Libav

libavcodec/vp3.c

Go to the documentation of this file.
00001 /*
00002  * Copyright (C) 2003-2004 the ffmpeg project
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00032 #include <stdio.h>
00033 #include <stdlib.h>
00034 #include <string.h>
00035 
00036 #include "avcodec.h"
00037 #include "dsputil.h"
00038 #include "get_bits.h"
00039 
00040 #include "vp3data.h"
00041 #include "xiph.h"
00042 
00043 #define FRAGMENT_PIXELS 8
00044 
00045 static av_cold int vp3_decode_end(AVCodecContext *avctx);
00046 
00047 //FIXME split things out into their own arrays
00048 typedef struct Vp3Fragment {
00049     int16_t dc;
00050     uint8_t coding_method;
00051     uint8_t qpi;
00052 } Vp3Fragment;
00053 
00054 #define SB_NOT_CODED        0
00055 #define SB_PARTIALLY_CODED  1
00056 #define SB_FULLY_CODED      2
00057 
00058 // This is the maximum length of a single long bit run that can be encoded
00059 // for superblock coding or block qps. Theora special-cases this to read a
00060 // bit instead of flipping the current bit to allow for runs longer than 4129.
00061 #define MAXIMUM_LONG_BIT_RUN 4129
00062 
00063 #define MODE_INTER_NO_MV      0
00064 #define MODE_INTRA            1
00065 #define MODE_INTER_PLUS_MV    2
00066 #define MODE_INTER_LAST_MV    3
00067 #define MODE_INTER_PRIOR_LAST 4
00068 #define MODE_USING_GOLDEN     5
00069 #define MODE_GOLDEN_MV        6
00070 #define MODE_INTER_FOURMV     7
00071 #define CODING_MODE_COUNT     8
00072 
00073 /* special internal mode */
00074 #define MODE_COPY             8
00075 
00076 /* There are 6 preset schemes, plus a free-form scheme */
00077 static const int ModeAlphabet[6][CODING_MODE_COUNT] =
00078 {
00079     /* scheme 1: Last motion vector dominates */
00080     {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
00081          MODE_INTER_PLUS_MV,    MODE_INTER_NO_MV,
00082          MODE_INTRA,            MODE_USING_GOLDEN,
00083          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00084 
00085     /* scheme 2 */
00086     {    MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
00087          MODE_INTER_NO_MV,      MODE_INTER_PLUS_MV,
00088          MODE_INTRA,            MODE_USING_GOLDEN,
00089          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00090 
00091     /* scheme 3 */
00092     {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
00093          MODE_INTER_PRIOR_LAST, MODE_INTER_NO_MV,
00094          MODE_INTRA,            MODE_USING_GOLDEN,
00095          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00096 
00097     /* scheme 4 */
00098     {    MODE_INTER_LAST_MV,    MODE_INTER_PLUS_MV,
00099          MODE_INTER_NO_MV,      MODE_INTER_PRIOR_LAST,
00100          MODE_INTRA,            MODE_USING_GOLDEN,
00101          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00102 
00103     /* scheme 5: No motion vector dominates */
00104     {    MODE_INTER_NO_MV,      MODE_INTER_LAST_MV,
00105          MODE_INTER_PRIOR_LAST, MODE_INTER_PLUS_MV,
00106          MODE_INTRA,            MODE_USING_GOLDEN,
00107          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00108 
00109     /* scheme 6 */
00110     {    MODE_INTER_NO_MV,      MODE_USING_GOLDEN,
00111          MODE_INTER_LAST_MV,    MODE_INTER_PRIOR_LAST,
00112          MODE_INTER_PLUS_MV,    MODE_INTRA,
00113          MODE_GOLDEN_MV,        MODE_INTER_FOURMV },
00114 
00115 };
00116 
00117 static const uint8_t hilbert_offset[16][2] = {
00118     {0,0}, {1,0}, {1,1}, {0,1},
00119     {0,2}, {0,3}, {1,3}, {1,2},
00120     {2,2}, {2,3}, {3,3}, {3,2},
00121     {3,1}, {2,1}, {2,0}, {3,0}
00122 };
00123 
00124 #define MIN_DEQUANT_VAL 2
00125 
00126 typedef struct Vp3DecodeContext {
00127     AVCodecContext *avctx;
00128     int theora, theora_tables;
00129     int version;
00130     int width, height;
00131     int chroma_x_shift, chroma_y_shift;
00132     AVFrame golden_frame;
00133     AVFrame last_frame;
00134     AVFrame current_frame;
00135     int keyframe;
00136     DSPContext dsp;
00137     int flipped_image;
00138     int last_slice_end;
00139 
00140     int qps[3];
00141     int nqps;
00142     int last_qps[3];
00143 
00144     int superblock_count;
00145     int y_superblock_width;
00146     int y_superblock_height;
00147     int y_superblock_count;
00148     int c_superblock_width;
00149     int c_superblock_height;
00150     int c_superblock_count;
00151     int u_superblock_start;
00152     int v_superblock_start;
00153     unsigned char *superblock_coding;
00154 
00155     int macroblock_count;
00156     int macroblock_width;
00157     int macroblock_height;
00158 
00159     int fragment_count;
00160     int fragment_width[2];
00161     int fragment_height[2];
00162 
00163     Vp3Fragment *all_fragments;
00164     int fragment_start[3];
00165     int data_offset[3];
00166 
00167     int8_t (*motion_val[2])[2];
00168 
00169     ScanTable scantable;
00170 
00171     /* tables */
00172     uint16_t coded_dc_scale_factor[64];
00173     uint32_t coded_ac_scale_factor[64];
00174     uint8_t base_matrix[384][64];
00175     uint8_t qr_count[2][3];
00176     uint8_t qr_size [2][3][64];
00177     uint16_t qr_base[2][3][64];
00178 
00196     int16_t *dct_tokens[3][64];
00197     int16_t *dct_tokens_base;
00198 #define TOKEN_EOB(eob_run)              ((eob_run) << 2)
00199 #define TOKEN_ZERO_RUN(coeff, zero_run) (((coeff) << 9) + ((zero_run) << 2) + 1)
00200 #define TOKEN_COEFF(coeff)              (((coeff) << 2) + 2)
00201 
00205     int num_coded_frags[3][64];
00206     int total_num_coded_frags;
00207 
00208     /* this is a list of indexes into the all_fragments array indicating
00209      * which of the fragments are coded */
00210     int *coded_fragment_list[3];
00211 
00212     VLC dc_vlc[16];
00213     VLC ac_vlc_1[16];
00214     VLC ac_vlc_2[16];
00215     VLC ac_vlc_3[16];
00216     VLC ac_vlc_4[16];
00217 
00218     VLC superblock_run_length_vlc;
00219     VLC fragment_run_length_vlc;
00220     VLC mode_code_vlc;
00221     VLC motion_vector_vlc;
00222 
00223     /* these arrays need to be on 16-byte boundaries since SSE2 operations
00224      * index into them */
00225     DECLARE_ALIGNED(16, int16_t, qmat)[3][2][3][64];     //<qmat[qpi][is_inter][plane]
00226 
00227     /* This table contains superblock_count * 16 entries. Each set of 16
00228      * numbers corresponds to the fragment indexes 0..15 of the superblock.
00229      * An entry will be -1 to indicate that no entry corresponds to that
00230      * index. */
00231     int *superblock_fragments;
00232 
00233     /* This is an array that indicates how a particular macroblock
00234      * is coded. */
00235     unsigned char *macroblock_coding;
00236 
00237     uint8_t edge_emu_buffer[9*2048]; //FIXME dynamic alloc
00238     int8_t qscale_table[2048]; //FIXME dynamic alloc (width+15)/16
00239 
00240     /* Huffman decode */
00241     int hti;
00242     unsigned int hbits;
00243     int entries;
00244     int huff_code_size;
00245     uint32_t huffman_table[80][32][2];
00246 
00247     uint8_t filter_limit_values[64];
00248     DECLARE_ALIGNED(8, int, bounding_values_array)[256+2];
00249 } Vp3DecodeContext;
00250 
00251 /************************************************************************
00252  * VP3 specific functions
00253  ************************************************************************/
00254 
00255 /*
00256  * This function sets up all of the various blocks mappings:
00257  * superblocks <-> fragments, macroblocks <-> fragments,
00258  * superblocks <-> macroblocks
00259  *
00260  * Returns 0 is successful; returns 1 if *anything* went wrong.
00261  */
00262 static int init_block_mapping(Vp3DecodeContext *s)
00263 {
00264     int sb_x, sb_y, plane;
00265     int x, y, i, j = 0;
00266 
00267     for (plane = 0; plane < 3; plane++) {
00268         int sb_width    = plane ? s->c_superblock_width  : s->y_superblock_width;
00269         int sb_height   = plane ? s->c_superblock_height : s->y_superblock_height;
00270         int frag_width  = s->fragment_width[!!plane];
00271         int frag_height = s->fragment_height[!!plane];
00272 
00273         for (sb_y = 0; sb_y < sb_height; sb_y++)
00274             for (sb_x = 0; sb_x < sb_width; sb_x++)
00275                 for (i = 0; i < 16; i++) {
00276                     x = 4*sb_x + hilbert_offset[i][0];
00277                     y = 4*sb_y + hilbert_offset[i][1];
00278 
00279                     if (x < frag_width && y < frag_height)
00280                         s->superblock_fragments[j++] = s->fragment_start[plane] + y*frag_width + x;
00281                     else
00282                         s->superblock_fragments[j++] = -1;
00283                 }
00284     }
00285 
00286     return 0;  /* successful path out */
00287 }
00288 
00289 /*
00290  * This function sets up the dequantization tables used for a particular
00291  * frame.
