libavcodec/imgresample.c

   1 /*
   2  * High quality image resampling with polyphase filters
   3  * Copyright (c) 2001 Fabrice Bellard.
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file imgresample.c
  24  * High quality image resampling with polyphase filters .
  25  */
  26
  27 #include "avcodec.h"
  28 #include "dsputil.h"
  29 #include "libswscale/swscale.h"
  30
  31 #ifdef HAVE_ALTIVEC
  32 #include "ppc/imgresample_altivec.h"
  33 #endif
  34
  35 #define NB_COMPONENTS 3
  36
  37 #define PHASE_BITS 4
  38 #define NB_PHASES  (1 << PHASE_BITS)
  39 #define NB_TAPS    4
  40 #define FCENTER    1  /* index of the center of the filter */
  41 //#define TEST    1  /* Test it */
  42
  43 #define POS_FRAC_BITS 16
  44 #define POS_FRAC      (1 << POS_FRAC_BITS)
  45 /* 6 bits precision is needed for MMX */
  46 #define FILTER_BITS   8
  47
  48 #define LINE_BUF_HEIGHT (NB_TAPS * 4)
  49
  50 struct SwsContext {
  51     const AVClass *av_class;
  52     struct ImgReSampleContext *resampling_ctx;
  53     enum PixelFormat src_pix_fmt, dst_pix_fmt;
  54 };
  55
  56 typedef struct ImgReSampleContext {
  57     int iwidth, iheight, owidth, oheight;
  58     int topBand, bottomBand, leftBand, rightBand;
  59     int padtop, padbottom, padleft, padright;
  60     int pad_owidth, pad_oheight;
  61     int h_incr, v_incr;
  62     DECLARE_ALIGNED_8(int16_t, h_filters[NB_PHASES][NB_TAPS]); /* horizontal filters */
  63     DECLARE_ALIGNED_8(int16_t, v_filters[NB_PHASES][NB_TAPS]); /* vertical filters */
  64     uint8_t *line_buf;
  65 } ImgReSampleContext;
  66
  67 void av_build_filter(int16_t *filter, double factor, int tap_count, int phase_count, int scale, int type);
  68
  69 static inline int get_phase(int pos)
  70 {
  71     return ((pos) >> (POS_FRAC_BITS - PHASE_BITS)) & ((1 << PHASE_BITS) - 1);
  72 }
  73
  74 /* This function must be optimized */
  75 static void h_resample_fast(uint8_t *dst, int dst_width, const uint8_t *src,
  76                             int src_width, int src_start, int src_incr,
  77                             int16_t *filters)
  78 {
  79     int src_pos, phase, sum, i;
  80     const uint8_t *s;
  81     int16_t *filter;
  82
  83     src_pos = src_start;
  84     for(i=0;i<dst_width;i++) {
  85 #ifdef TEST
  86         /* test */
  87         if ((src_pos >> POS_FRAC_BITS) < 0 ||
  88             (src_pos >> POS_FRAC_BITS) > (src_width - NB_TAPS))
  89             av_abort();
  90 #endif
  91         s = src + (src_pos >> POS_FRAC_BITS);
  92         phase = get_phase(src_pos);
  93         filter = filters + phase * NB_TAPS;
  94 #if NB_TAPS == 4
  95         sum = s[0] * filter[0] +
  96             s[1] * filter[1] +
  97             s[2] * filter[2] +
  98             s[3] * filter[3];
  99 #else
 100         {
 101             int j;
 102             sum = 0;
 103             for(j=0;j<NB_TAPS;j++)
 104                 sum += s[j] * filter[j];
 105         }
 106 #endif
 107         sum = sum >> FILTER_BITS;
 108         if (sum < 0)
 109             sum = 0;
 110         else if (sum > 255)
 111             sum = 255;
 112         dst[0] = sum;
 113         src_pos += src_incr;
 114         dst++;
 115     }
 116 }
 117
 118 /* This function must be optimized */
 119 static void v_resample(uint8_t *dst, int dst_width, const uint8_t *src,
 120                        int wrap, int16_t *filter)
 121 {
 122     int sum, i;
 123     const uint8_t *s;
 124
 125     s = src;
 126     for(i=0;i<dst_width;i++) {
 127 #if NB_TAPS == 4
 128         sum = s[0 * wrap] * filter[0] +
 129             s[1 * wrap] * filter[1] +
