view indeo3.c @ 2207:22b768f1261a libavcodec

10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster now

/*
 * Intel Indeo 3 (IV31, IV32, etc.) video decoder for ffmpeg
 * written, produced, and directed by Alan Smithee
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include "common.h"
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"

#include "indeo3data.h"

typedef struct
{
  unsigned char *Ybuf;
  unsigned char *Ubuf;
  unsigned char *Vbuf;
  unsigned char *the_buf;
  unsigned int the_buf_size;
  unsigned short y_w, y_h;
  unsigned short uv_w, uv_h;
} YUVBufs;

typedef struct Indeo3DecodeContext {
    AVCodecContext *avctx;
    int width, height;
    AVFrame frame;

    YUVBufs iv_frame[2];
    YUVBufs *cur_frame;
    YUVBufs *ref_frame;

    unsigned char *ModPred;
    unsigned short *corrector_type;
} Indeo3DecodeContext;

static int corrector_type_0[24] = {
  195, 159, 133, 115, 101,  93,  87,  77,
  195, 159, 133, 115, 101,  93,  87,  77,
  128,  79,  79,  79,  79,  79,  79,  79
};

static int corrector_type_2[8] = { 9, 7, 6, 8, 5, 4, 3, 2 };

static void build_modpred(Indeo3DecodeContext *s) 
{
  int i, j;

  s->ModPred = (unsigned char *) av_malloc (8 * 128);

  for (i=0; i < 128; ++i) {
    s->ModPred[i+0*128] = (i > 126) ? 254 : 2*((i + 1) - ((i + 1) % 2));
    s->ModPred[i+1*128] = (i == 7)  ?  20 : ((i == 119 || i == 120)
				 ? 236 : 2*((i + 2) - ((i + 1) % 3)));
    s->ModPred[i+2*128] = (i > 125) ? 248 : 2*((i + 2) - ((i + 2) % 4));
    s->ModPred[i+3*128] =			 2*((i + 1) - ((i - 3) % 5));
    s->ModPred[i+4*128] = (i == 8)  ?  20 : 2*((i + 1) - ((i - 3) % 6));
    s->ModPred[i+5*128] =			 2*((i + 4) - ((i + 3) % 7));
    s->ModPred[i+6*128] = (i > 123) ? 240 : 2*((i + 4) - ((i + 4) % 8));
    s->ModPred[i+7*128] =			 2*((i + 5) - ((i + 4) % 9));
  }

  s->corrector_type = (unsigned short *) av_malloc (24 * 256 * sizeof(unsigned short));

  for (i=0; i < 24; ++i) {
    for (j=0; j < 256; ++j) {
      s->corrector_type[i*256+j] = (j < corrector_type_0[i])
				? 1 : ((j < 248 || (i == 16 && j == 248))
				       ? 0 : corrector_type_2[j - 248]);
    }
  }
}

static void iv_Decode_Chunk(Indeo3DecodeContext *s, unsigned char *cur, 
  unsigned char *ref, int width, int height, unsigned char *buf1, 
  long fflags2, unsigned char *hdr,
  unsigned char *buf2, int min_width_160);

#define min(a,b) ((a) < (b) ? (a) : (b))

/* ---------------------------------------------------------------------- */
static void iv_alloc_frames(Indeo3DecodeContext *s) 
{
  int luma_width, luma_height, luma_pixels, chroma_width, chroma_height,
    chroma_pixels, bufsize, i;

  luma_width   = (s->width  + 3) & (~3);
  luma_height  = (s->height + 3) & (~3);

  s->iv_frame[0].y_w = s->iv_frame[0].y_h = 
    s->iv_frame[0].the_buf_size = 0;
  s->iv_frame[1].y_w = s->iv_frame[1].y_h = 
    s->iv_frame[1].the_buf_size = 0;
  s->iv_frame[1].the_buf = NULL;

  chroma_width  = ((luma_width >> 2) + 3) & (~3);
  chroma_height = ((luma_height>> 2) + 3) & (~3);
  luma_pixels = luma_width * luma_height;
  chroma_pixels = chroma_width * chroma_height;

  bufsize = luma_pixels * 2 + luma_width * 3 + 
    (chroma_pixels + chroma_width) * 4;

  if((s->iv_frame[0].the_buf = 
    (s->iv_frame[0].the_buf_size == 0 ? av_malloc(bufsize) : 
      av_realloc(s->iv_frame[0].the_buf, bufsize))) == NULL)
    return;
  s->iv_frame[0].y_w = s->iv_frame[1].y_w = luma_width;
  s->iv_frame[0].y_h = s->iv_frame[1].y_h = luma_height;
  s->iv_frame[0].uv_w = s->iv_frame[1].uv_w = chroma_width;
  s->iv_frame[0].uv_h = s->iv_frame[1].uv_h = chroma_height;
  s->iv_frame[0].the_buf_size = bufsize;

