p4, p5, p6, p7; 455 d64_2x32 dd; 456 DEF_VARS(DTYPE); 457 mlib_s32 chan2; 458 mlib_s32 *buffo, *buffi; 459 mlib_status status = MLIB_SUCCESS; 460 461 GET_SRC_DST_PARAMETERS(DTYPE); 462 463 if (scale > 30) { 464 fscale *= 1.0/(1 << 30); 465 scale -= 30; 466 } 467 468 fscale /= (1 << scale); 469 470 mn = m*n; 471 472 if (mn > 256) { 473 k = mlib_malloc(mn*sizeof(mlib_d64)); 474 475 if (k == NULL) return MLIB_FAILURE; 476 } 477 478 for (i = 0; i < mn; i++) { 479 k[i] = kernel[i]*fscale; 480 } 481 482 if (m == 1) { 483 status = mlib_ImageConv1xN(dst, src, k, n, dn, cmask); 484 FREE_AND_RETURN_STATUS; 485 } 486 487 bsize = (n + 3)*wid; 488 489 if ((bsize > BUFF_SIZE) || (n > MAX_N)) { 490 pbuff = mlib_malloc(sizeof(FTYPE)*bsize + sizeof(FTYPE *)*2*(n + 1)); 491 492 if (pbuff == NULL) { 493 status = MLIB_FAILURE; 494 FREE_AND_RETURN_STATUS; 495 } 496 buffs = (FTYPE **)(pbuff + bsize); 497 } 498 499 for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*wid; 500 for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l]; 501 buffd = buffs[n] + wid; 502 buffo = (mlib_s32*)(buffd + wid); 503 buffi = buffo + (wid &~ 1); 504 505 chan1 = nchannel; 506 chan2 = chan1 + chan1; 507 508 wid -= (m - 1); 509 hgt -= (n - 1); 510 adr_dst += dn*dll + dm*nchannel; 511 512 for (c = 0; c < nchannel; c++) { 513 if (!(cmask & (1 << (chan1 - 1 - c)))) continue; 514 515 sl = adr_src + c; 516 dl = adr_dst + c; 517 518 for (l = 0; l < n; l++) { 519 FTYPE *buff = buffs[l]; 520 521 #ifdef __SUNPRO_C 522 #pragma pipeloop(0) 523 #endif /* __SUNPRO_C */ 524 for (i = 0; i < wid + (m - 1); i++) { 525 buff[i] = (FTYPE)sl[i*chan1]; 526 } 527 528 sl += sll; 529 } 530 531 buff_ind = 0; 532 533 #ifdef __SUNPRO_C 534 #pragma pipeloop(0) 535 #endif /* __SUNPRO_C */ 536 for (i = 0; i < wid; i++) buffd[i] = 0.0; 537 538 for (j = 0; j < hgt; j++) { 539 FTYPE **buffc = buffs + buff_ind; 540 FTYPE *buffn = buffc[n]; 541 FTYPE *pk = k; 542 543 for (l = 0; l < n; l++) { 544 FTYPE *buff_l = buffc[l]; 545 546 for (off = 0; off < m;) { 547 FTYPE *buff = buff_l + off; 548 549 kw = m - off; 550 551 if (kw > 2*MAX_KER) kw = MAX_KER; else 552 if (kw > MAX_KER) kw = kw/2; 553 off += kw; 554 555 sp = sl; 556 dp = dl; 557 558 p2 = buff[0]; p3 = buff[1]; p4 = buff[2]; 559 p5 = buff[3]; p6 = buff[4]; p7 = buff[5]; 560 561 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; 562 k4 = pk[4]; k5 = pk[5]; k6 = pk[6]; 563 pk += kw; 564 565 if (kw == 7) { 566 567 if (l < (n - 1) || off < m) { 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; p3 = p5; p4 = p6; p5 = p7; 573 574 p6 = buff[i + 6]; p7 = buff[i + 7]; 575 576 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6; 577 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6; 578 } 579 580 } else { 581 #ifdef __SUNPRO_C 582 #pragma pipeloop(0) 583 #endif /* __SUNPRO_C */ 584 for (i = 0; i <= (wid - 2); i += 2) { 585 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7; 586 