 130             s[2 * wrap] * filter[2] +
 131             s[3 * wrap] * filter[3];
 132 #else
 133         {
 134             int j;
 135             uint8_t *s1 = s;
 136
 137             sum = 0;
 138             for(j=0;j<NB_TAPS;j++) {
 139                 sum += s1[0] * filter[j];
 140                 s1 += wrap;
 141             }
 142         }
 143 #endif
 144         sum = sum >> FILTER_BITS;
 145         if (sum < 0)
 146             sum = 0;
 147         else if (sum > 255)
 148             sum = 255;
 149         dst[0] = sum;
 150         dst++;
 151         s++;
 152     }
 153 }
 154
 155 #ifdef HAVE_MMX
 156
 157 #include "i386/mmx.h"
 158
 159 #define FILTER4(reg) \
 160 {\
 161         s = src + (src_pos >> POS_FRAC_BITS);\
 162         phase = get_phase(src_pos);\
 163         filter = filters + phase * NB_TAPS;\
 164         movq_m2r(*s, reg);\
 165         punpcklbw_r2r(mm7, reg);\
 166         movq_m2r(*filter, mm6);\
 167         pmaddwd_r2r(reg, mm6);\
 168         movq_r2r(mm6, reg);\
 169         psrlq_i2r(32, reg);\
 170         paddd_r2r(mm6, reg);\
 171         psrad_i2r(FILTER_BITS, reg);\
 172         src_pos += src_incr;\
 173 }
 174
 175 #define DUMP(reg) movq_r2m(reg, tmp); printf(#reg "=%016"PRIx64"\n", tmp.uq);
 176
 177 /* XXX: do four pixels at a time */
 178 static void h_resample_fast4_mmx(uint8_t *dst, int dst_width,
 179                                  const uint8_t *src, int src_width,
 180                                  int src_start, int src_incr, int16_t *filters)
 181 {
 182     int src_pos, phase;
 183     const uint8_t *s;
 184     int16_t *filter;
 185     mmx_t tmp;
 186
 187     src_pos = src_start;
 188     pxor_r2r(mm7, mm7);
 189
 190     while (dst_width >= 4) {
 191
 192         FILTER4(mm0);
 193         FILTER4(mm1);
 194         FILTER4(mm2);
 195         FILTER4(mm3);
 196
 197         packuswb_r2r(mm7, mm0);
 198         packuswb_r2r(mm7, mm1);
 199         packuswb_r2r(mm7, mm3);
 200         packuswb_r2r(mm7, mm2);
 201         movq_r2m(mm0, tmp);
 202         dst[0] = tmp.ub[0];
 203         movq_r2m(mm1, tmp);
 204         dst[1] = tmp.ub[0];
 205         movq_r2m(mm2, tmp);
 206         dst[2] = tmp.ub[0];
 207         movq_r2m(mm3, tmp);
 208         dst[3] = tmp.ub[0];
 209         dst += 4;
 210         dst_width -= 4;
 211     }
 212     while (dst_width > 0) {
 213         FILTER4(mm0);
 214         packuswb_r2r(mm7, mm0);
 215         movq_r2m(mm0, tmp);
 216         dst[0] = tmp.ub[0];
 217         dst++;
 218         dst_width--;
 219     }
 220     emms();
 221 }
 222
 223 static void v_resample4_mmx(uint8_t *dst, int dst_width, const uint8_t *src,
 224                             int wrap, int16_t *filter)
 225 {
 226     int sum, i, v;
 227     const uint8_t *s;
 228     mmx_t tmp;
 229     mmx_t coefs[4];
 230
 231     for(i=0;i<4;i++) {
 232         v = filter[i];
 233         coefs[i].uw[0] = v;
 234         coefs[i].uw[1] = v;
 235         coefs[i].uw[2] = v;
 236         coefs[i].uw[3] = v;
 237     }
 238
 239     pxor_r2r(mm7, mm7);
 240     s = src;
 241     while (dst_width >= 4) {
 242         movq_m2r(s[0 * wrap], mm0);
 243         punpcklbw_r2r(mm7, mm0);
 244         movq_m2r(s[1 * wrap], mm1);
 245         punpcklbw_r2r(mm7, mm1);
 246         movq_m2r(s[2 * wrap], mm2);
 247         punpcklbw_r2r(mm7, mm2);
 248         movq_m2r(s[3 * wrap], mm3);
 249         punpcklbw_r2r(mm7, mm3);
 250
 251         pmullw_m2r(coefs[0], mm0);
 252         pmullw_m2r(coefs[1], mm1);
 253         pmullw_m2r(coefs[2], mm2);
 254         pmullw_m2r(coefs[3], mm3);