  s->iv_frame[0].Ybuf = s->iv_frame[0].the_buf + luma_width;
  i = luma_pixels + luma_width * 2;
  s->iv_frame[1].Ybuf = s->iv_frame[0].the_buf + i;
  i += (luma_pixels + luma_width);
  s->iv_frame[0].Ubuf = s->iv_frame[0].the_buf + i;
  i += (chroma_pixels + chroma_width);
  s->iv_frame[1].Ubuf = s->iv_frame[0].the_buf + i;
  i += (chroma_pixels + chroma_width);
  s->iv_frame[0].Vbuf = s->iv_frame[0].the_buf + i;
  i += (chroma_pixels + chroma_width);
  s->iv_frame[1].Vbuf = s->iv_frame[0].the_buf + i;

  for(i = 1; i <= luma_width; i++)
    s->iv_frame[0].Ybuf[-i] = s->iv_frame[1].Ybuf[-i] = 
      s->iv_frame[0].Ubuf[-i] = 0x80;

  for(i = 1; i <= chroma_width; i++) {
    s->iv_frame[1].Ubuf[-i] = 0x80;
    s->iv_frame[0].Vbuf[-i] = 0x80;
    s->iv_frame[1].Vbuf[-i] = 0x80;
    s->iv_frame[1].Vbuf[chroma_pixels+i-1] = 0x80;
  }
}

/* ---------------------------------------------------------------------- */
static void iv_free_func(Indeo3DecodeContext *s) 
{
  int i;

  for(i = 0 ; i < 2 ; i++) {
    if(s->iv_frame[i].the_buf != NULL) 
      av_free(s->iv_frame[i].the_buf);
    s->iv_frame[i].Ybuf = s->iv_frame[i].Ubuf = 
      s->iv_frame[i].Vbuf = NULL;
    s->iv_frame[i].the_buf = NULL;
    s->iv_frame[i].the_buf_size = 0;
    s->iv_frame[i].y_w = s->iv_frame[i].y_h = 0;
    s->iv_frame[i].uv_w = s->iv_frame[i].uv_h = 0;
  }

  av_free(s->ModPred);
  av_free(s->corrector_type);
}

/* ---------------------------------------------------------------------- */
static unsigned long iv_decode_frame(Indeo3DecodeContext *s, 
                                     unsigned char *buf, int buf_size) 
{
  unsigned int hdr_width, hdr_height,
    chroma_width, chroma_height;
  unsigned long fflags1, fflags2, fflags3, offs1, offs2, offs3, offs;
  unsigned char *hdr_pos, *buf_pos;

  buf_pos = buf;
  buf_pos += 18;

  fflags1 = le2me_16(*(uint16_t *)buf_pos);
  buf_pos += 2;
  fflags3 = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;
  fflags2 = *buf_pos++;
  buf_pos += 3;
  hdr_height = le2me_16(*(uint16_t *)buf_pos);
  buf_pos += 2;
  hdr_width = le2me_16(*(uint16_t *)buf_pos);
  buf_pos += 2;
  chroma_height = ((hdr_height >> 2) + 3) & 0x7ffc;
  chroma_width = ((hdr_width >> 2) + 3) & 0x7ffc;
  offs1 = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;
  offs2 = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;
  offs3 = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 8;
  hdr_pos = buf_pos;
  if(fflags3 == 0x80) return 4;

  if(fflags1 & 0x200) {
    s->cur_frame = s->iv_frame + 1;
    s->ref_frame = s->iv_frame;
  } else {
    s->cur_frame = s->iv_frame;
    s->ref_frame = s->iv_frame + 1;
  }

  buf_pos = buf + 16 + offs1;
  offs = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;

  iv_Decode_Chunk(s, s->cur_frame->Ybuf, s->ref_frame->Ybuf, hdr_width, 
    hdr_height, buf_pos + offs * 2, fflags2, hdr_pos, buf_pos, 
    min(hdr_width, 160));

  buf_pos = buf + 16 + offs2;
  offs = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;

  iv_Decode_Chunk(s, s->cur_frame->Vbuf, s->ref_frame->Vbuf, chroma_width, 
    chroma_height, buf_pos + offs * 2, fflags2, hdr_pos, buf_pos, 
    min(chroma_width, 40));

  buf_pos = buf + 16 + offs3;
  offs = le2me_32(*(uint32_t *)buf_pos);
  buf_pos += 4;

  iv_Decode_Chunk(s, s->cur_frame->Ubuf, s->ref_frame->Ubuf, chroma_width, 
    chroma_height, buf_pos + offs * 2, fflags2, hdr_pos, buf_pos, 
    min(chroma_width, 40));

  return 8;
}

typedef struct {
  long xpos;
  long ypos;
  long width;
  long height;
  long split_flag;
  long split_direction;
  long usl7;
} ustr_t;

/* ---------------------------------------------------------------------- */

#define LV1_CHECK(buf1,rle_v3,lv1,lp2)  \
  if((lv1 & 0x80) != 0) {   \
    if(rle_v3 != 0)         \
      rle_v3 = 0;           \
    else {                  \
      rle_v3 = 1;           \
      buf1 -= 2;            \
    }                       \
  }                         \
  lp2 = 4;


