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