00292  */
00293 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
00294 {
00295     int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
00296     int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
00297     int i, plane, inter, qri, bmi, bmj, qistart;
00298 
00299     for(inter=0; inter<2; inter++){
00300         for(plane=0; plane<3; plane++){
00301             int sum=0;
00302             for(qri=0; qri<s->qr_count[inter][plane]; qri++){
00303                 sum+= s->qr_size[inter][plane][qri];
00304                 if(s->qps[qpi] <= sum)
00305                     break;
00306             }
00307             qistart= sum - s->qr_size[inter][plane][qri];
00308             bmi= s->qr_base[inter][plane][qri  ];
00309             bmj= s->qr_base[inter][plane][qri+1];
00310             for(i=0; i<64; i++){
00311                 int coeff= (  2*(sum    -s->qps[qpi])*s->base_matrix[bmi][i]
00312                             - 2*(qistart-s->qps[qpi])*s->base_matrix[bmj][i]
00313                             + s->qr_size[inter][plane][qri])
00314                            / (2*s->qr_size[inter][plane][qri]);
00315 
00316                 int qmin= 8<<(inter + !i);
00317                 int qscale= i ? ac_scale_factor : dc_scale_factor;
00318 
00319                 s->qmat[qpi][inter][plane][s->dsp.idct_permutation[i]]= av_clip((qscale * coeff)/100 * 4, qmin, 4096);
00320             }
00321             // all DC coefficients use the same quant so as not to interfere with DC prediction
00322             s->qmat[qpi][inter][plane][0] = s->qmat[0][inter][plane][0];
00323         }
00324     }
00325 
00326     memset(s->qscale_table, (FFMAX(s->qmat[0][0][0][1], s->qmat[0][0][1][1])+8)/16, 512); //FIXME finetune
00327 }
00328 
00329 /*
00330  * This function initializes the loop filter boundary limits if the frame's
00331  * quality index is different from the previous frame's.
00332  *
00333  * The filter_limit_values may not be larger than 127.
00334  */
00335 static void init_loop_filter(Vp3DecodeContext *s)
00336 {
00337     int *bounding_values= s->bounding_values_array+127;
00338     int filter_limit;
00339     int x;
00340     int value;
00341 
00342     filter_limit = s->filter_limit_values[s->qps[0]];
00343 
00344     /* set up the bounding values */
00345     memset(s->bounding_values_array, 0, 256 * sizeof(int));
00346     for (x = 0; x < filter_limit; x++) {
00347         bounding_values[-x] = -x;
00348         bounding_values[x] = x;
00349     }
00350     for (x = value = filter_limit; x < 128 && value; x++, value--) {
00351         bounding_values[ x] =  value;
00352         bounding_values[-x] = -value;
00353     }
00354     if (value)
00355         bounding_values[128] = value;
00356     bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202;
00357 }
00358 
00359 /*
00360  * This function unpacks all of the superblock/macroblock/fragment coding
00361  * information from the bitstream.
00362  */
00363 static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
00364 {
00365     int superblock_starts[3] = { 0, s->u_superblock_start, s->v_superblock_start };
00366     int bit = 0;
00367     int current_superblock = 0;
00368     int current_run = 0;
00369     int num_partial_superblocks = 0;
00370 
00371     int i, j;
00372     int current_fragment;
00373     int plane;
00374 
00375     if (s->keyframe) {
00376         memset(s->superblock_coding, SB_FULLY_CODED, s->superblock_count);
00377 
00378     } else {
00379 
00380         /* unpack the list of partially-coded superblocks */
00381         bit = get_bits1(gb);
00382         while (current_superblock < s->superblock_count && get_bits_left(gb) > 0) {
00383                 current_run = get_vlc2(gb,
00384                     s->superblock_run_length_vlc.table, 6, 2) + 1;
00385                 if (current_run == 34)
00386                     current_run += get_bits(gb, 12);
00387 
00388             if (current_superblock + current_run > s->superblock_count) {
00389                 av_log(s->avctx, AV_LOG_ERROR, "Invalid partially coded superblock run length\n");
00390                 return -1;
00391             }
00392 
00393             memset(s->superblock_coding + current_superblock, bit, current_run);
00394 
00395             current_superblock += current_run;
00396             if (bit)
00397                 num_partial_superblocks += current_run;
00398 
00399             if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
00400                 bit = get_bits1(gb);
00401             else
00402                 bit ^= 1;
00403         }
00404 
00405         /* unpack the list of fully coded superblocks if any of the blocks were
00406          * not marked as partially coded in the previous step */
00407         if (num_partial_superblocks < s->superblock_count) {
00408             int superblocks_decoded = 0;
00409 
00410             current_superblock = 0;
00411             bit = get_bits1(gb);
00412             while (superblocks_decoded < s->superblock_count - num_partial_superblocks
00413                    && get_bits_left(gb) > 0) {
00414                         current_run = get_vlc2(gb,
00415                             s->superblock_run_length_vlc.table, 6, 2) + 1;
00416                         if (current_run == 34)
00417                             current_run += get_bits(gb, 12);
00418 
00419                 for (j = 0; j < current_run; current_superblock++) {
00420                     if (current_superblock >= s->superblock_count) {
00421                         av_log(s->avctx, AV_LOG_ERROR, "Invalid fully coded superblock run length\n");
00422                         return -1;
00423                     }
00424 
00425                 /* skip any superblocks already marked as partially coded */
00426                 if (s->superblock_coding[current_superblock] == SB_NOT_CODED) {
00427                     s->superblock_coding[current_superblock] = 2*bit;
00428                     j++;
00429                 }
00430                 }
00431                 superblocks_decoded += current_run;
00432 
00433                 if (s->theora && current_run == MAXIMUM_LONG_BIT_RUN)
00434                     bit = get_bits1(gb);
00435                 else
00436                     bit ^= 1;
00437             }
00438         }
00439 
00440         /* if there were partial blocks, initialize bitstream for
00441          * unpacking fragment codings */
00442         if (num_partial_superblocks) {
00443 
00444             current_run = 0;
00445             bit = get_bits1(gb);
00446             /* toggle the bit because as soon as the first run length is
00447              * fetched the bit will be toggled again */
00448             bit ^= 1;
00449         }
00450     }
00451 
00452     /* figure out which fragments are coded; iterate through each
00453      * superblock (all planes) */
00454     s->total_num_coded_frags = 0;
00455     memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
00456 
00457     for (plane = 0; plane < 3; plane++) {
00458         int sb_start = superblock_starts[plane];
00459         int sb_end = sb_start + (plane ? s->c_superblock_count : s->y_superblock_count);
00460         int num_coded_frags = 0;
00461 
00462     for (i = sb_start; i < sb_end && get_bits_left(gb) > 0; i++) {
00463 
00464         /* iterate through all 16 fragments in a superblock */
00465         for (j = 0; j < 16; j++) {
00466 
00467             /* if the fragment is in bounds, check its coding status */
00468             current_fragment = s->superblock_fragments[i * 16 + j];
00469             if (current_fragment != -1) {
00470                 int coded = s->superblock_coding[i];
00471 
00472                 if (s->superblock_coding[i] == SB_PARTIALLY_CODED) {
00473 
00474                     /* fragment may or may not be coded; this is the case
00475                      * that cares about the fragment coding runs */
00476                     if (current_run-- == 0) {
00477                         bit ^= 1;
00478                         current_run = get_vlc2(gb,
00479                             s->fragment_run_length_vlc.table, 5, 2);
00480                     }
00481                     coded = bit;
00482                 }
00483 
00484                     if (coded) {
00485                         /* default mode; actual mode will be decoded in
00486                          * the next phase */
00487                         s->all_fragments[current_fragment].coding_method =
00488                             MODE_INTER_NO_MV;
00489                         s->coded_fragment_list[plane][num_coded_frags++] =
00490                             current_fragment;
00491                     } else {
00492                         /* not coded; copy this fragment from the prior frame */
00493                         s->all_fragments[current_fragment].coding_method =
00494                             MODE_COPY;
00495                     }
00496             }
00497         }
00498     }
00499         s->total_num_coded_frags += num_coded_frags;
00500         for (i = 0; i < 64; i++)
00501             s->num_coded_frags[plane][i] = num_coded_frags;
00502         if (plane < 2)
00503             s->coded_fragment_list[plane+1] = s->coded_fragment_list[plane] + num_coded_frags;
00504     }
00505     return 0;
00506 }
00507 
00508 /*
00509  * This function unpacks all the coding mode data for individual macroblocks
00510  * from the bitstream.