#define RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)  \
  if(rle_v3 == 0) {         \
    rle_v2 = *buf1;         \
    rle_v1 = 1;             \
    if(rle_v2 > 32) {       \
      rle_v2 -= 32;         \
      rle_v1 = 0;           \
    }                       \
    rle_v3 = 1;             \
  }                         \
  buf1--;


#define LP2_CHECK(buf1,rle_v3,lp2)  \
  if(lp2 == 0 && rle_v3 != 0)     \
    rle_v3 = 0;           \
  else {                  \
    buf1--;               \
    rle_v3 = 1;           \
  }


#define RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2) \
  rle_v2--;             \
  if(rle_v2 == 0) {     \
    rle_v3 = 0;         \
    buf1 += 2;          \
  }                     \
  lp2 = 4;

static void iv_Decode_Chunk(Indeo3DecodeContext *s,
  unsigned char *cur, unsigned char *ref, int width, int height, 
  unsigned char *buf1, long fflags2, unsigned char *hdr,
  unsigned char *buf2, int min_width_160)
{
  unsigned char bit_buf;
  unsigned long bit_pos, lv, lv1, lv2;
  long *width_tbl, width_tbl_arr[10];
  char *ref_vectors;
  unsigned char *cur_frm_pos, *ref_frm_pos, *cp, *cp2;
  uint32_t *cur_lp, *ref_lp;
  const uint32_t *correction_lp[2], *correctionloworder_lp[2], *correctionhighorder_lp[2];
  unsigned short *correction_type_sp[2];
  ustr_t strip_tbl[20], *strip;
  int i, j, k, lp1, lp2, flag1, cmd, blks_width, blks_height, region_160_width,
    rle_v1, rle_v2, rle_v3;

  bit_buf = 0;
  ref_vectors = NULL;

  width_tbl = width_tbl_arr + 1;
  i = (width < 0 ? width + 3 : width)/4;
  for(j = -1; j < 8; j++) 
    width_tbl[j] = i * j;

  strip = strip_tbl;

  for(region_160_width = 0; region_160_width < (width - min_width_160); region_160_width += min_width_160);

  strip->ypos = strip->xpos = 0;
  for(strip->width = min_width_160; width > strip->width; strip->width *= 2);
  strip->height = height;
  strip->split_direction = 0;
  strip->split_flag = 0;
  strip->usl7 = 0;

  bit_pos = 0;

  rle_v1 = rle_v2 = rle_v3 = 0;

  while(strip >= strip_tbl) {
    if(bit_pos <= 0) {
      bit_pos = 8;
      bit_buf = *buf1++;
    }

    bit_pos -= 2;
    cmd = (bit_buf >> bit_pos) & 0x03;

    if(cmd == 0) {
      strip++;
      memcpy(strip, strip-1, sizeof(ustr_t));
      strip->split_flag = 1;
      strip->split_direction = 0;
      strip->height = (strip->height > 8 ? ((strip->height+8)>>4)<<3 : 4);
      continue;
    } else if(cmd == 1) {
      strip++;
      memcpy(strip, strip-1, sizeof(ustr_t));
      strip->split_flag = 1;
      strip->split_direction = 1;
      strip->width = (strip->width > 8 ? ((strip->width+8)>>4)<<3 : 4);
      continue;
    } else if(cmd == 2) {
      if(strip->usl7 == 0) {
        strip->usl7 = 1;
        ref_vectors = NULL;
        continue;
      }
    } else if(cmd == 3) {
      if(strip->usl7 == 0) {
        strip->usl7 = 1;
        ref_vectors = buf2 + (*buf1 * 2);
        buf1++;
        continue;
      }
    }

    cur_frm_pos = cur + width * strip->ypos + strip->xpos;

    if((blks_width = strip->width) < 0) 
      blks_width += 3;
    blks_width >>= 2;
    blks_height = strip->height;

    if(ref_vectors != NULL) {
      ref_frm_pos = ref + (ref_vectors[0] + strip->ypos) * width +
        ref_vectors[1] + strip->xpos;
    } else 
      ref_frm_pos = cur_frm_pos - width_tbl[4];

    if(cmd == 2) {
      if(bit_pos <= 0) {
        bit_pos = 8;
        bit_buf = *buf1++;
      }

      bit_pos -= 2;
      cmd = (bit_buf >> bit_pos) & 0x03;

      if(cmd == 0 || ref_vectors != NULL) {
        for(lp1 = 0; lp1 < blks_width; lp1++) {
          for(i = 0, j = 0; i < blks_height; i++, j += width_tbl[1])
            ((uint32_t *)cur_frm_pos)[j] = ((uint32_t *)ref_frm_pos)[j];
          cur_frm_pos += 4;
          ref_frm_pos += 4;
        }
      } else if(cmd != 1) 
        return;
    } else {
      k = *buf1 >> 4;
      j = *buf1 & 0x0f;
      buf1++;
      lv = j + fflags2;