 255
 256         paddw_r2r(mm1, mm0);
 257         paddw_r2r(mm3, mm2);
 258         paddw_r2r(mm2, mm0);
 259         psraw_i2r(FILTER_BITS, mm0);
 260
 261         packuswb_r2r(mm7, mm0);
 262         movq_r2m(mm0, tmp);
 263
 264         *(uint32_t *)dst = tmp.ud[0];
 265         dst += 4;
 266         s += 4;
 267         dst_width -= 4;
 268     }
 269     while (dst_width > 0) {
 270         sum = s[0 * wrap] * filter[0] +
 271             s[1 * wrap] * filter[1] +
 272             s[2 * wrap] * filter[2] +
 273             s[3 * wrap] * filter[3];
 274         sum = sum >> FILTER_BITS;
 275         if (sum < 0)
 276             sum = 0;
 277         else if (sum > 255)
 278             sum = 255;
 279         dst[0] = sum;
 280         dst++;
 281         s++;
 282         dst_width--;
 283     }
 284     emms();
 285 }
 286 #endif /* HAVE_MMX */
 287
 288 /* slow version to handle limit cases. Does not need optimization */
 289 static void h_resample_slow(uint8_t *dst, int dst_width,
 290                             const uint8_t *src, int src_width,
 291                             int src_start, int src_incr, int16_t *filters)
 292 {
 293     int src_pos, phase, sum, j, v, i;
 294     const uint8_t *s, *src_end;
 295     int16_t *filter;
 296
 297     src_end = src + src_width;
 298     src_pos = src_start;
 299     for(i=0;i<dst_width;i++) {
 300         s = src + (src_pos >> POS_FRAC_BITS);
 301         phase = get_phase(src_pos);
 302         filter = filters + phase * NB_TAPS;
 303         sum = 0;
 304         for(j=0;j<NB_TAPS;j++) {
 305             if (s < src)
 306                 v = src[0];
 307             else if (s >= src_end)
 308                 v = src_end[-1];
 309             else
 310                 v = s[0];
 311             sum += v * filter[j];
 312             s++;
 313         }
 314         sum = sum >> FILTER_BITS;
 315         if (sum < 0)
 316             sum = 0;
 317         else if (sum > 255)
 318             sum = 255;
 319         dst[0] = sum;
 320         src_pos += src_incr;
 321         dst++;
 322     }
 323 }
 324
 325 static void h_resample(uint8_t *dst, int dst_width, const uint8_t *src,
 326                        int src_width, int src_start, int src_incr,
 327                        int16_t *filters)
 328 {
 329     int n, src_end;
 330
 331     if (src_start < 0) {
 332         n = (0 - src_start + src_incr - 1) / src_incr;
 333         h_resample_slow(dst, n, src, src_width, src_start, src_incr, filters);
 334         dst += n;
 335         dst_width -= n;
 336         src_start += n * src_incr;
 337     }
 338     src_end = src_start + dst_width * src_incr;
 339     if (src_end > ((src_width - NB_TAPS) << POS_FRAC_BITS)) {
 340         n = (((src_width - NB_TAPS + 1) << POS_FRAC_BITS) - 1 - src_start) /
 341             src_incr;
 342     } else {
 343         n = dst_width;
 344     }
 345 #ifdef HAVE_MMX
 346     if ((mm_flags & MM_MMX) && NB_TAPS == 4)
 347         h_resample_fast4_mmx(dst, n,
 348                              src, src_width, src_start, src_incr, filters);
 349     else
 350 #endif
 351         h_resample_fast(dst, n,
 352                         src, src_width, src_start, src_incr, filters);
 353     if (n < dst_width) {
 354         dst += n;
 355         dst_width -= n;
 356         src_start += n * src_incr;
 357         h_resample_slow(dst, dst_width,
 358                         src, src_width, src_start, src_incr, filters);
 359     }
 360 }
 361
 362 static void component_resample(ImgReSampleContext *s,
 363                                uint8_t *output, int owrap, int owidth, int oheight,