587 p6 = buff[i + 6]; p7 = buff[i + 7]; 588 589 LOAD_BUFF(buffi); 590 591 dd.d64 = *(FTYPE *)(buffi + i); 592 buffn[i ] = (FTYPE)dd.i32s.i0; 593 buffn[i + 1] = (FTYPE)dd.i32s.i1; 594 595 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]); 596 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]); 597 598 dp[0 ] = FROM_S32(d0); 599 dp[chan1] = FROM_S32(d1); 600 601 buffd[i ] = 0.0; 602 buffd[i + 1] = 0.0; 603 604 sp += chan2; 605 dp += chan2; 606 } 607 } 608 609 } else if (kw == 6) { 610 611 if (l < (n - 1) || off < m) { 612 #ifdef __SUNPRO_C 613 #pragma pipeloop(0) 614 #endif /* __SUNPRO_C */ 615 for (i = 0; i <= (wid - 2); i += 2) { 616 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; 617 618 p5 = buff[i + 5]; p6 = buff[i + 6]; 619 620 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5; 621 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5; 622 } 623 624 } else { 625 #ifdef __SUNPRO_C 626 #pragma pipeloop(0) 627 #endif /* __SUNPRO_C */ 628 for (i = 0; i <= (wid - 2); i += 2) { 629 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; 630 631 p5 = buff[i + 5]; p6 = buff[i + 6]; 632 633 buffn[i ] = (FTYPE)sp[0]; 634 buffn[i + 1] = (FTYPE)sp[chan1]; 635 636 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]); 637 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]); 638 639 dp[0 ] = FROM_S32(d0); 640 dp[chan1] = FROM_S32(d1); 641 642 buffd[i ] = 0.0; 643 buffd[i + 1] = 0.0; 644 645 sp += chan2; 646 dp += chan2; 647 } 648 } 649 650 } else if (kw == 5) { 651 652 if (l < (n - 1) || off < m) { 653 #ifdef __SUNPRO_C 654 #pragma pipeloop(0) 655 #endif /* __SUNPRO_C */ 656 for (i = 0; i <= (wid - 2); i += 2) { 657 p0 = p2; p1 = p3; p2 = p4; p3 = p5; 658 659 p4 = buff[i + 4]; p5 = buff[i + 5]; 660 661 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4; 662 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4; 663 } 664 665 } else { 666 #ifdef __SUNPRO_C 667 #pragma pipeloop(0) 668 #endif /* __SUNPRO_C */ 669 for (i = 0; i <= (wid - 2); i += 2) { 670 p0 = p2; p1 = p3; p2 = p4; p3 = p5; 671 672 p4 = buff[i + 4]; p5 = buff[i + 5]; 673 674 buffn[i ] = (FTYPE)sp[0]; 675 buffn[i + 1] = (FTYPE)sp[chan1]; 676 677 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]); 678 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]); 679 680 dp[0 ] = FROM_S32(d0); 681 dp[chan1] = FROM_S32(d1); 682 683 buffd[i ] = 0.0; 684 buffd[i + 1] = 0.0; 685 686 sp += chan2; 687 dp += chan2; 688 } 689 } 690 691 } else if (kw == 4) { 692 693 if (l < (n - 1) || off < m) { 694 #ifdef __SUNPRO_C 695 #pragma pipeloop(0) 696 #endif /* __SUNPRO_C */ 697 for (i = 0; i <= (wid - 2); i += 2) { 698 p0 = p2; p1 = p3; p2 = p4; 699 700 p3 = buff[i + 3]; p4 = buff[i + 4]; 701 702 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3; 703 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3; 704 } 705 706 } else { 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; 712 713 p3 = buff[i + 3]; p4 = buff[i + 4]; 714 715 buffn[i ] = (FTYPE)sp[0]; 716 buffn[i + 1] = (FTYPE)sp[chan1]; 717 718 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]); 719 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]); 720 721 dp[0 ] = FROM_S32(d0); 722 dp[chan1] = FROM_S32(d1); 723 724 buffd[i ] = 0.0; 725 buffd[i + 1] = 0.0; 726 727 sp += chan2; 728 dp += chan2; 729 } 730 } 731 732 } else if (kw == 3) { 733 734 if (l < (n - 1) || off < m) { 735 #ifdef __SUNPRO_C 736 #pragma pipeloop(0) 737 #endif /* __SUNPRO_C */ 738 for (i = 0; i <= (wid - 2); i += 2) { 739 p0 = p2; p1 = p3; 740 741 p2 = buff[i + 2]; p3 = buff[i + 3]; 742 743 buffd[i ] += p0*k0 + p1*k1 + p2*k2; 744 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2; 745 } 746 747 } else { 748 #ifdef __SUNPRO_C 749 #pragma pipeloop(0) 750 #endif /* __SUNPRO_C */ 751 for (i = 0; i <= (wid - 2); i += 2) { 752 p0 = p2; p1 = p3; 753 754 p2 = buff[i + 2]; p3 = buff[i + 3]; 755 756 buffn[i ] = (FTYPE)sp[0]; 757 buffn[i + 1] = (FTYPE)sp[chan1]; 758 759 d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]); 760 d1 = D2I(p1*k0 + p2*k1 + p3*k2 + 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 == 2)*/ { 774 775 if (l < (n - 1) || off < m) { 776 #ifdef __SUNPRO_C 777 #pragma pipeloop(0) 778 #endif /* __SUNPRO_C */ 779 for (i = 0; i <= (wid - 2); i += 2) { 780 p0 = p2; 781 782 p1 = buff[i + 1]; p2 = buff[i + 2]; 783 784 buffd[i ] += p0*k0 + p1*k1; 785 buffd[i + 1] += p1*k0 + p2*k1; 786 } 787 788 } else { 789 #ifdef __SUNPRO_C 790 #pragma pipeloop(0) 791 #endif /* __SUNPRO_C */ 792 for (i = 0; i <= (wid - 2); i += 2) { 793 p0 = p2; 794 795 p1 = buff[i + 1]; p2 = buff[i + 2]; 796 797 buffn[i ] = (FTYPE)sp[0]; 798 buffn[i + 1] = (FTYPE)sp[chan1]; 799 800 d0 = D2I(p0*k0 + p1*k1 + buffd[i ]); 801 d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]); 802 803 dp[0 ] = FROM_S32(d0); 804 dp[chan1] = FROM_S32(d1); 805 806 buffd[i ] = 0.0; 807 buffd[i + 1] = 0.0; 808 809 sp += chan2; 810 dp += chan2; 811 } 812 } 813 } 814 } 815 } 816 817 /* last pixels */ 818 for (; i < wid; i++) { 819 FTYPE *pk = k, s = 0; 820 mlib_s32 x, d0; 821 822 for (l = 0; l < n; l++) { 823 FTYPE *buff = buffc[l] + i; 824 825 for (x = 0; x < m; x++) s += buff[x] * (*pk++); 826 } 827 828 d0 = D2I(s); 829 dp[0] = FROM_S32(d0); 830 831 buffn[i] = (FTYPE)sp[0]; 832 833 sp += chan1; 834 dp += chan1; 835 } 836 837 for (l = 0; l < (m - 1); l++) buffn[wid + l] = sp[l*chan1]; 838 839 /* next line */ 840 sl += sll; 841 dl += dll; 842 843 buff_ind++; 844 845 if (buff_ind >= n + 1) buff_ind = 0; 846 } 847 } 848 849 FREE_AND_RETURN_STATUS; 850 } 851 852 /***************************************************************/ 853 /* for x86, using integer multiplies is faster */ 854 855 #define STORE_RES(res, x) \ 856 x >>= shift2; \ 857 CLAMP_STORE(res, x) 858 859 mlib_status CONV_FUNC_I(MxN)(mlib_image *dst, 860 const mlib_image *src, 861 const mlib_s32 *kernel, 862 