      if((lv - 8) <= 7 && (k == 0 || k == 3 || k == 10)) {
        cp2 = s->ModPred + ((lv - 8) << 7);
        cp = ref_frm_pos;
        for(i = 0; i < blks_width << 2; i++) { 
            int v = *cp >> 1;
            *(cp++) = cp2[v]; 
        }
      }

      if(k == 1 || k == 4) {
        lv = (hdr[j] & 0xf) + fflags2;
        correction_type_sp[0] = s->corrector_type + (lv << 8);
        correction_lp[0] = correction + (lv << 8);
        lv = (hdr[j] >> 4) + fflags2;
        correction_lp[1] = correction + (lv << 8);
        correction_type_sp[1] = s->corrector_type + (lv << 8);
      } else {
        correctionloworder_lp[0] = correctionloworder_lp[1] = correctionloworder + (lv << 8);
        correctionhighorder_lp[0] = correctionhighorder_lp[1] = correctionhighorder + (lv << 8);
        correction_type_sp[0] = correction_type_sp[1] = s->corrector_type + (lv << 8);
        correction_lp[0] = correction_lp[1] = correction + (lv << 8);
      }

      switch(k) {
        case 1:
        case 0:                    /********** CASE 0 **********/
          for( ; blks_height > 0; blks_height -= 4) {
            for(lp1 = 0; lp1 < blks_width; lp1++) {
              for(lp2 = 0; lp2 < 4; ) {
                k = *buf1++;
                cur_lp = ((uint32_t *)cur_frm_pos) + width_tbl[lp2];
                ref_lp = ((uint32_t *)ref_frm_pos) + width_tbl[lp2];

                switch(correction_type_sp[0][k]) {
                  case 0:
                    *cur_lp = ((*ref_lp >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    lp2++;
                    break;
                  case 1:
                    ((unsigned short *)cur_lp)[0] = ((((unsigned short *)(ref_lp))[0] >> 1)
                      + correction_lp[lp2 & 0x01][*buf1++]) << 1;
                    ((unsigned short *)cur_lp)[1] = ((((unsigned short *)(ref_lp))[1] >> 1)
                      + correction_lp[lp2 & 0x01][k]) << 1;
                    lp2++;
                    break;
                  case 2:
                    if(lp2 == 0) {
                      for(i = 0, j = 0; i < 2; i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];
                      lp2 += 2;
                    }
                    break;
                  case 3:
                    if(lp2 < 2) {
                      for(i = 0, j = 0; i < (3 - lp2); i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];
                      lp2 = 3;
                    }
                    break;
                  case 8:
                    if(lp2 == 0) {
                      RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)

                      if(rle_v1 == 1 || ref_vectors != NULL) {
                        for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                          cur_lp[j] = ref_lp[j];
                      }

                      RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2)
                      break;
                    } else {
                      rle_v1 = 1;
                      rle_v2 = *buf1 - 1;
                    }
                  case 5:
                      LP2_CHECK(buf1,rle_v3,lp2)
                  case 4:
                    for(i = 0, j = 0; i < (4 - lp2); i++, j += width_tbl[1])
                      cur_lp[j] = ref_lp[j];
                    lp2 = 4;
                    break;

                  case 7:
                    if(rle_v3 != 0) 
                      rle_v3 = 0;
                    else {
                      buf1--;
                      rle_v3 = 1;
                    }
                  case 6:
                    if(ref_vectors != NULL) {
                      for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];
                    }
                    lp2 = 4;
                    break;

                  case 9:
                    lv1 = *buf1++;
                    lv = (lv1 & 0x7F) << 1;
                    lv += (lv << 8);
                    lv += (lv << 16);
                    for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                      cur_lp[j] = lv;

                    LV1_CHECK(buf1,rle_v3,lv1,lp2)
                    break;
                  default: 
                    return;
                }
              }

              cur_frm_pos += 4;
              ref_frm_pos += 4;
            }

            cur_frm_pos += ((width - blks_width) * 4);
            ref_frm_pos += ((width - blks_width) * 4);
          }
          break;

        case 4:
        case 3:                    /********** CASE 3 **********/
          if(ref_vectors != NULL) 
            return;
          flag1 = 1;

          for( ; blks_height > 0; blks_height -= 8) {
            for(lp1 = 0; lp1 < blks_width; lp1++) {
              for(lp2 = 0; lp2 < 4; ) {
                k = *buf1++;

                cur_lp = ((uint32_t *)cur_frm_pos) + width_tbl[lp2 * 2];
                ref_lp = ((uint32_t *)cur_frm_pos) + width_tbl[(lp2 * 2) - 1];

                switch(correction_type_sp[lp2 & 0x01][k]) {
                  case 0:
                    cur_lp[width_tbl[1]] = ((*ref_lp >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    if(lp2 > 0 || flag1 == 0 || strip->ypos != 0)
                      cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                    else
                      cur_lp[0] = ((*ref_lp >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    lp2++;
                    break;