 364                                uint8_t *input, int iwrap, int iwidth, int iheight)
 365 {
 366     int src_y, src_y1, last_src_y, ring_y, phase_y, y1, y;
 367     uint8_t *new_line, *src_line;
 368
 369     last_src_y = - FCENTER - 1;
 370     /* position of the bottom of the filter in the source image */
 371     src_y = (last_src_y + NB_TAPS) * POS_FRAC;
 372     ring_y = NB_TAPS; /* position in ring buffer */
 373     for(y=0;y<oheight;y++) {
 374         /* apply horizontal filter on new lines from input if needed */
 375         src_y1 = src_y >> POS_FRAC_BITS;
 376         while (last_src_y < src_y1) {
 377             if (++ring_y >= LINE_BUF_HEIGHT + NB_TAPS)
 378                 ring_y = NB_TAPS;
 379             last_src_y++;
 380             /* handle limit conditions : replicate line (slightly
 381                inefficient because we filter multiple times) */
 382             y1 = last_src_y;
 383             if (y1 < 0) {
 384                 y1 = 0;
 385             } else if (y1 >= iheight) {
 386                 y1 = iheight - 1;
 387             }
 388             src_line = input + y1 * iwrap;
 389             new_line = s->line_buf + ring_y * owidth;
 390             /* apply filter and handle limit cases correctly */
 391             h_resample(new_line, owidth,
 392                        src_line, iwidth, - FCENTER * POS_FRAC, s->h_incr,
 393                        &s->h_filters[0][0]);
 394             /* handle ring buffer wrapping */
 395             if (ring_y >= LINE_BUF_HEIGHT) {
 396                 memcpy(s->line_buf + (ring_y - LINE_BUF_HEIGHT) * owidth,
 397                        new_line, owidth);
 398             }
 399         }
 400         /* apply vertical filter */
 401         phase_y = get_phase(src_y);
 402 #ifdef HAVE_MMX
 403         /* desactivated MMX because loss of precision */
 404         if ((mm_flags & MM_MMX) && NB_TAPS == 4 && 0)
 405             v_resample4_mmx(output, owidth,
 406                             s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth,
 407                             &s->v_filters[phase_y][0]);
 408         else
 409 #endif
 410 #ifdef HAVE_ALTIVEC
 411         if ((mm_flags & MM_ALTIVEC) && NB_TAPS == 4 && FILTER_BITS <= 6)
 412             v_resample16_altivec(output, owidth,
 413                                  s->line_buf + (ring_y - NB_TAPS + 1) * owidth,
 414                                  owidth, &s->v_filters[phase_y][0]);
 415         else
 416 #endif
 417             v_resample(output, owidth,
 418                        s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth,
 419                        &s->v_filters[phase_y][0]);
 420
 421         src_y += s->v_incr;
 422
 423         output += owrap;
 424     }
 425 }
 426
 427 ImgReSampleContext *img_resample_full_init(int owidth, int oheight,
 428                                       int iwidth, int iheight,
 429                                       int topBand, int bottomBand,
 430         int leftBand, int rightBand,
 431         int padtop, int padbottom,
 432         int padleft, int padright)
 433 {
 434     ImgReSampleContext *s;
 435
 436     if (!owidth || !oheight || !iwidth || !iheight)
 437         return NULL;
 438
 439     s = av_mallocz(sizeof(ImgReSampleContext));
 440     if (!s)
 441         return NULL;
 442     if((unsigned)owidth >= UINT_MAX / (LINE_BUF_HEIGHT + NB_TAPS))
 443         goto fail;
 444     s->line_buf = av_mallocz(owidth * (LINE_BUF_HEIGHT + NB_TAPS));
 445     if (!s->line_buf)
 446         goto fail;
 447