00511  */
00512 static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
00513 {
00514     int i, j, k, sb_x, sb_y;
00515     int scheme;
00516     int current_macroblock;
00517     int current_fragment;
00518     int coding_mode;
00519     int custom_mode_alphabet[CODING_MODE_COUNT];
00520     const int *alphabet;
00521     Vp3Fragment *frag;
00522 
00523     if (s->keyframe) {
00524         for (i = 0; i < s->fragment_count; i++)
00525             s->all_fragments[i].coding_method = MODE_INTRA;
00526 
00527     } else {
00528 
00529         /* fetch the mode coding scheme for this frame */
00530         scheme = get_bits(gb, 3);
00531 
00532         /* is it a custom coding scheme? */
00533         if (scheme == 0) {
00534             for (i = 0; i < 8; i++)
00535                 custom_mode_alphabet[i] = MODE_INTER_NO_MV;
00536             for (i = 0; i < 8; i++)
00537                 custom_mode_alphabet[get_bits(gb, 3)] = i;
00538             alphabet = custom_mode_alphabet;
00539         } else
00540             alphabet = ModeAlphabet[scheme-1];
00541 
00542         /* iterate through all of the macroblocks that contain 1 or more
00543          * coded fragments */
00544         for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
00545             for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
00546                 if (get_bits_left(gb) <= 0)
00547                     return -1;
00548 
00549             for (j = 0; j < 4; j++) {
00550                 int mb_x = 2*sb_x +   (j>>1);
00551                 int mb_y = 2*sb_y + (((j>>1)+j)&1);
00552                 current_macroblock = mb_y * s->macroblock_width + mb_x;
00553 
00554                 if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height)
00555                     continue;
00556 
00557 #define BLOCK_X (2*mb_x + (k&1))
00558 #define BLOCK_Y (2*mb_y + (k>>1))
00559                 /* coding modes are only stored if the macroblock has at least one
00560                  * luma block coded, otherwise it must be INTER_NO_MV */
00561                 for (k = 0; k < 4; k++) {
00562                     current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
00563                     if (s->all_fragments[current_fragment].coding_method != MODE_COPY)
00564                         break;
00565                 }
00566                 if (k == 4) {
00567                     s->macroblock_coding[current_macroblock] = MODE_INTER_NO_MV;
00568                     continue;
00569                 }
00570 
00571                 /* mode 7 means get 3 bits for each coding mode */
00572                 if (scheme == 7)
00573                     coding_mode = get_bits(gb, 3);
00574                 else
00575                     coding_mode = alphabet
00576                         [get_vlc2(gb, s->mode_code_vlc.table, 3, 3)];
00577 
00578                 s->macroblock_coding[current_macroblock] = coding_mode;
00579                 for (k = 0; k < 4; k++) {
00580                     frag = s->all_fragments + BLOCK_Y*s->fragment_width[0] + BLOCK_X;
00581                     if (frag->coding_method != MODE_COPY)
00582                         frag->coding_method = coding_mode;
00583                 }
00584 
00585 #define SET_CHROMA_MODES \
00586     if (frag[s->fragment_start[1]].coding_method != MODE_COPY) \
00587         frag[s->fragment_start[1]].coding_method = coding_mode;\
00588     if (frag[s->fragment_start[2]].coding_method != MODE_COPY) \
00589         frag[s->fragment_start[2]].coding_method = coding_mode;
00590 
00591                 if (s->chroma_y_shift) {
00592                     frag = s->all_fragments + mb_y*s->fragment_width[1] + mb_x;
00593                     SET_CHROMA_MODES
00594                 } else if (s->chroma_x_shift) {
00595                     frag = s->all_fragments + 2*mb_y*s->fragment_width[1] + mb_x;
00596                     for (k = 0; k < 2; k++) {
00597                         SET_CHROMA_MODES
00598                         frag += s->fragment_width[1];
00599                     }
00600                 } else {
00601                     for (k = 0; k < 4; k++) {
00602                         frag = s->all_fragments + BLOCK_Y*s->fragment_width[1] + BLOCK_X;
00603                         SET_CHROMA_MODES
00604                     }
00605                 }
00606             }
00607             }
00608         }
00609     }
00610 
00611     return 0;
00612 }
00613 
00614 /*
00615  * This function unpacks all the motion vectors for the individual
00616  * macroblocks from the bitstream.
00617  */
00618 static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
00619 {
00620     int j, k, sb_x, sb_y;
00621     int coding_mode;
00622     int motion_x[4];
00623     int motion_y[4];
00624     int last_motion_x = 0;
00625     int last_motion_y = 0;
00626     int prior_last_motion_x = 0;
00627     int prior_last_motion_y = 0;
00628     int current_macroblock;
00629     int current_fragment;
00630     int frag;
00631 
00632     if (s->keyframe)
00633         return 0;
00634 
00635     /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
00636     coding_mode = get_bits1(gb);
00637 
00638     /* iterate through all of the macroblocks that contain 1 or more
00639      * coded fragments */
00640     for (sb_y = 0; sb_y < s->y_superblock_height; sb_y++) {
00641         for (sb_x = 0; sb_x < s->y_superblock_width; sb_x++) {
00642             if (get_bits_left(gb) <= 0)
00643                 return -1;
00644 
00645         for (j = 0; j < 4; j++) {
00646             int mb_x = 2*sb_x +   (j>>1);
00647             int mb_y = 2*sb_y + (((j>>1)+j)&1);
00648             current_macroblock = mb_y * s->macroblock_width + mb_x;
00649 
00650             if (mb_x >= s->macroblock_width || mb_y >= s->macroblock_height ||
00651                 (s->macroblock_coding[current_macroblock] == MODE_COPY))
00652                 continue;
00653 
00654             switch (s->macroblock_coding[current_macroblock]) {
00655 
00656             case MODE_INTER_PLUS_MV:
00657             case MODE_GOLDEN_MV:
00658                 /* all 6 fragments use the same motion vector */
00659                 if (coding_mode == 0) {
00660                     motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
00661                     motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
00662                 } else {
00663                     motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
00664                     motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
00665                 }
00666 
00667                 /* vector maintenance, only on MODE_INTER_PLUS_MV */
00668                 if (s->macroblock_coding[current_macroblock] ==
00669                     MODE_INTER_PLUS_MV) {
00670                     prior_last_motion_x = last_motion_x;
00671                     prior_last_motion_y = last_motion_y;
00672                     last_motion_x = motion_x[0];
00673                     last_motion_y = motion_y[0];
00674                 }
00675                 break;
00676 
00677             case MODE_INTER_FOURMV:
00678                 /* vector maintenance */
00679                 prior_last_motion_x = last_motion_x;
00680                 prior_last_motion_y = last_motion_y;
00681 
00682                 /* fetch 4 vectors from the bitstream, one for each
00683                  * Y fragment, then average for the C fragment vectors */
00684                 for (k = 0; k < 4; k++) {
00685                     current_fragment = BLOCK_Y*s->fragment_width[0] + BLOCK_X;
00686                     if (s->all_fragments[current_fragment].coding_method != MODE_COPY) {
00687                         if (coding_mode == 0) {
00688                             motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
00689                             motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
00690                         } else {
00691                             motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
00692                             motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
00693                         }
00694                         last_motion_x = motion_x[k];
00695                         last_motion_y = motion_y[k];
00696                     } else {
00697                         motion_x[k] = 0;
00698                         motion_y[k] = 0;
00699                     }
00700                 }
00701                 break;
00702 
00703             case MODE_INTER_LAST_MV:
00704                 /* all 6 fragments use the last motion vector */
00705                 motion_x[0] = last_motion_x;
00706                 motion_y[0] = last_motion_y;
00707 
00708                 /* no vector maintenance (last vector remains the
00709                  * last vector) */
00710                 break;
00711 
00712             case MODE_INTER_PRIOR_LAST:
00713                 /* all 6 fragments use the motion vector prior to the
00714                  * last motion vector */
00715                 motion_x[0] = prior_last_motion_x;
00716                 motion_y[0] = prior_last_motion_y;
00717 
00718                 /* vector maintenance */
00719                 prior_last_motion_x = last_motion_x;
00720                 prior_last_motion_y = last_motion_y;
00721                 last_motion_x = motion_x[0];
00722                 last_motion_y = motion_y[0];
00723                 break;
00724 
00725             default:
00726                 /* covers intra, inter without MV, golden without MV */
00727                 motion_x[0] = 0;
00728                 motion_y[0] = 0;
00729 
00730                 /* no vector maintenance */
00731                 break;
00732             }
00733 
00734             /* assign the motion vectors to the correct fragments */
00735             for (k = 0; k < 4; k++) {
00736                 current_fragment =
00737                     BLOCK_Y*s->fragment_width[0] + BLOCK_X;
00738                 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
00739                     s->motion_val[0][current_fragment][0] = motion_x[k];
00740                     s->motion_val[0][current_fragment][1] = motion_y[k];
00741                 } else {
00742                     s->motion_val[0][current_fragment][0] = motion_x[0];
00743                     s->motion_val[0][current_fragment][1] = motion_y[0];