                  case 1:
                    ((unsigned short *)cur_lp)[width_tbl[2]] =
                      ((((unsigned short *)ref_lp)[0] >> 1) + correction_lp[lp2 & 0x01][*buf1++]) << 1;
                    ((unsigned short *)cur_lp)[width_tbl[2]+1] =
                      ((((unsigned short *)ref_lp)[1] >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    if(lp2 > 0 || flag1 == 0 || strip->ypos != 0)
                      cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                    else
                      cur_lp[0] = cur_lp[width_tbl[1]];
                    lp2++;
                    break;

                  case 2:
                    if(lp2 == 0) {
                      for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                        cur_lp[j] = *ref_lp;
                      lp2 += 2;
                    }
                    break;

                  case 3:
                    if(lp2 < 2) {
                      for(i = 0, j = 0; i < 6 - (lp2 * 2); i++, j += width_tbl[1])
                        cur_lp[j] = *ref_lp;
                      lp2 = 3;
                    }
                    break;

                  case 6:
                    lp2 = 4;
                    break;

                  case 7:
                    if(rle_v3 != 0) 
                      rle_v3 = 0;
                    else {
                      buf1--;
                      rle_v3 = 1;
                    }
                    lp2 = 4;
                    break;

                  case 8:
                    if(lp2 == 0) {
                      RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)

                      if(rle_v1 == 1) {
                        for(i = 0, j = 0; i < 8; i++, j += width_tbl[1])
                          cur_lp[j] = ref_lp[j];
                      }

                      RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2)
                      break;
                    } else {
                      rle_v2 = (*buf1) - 1;
                      rle_v1 = 1;
                    }
                  case 5:
                      LP2_CHECK(buf1,rle_v3,lp2)
                  case 4:
                    for(i = 0, j = 0; i < 8 - (lp2 * 2); i++, j += width_tbl[1])
                      cur_lp[j] = *ref_lp;
                    lp2 = 4;
                    break;

                  case 9:
                    av_log(s->avctx, AV_LOG_ERROR, "UNTESTED.\n");
                    lv1 = *buf1++;
                    lv = (lv1 & 0x7F) << 1;
                    lv += (lv << 8);
                    lv += (lv << 16);

                    for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                      cur_lp[j] = lv;

                    LV1_CHECK(buf1,rle_v3,lv1,lp2)
                    break;

                  default: 
                    return;
                }
              }

              cur_frm_pos += 4;
            }

            cur_frm_pos += (((width * 2) - blks_width) * 4);
            flag1 = 0;
          }
          break;

        case 10:                    /********** CASE 10 **********/
          if(ref_vectors == NULL) {
            flag1 = 1;

            for( ; blks_height > 0; blks_height -= 8) {
              for(lp1 = 0; lp1 < blks_width; lp1 += 2) {
                for(lp2 = 0; lp2 < 4; ) {
                  k = *buf1++;
                  cur_lp = ((uint32_t *)cur_frm_pos) + width_tbl[lp2 * 2];
                  ref_lp = ((uint32_t *)cur_frm_pos) + width_tbl[(lp2 * 2) - 1];
                  lv1 = ref_lp[0];
                  lv2 = ref_lp[1];
                  if(lp2 == 0 && flag1 != 0) {
                    lv1 = lv1 & 0x00FF00FF;
                    lv1 = (lv1 << 8) | lv1;
                    lv2 = lv2 & 0x00FF00FF;
                    lv2 = (lv2 << 8) | lv2;
                  }

                  switch(correction_type_sp[lp2 & 0x01][k]) {
                    case 0:
                      cur_lp[width_tbl[1]] = ((lv1 >> 1) + correctionloworder_lp[lp2 & 0x01][k]) << 1;
                      cur_lp[width_tbl[1]+1] = ((lv2 >> 1) + correctionhighorder_lp[lp2 & 0x01][k]) << 1;
                      if(lp2 > 0 || strip->ypos != 0 || flag1 == 0) {
                        cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                        cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                      } else {
                        cur_lp[0] = cur_lp[width_tbl[1]];
                        cur_lp[1] = cur_lp[width_tbl[1]+1];
                      }
                      lp2++;
                      break;

                    case 1:
                      cur_lp[width_tbl[1]] = ((lv1 >> 1) + correctionloworder_lp[lp2 & 0x01][*buf1++]) << 1;
                      cur_lp[width_tbl[1]+1] = ((lv2 >> 1) + correctionloworder_lp[lp2 & 0x01][k]) << 1;
                      if(lp2 > 0 || strip->ypos != 0 || flag1 == 0) {
                        cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                        cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                      } else {
                        cur_lp[0] = cur_lp[width_tbl[1]];
                        cur_lp[1] = cur_lp[width_tbl[1]+1];
                      }
                      lp2++;
                      break;