 448     s->owidth = owidth;
 449     s->oheight = oheight;
 450     s->iwidth = iwidth;
 451     s->iheight = iheight;
 452
 453     s->topBand = topBand;
 454     s->bottomBand = bottomBand;
 455     s->leftBand = leftBand;
 456     s->rightBand = rightBand;
 457
 458     s->padtop = padtop;
 459     s->padbottom = padbottom;
 460     s->padleft = padleft;
 461     s->padright = padright;
 462
 463     s->pad_owidth = owidth - (padleft + padright);
 464     s->pad_oheight = oheight - (padtop + padbottom);
 465
 466     s->h_incr = ((iwidth - leftBand - rightBand) * POS_FRAC) / s->pad_owidth;
 467     s->v_incr = ((iheight - topBand - bottomBand) * POS_FRAC) / s->pad_oheight;
 468
 469     av_build_filter(&s->h_filters[0][0], (float) s->pad_owidth  /
 470             (float) (iwidth - leftBand - rightBand), NB_TAPS, NB_PHASES, 1<<FILTER_BITS, 0);
 471     av_build_filter(&s->v_filters[0][0], (float) s->pad_oheight /
 472             (float) (iheight - topBand - bottomBand), NB_TAPS, NB_PHASES, 1<<FILTER_BITS, 0);
 473
 474     return s;
 475 fail:
 476     av_free(s);
 477     return NULL;
 478 }
 479
 480 ImgReSampleContext *img_resample_init(int owidth, int oheight,
 481                                       int iwidth, int iheight)
 482 {
 483     return img_resample_full_init(owidth, oheight, iwidth, iheight,
 484             0, 0, 0, 0, 0, 0, 0, 0);
 485 }
 486
 487 void img_resample(ImgReSampleContext *s,
 488                   AVPicture *output, const AVPicture *input)
 489 {
 490     int i, shift;
 491     uint8_t* optr;
 492
 493     for (i=0;i<3;i++) {
 494         shift = (i == 0) ? 0 : 1;
 495
 496         optr = output->data[i] + (((output->linesize[i] *
 497                         s->padtop) + s->padleft) >> shift);
 498
 499         component_resample(s, optr, output->linesize[i],
 500                 s->pad_owidth >> shift, s->pad_oheight >> shift,
 501                 input->data[i] + (input->linesize[i] *
 502                     (s->topBand >> shift)) + (s->leftBand >> shift),
 503                 input->linesize[i], ((s->iwidth - s->leftBand -
 504                         s->rightBand) >> shift),
 505                            (s->iheight - s->topBand - s->bottomBand) >> shift);
 506     }
 507 }
 508
 509 void img_resample_close(ImgReSampleContext *s)
 510 {
 511     av_free(s->line_buf);
 512     av_free(s);
 513 }
 514
 515 static const char *context_to_name(void* ptr)
 516 {
 517     return "imgconvert";
 518 }
 519
 520 static const AVClass context_class = { "imgresample", context_to_name, NULL };
 521
 522 struct SwsContext *sws_getContext(int srcW, int srcH, int srcFormat,
 523                                   int dstW, int dstH, int dstFormat,
 524                                   int flags, SwsFilter *srcFilter,
 525                                   SwsFilter *dstFilter, double *param)
 526 {
 527     struct SwsContext *ctx;
 528
 529     ctx = av_malloc(sizeof(struct SwsContext));
 530     if (!ctx) {
 531         av_log(NULL, AV_LOG_ERROR, "Cannot allocate a resampling context!\n");
 532
 533         return NULL;
 534     }
 535     ctx->av_class = &context_class;
 536
 537     if ((srcH != dstH) || (srcW != dstW)) {
 538         if ((srcFormat != PIX_FMT_YUV420P) || (dstFormat != PIX_FMT_YUV420P)) {
 539             av_log(ctx, AV_LOG_INFO, "PIX_FMT_YUV420P will be used as an intermediate format for rescaling\n");
 540         }
 541         ctx->resampling_ctx = img_resample_init(dstW, dstH, srcW, srcH);
 542     } else {
 543         ctx->resampling_ctx = av_malloc(sizeof(ImgReSampleContext));