mlib_s32 m, 863 mlib_s32 n, 864 mlib_s32 dm, 865 mlib_s32 dn, 866 mlib_s32 scale, 867 mlib_s32 cmask) 868 { 869 mlib_s32 buff[BUFF_SIZE], *buffd = buff; 870 mlib_s32 l, off, kw; 871 mlib_s32 d0, d1, shift1, shift2; 872 mlib_s32 k0, k1, k2, k3, k4, k5, k6; 873 mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7; 874 DTYPE *adr_src, *sl, *sp = NULL; 875 DTYPE *adr_dst, *dl, *dp = NULL; 876 mlib_s32 wid, hgt, sll, dll; 877 mlib_s32 nchannel, chan1; 878 mlib_s32 i, j, c; 879 mlib_s32 chan2; 880 mlib_s32 k_locl[MAX_N*MAX_N], *k = k_locl; 881 GET_SRC_DST_PARAMETERS(DTYPE); 882 883 #if IMG_TYPE != 1 884 shift1 = 16; 885 #else 886 shift1 = 8; 887 #endif /* IMG_TYPE != 1 */ 888 shift2 = scale - shift1; 889 890 chan1 = nchannel; 891 chan2 = chan1 + chan1; 892 893 wid -= (m - 1); 894 hgt -= (n - 1); 895 adr_dst += dn*dll + dm*nchannel; 896 897 if (wid > BUFF_SIZE) { 898 buffd = mlib_malloc(sizeof(mlib_s32)*wid); 899 900 if (buffd == NULL) return MLIB_FAILURE; 901 } 902 903 if (m*n > MAX_N*MAX_N) { 904 k = mlib_malloc(sizeof(mlib_s32)*(m*n)); 905 906 if (k == NULL) { 907 if (buffd != buff) mlib_free(buffd); 908 return MLIB_FAILURE; 909 } 910 } 911 912 for (i = 0; i < m*n; i++) { 913 k[i] = kernel[i] >> shift1; 914 } 915 916 for (c = 0; c < nchannel; c++) { 917 if (!(cmask & (1 << (nchannel - 1 - c)))) continue; 918 919 sl = adr_src + c; 920 dl = adr_dst + c; 921 922 #ifdef __SUNPRO_C 923 #pragma pipeloop(0) 924 #endif /* __SUNPRO_C */ 925 for (i = 0; i < wid; i++) buffd[i] = 0; 926 927 for (j = 0; j < hgt; j++) { 928 mlib_s32 *pk = k; 929 930 for (l = 0; l < n; l++) { 931 DTYPE *sp0 = sl + l*sll; 932 933 for (off = 0; off < m;) { 934 sp = sp0 + off*chan1; 935 dp = dl; 936 937 kw = m - off; 938 939 if (kw > 2*MAX_KER) kw = MAX_KER; else 940 if (kw > MAX_KER) kw = kw/2; 941 off += kw; 942 943 p2 = sp[0]; p3 = sp[chan1]; p4 = sp[chan2]; 944 p5 = sp[chan2 + chan1]; p6 = sp[chan2 + chan2]; p7 = sp[5*chan1]; 945 946 k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; 947 k4 = pk[4]; k5 = pk[5]; k6 = pk[6]; 948 pk += kw; 949 950 sp += (kw - 1)*chan1; 951 952 if (kw == 7) { 953 954 if (l < (n - 1) || off < m) { 955 #ifdef __SUNPRO_C 956 #pragma pipeloop(0) 957 #endif /* __SUNPRO_C */ 958 for (i = 0; i <= (wid - 2); i += 2) { 959 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7; 960 p6 = sp[0]; 961 p7 = sp[chan1]; 962 963 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6; 964 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6; 965 966 sp += chan2; 967 } 968 969 } else { 970 #ifdef __SUNPRO_C 971 #pragma pipeloop(0) 972 #endif /* __SUNPRO_C */ 973 for (i = 0; i <= (wid - 2); i += 2) { 974 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7; 975 p6 = sp[0]; 976 p7 = sp[chan1]; 977 978 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]); 979 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]); 980 981 STORE_RES(dp[0 ], d0); 982 STORE_RES(dp[chan1], d1); 983 984 