                    case 2:
                      if(lp2 == 0) {
                        if(flag1 != 0) {
                          for(i = 0, j = width_tbl[1]; i < 3; i++, j += width_tbl[1]) {
                            cur_lp[j] = lv1;
                            cur_lp[j+1] = lv2;
                          }
                          cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                          cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                        } else {
                          for(i = 0, j = 0; i < 4; i++, j += width_tbl[1]) {
                            cur_lp[j] = lv1;
                            cur_lp[j+1] = lv2;
                          }
                        }
                        lp2 += 2;
                      }
                      break;

                    case 3:
                      if(lp2 < 2) {
                        if(lp2 == 0 && flag1 != 0) {
                          for(i = 0, j = width_tbl[1]; i < 5; i++, j += width_tbl[1]) {
                            cur_lp[j] = lv1;
                            cur_lp[j+1] = lv2;
                          }
                          cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                          cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                        } else {
                          for(i = 0, j = 0; i < 6 - (lp2 * 2); i++, j += width_tbl[1]) {
                            cur_lp[j] = lv1;
                            cur_lp[j+1] = lv2;
                          }
                        }
                        lp2 = 3;
                      }
                      break;

                    case 8:
                      if(lp2 == 0) {
                        RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)
                        if(rle_v1 == 1) {
                          if(flag1 != 0) {
                            for(i = 0, j = width_tbl[1]; i < 7; i++, j += width_tbl[1]) {
                              cur_lp[j] = lv1;
                              cur_lp[j+1] = lv2;
                            }
                            cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                            cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                          } else {
                            for(i = 0, j = 0; i < 8; i++, j += width_tbl[1]) {
                              cur_lp[j] = lv1;
                              cur_lp[j+1] = lv2;
                            }
                          }
                        }
                        RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2)
                        break;
                      } else {
                        rle_v1 = 1;
                        rle_v2 = (*buf1) - 1;
                      }
                    case 5:
                        LP2_CHECK(buf1,rle_v3,lp2)
                    case 4:
                      if(lp2 == 0 && flag1 != 0) {
                        for(i = 0, j = width_tbl[1]; i < 7; i++, j += width_tbl[1]) {
                          cur_lp[j] = lv1;
                          cur_lp[j+1] = lv2;
                        }
                        cur_lp[0] = ((cur_lp[-width_tbl[1]] >> 1) + (cur_lp[width_tbl[1]] >> 1)) & 0xFEFEFEFE;
                        cur_lp[1] = ((cur_lp[-width_tbl[1]+1] >> 1) + (cur_lp[width_tbl[1]+1] >> 1)) & 0xFEFEFEFE;
                      } else {
                        for(i = 0, j = 0; i < 8 - (lp2 * 2); i++, j += width_tbl[1]) {
                          cur_lp[j] = lv1;
                          cur_lp[j+1] = lv2;
                        }
                      }
                      lp2 = 4;
                      break;

                    case 6:
                      lp2 = 4;
                      break;

                    case 7:
                      if(lp2 == 0) {
                        if(rle_v3 != 0) 
                          rle_v3 = 0;
                        else {
                          buf1--;
                          rle_v3 = 1;
                        }
                        lp2 = 4;
                      }
                      break;

                    case 9:
                      av_log(s->avctx, AV_LOG_ERROR, "UNTESTED.\n");
                      lv1 = *buf1;
                      lv = (lv1 & 0x7F) << 1;
                      lv += (lv << 8);
                      lv += (lv << 16);
                      for(i = 0, j = 0; i < 8; i++, j += width_tbl[1])
                        cur_lp[j] = lv;
                      LV1_CHECK(buf1,rle_v3,lv1,lp2)
                      break;

                    default: 
                      return;
                  }
                }

                cur_frm_pos += 8;
              }

              cur_frm_pos += (((width * 2) - blks_width) * 4);
              flag1 = 0;
            }
          } else {
            for( ; blks_height > 0; blks_height -= 8) {
              for(lp1 = 0; lp1 < blks_width; lp1 += 2) {
                for(lp2 = 0; lp2 < 4; ) {
                  k = *buf1++;
                  cur_lp = ((uint32_t *)cur_frm_pos) + width_tbl[lp2 * 2];
                  ref_lp = ((uint32_t *)ref_frm_pos) + width_tbl[lp2 * 2];

                  switch(correction_type_sp[lp2 & 0x01][k]) {
                    case 0:
                      lv1 = correctionloworder_lp[lp2 & 0x01][k];
                      lv2 = correctionhighorder_lp[lp2 & 0x01][k];
                      cur_lp[0] = ((ref_lp[0] >> 1) + lv1) << 1;
                      cur_lp[1] = ((ref_lp[1] >> 1) + lv2) << 1;
                      cur_lp[width_tbl[1]] = ((ref_lp[width_tbl[1]] >> 1) + lv1) << 1;
                      cur_lp[width_tbl[1]+1] = ((ref_lp[width_tbl[1]+1] >> 1) + lv2) << 1;
                      lp2++;
                      break;