 544         ctx->resampling_ctx->iheight = srcH;
 545         ctx->resampling_ctx->iwidth = srcW;
 546         ctx->resampling_ctx->oheight = dstH;
 547         ctx->resampling_ctx->owidth = dstW;
 548     }
 549     ctx->src_pix_fmt = srcFormat;
 550     ctx->dst_pix_fmt = dstFormat;
 551
 552     return ctx;
 553 }
 554
 555 void sws_freeContext(struct SwsContext *ctx)
 556 {
 557     if (!ctx)
 558         return;
 559     if ((ctx->resampling_ctx->iwidth != ctx->resampling_ctx->owidth) ||
 560         (ctx->resampling_ctx->iheight != ctx->resampling_ctx->oheight)) {
 561         img_resample_close(ctx->resampling_ctx);
 562     } else {
 563         av_free(ctx->resampling_ctx);
 564     }
 565     av_free(ctx);
 566 }
 567
 568
 569 /**
 570  * Checks if context is valid or reallocs a new one instead.
 571  * If context is NULL, just calls sws_getContext() to get a new one.
 572  * Otherwise, checks if the parameters are the same already saved in context.
 573  * If that is the case, returns the current context.
 574  * Otherwise, frees context and gets a new one.
 575  *
 576  * Be warned that srcFilter, dstFilter are not checked, they are
 577  * asumed to remain valid.
 578  */
 579 struct SwsContext *sws_getCachedContext(struct SwsContext *ctx,
 580                         int srcW, int srcH, int srcFormat,
 581                         int dstW, int dstH, int dstFormat, int flags,
 582                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
 583 {
 584     if (ctx != NULL) {
 585         if ((ctx->resampling_ctx->iwidth != srcW) ||
 586                         (ctx->resampling_ctx->iheight != srcH) ||
 587                         (ctx->src_pix_fmt != srcFormat) ||
 588                         (ctx->resampling_ctx->owidth != dstW) ||
 589                         (ctx->resampling_ctx->oheight != dstH) ||
 590                         (ctx->dst_pix_fmt != dstFormat))
 591         {
 592             sws_freeContext(ctx);
 593             ctx = NULL;
 594         }
 595     }
 596     if (ctx == NULL) {
 597         return sws_getContext(srcW, srcH, srcFormat,
 598                         dstW, dstH, dstFormat, flags,
 599                         srcFilter, dstFilter, param);
 600     }
 601     return ctx;
 602 }
 603
 604 int sws_scale(struct SwsContext *ctx, uint8_t* src[], int srcStride[],
 605               int srcSliceY, int srcSliceH, uint8_t* dst[], int dstStride[])
 606 {
 607     AVPicture src_pict, dst_pict;
 608     int i, res = 0;
 609     AVPicture picture_format_temp;
 610     AVPicture picture_resample_temp, *formatted_picture, *resampled_picture;
 611     uint8_t *buf1 = NULL, *buf2 = NULL;
 612     enum PixelFormat current_pix_fmt;
 613
 614     for (i = 0; i < 4; i++) {
 615         src_pict.data[i] = src[i];
 616         src_pict.linesize[i] = srcStride[i];
 617         dst_pict.data[i] = dst[i];
 618         dst_pict.linesize[i] = dstStride[i];
 619     }
 620     if ((ctx->resampling_ctx->iwidth != ctx->resampling_ctx->owidth) ||
 621         (ctx->resampling_ctx->iheight != ctx->resampling_ctx->oheight)) {
 622         /* We have to rescale the picture, but only YUV420P rescaling is supported... */
 623
 624         if (ctx->src_pix_fmt != PIX_FMT_YUV420P) {
 625             int size;
 626
 627             /* create temporary picture for rescaling input*/
 628             size = avpicture_get_size(PIX_FMT_YUV420P, ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight);
 629             buf1 = av_malloc(size);
 630             if (!buf1) {
 631                 res = -1;
 632                 goto the_end;
 633             }