buffd[i ] = 0; 985 buffd[i + 1] = 0; 986 987 sp += chan2; 988 dp += chan2; 989 } 990 } 991 992 } else if (kw == 6) { 993 994 if (l < (n - 1) || off < m) { 995 #ifdef __SUNPRO_C 996 #pragma pipeloop(0) 997 #endif /* __SUNPRO_C */ 998 for (i = 0; i <= (wid - 2); i += 2) { 999 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; 1000 p5 = sp[0]; 1001 p6 = sp[chan1]; 1002 1003 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5; 1004 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5; 1005 1006 sp += chan2; 1007 } 1008 1009 } else { 1010 #ifdef __SUNPRO_C 1011 #pragma pipeloop(0) 1012 #endif /* __SUNPRO_C */ 1013 for (i = 0; i <= (wid - 2); i += 2) { 1014 p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; 1015 p5 = sp[0]; 1016 p6 = sp[chan1]; 1017 1018 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]); 1019 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]); 1020 1021 STORE_RES(dp[0 ], d0); 1022 STORE_RES(dp[chan1], d1); 1023 1024 buffd[i ] = 0; 1025 buffd[i + 1] = 0; 1026 1027 sp += chan2; 1028 dp += chan2; 1029 } 1030 } 1031 1032 } else if (kw == 5) { 1033 1034 if (l < (n - 1) || off < m) { 1035 #ifdef __SUNPRO_C 1036 #pragma pipeloop(0) 1037 #endif /* __SUNPRO_C */ 1038 for (i = 0; i <= (wid - 2); i += 2) { 1039 p0 = p2; p1 = p3; p2 = p4; p3 = p5; 1040 p4 = sp[0]; 1041 p5 = sp[chan1]; 1042 1043 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4; 1044 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4; 1045 1046 sp += chan2; 1047 } 1048 1049 } else { 1050 #ifdef __SUNPRO_C 1051 #pragma pipeloop(0) 1052 #endif /* __SUNPRO_C */ 1053 for (i = 0; i <= (wid - 2); i += 2) { 1054 p0 = p2; p1 = p3; p2 = p4; p3 = p5; 1055 p4 = sp[0]; 1056 p5 = sp[chan1]; 1057 1058 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]); 1059 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]); 1060 1061 STORE_RES(dp[0 ], d0); 1062 STORE_RES(dp[chan1], d1); 1063 1064 buffd[i ] = 0; 1065 buffd[i + 1] = 0; 1066 1067 sp += chan2; 1068 dp += chan2; 1069 } 1070 } 1071 1072 } else if (kw == 4) { 1073 1074 if (l < (n - 1) || off < m) { 1075 #ifdef __SUNPRO_C 1076 #pragma pipeloop(0) 1077 #endif /* __SUNPRO_C */ 1078 for (i = 0; i <= (wid - 2); i += 2) { 1079 p0 = p2; p1 = p3; p2 = p4; 1080 p3 = sp[0]; 1081 p4 = sp[chan1]; 1082 1083 buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3; 1084 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3; 1085 1086 sp += chan2; 1087 } 1088 1089 } else { 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; 1095 p3 = sp[0]; 1096 p4 = sp[chan1]; 1097 1098 d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]); 1099 d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]); 1100 1101 STORE_RES(dp[0 ], d0); 1102 STORE_RES(dp[chan1], d1); 1103 1104 buffd[i ] = 0; 1105 buffd[i + 1] = 0; 1106 1107 sp += chan2; 1108 dp += chan2; 1109 } 1110 } 1111 1112 } else if (kw == 3) { 1113 1114 if (l < (n - 1) || off < m) { 1115 #ifdef __SUNPRO_C 