                    case 1:
                      lv1 = correctionloworder_lp[lp2 & 0x01][*buf1++];
                      lv2 = correctionloworder_lp[lp2 & 0x01][k];
                      cur_lp[0] = ((ref_lp[0] >> 1) + lv1) << 1;
                      cur_lp[1] = ((ref_lp[1] >> 1) + lv2) << 1;
                      cur_lp[width_tbl[1]] = ((ref_lp[width_tbl[1]] >> 1) + lv1) << 1;
                      cur_lp[width_tbl[1]+1] = ((ref_lp[width_tbl[1]+1] >> 1) + lv2) << 1;
                      lp2++;
                      break;

                    case 2:
                      if(lp2 == 0) {
                        for(i = 0, j = 0; i < 4; i++, j += width_tbl[1]) {
                          cur_lp[j] = ref_lp[j];
                          cur_lp[j+1] = ref_lp[j+1];
                        }
                        lp2 += 2;
                      }
                      break;

                    case 3:
                      if(lp2 < 2) {
                        for(i = 0, j = 0; i < 6 - (lp2 * 2); i++, j += width_tbl[1]) {
                          cur_lp[j] = ref_lp[j];
                          cur_lp[j+1] = ref_lp[j+1];
                        }
                        lp2 = 3;
                      }
                      break;

                    case 8:
                      if(lp2 == 0) {
                        RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)
                        for(i = 0, j = 0; i < 8; i++, j += width_tbl[1]) {
                          ((uint32_t *)cur_frm_pos)[j] = ((uint32_t *)ref_frm_pos)[j];
                          ((uint32_t *)cur_frm_pos)[j+1] = ((uint32_t *)ref_frm_pos)[j+1];
                        }
                        RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2)
                        break;
                      } else {
                        rle_v1 = 1;
                        rle_v2 = (*buf1) - 1;
                      }
                    case 5:
                    case 7:
                        LP2_CHECK(buf1,rle_v3,lp2)
                    case 6:
                    case 4:
                      for(i = 0, j = 0; i < 8 - (lp2 * 2); i++, j += width_tbl[1]) {
                        cur_lp[j] = ref_lp[j];
                        cur_lp[j+1] = ref_lp[j+1];
                      }
                      lp2 = 4;
                      break;

                    case 9:
                      av_log(s->avctx, AV_LOG_ERROR, "UNTESTED.\n");
                      lv1 = *buf1;
                      lv = (lv1 & 0x7F) << 1;
                      lv += (lv << 8);
                      lv += (lv << 16);
                      for(i = 0, j = 0; i < 8; i++, j += width_tbl[1])
                        ((uint32_t *)cur_frm_pos)[j] = ((uint32_t *)cur_frm_pos)[j+1] = lv;
                      LV1_CHECK(buf1,rle_v3,lv1,lp2)
                      break;

                    default: 
                      return;
                  }
                }

                cur_frm_pos += 8;
                ref_frm_pos += 8;
              }

              cur_frm_pos += (((width * 2) - blks_width) * 4);
              ref_frm_pos += (((width * 2) - blks_width) * 4);
            }
          }
          break;

        case 11:                    /********** CASE 11 **********/
          if(ref_vectors == NULL) 
            return;

          for( ; blks_height > 0; blks_height -= 8) {
            for(lp1 = 0; lp1 < blks_width; lp1++) {
              for(lp2 = 0; lp2 < 4; ) {
                k = *buf1++;
                cur_lp = ((uint32_t *)cur_frm_pos) + width_tbl[lp2 * 2];
                ref_lp = ((uint32_t *)ref_frm_pos) + width_tbl[lp2 * 2];

                switch(correction_type_sp[lp2 & 0x01][k]) {
                  case 0:
                    cur_lp[0] = ((*ref_lp >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    cur_lp[width_tbl[1]] = ((ref_lp[width_tbl[1]] >> 1) + correction_lp[lp2 & 0x01][k]) << 1;
                    lp2++;
                    break;

                  case 1:
                    lv1 = (unsigned short)(correction_lp[lp2 & 0x01][*buf1++]);
                    lv2 = (unsigned short)(correction_lp[lp2 & 0x01][k]);
                    ((unsigned short *)cur_lp)[0] = ((((unsigned short *)ref_lp)[0] >> 1) + lv1) << 1;
                    ((unsigned short *)cur_lp)[1] = ((((unsigned short *)ref_lp)[1] >> 1) + lv2) << 1;
                    ((unsigned short *)cur_lp)[width_tbl[2]] = ((((unsigned short *)ref_lp)[width_tbl[2]] >> 1) + lv1) << 1;
                    ((unsigned short *)cur_lp)[width_tbl[2]+1] = ((((unsigned short *)ref_lp)[width_tbl[2]+1] >> 1) + lv2) << 1;
                    lp2++;
                    break;

                  case 2:
                    if(lp2 == 0) {
                      for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];
                      lp2 += 2;
                    }
                    break;

                  case 3:
                    if(lp2 < 2) {
                      for(i = 0, j = 0; i < 6 - (lp2 * 2); i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];
                      lp2 = 3;
                    }
                    break;

                  case 8:
                    if(lp2 == 0) {
                      RLE_V3_CHECK(buf1,rle_v1,rle_v2,rle_v3)

                      for(i = 0, j = 0; i < 8; i++, j += width_tbl[1])
                        cur_lp[j] = ref_lp[j];