00744                 }
00745             }
00746 
00747             if (s->chroma_y_shift) {
00748                 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
00749                     motion_x[0] = RSHIFT(motion_x[0] + motion_x[1] + motion_x[2] + motion_x[3], 2);
00750                     motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] + motion_y[2] + motion_y[3], 2);
00751                 }
00752                 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
00753                 motion_y[0] = (motion_y[0]>>1) | (motion_y[0]&1);
00754                 frag = mb_y*s->fragment_width[1] + mb_x;
00755                 s->motion_val[1][frag][0] = motion_x[0];
00756                 s->motion_val[1][frag][1] = motion_y[0];
00757             } else if (s->chroma_x_shift) {
00758                 if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
00759                     motion_x[0] = RSHIFT(motion_x[0] + motion_x[1], 1);
00760                     motion_y[0] = RSHIFT(motion_y[0] + motion_y[1], 1);
00761                     motion_x[1] = RSHIFT(motion_x[2] + motion_x[3], 1);
00762                     motion_y[1] = RSHIFT(motion_y[2] + motion_y[3], 1);
00763                 } else {
00764                     motion_x[1] = motion_x[0];
00765                     motion_y[1] = motion_y[0];
00766                 }
00767                 motion_x[0] = (motion_x[0]>>1) | (motion_x[0]&1);
00768                 motion_x[1] = (motion_x[1]>>1) | (motion_x[1]&1);
00769 
00770                 frag = 2*mb_y*s->fragment_width[1] + mb_x;
00771                 for (k = 0; k < 2; k++) {
00772                     s->motion_val[1][frag][0] = motion_x[k];
00773                     s->motion_val[1][frag][1] = motion_y[k];
00774                     frag += s->fragment_width[1];
00775                 }
00776             } else {
00777                 for (k = 0; k < 4; k++) {
00778                     frag = BLOCK_Y*s->fragment_width[1] + BLOCK_X;
00779                     if (s->macroblock_coding[current_macroblock] == MODE_INTER_FOURMV) {
00780                         s->motion_val[1][frag][0] = motion_x[k];
00781                         s->motion_val[1][frag][1] = motion_y[k];
00782                     } else {
00783                         s->motion_val[1][frag][0] = motion_x[0];
00784                         s->motion_val[1][frag][1] = motion_y[0];
00785                     }
00786                 }
00787             }
00788         }
00789         }
00790     }
00791 
00792     return 0;
00793 }
00794 
00795 static int unpack_block_qpis(Vp3DecodeContext *s, GetBitContext *gb)
00796 {
00797     int qpi, i, j, bit, run_length, blocks_decoded, num_blocks_at_qpi;
00798     int num_blocks = s->total_num_coded_frags;
00799 
00800     for (qpi = 0; qpi < s->nqps-1 && num_blocks > 0; qpi++) {
00801         i = blocks_decoded = num_blocks_at_qpi = 0;
00802 
00803         bit = get_bits1(gb);
00804 
00805         do {
00806             run_length = get_vlc2(gb, s->superblock_run_length_vlc.table, 6, 2) + 1;
00807             if (run_length == 34)
00808                 run_length += get_bits(gb, 12);
00809             blocks_decoded += run_length;
00810 
00811             if (!bit)
00812                 num_blocks_at_qpi += run_length;
00813 
00814             for (j = 0; j < run_length; i++) {
00815                 if (i >= s->total_num_coded_frags)
00816                     return -1;
00817 
00818                 if (s->all_fragments[s->coded_fragment_list[0][i]].qpi == qpi) {
00819                     s->all_fragments[s->coded_fragment_list[0][i]].qpi += bit;
00820                     j++;
00821                 }
00822             }
00823 
00824             if (run_length == MAXIMUM_LONG_BIT_RUN)
00825                 bit = get_bits1(gb);
00826             else
00827                 bit ^= 1;
00828         } while (blocks_decoded < num_blocks && get_bits_left(gb) > 0);
00829 
00830         num_blocks -= num_blocks_at_qpi;
00831     }
00832 
00833     return 0;
00834 }
00835 
00836 /*
00837  * This function is called by unpack_dct_coeffs() to extract the VLCs from
00838  * the bitstream. The VLCs encode tokens which are used to unpack DCT
00839  * data. This function unpacks all the VLCs for either the Y plane or both
00840  * C planes, and is called for DC coefficients or different AC coefficient
00841  * levels (since different coefficient types require different VLC tables.
00842  *
00843  * This function returns a residual eob run. E.g, if a particular token gave
00844  * instructions to EOB the next 5 fragments and there were only 2 fragments
00845  * left in the current fragment range, 3 would be returned so that it could
00846  * be passed into the next call to this same function.
00847  */
00848 static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
00849                         VLC *table, int coeff_index,
00850                         int plane,
00851                         int eob_run)
00852 {
00853     int i, j = 0;
00854     int token;
00855     int zero_run = 0;
00856     DCTELEM coeff = 0;
00857     int bits_to_get;
00858     int blocks_ended;
00859     int coeff_i = 0;
00860     int num_coeffs = s->num_coded_frags[plane][coeff_index];
00861     int16_t *dct_tokens = s->dct_tokens[plane][coeff_index];
00862 
00863     /* local references to structure members to avoid repeated deferences */
00864     int *coded_fragment_list = s->coded_fragment_list[plane];
00865     Vp3Fragment *all_fragments = s->all_fragments;
00866     VLC_TYPE (*vlc_table)[2] = table->table;
00867 
00868     if (num_coeffs < 0)
00869         av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coefficents at level %d\n", coeff_index);
00870 
00871     if (eob_run > num_coeffs) {
00872         coeff_i = blocks_ended = num_coeffs;
00873         eob_run -= num_coeffs;
00874     } else {
00875         coeff_i = blocks_ended = eob_run;
00876         eob_run = 0;
00877     }
00878 
00879     // insert fake EOB token to cover the split between planes or zzi
00880     if (blocks_ended)
00881         dct_tokens[j++] = blocks_ended << 2;
00882 
00883     while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
00884             /* decode a VLC into a token */
00885             token = get_vlc2(gb, vlc_table, 11, 3);
00886             /* use the token to get a zero run, a coefficient, and an eob run */
00887             if ((unsigned) token <= 6U) {
00888                 eob_run = eob_run_base[token];
00889                 if (eob_run_get_bits[token])
00890                     eob_run += get_bits(gb, eob_run_get_bits[token]);
00891 
00892                 // record only the number of blocks ended in this plane,
00893                 // any spill will be recorded in the next plane.
00894                 if (eob_run > num_coeffs - coeff_i) {
00895                     dct_tokens[j++] = TOKEN_EOB(num_coeffs - coeff_i);
00896                     blocks_ended   += num_coeffs - coeff_i;
00897                     eob_run        -= num_coeffs - coeff_i;
00898                     coeff_i         = num_coeffs;
00899                 } else {
00900                     dct_tokens[j++] = TOKEN_EOB(eob_run);
00901                     blocks_ended   += eob_run;
00902                     coeff_i        += eob_run;
00903                     eob_run = 0;
00904                 }
00905             } else if (token >= 0) {
00906                 bits_to_get = coeff_get_bits[token];
00907                 if (bits_to_get)
00908                     bits_to_get = get_bits(gb, bits_to_get);
00909                 coeff = coeff_tables[token][bits_to_get];
00910 
00911                 zero_run = zero_run_base[token];
00912                 if (zero_run_get_bits[token])
00913                     zero_run += get_bits(gb, zero_run_get_bits[token]);
00914 
00915                 if (zero_run) {
00916                     dct_tokens[j++] = TOKEN_ZERO_RUN(coeff, zero_run);
00917                 } else {
00918                     // Save DC into the fragment structure. DC prediction is
00919                     // done in raster order, so the actual DC can't be in with
00920                     // other tokens. We still need the token in dct_tokens[]
00921                     // however, or else the structure collapses on itself.
00922                     if (!coeff_index)
00923                         all_fragments[coded_fragment_list[coeff_i]].dc = coeff;
00924 
00925                     dct_tokens[j++] = TOKEN_COEFF(coeff);
00926                 }
00927 
00928                 if (coeff_index + zero_run > 64) {
00929                     av_log(s->avctx, AV_LOG_DEBUG, "Invalid zero run of %d with"
00930                            " %d coeffs left\n", zero_run, 64-coeff_index);
00931                     zero_run = 64 - coeff_index;
00932                 }
00933 
00934                 // zero runs code multiple coefficients,
00935                 // so don't try to decode coeffs for those higher levels
00936                 for (i = coeff_index+1; i <= coeff_index+zero_run; i++)
00937                     s->num_coded_frags[plane][i]--;
00938                 coeff_i++;
00939             } else {
00940                 av_log(s->avctx, AV_LOG_ERROR,
00941                        "Invalid token %d\n", token);
00942                 return -1;
00943             }
00944     }
00945 
00946     if (blocks_ended > s->num_coded_frags[plane][coeff_index])
00947         av_log(s->avctx, AV_LOG_ERROR, "More blocks ended than coded!\n");
00948 
00949     // decrement the number of blocks that have higher coeffecients for each
00950     // EOB run at this level
00951     if (blocks_ended)
00952         for (i = coeff_index+1; i < 64; i++)
00953             s->num_coded_frags[plane][i] -= blocks_ended;
00954 
00955     // setup the next buffer
00956     if (plane < 2)
00957         s->dct_tokens[plane+1][coeff_index] = dct_tokens + j;
00958     else if (coeff_index < 63)
00959         s->dct_tokens[0][coeff_index+1] = dct_tokens + j;
00960 
00961     return eob_run;
00962 }
00963 
00964 static void reverse_dc_prediction(Vp3DecodeContext *s,
00965                                   int first_fragment,
00966                                   int fragment_width,
00967                                   int fragment_height);
00968 /*
00969  * This function unpacks all of the DCT coefficient data from the
00970  * bitstream.