 634             formatted_picture = &picture_format_temp;
 635             avpicture_fill((AVPicture*)formatted_picture, buf1,
 636                            PIX_FMT_YUV420P, ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight);
 637
 638             if (img_convert((AVPicture*)formatted_picture, PIX_FMT_YUV420P,
 639                             &src_pict, ctx->src_pix_fmt,
 640                             ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight) < 0) {
 641
 642                 av_log(ctx, AV_LOG_ERROR, "pixel format conversion not handled\n");
 643                 res = -1;
 644                 goto the_end;
 645             }
 646         } else {
 647             formatted_picture = &src_pict;
 648         }
 649
 650         if (ctx->dst_pix_fmt != PIX_FMT_YUV420P) {
 651             int size;
 652
 653             /* create temporary picture for rescaling output*/
 654             size = avpicture_get_size(PIX_FMT_YUV420P, ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight);
 655             buf2 = av_malloc(size);
 656             if (!buf2) {
 657                 res = -1;
 658                 goto the_end;
 659             }
 660             resampled_picture = &picture_resample_temp;
 661             avpicture_fill((AVPicture*)resampled_picture, buf2,
 662                            PIX_FMT_YUV420P, ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight);
 663
 664         } else {
 665             resampled_picture = &dst_pict;
 666         }
 667
 668         /* ...and finally rescale!!! */
 669         img_resample(ctx->resampling_ctx, resampled_picture, formatted_picture);
 670         current_pix_fmt = PIX_FMT_YUV420P;
 671     } else {
 672         resampled_picture = &src_pict;
 673         current_pix_fmt = ctx->src_pix_fmt;
 674     }
 675
 676     if (current_pix_fmt != ctx->dst_pix_fmt) {
 677         if (img_convert(&dst_pict, ctx->dst_pix_fmt,
 678                         resampled_picture, current_pix_fmt,
 679                         ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight) < 0) {
 680
 681             av_log(ctx, AV_LOG_ERROR, "pixel format conversion not handled\n");
 682
 683             res = -1;
 684             goto the_end;
 685         }
 686     } else if (resampled_picture != &dst_pict) {
 687         av_picture_copy(&dst_pict, resampled_picture, current_pix_fmt,
 688                         ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight);
 689     }
 690
 691 the_end:
 692     av_free(buf1);
 693     av_free(buf2);
 694     return res;
 695 }
 696
 697
 698 #ifdef TEST
 699 #include <stdio.h>
 700 #undef exit
 701
 702 /* input */
 703 #define XSIZE 256
 704 #define YSIZE 256
 705 uint8_t img[XSIZE * YSIZE];
 706
 707 /* output */
 708 #define XSIZE1 512
 709 #define YSIZE1 512
 710 uint8_t img1[XSIZE1 * YSIZE1];
 711 uint8_t img2[XSIZE1 * YSIZE1];
 712
 713 void save_pgm(const char *filename, uint8_t *img, int xsize, int ysize)
 714 {
 715 #undef fprintf
 716     FILE *f;
 717     f=fopen(filename,"w");
 718     fprintf(f,"P5\n%d %d\n%d\n", xsize, ysize, 255);
 719     fwrite(img,1, xsize * ysize,f);
 720     fclose(f);
 721 #define fprintf please_use_av_log
 722 }
 723
 724 static void dump_filter(int16_t *filter)
 725 {
 726     int i, ph;
 727
 728     for(ph=0;ph<NB_PHASES;ph++) {
 729         av_log(NULL, AV_LOG_INFO, "%2d: ", ph);
 730         for(i=0;i<NB_TAPS;i++) {
 731             av_log(NULL, AV_LOG_INFO, " %5.2f", filter[ph * NB_TAPS + i] / 256.0);
 732         }
 733         av_log(NULL, AV_LOG_INFO, "\n");
 734     }
 735 }
 736