                      RLE_V2_CHECK(buf1,rle_v2, rle_v3,lp2)
                      break;
                    } else {
                      rle_v1 = 1;
                      rle_v2 = (*buf1) - 1;
                    }
                  case 5:
                  case 7:
                      LP2_CHECK(buf1,rle_v3,lp2)
                  case 4:
                  case 6:
                    for(i = 0, j = 0; i < 8 - (lp2 * 2); i++, j += width_tbl[1])
                      cur_lp[j] = ref_lp[j];
                    lp2 = 4;
                    break;

                case 9:
                  av_log(s->avctx, AV_LOG_ERROR, "UNTESTED.\n");
                  lv1 = *buf1++;
                  lv = (lv1 & 0x7F) << 1;
                  lv += (lv << 8);
                  lv += (lv << 16);
                  for(i = 0, j = 0; i < 4; i++, j += width_tbl[1])
                    cur_lp[j] = lv;
                  LV1_CHECK(buf1,rle_v3,lv1,lp2)
                  break;

                  default: 
                    return;
                }
              }

              cur_frm_pos += 4;
              ref_frm_pos += 4;
            }

            cur_frm_pos += (((width * 2) - blks_width) * 4);
            ref_frm_pos += (((width * 2) - blks_width) * 4);
          }
          break;

        default: 
          return;
      }
    }

    if(strip < strip_tbl) 
      return;

    for( ; strip >= strip_tbl; strip--) {
      if(strip->split_flag != 0) {
        strip->split_flag = 0;
        strip->usl7 = (strip-1)->usl7;

        if(strip->split_direction) {
          strip->xpos += strip->width;
          strip->width = (strip-1)->width - strip->width;
          if(region_160_width <= strip->xpos && width < strip->width + strip->xpos)
            strip->width = width - strip->xpos;
        } else {
          strip->ypos += strip->height;
          strip->height = (strip-1)->height - strip->height;
        }
        break;
      }
    }
  }
}

static int indeo3_decode_init(AVCodecContext *avctx)
{
    Indeo3DecodeContext *s = avctx->priv_data;

    s->avctx = avctx;
    s->width = avctx->width;
    s->height = avctx->height;
    avctx->pix_fmt = PIX_FMT_YUV410P;
    avctx->has_b_frames = 0;

    build_modpred(s);
    iv_alloc_frames(s);

    return 0;
}

static int indeo3_decode_frame(AVCodecContext *avctx,
                               void *data, int *data_size,
                               unsigned char *buf, int buf_size)
{
    Indeo3DecodeContext *s=avctx->priv_data;
    unsigned char *src, *dest;
    int y;

    /* no supplementary picture */
    if (buf_size == 0) {
        return 0;
    }

    iv_decode_frame(s, buf, buf_size);

    if(s->frame.data[0])
        avctx->release_buffer(avctx, &s->frame);

    s->frame.reference = 0;
    if(avctx->get_buffer(avctx, &s->frame) < 0) {
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return -1;
    }

    src = s->cur_frame->Ybuf;
    dest = s->frame.data[0];
    for (y = 0; y < s->height; y++) {
      memcpy(dest, src, s->cur_frame->y_w);
      src += s->cur_frame->y_w;
      dest += s->frame.linesize[0];
    }

    src = s->cur_frame->Ubuf;
    dest = s->frame.data[1];
    for (y = 0; y < s->height / 4; y++) {
      memcpy(dest, src, s->cur_frame->uv_w);
      src += s->cur_frame->uv_w;
      dest += s->frame.linesize[1];
    }

    src = s->cur_frame->Vbuf;
    dest = s->frame.data[2];
    for (y = 0; y < s->height / 4; y++) {
      memcpy(dest, src, s->cur_frame->uv_w);
      src += s->cur_frame->uv_w;
      dest += s->frame.linesize[2];
    }

    *data_size=sizeof(AVFrame);
    *(AVFrame*)data= s->frame;

    return buf_size;
}

static int indeo3_decode_end(AVCodecContext *avctx)
{
    Indeo3DecodeContext *s = avctx->priv_data;

    iv_free_func(s);

    return 0;
}

AVCodec indeo3_decoder = {
    "indeo3",
    CODEC_TYPE_VIDEO,
    CODEC_ID_INDEO3,
    sizeof(Indeo3DecodeContext),
    indeo3_decode_init,
    NULL,
    indeo3_decode_end,
    indeo3_decode_frame,
    0,
    NULL
};