00971  */
00972 static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
00973 {
00974     int i;
00975     int dc_y_table;
00976     int dc_c_table;
00977     int ac_y_table;
00978     int ac_c_table;
00979     int residual_eob_run = 0;
00980     VLC *y_tables[64];
00981     VLC *c_tables[64];
00982 
00983     s->dct_tokens[0][0] = s->dct_tokens_base;
00984 
00985     /* fetch the DC table indexes */
00986     dc_y_table = get_bits(gb, 4);
00987     dc_c_table = get_bits(gb, 4);
00988 
00989     /* unpack the Y plane DC coefficients */
00990     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_y_table], 0,
00991         0, residual_eob_run);
00992     if (residual_eob_run < 0)
00993         return residual_eob_run;
00994 
00995     /* reverse prediction of the Y-plane DC coefficients */
00996     reverse_dc_prediction(s, 0, s->fragment_width[0], s->fragment_height[0]);
00997 
00998     /* unpack the C plane DC coefficients */
00999     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
01000         1, residual_eob_run);
01001     if (residual_eob_run < 0)
01002         return residual_eob_run;
01003     residual_eob_run = unpack_vlcs(s, gb, &s->dc_vlc[dc_c_table], 0,
01004         2, residual_eob_run);
01005     if (residual_eob_run < 0)
01006         return residual_eob_run;
01007 
01008     /* reverse prediction of the C-plane DC coefficients */
01009     if (!(s->avctx->flags & CODEC_FLAG_GRAY))
01010     {
01011         reverse_dc_prediction(s, s->fragment_start[1],
01012             s->fragment_width[1], s->fragment_height[1]);
01013         reverse_dc_prediction(s, s->fragment_start[2],
01014             s->fragment_width[1], s->fragment_height[1]);
01015     }
01016 
01017     /* fetch the AC table indexes */
01018     ac_y_table = get_bits(gb, 4);
01019     ac_c_table = get_bits(gb, 4);
01020 
01021     /* build tables of AC VLC tables */
01022     for (i = 1; i <= 5; i++) {
01023         y_tables[i] = &s->ac_vlc_1[ac_y_table];
01024         c_tables[i] = &s->ac_vlc_1[ac_c_table];
01025     }
01026     for (i = 6; i <= 14; i++) {
01027         y_tables[i] = &s->ac_vlc_2[ac_y_table];
01028         c_tables[i] = &s->ac_vlc_2[ac_c_table];
01029     }
01030     for (i = 15; i <= 27; i++) {
01031         y_tables[i] = &s->ac_vlc_3[ac_y_table];
01032         c_tables[i] = &s->ac_vlc_3[ac_c_table];
01033     }
01034     for (i = 28; i <= 63; i++) {
01035         y_tables[i] = &s->ac_vlc_4[ac_y_table];
01036         c_tables[i] = &s->ac_vlc_4[ac_c_table];
01037     }
01038 
01039     /* decode all AC coefficents */
01040     for (i = 1; i <= 63; i++) {
01041             residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
01042                 0, residual_eob_run);
01043             if (residual_eob_run < 0)
01044                 return residual_eob_run;
01045 
01046             residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
01047                 1, residual_eob_run);
01048             if (residual_eob_run < 0)
01049                 return residual_eob_run;
01050             residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
01051                 2, residual_eob_run);
01052             if (residual_eob_run < 0)
01053                 return residual_eob_run;
01054     }
01055 
01056     return 0;
01057 }
01058 
01059 /*
01060  * This function reverses the DC prediction for each coded fragment in
01061  * the frame. Much of this function is adapted directly from the original
01062  * VP3 source code.
01063  */
01064 #define COMPATIBLE_FRAME(x) \
01065   (compatible_frame[s->all_fragments[x].coding_method] == current_frame_type)
01066 #define DC_COEFF(u) s->all_fragments[u].dc
01067 
01068 static void reverse_dc_prediction(Vp3DecodeContext *s,
01069                                   int first_fragment,
01070                                   int fragment_width,
01071                                   int fragment_height)
01072 {
01073 
01074 #define PUL 8
01075 #define PU 4
01076 #define PUR 2
01077 #define PL 1
01078 
01079     int x, y;
01080     int i = first_fragment;
01081 
01082     int predicted_dc;
01083 
01084     /* DC values for the left, up-left, up, and up-right fragments */
01085     int vl, vul, vu, vur;
01086 
01087     /* indexes for the left, up-left, up, and up-right fragments */
01088     int l, ul, u, ur;
01089 
01090     /*
01091      * The 6 fields mean:
01092      *   0: up-left multiplier
01093      *   1: up multiplier
01094      *   2: up-right multiplier
01095      *   3: left multiplier
01096      */
01097     static const int predictor_transform[16][4] = {
01098         {  0,  0,  0,  0},
01099         {  0,  0,  0,128},        // PL
01100         {  0,  0,128,  0},        // PUR
01101         {  0,  0, 53, 75},        // PUR|PL
01102         {  0,128,  0,  0},        // PU
01103         {  0, 64,  0, 64},        // PU|PL
01104         {  0,128,  0,  0},        // PU|PUR
01105         {  0,  0, 53, 75},        // PU|PUR|PL
01106         {128,  0,  0,  0},        // PUL
01107         {  0,  0,  0,128},        // PUL|PL
01108         { 64,  0, 64,  0},        // PUL|PUR
01109         {  0,  0, 53, 75},        // PUL|PUR|PL
01110         {  0,128,  0,  0},        // PUL|PU
01111        {-104,116,  0,116},        // PUL|PU|PL
01112         { 24, 80, 24,  0},        // PUL|PU|PUR
01113        {-104,116,  0,116}         // PUL|PU|PUR|PL
01114     };
01115 
01116     /* This table shows which types of blocks can use other blocks for
01117      * prediction. For example, INTRA is the only mode in this table to
01118      * have a frame number of 0. That means INTRA blocks can only predict
01119      * from other INTRA blocks. There are 2 golden frame coding types;
01120      * blocks encoding in these modes can only predict from other blocks
01121      * that were encoded with these 1 of these 2 modes. */
01122     static const unsigned char compatible_frame[9] = {
01123         1,    /* MODE_INTER_NO_MV */
01124         0,    /* MODE_INTRA */
01125         1,    /* MODE_INTER_PLUS_MV */
01126         1,    /* MODE_INTER_LAST_MV */
01127         1,    /* MODE_INTER_PRIOR_MV */
01128         2,    /* MODE_USING_GOLDEN */
01129         2,    /* MODE_GOLDEN_MV */
01130         1,    /* MODE_INTER_FOUR_MV */
01131         3     /* MODE_COPY */
01132     };
01133     int current_frame_type;
01134 
01135     /* there is a last DC predictor for each of the 3 frame types */
01136     short last_dc[3];
01137 
01138     int transform = 0;
01139 
01140     vul = vu = vur = vl = 0;
01141     last_dc[0] = last_dc[1] = last_dc[2] = 0;
01142 
01143     /* for each fragment row... */
01144     for (y = 0; y < fragment_height; y++) {
01145 
01146         /* for each fragment in a row... */
01147         for (x = 0; x < fragment_width; x++, i++) {
01148 
01149             /* reverse prediction if this block was coded */
01150             if (s->all_fragments[i].coding_method != MODE_COPY) {
01151 
01152                 current_frame_type =
01153                     compatible_frame[s->all_fragments[i].coding_method];
01154 
01155                 transform= 0;
01156                 if(x){
01157                     l= i-1;
01158                     vl = DC_COEFF(l);
01159                     if(COMPATIBLE_FRAME(l))
01160                         transform |= PL;
01161                 }
01162                 if(y){
01163                     u= i-fragment_width;
01164                     vu = DC_COEFF(u);
01165                     if(COMPATIBLE_FRAME(u))
01166                         transform |= PU;
01167                     if(x){
01168                         ul= i-fragment_width-1;
01169                         vul = DC_COEFF(ul);
01170                         if(COMPATIBLE_FRAME(ul))
01171                             transform |= PUL;
01172                     }
01173                     if(x + 1 < fragment_width){
01174                         ur= i-fragment_width+1;
01175                         vur = DC_COEFF(ur);
01176                         if(COMPATIBLE_FRAME(ur))
01177                             transform |= PUR;
01178                     }
01179                 }
01180 
01181                 if (transform == 0) {
01182 
01183                     /* if there were no fragments to predict from, use last
01184                      * DC saved */
01185                     predicted_dc = last_dc[current_frame_type];
01186                 } else {
01187 
01188                     /* apply the appropriate predictor transform */
01189                     predicted_dc =
01190                         (predictor_transform[transform][0] * vul) +
01191                         (predictor_transform[transform][1] * vu) +
01192                         (predictor_transform[transform][2] * vur) +
01193                         (predictor_transform[transform][3] * vl);
01194 
01195                     predicted_dc /= 128;
01196 
01197                     /* check for outranging on the [ul u l] and
01198                      * [ul u ur l] predictors */
01199                     if ((transform == 15) || (transform == 13)) {
01200                         if (FFABS(predicted_dc - vu) > 128)
01201                             predicted_dc = vu;
01202                         else if (FFABS(predicted_dc - vl) > 128)
01203                             predicted_dc = vl;
01204                         else if (FFABS(predicted_dc - vul) > 128)
01205                             predicted_dc = vul;
01206                     }
01207                 }
01208 
01209                 /* at long last, apply the predictor */
01210                 DC_COEFF(i) += predicted_dc;
01211                 /* save the DC */
01212                 last_dc[current_frame_type] = DC_COEFF(i);
01213             }
01214         }
01215     }
01216 }
01217 
01218 static void apply_loop_filter(Vp3DecodeContext *s, int plane, int ystart, int yend)
01219 {
01220     int x, y;
01221     int *bounding_values= s->bounding_values_array+127;
01222 
01223     int width           = s->fragment_width[!!plane];
01224     int height          = s->fragment_height[!!plane];
01225     int fragment        = s->fragment_start        [plane] + ystart * width;
01226     int stride          = s->current_frame.linesize[plane];
01227     uint8_t *plane_data = s->current_frame.data    [plane];
01228     if (!s->flipped_image) stride = -stride;
01229     plane_data += s->data_offset[plane] + 8*ystart*stride;
01230 
01231     for (y = ystart; y < yend; y++) {
01232 
01233         for (x = 0; x < width; x++) {
01234             /* This code basically just deblocks on the edges of coded blocks.