 737 #ifdef HAVE_MMX
 738 int mm_flags;
 739 #endif
 740
 741 int main(int argc, char **argv)
 742 {
 743     int x, y, v, i, xsize, ysize;
 744     ImgReSampleContext *s;
 745     float fact, factors[] = { 1/2.0, 3.0/4.0, 1.0, 4.0/3.0, 16.0/9.0, 2.0 };
 746     char buf[256];
 747
 748     /* build test image */
 749     for(y=0;y<YSIZE;y++) {
 750         for(x=0;x<XSIZE;x++) {
 751             if (x < XSIZE/2 && y < YSIZE/2) {
 752                 if (x < XSIZE/4 && y < YSIZE/4) {
 753                     if ((x % 10) <= 6 &&
 754                         (y % 10) <= 6)
 755                         v = 0xff;
 756                     else
 757                         v = 0x00;
 758                 } else if (x < XSIZE/4) {
 759                     if (x & 1)
 760                         v = 0xff;
 761                     else
 762                         v = 0;
 763                 } else if (y < XSIZE/4) {
 764                     if (y & 1)
 765                         v = 0xff;
 766                     else
 767                         v = 0;
 768                 } else {
 769                     if (y < YSIZE*3/8) {
 770                         if ((y+x) & 1)
 771                             v = 0xff;
 772                         else
 773                             v = 0;
 774                     } else {
 775                         if (((x+3) % 4) <= 1 &&
 776                             ((y+3) % 4) <= 1)
 777                             v = 0xff;
 778                         else
 779                             v = 0x00;
 780                     }
 781                 }
 782             } else if (x < XSIZE/2) {
 783                 v = ((x - (XSIZE/2)) * 255) / (XSIZE/2);
 784             } else if (y < XSIZE/2) {
 785                 v = ((y - (XSIZE/2)) * 255) / (XSIZE/2);
 786             } else {
 787                 v = ((x + y - XSIZE) * 255) / XSIZE;
 788             }
 789             img[(YSIZE - y) * XSIZE + (XSIZE - x)] = v;
 790         }
 791     }
 792     save_pgm("/tmp/in.pgm", img, XSIZE, YSIZE);
 793     for(i=0;i<sizeof(factors)/sizeof(float);i++) {
 794         fact = factors[i];
 795         xsize = (int)(XSIZE * fact);
 796         ysize = (int)((YSIZE - 100) * fact);
 797         s = img_resample_full_init(xsize, ysize, XSIZE, YSIZE, 50 ,50, 0, 0, 0, 0, 0, 0);
 798         av_log(NULL, AV_LOG_INFO, "Factor=%0.2f\n", fact);
 799         dump_filter(&s->h_filters[0][0]);
 800         component_resample(s, img1, xsize, xsize, ysize,
 801                            img + 50 * XSIZE, XSIZE, XSIZE, YSIZE - 100);
 802         img_resample_close(s);
 803
 804         snprintf(buf, sizeof(buf), "/tmp/out%d.pgm", i);
 805         save_pgm(buf, img1, xsize, ysize);
 806     }
 807
 808     /* mmx test */
 809 #ifdef HAVE_MMX
 810     av_log(NULL, AV_LOG_INFO, "MMX test\n");
 811     fact = 0.72;
 812     xsize = (int)(XSIZE * fact);
 813     ysize = (int)(YSIZE * fact);
 814     mm_flags = MM_MMX;
 815     s = img_resample_init(xsize, ysize, XSIZE, YSIZE);
 816     component_resample(s, img1, xsize, xsize, ysize,
 817                        img, XSIZE, XSIZE, YSIZE);
 818
 819     mm_flags = 0;
 820     s = img_resample_init(xsize, ysize, XSIZE, YSIZE);
 821     component_resample(s, img2, xsize, xsize, ysize,
 822                        img, XSIZE, XSIZE, YSIZE);
 823     if (memcmp(img1, img2, xsize * ysize) != 0) {
 824         av_log(NULL, AV_LOG_ERROR, "mmx error\n");
 825         exit(1);
 826     }
 827     av_log(NULL, AV_LOG_INFO, "MMX OK\n");
 828 #endif /* HAVE_MMX */
 829     return 0;
 830 }
 831
 832 #endif /* TEST */