1116 #pragma pipeloop(0) 1117 #endif /* __SUNPRO_C */ 1118 for (i = 0; i <= (wid - 2); i += 2) { 1119 p0 = p2; p1 = p3; 1120 p2 = sp[0]; 1121 p3 = sp[chan1]; 1122 1123 buffd[i ] += p0*k0 + p1*k1 + p2*k2; 1124 buffd[i + 1] += p1*k0 + p2*k1 + p3*k2; 1125 1126 sp += chan2; 1127 } 1128 1129 } else { 1130 #ifdef __SUNPRO_C 1131 #pragma pipeloop(0) 1132 #endif /* __SUNPRO_C */ 1133 for (i = 0; i <= (wid - 2); i += 2) { 1134 p0 = p2; p1 = p3; 1135 p2 = sp[0]; 1136 p3 = sp[chan1]; 1137 1138 d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]); 1139 d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]); 1140 1141 STORE_RES(dp[0 ], d0); 1142 STORE_RES(dp[chan1], d1); 1143 1144 buffd[i ] = 0; 1145 buffd[i + 1] = 0; 1146 1147 sp += chan2; 1148 dp += chan2; 1149 } 1150 } 1151 1152 } else if (kw == 2) { 1153 1154 if (l < (n - 1) || off < m) { 1155 #ifdef __SUNPRO_C 1156 #pragma pipeloop(0) 1157 #endif /* __SUNPRO_C */ 1158 for (i = 0; i <= (wid - 2); i += 2) { 1159 p0 = p2; 1160 p1 = sp[0]; 1161 p2 = sp[chan1]; 1162 1163 buffd[i ] += p0*k0 + p1*k1; 1164 buffd[i + 1] += p1*k0 + p2*k1; 1165 1166 sp += chan2; 1167 } 1168 1169 } else { 1170 #ifdef __SUNPRO_C 1171 #pragma pipeloop(0) 1172 #endif /* __SUNPRO_C */ 1173 for (i = 0; i <= (wid - 2); i += 2) { 1174 p0 = p2; 1175 p1 = sp[0]; 1176 p2 = sp[chan1]; 1177 1178 d0 = (p0*k0 + p1*k1 + buffd[i ]); 1179 d1 = (p1*k0 + p2*k1 + buffd[i + 1]); 1180 1181 STORE_RES(dp[0 ], d0); 1182 STORE_RES(dp[chan1], d1); 1183 1184 buffd[i ] = 0; 1185 buffd[i + 1] = 0; 1186 1187 sp += chan2; 1188 dp += chan2; 1189 } 1190 } 1191 1192 } else /*if (kw == 1)*/ { 1193 1194 if (l < (n - 1) || off < m) { 1195 #ifdef __SUNPRO_C 1196 #pragma pipeloop(0) 1197 #endif /* __SUNPRO_C */ 1198 for (i = 0; i <= (wid - 2); i += 2) { 1199 p0 = sp[0]; 1200 p1 = sp[chan1]; 1201 1202 buffd[i ] += p0*k0; 1203 buffd[i + 1] += p1*k0; 1204 1205 sp += chan2; 1206 } 1207 1208 } else { 1209 #ifdef __SUNPRO_C 1210 #pragma pipeloop(0) 1211 #endif /* __SUNPRO_C */ 1212 for (i = 0; i <= (wid - 2); i += 2) { 1213 p0 = sp[0]; 1214 p1 = sp[chan1]; 1215 1216 d0 = (p0*k0 + buffd[i ]); 1217 d1 = (p1*k0 + buffd[i + 1]); 1218 1219 STORE_RES(dp[0 ], d0); 1220 STORE_RES(dp[chan1], d1); 1221 1222 buffd[i ] = 0; 1223 buffd[i + 1] = 0; 1224 1225 sp += chan2; 1226 dp += chan2; 1227 } 1228 } 1229 } 1230 } 1231 } 1232 1233 /* last pixels */ 1234 for (; i < wid; i++) { 1235 mlib_s32 *pk = k, s = 0; 1236 mlib_s32 x; 1237 1238 for (l = 0; l < n; l++) { 1239 sp = sl + l*sll + i*chan1; 1240 1241 for (x = 0; x < m; x++) { 1242 s += sp[0] * pk[0]; 1243 sp += chan1; 1244 pk ++; 1245 } 1246 } 1247 1248 STORE_RES(dp[0], s); 1249 1250 sp += chan1; 1251 dp += chan1; 1252 } 1253 1254 sl += sll; 1255 dl += dll; 1256 } 1257 } 1258 1259 if (buffd != buff) mlib_free(buffd); 1260 if (k != k_locl) mlib_free(k); 1261 1262 return MLIB_SUCCESS; 1263 } 1264 1265 /***************************************************************/