01235              * However, it has to be much more complicated because of the
01236              * braindamaged deblock ordering used in VP3/Theora. Order matters
01237              * because some pixels get filtered twice. */
01238             if( s->all_fragments[fragment].coding_method != MODE_COPY )
01239             {
01240                 /* do not perform left edge filter for left columns frags */
01241                 if (x > 0) {
01242                     s->dsp.vp3_h_loop_filter(
01243                         plane_data + 8*x,
01244                         stride, bounding_values);
01245                 }
01246 
01247                 /* do not perform top edge filter for top row fragments */
01248                 if (y > 0) {
01249                     s->dsp.vp3_v_loop_filter(
01250                         plane_data + 8*x,
01251                         stride, bounding_values);
01252                 }
01253 
01254                 /* do not perform right edge filter for right column
01255                  * fragments or if right fragment neighbor is also coded
01256                  * in this frame (it will be filtered in next iteration) */
01257                 if ((x < width - 1) &&
01258                     (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
01259                     s->dsp.vp3_h_loop_filter(
01260                         plane_data + 8*x + 8,
01261                         stride, bounding_values);
01262                 }
01263 
01264                 /* do not perform bottom edge filter for bottom row
01265                  * fragments or if bottom fragment neighbor is also coded
01266                  * in this frame (it will be filtered in the next row) */
01267                 if ((y < height - 1) &&
01268                     (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
01269                     s->dsp.vp3_v_loop_filter(
01270                         plane_data + 8*x + 8*stride,
01271                         stride, bounding_values);
01272                 }
01273             }
01274 
01275             fragment++;
01276         }
01277         plane_data += 8*stride;
01278     }
01279 }
01280 
01285 static inline int vp3_dequant(Vp3DecodeContext *s, Vp3Fragment *frag,
01286                               int plane, int inter, DCTELEM block[64])
01287 {
01288     int16_t *dequantizer = s->qmat[frag->qpi][inter][plane];
01289     uint8_t *perm = s->scantable.permutated;
01290     int i = 0;
01291 
01292     do {
01293         int token = *s->dct_tokens[plane][i];
01294         switch (token & 3) {
01295         case 0: // EOB
01296             if (--token < 4) // 0-3 are token types, so the EOB run must now be 0
01297                 s->dct_tokens[plane][i]++;
01298             else
01299                 *s->dct_tokens[plane][i] = token & ~3;
01300             goto end;
01301         case 1: // zero run
01302             s->dct_tokens[plane][i]++;
01303             i += (token >> 2) & 0x7f;
01304             if (i > 63) {
01305                 av_log(s->avctx, AV_LOG_ERROR, "Coefficient index overflow\n");
01306                 return i;
01307             }
01308             block[perm[i]] = (token >> 9) * dequantizer[perm[i]];
01309             i++;
01310             break;
01311         case 2: // coeff
01312             block[perm[i]] = (token >> 2) * dequantizer[perm[i]];
01313             s->dct_tokens[plane][i++]++;
01314             break;
01315         default: // shouldn't happen
01316             return i;
01317         }
01318     } while (i < 64);
01319     // return value is expected to be a valid level
01320     i--;
01321 end:
01322     // the actual DC+prediction is in the fragment structure
01323     block[0] = frag->dc * s->qmat[0][inter][plane][0];
01324     return i;
01325 }
01326 
01330 static void vp3_draw_horiz_band(Vp3DecodeContext *s, int y)
01331 {
01332     int h, cy;
01333     int offset[4];
01334 
01335     if(s->avctx->draw_horiz_band==NULL)
01336         return;
01337 
01338     h= y - s->last_slice_end;
01339     y -= h;
01340 
01341     if (!s->flipped_image) {
01342         if (y == 0)
01343             h -= s->height - s->avctx->height;  // account for non-mod16
01344         y = s->height - y - h;
01345     }
01346 
01347     cy = y >> 1;
01348     offset[0] = s->current_frame.linesize[0]*y;
01349     offset[1] = s->current_frame.linesize[1]*cy;
01350     offset[2] = s->current_frame.linesize[2]*cy;
01351     offset[3] = 0;
01352 
01353     emms_c();
01354     s->avctx->draw_horiz_band(s->avctx, &s->current_frame, offset, y, 3, h);
01355     s->last_slice_end= y + h;
01356 }
01357 
01358 /*
01359  * Perform the final rendering for a particular slice of data.
01360  * The slice number ranges from 0..(c_superblock_height - 1).
01361  */
01362 static void render_slice(Vp3DecodeContext *s, int slice)
01363 {
01364     int x, y, i, j;
01365     LOCAL_ALIGNED_16(DCTELEM, block, [64]);
01366     int motion_x = 0xdeadbeef, motion_y = 0xdeadbeef;
01367     int motion_halfpel_index;
01368     uint8_t *motion_source;
01369     int plane, first_pixel;
01370 
01371     if (slice >= s->c_superblock_height)
01372         return;
01373 
01374     for (plane = 0; plane < 3; plane++) {
01375         uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
01376         uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
01377         uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
01378         int stride            = s->current_frame.linesize[plane];
01379         int plane_width       = s->width  >> (plane && s->chroma_x_shift);
01380         int plane_height      = s->height >> (plane && s->chroma_y_shift);
01381         int8_t (*motion_val)[2] = s->motion_val[!!plane];
01382 
01383         int sb_x, sb_y        = slice << (!plane && s->chroma_y_shift);
01384         int slice_height      = sb_y + 1 + (!plane && s->chroma_y_shift);
01385         int slice_width       = plane ? s->c_superblock_width : s->y_superblock_width;
01386 
01387         int fragment_width    = s->fragment_width[!!plane];
01388         int fragment_height   = s->fragment_height[!!plane];
01389         int fragment_start    = s->fragment_start[plane];
01390 
01391         if (!s->flipped_image) stride = -stride;
01392         if (CONFIG_GRAY && plane && (s->avctx->flags & CODEC_FLAG_GRAY))
01393             continue;
01394 
01395 
01396         if(FFABS(stride) > 2048)
01397             return; //various tables are fixed size
01398 
01399         /* for each superblock row in the slice (both of them)... */
01400         for (; sb_y < slice_height; sb_y++) {
01401 
01402             /* for each superblock in a row... */
01403             for (sb_x = 0; sb_x < slice_width; sb_x++) {
01404 
01405                 /* for each block in a superblock... */
01406                 for (j = 0; j < 16; j++) {
01407                     x = 4*sb_x + hilbert_offset[j][0];
01408                     y = 4*sb_y + hilbert_offset[j][1];
01409 
01410                     i = fragment_start + y*fragment_width + x;
01411 
01412                     // bounds check
01413                     if (x >= fragment_width || y >= fragment_height)
01414                         continue;
01415 
01416                 first_pixel = 8*y*stride + 8*x;
01417 
01418                 /* transform if this block was coded */
01419                 if (s->all_fragments[i].coding_method != MODE_COPY) {
01420                     if ((s->all_fragments[i].coding_method == MODE_USING_GOLDEN) ||
01421                         (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
01422                         motion_source= golden_plane;
01423                     else
01424                         motion_source= last_plane;
01425 
01426                     motion_source += first_pixel;
01427                     motion_halfpel_index = 0;
01428 
01429                     /* sort out the motion vector if this fragment is coded
01430                      * using a motion vector method */
01431                     if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
01432                         (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
01433                         int src_x, src_y;
01434                         motion_x = motion_val[y*fragment_width + x][0];
01435                         motion_y = motion_val[y*fragment_width + x][1];
01436 
01437                         src_x= (motion_x>>1) + 8*x;
01438                         src_y= (motion_y>>1) + 8*y;
01439 
01440                         motion_halfpel_index = motion_x & 0x01;
01441                         motion_source += (motion_x >> 1);
01442 
01443                         motion_halfpel_index |= (motion_y & 0x01) << 1;
01444                         motion_source += ((motion_y >> 1) * stride);
01445 
01446                         if(src_x<0 || src_y<0 || src_x + 9 >= plane_width || src_y + 9 >= plane_height){
01447                             uint8_t *temp= s->edge_emu_buffer;
01448                             if(stride<0) temp -= 9*stride;
01449                             else temp += 9*stride;
01450 
01451                             ff_emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height);
01452                             motion_source= temp;
01453                         }
01454                     }
01455 
01456 
01457                     /* first, take care of copying a block from either the
01458                      * previous or the golden frame */
01459                     if (s->all_fragments[i].coding_method != MODE_INTRA) {
01460                         /* Note, it is possible to implement all MC cases with
01461                            put_no_rnd_pixels_l2 which would look more like the
01462                            VP3 source but this would be slower as
01463                            put_no_rnd_pixels_tab is better optimzed */
01464                         if(motion_halfpel_index != 3){
01465                             s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
01466                                 output_plane + first_pixel,
01467                                 motion_source, stride, 8);
01468                         }else{
01469                             int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
01470                             s->dsp.put_no_rnd_pixels_l2[1](
01471                                 output_plane + first_pixel,
01472                                 motion_source - d,
01473                                 motion_source + stride + 1 + d,
01474                                 stride, 8);
01475                         }
01476                     }
01477 
01478                         s->dsp.clear_block(block);
01479 
01480                     /* invert DCT and place (or add) in final output */
01481 
01482                     if (s->all_fragments[i].coding_method == MODE_INTRA) {
01483                         int index;
01484                         index = vp3_dequant(s, s->all_fragments + i, plane, 0, block);
01485                         if (index > 63)
01486                             continue;
01487                         if(s->avctx->idct_algo!=FF_IDCT_VP3)
01488                             block[0] += 128<<3;
01489                         s->dsp.idct_put(
01490                             output_plane + first_pixel,
01491                             stride,
01492                             block);
01493                     } else {
01494                         int index = vp3_dequant(s, s->all_fragments + i, plane, 1, block);
01495                         if (index > 63)
01496                             continue;
01497                         if (index > 0) {
01498                         s->dsp.idct_add(
01499                             output_plane + first_pixel,
01500                             stride,
01501                             block);
01502                         } else {
01503                             s->dsp.vp3_idct_dc_add(output_plane + first_pixel, stride, block);
01504                         }
01505                     }
01506                 } else {
01507 
01508                     /* copy directly from the previous frame */
01509                     s->dsp.put_pixels_tab[1][0](
01510                         output_plane + first_pixel,
01511                         last_plane + first_pixel,
01512                         stride, 8);
01513 
01514                 }
01515                 }
01516             }
01517 
01518             // Filter up to the last row in the superblock row
01519             apply_loop_filter(s, plane, 4*sb_y - !!sb_y, FFMIN(4*sb_y+3, fragment_height-1));
01520         }
01521     }
01522 
01523      /* this looks like a good place for slice dispatch... */
01524      /* algorithm:
01525       *   if (slice == s->macroblock_height - 1)
01526       *     dispatch (both last slice & 2nd-to-last slice);
01527       *   else if (slice > 0)
01528       *     dispatch (slice - 1);
01529       */
01530 
01531     vp3_draw_horiz_band(s, FFMIN(64*slice + 64-16, s->height-16));
01532 }
01533 
01534 /*
01535  * This is the ffmpeg/libavcodec API init function.
01536  */
01537 static av_cold int vp3_decode_init(AVCodecContext *avctx)
01538 {
01539     Vp3DecodeContext *s = avctx->priv_data;
01540     int i, inter, plane;
01541     int c_width;
01542     int c_height;
01543     int y_fragment_count, c_fragment_count;
01544 
01545     if (avctx->codec_tag == MKTAG('V','P','3','0'))
01546         s->version = 0;
01547     else
01548         s->version = 1;
01549 
01550     s->avctx = avctx;
01551     s->width = FFALIGN(avctx->width, 16);
01552     s->height = FFALIGN(avctx->height, 16);
01553     if (avctx->pix_fmt == PIX_FMT_NONE)
01554         avctx->pix_fmt = PIX_FMT_YUV420P;
01555     avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
01556     if(avctx->idct_algo==FF_IDCT_AUTO)
01557         avctx->idct_algo=FF_IDCT_VP3;
01558     dsputil_init(&s->dsp, avctx);
01559 
01560     ff_init_scantable(s->dsp.idct_permutation, &s->scantable, ff_zigzag_direct);
01561 
01562     /* initialize to an impossible value which will force a recalculation
01563      * in the first frame decode */
01564     for (i = 0; i < 3; i++)
01565         s->qps[i] = -1;
01566 
01567     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
01568 
01569     s->y_superblock_width = (s->width + 31) / 32;
01570     s->y_superblock_height = (s->height + 31) / 32;
01571     s->y_superblock_count = s->y_superblock_width * s->y_superblock_height;
01572 
01573     /* work out the dimensions for the C planes */
01574     c_width = s->width >> s->chroma_x_shift;
01575     c_height = s->height >> s->chroma_y_shift;
01576     s->c_superblock_width = (c_width + 31) / 32;
01577     s->c_superblock_height = (c_height + 31) / 32;
01578     s->c_superblock_count = s->c_superblock_width * s->c_superblock_height;
01579 
01580     s->superblock_count = s->y_superblock_count + (s->c_superblock_count * 2);
01581     s->u_superblock_start = s->y_superblock_count;
01582     s->v_superblock_start = s->u_superblock_start + s->c_superblock_count;
01583     s->superblock_coding = av_malloc(s->superblock_count);
01584 
01585     s->macroblock_width = (s->width + 15) / 16;
01586     s->macroblock_height = (s->height + 15) / 16;
01587     s->macroblock_count = s->macroblock_width * s->macroblock_height;
01588 
01589     s->fragment_width[0] = s->width / FRAGMENT_PIXELS;
01590     s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
01591     s->fragment_width[1]  = s->fragment_width[0]  >> s->chroma_x_shift;
01592     s->fragment_height[1] = s->fragment_height[0] >> s->chroma_y_shift;
01593 
01594     /* fragment count covers all 8x8 blocks for all 3 planes */
01595     y_fragment_count     = s->fragment_width[0] * s->fragment_height[0];
01596     c_fragment_count     = s->fragment_width[1] * s->fragment_height[1];
01597     s->fragment_count    = y_fragment_count + 2*c_fragment_count;
01598     s->fragment_start[1] = y_fragment_count;
01599     s->fragment_start[2] = y_fragment_count + c_fragment_count;
01600 
01601     s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
01602     s->coded_fragment_list[0] = av_malloc(s->fragment_count * sizeof(int));
01603     s->dct_tokens_base = av_malloc(64*s->fragment_count * sizeof(*s->dct_tokens_base));
01604     s->motion_val[0] = av_malloc(y_fragment_count * sizeof(*s->motion_val[0]));
01605     s->motion_val[1] = av_malloc(c_fragment_count * sizeof(*s->motion_val[1]));
01606 
01607     if (!s->superblock_coding || !s->all_fragments || !s->dct_tokens_base ||
01608         !s->coded_fragment_list[0] || !s->motion_val[0] || !s->motion_val[1]) {
01609         vp3_decode_end(avctx);
01610         return -1;
01611     }
01612 
01613     if (!s->theora_tables)
01614     {
01615         for (i = 0; i < 64; i++) {
01616             s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
01617             s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
01618             s->base_matrix[0][i] = vp31_intra_y_dequant[i];
01619             s->base_matrix[1][i] = vp31_intra_c_dequant[i];
01620             s->base_matrix[2][i] = vp31_inter_dequant[i];
01621             s->filter_limit_values[i] = vp31_filter_limit_values[i];
01622         }
01623 
01624         for(inter=0; inter<2; inter++){
01625             for(plane=0; plane<3; plane++){
01626                 s->qr_count[inter][plane]= 1;
01627                 s->qr_size [inter][plane][0]= 63;
01628                 s->qr_base [inter][plane][0]=
01629                 s->qr_base [inter][plane][1]= 2*inter + (!!plane)*!inter;
01630             }
01631         }
01632 
01633         /* init VLC tables */
01634         for (i = 0; i < 16; i++) {
01635 
01636             /* DC histograms */
01637             init_vlc(&s->dc_vlc[i], 11, 32,
01638                 &dc_bias[i][0][1], 4, 2,
01639                 &dc_bias[i][0][0], 4, 2, 0);
01640 
01641             /* group 1 AC histograms */
01642             init_vlc(&s->ac_vlc_1[i], 11, 32,
01643                 &ac_bias_0[i][0][1], 4, 2,
01644                 &ac_bias_0[i][0][0], 4, 2, 0);
01645 
01646             /* group 2 AC histograms */
01647             init_vlc(&s->ac_vlc_2[i], 11, 32,
01648                 &ac_bias_1[i][0][1], 4, 2,
01649                 &ac_bias_1[i][0][0], 4, 2, 0);
01650 
01651             /* group 3 AC histograms */
01652             init_vlc(&s->ac_vlc_3[i], 11, 32,
01653                 &ac_bias_2[i][0][1], 4, 2,
01654                 &ac_bias_2[i][0][0], 4, 2, 0);
01655 
01656             /* group 4 AC histograms */
01657             init_vlc(&s->ac_vlc_4[i], 11, 32,
01658                 &ac_bias_3[i][0][1], 4, 2,
01659                 &ac_bias_3[i][0][0], 4, 2, 0);
01660         }
01661     } else {
01662 
01663         for (i = 0; i < 16; i++) {
01664             /* DC histograms */
01665             if (init_vlc(&s->dc_vlc[i], 11, 32,
01666                 &s->huffman_table[i][0][1], 8, 4,
01667                 &s->huffman_table[i][0][0], 8, 4, 0) < 0)
01668                 goto vlc_fail;
01669 
01670             /* group 1 AC histograms */
01671             if (init_vlc(&s->ac_vlc_1[i], 11, 32,
01672                 &s->huffman_table[i+16][0][1], 8, 4,
01673                 &s->huffman_table[i+16][0][0], 8, 4, 0) < 0)
01674                 goto vlc_fail;
01675 
01676             /* group 2 AC histograms */
01677             if (init_vlc(&s->ac_vlc_2[i], 11, 32,
01678                 &s->huffman_table[i+16*2][0][1], 8, 4,
01679                 &s->huffman_table[i+16*2][0][0], 8, 4, 0) < 0)
01680                 goto vlc_fail;
01681 
01682             /* group 3 AC histograms */
01683             if (init_vlc(&s->ac_vlc_3[i], 11, 32,
01684                 &s->huffman_table[i+16*3][0][1], 8, 4,
01685                 &s->huffman_table[i+16*3][0][0], 8, 4, 0) < 0)
01686                 goto vlc_fail;
01687 
01688             /* group 4 AC histograms */
01689             if (init_vlc(&s->ac_vlc_4[i], 11, 32,
01690                 &s->huffman_table[i+16*4][0][1], 8, 4,
01691                 &s->huffman_table[i+16*4][0][0], 8, 4, 0) < 0)
01692                 goto vlc_fail;
01693         }
01694     }
01695 
01696     init_vlc(&s->superblock_run_length_vlc, 6, 34,
01697         &superblock_run_length_vlc_table[0][1], 4, 2,
01698         &superblock_run_length_vlc_table[0][0], 4, 2, 0);
01699 
01700     init_vlc(&s->fragment_run_length_vlc, 5, 30,
01701         &fragment_run_length_vlc_table[0][1], 4, 2,
01702         &fragment_run_length_vlc_table[0][0], 4, 2, 0);
01703 
01704     init_vlc(&s->mode_code_vlc, 3, 8,
01705         &mode_code_vlc_table[0][1], 2, 1,
01706         &mode_code_vlc_table[0][0], 2, 1, 0);
01707 
01708     init_vlc(&s->motion_vector_vlc, 6, 63,
01709         &motion_vector_vlc_table[0][1], 2, 1,
01710         &motion_vector_vlc_table[0][0], 2, 1, 0);
01711 
01712     /* work out the block mapping tables */
01713     s->superblock_fragments = av_malloc(s->superblock_count * 16 * sizeof(int));
01714     s->macroblock_coding = av_malloc(s->macroblock_count + 1);
01715     if (!s->superblock_fragments || !s->macroblock_coding) {
01716         vp3_decode_end(avctx);
01717         return -1;
01718     }
01719     init_block_mapping(s);
01720 
01721     for (i = 0; i < 3; i++) {
01722         s->current_frame.data[i] = NULL;
01723         s->last_frame.data[i] = NULL;
01724         s->golden_frame.data[i] = NULL;
01725     }
01726 
01727     return 0;
01728 
01729 vlc_fail:
01730     av_log(avctx, AV_LOG_FATAL, "Invalid huffman table\n");
01731     return -1;
01732 }
01733 
01734 /*
01735  * This is the ffmpeg/libavcodec API frame decode function.
01736  */
01737 static int vp3_decode_frame(AVCodecContext *avctx,
01738                             void *data, int *data_size,
01739                             AVPacket *avpkt)
01740 {
01741     const uint8_t *buf = avpkt->data;
01742     int buf_size = avpkt->size;
01743     Vp3DecodeContext *s = avctx->priv_data;
01744     GetBitContext gb;
01745     static int counter = 0;
01746     int i;
01747 
01748     init_get_bits(&gb, buf, buf_size * 8);
01749 
01750     if (s->theora && get_bits1(&gb))
01751     {
01752         av_log(avctx, AV_LOG_ERROR, "Header packet passed to frame decoder, skipping\n");
01753         return -1;
01754     }
01755 
01756     s->keyframe = !get_bits1(&gb);
01757     if (!s->theora)
01758         skip_bits(&gb, 1);
01759     for (i = 0; i < 3; i++)
01760         s->last_qps[i] = s->qps[i];
01761 
01762     s->nqps=0;
01763     do{
01764         s->qps[s->nqps++]= get_bits(&gb, 6);
01765     } while(s->theora >= 0x030200 && s->nqps<3 && get_bits1(&gb));
01766     for (i = s->nqps; i < 3; i++)
01767         s->qps[i] = -1;
01768 
01769     if (s->avctx->debug & FF_DEBUG_PICT_INFO)
01770         av_log(s->avctx, AV_LOG_INFO, " VP3 %sframe #%d: Q index = %d\n",
01771             s->keyframe?"key":"", counter, s->qps[0]);
01772     counter++;
01773 
01774     if (s->qps[0] != s->last_qps[0])
01775         init_loop_filter(s);
01776 
01777     for (i = 0; i < s->nqps; i++)
01778         // reinit all dequantizers if the first one changed, because
01779         // the DC of the first quantizer must be used for all matrices
01780         if (s->qps[i] != s->last_qps[i] || s->qps[0] != s->last_qps[0])
01781             init_dequantizer(s, i);
01782 
01783     if (avctx->skip_frame >= AVDISCARD_NONKEY && !s->keyframe)
01784         return buf_size;
01785 
01786     s->current_frame.reference = 3;
01787     s->current_frame.pict_type = s->keyframe ? FF_I_TYPE : FF_P_TYPE;
01788     if (avctx->get_buffer(avctx, &s->current_frame) < 0) {
01789         av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
01790         goto error;
01791     }
01792 
01793     if (s->keyframe) {
01794         if (!s->theora)
01795         {
01796             skip_bits(&gb, 4); /* width code */
01797             skip_bits(&gb, 4); /* height code */
01798             if (s->version)
01799             {
01800                 s->version = get_bits(&gb, 5);
01801                 if (counter == 1)
01802                     av_log(s->avctx, AV_LOG_DEBUG, "VP version: %d\n", s->version);
01803             }
01804         }
01805         if (s->version || s->theora)
01806         {
01807                 if (get_bits1(&gb))
01808                     av_log(s->avctx, AV_LOG_ERROR, "Warning, unsupported keyframe coding type?!\n");
01809             skip_bits(&gb, 2); /* reserved? */
01810         }
01811     } else {
01812         if (!s->golden_frame.data[0]) {
01813             av_log(s->avctx, AV_LOG_WARNING, "vp3: first frame not a keyframe\n");
01814 
01815             s->golden_frame.reference = 3;
01816             s->golden_frame.pict_type = FF_I_TYPE;
01817             if (avctx->get_buffer(avctx, &s->golden_frame) < 0) {
01818                 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
01819                 goto error;
01820             }
01821             s->last_frame = s->golden_frame;
01822             s->last_frame.type = FF_BUFFER_TYPE_COPY;
01823         }
01824     }
01825 
01826     s->current_frame.qscale_table= s->qscale_table; //FIXME allocate individual tables per AVFrame
01827     s->current_frame.qstride= 0;
01828 
01829     memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
01830 
01831     if (unpack_superblocks(s, &gb)){
01832         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
01833         goto error;
01834     }
01835     if (unpack_modes(s, &gb)){
01836         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
01837         goto error;
01838     }
01839     if (unpack_vectors(s, &gb)){
01840         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_vectors\n");
01841         goto error;
01842     }
01843     if (unpack_block_qpis(s, &gb)){
01844         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
01845         goto error;
01846     }
01847     if (unpack_dct_coeffs(s, &gb)){
01848         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
01849         goto error;
01850     }
01851 
01852     for (i = 0; i < 3; i++) {
01853         int height = s->height >> (i && s->chroma_y_shift);
01854         if (s->flipped_image)
01855             s->data_offset[i] = 0;
01856         else
01857             s->data_offset[i] = (height-1) * s->current_frame.linesize[i];
01858     }
01859 
01860     s->last_slice_end = 0;
01861     for (i = 0; i < s->c_superblock_height; i++)
01862         render_slice(s, i);
01863 
01864     // filter the last row
01865     for (i = 0; i < 3; i++) {
01866         int row = (s->height >> (3+(i && s->chroma_y_shift))) - 1;
01867         apply_loop_filter(s, i, row, row+1);
01868     }
01869     vp3_draw_horiz_band(s, s->height);
01870 
01871     *data_size=sizeof(AVFrame);
01872     *(AVFrame*)data= s->current_frame;
01873 
01874     /* release the last frame, if it is allocated and if it is not the
01875      * golden frame */
01876     if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
01877         avctx->release_buffer(avctx, &s->last_frame);
01878 
01879     /* shuffle frames (last = current) */
01880     s->last_frame= s->current_frame;
01881 
01882     if (s->keyframe) {
01883         if (s->golden_frame.data[0])
01884             avctx->release_buffer(avctx, &s->golden_frame);
01885         s->golden_frame = s->current_frame;
01886         s->last_frame.type = FF_BUFFER_TYPE_COPY;
01887     }
01888 
01889     s->current_frame.data[0]= NULL; /* ensure that we catch any access to this released frame */
01890 
01891     return buf_size;
01892 
01893 error:
01894     if (s->current_frame.data[0])
01895         avctx->release_buffer(avctx, &s->current_frame);
01896     return -1;
01897 }
01898 
01899 /*
01900  * This is the ffmpeg/libavcodec API module cleanup function.
01901  */
01902 static av_cold int vp3_decode_end(AVCodecContext *avctx)
01903 {
01904     Vp3DecodeContext *s = avctx->priv_data;
01905     int i;
01906 
01907     av_free(s->superblock_coding);
01908     av_free(s->all_fragments);
01909     av_free(s->coded_fragment_list[0]);
01910     av_free(s->dct_tokens_base);
01911     av_free(s->superblock_fragments);
01912     av_free(s->macroblock_coding);
01913     av_free(s->motion_val[0]);
01914     av_free(s->motion_val[1]);
01915 
01916     for (i = 0; i < 16; i++) {
01917         free_vlc(&s->dc_vlc[i]);
01918         free_vlc(&s->ac_vlc_1[i]);
01919         free_vlc(&s->ac_vlc_2[i]);
01920         free_vlc(&s->ac_vlc_3[i]);
01921         free_vlc(&s->ac_vlc_4[i]);
01922     }
01923 
01924     free_vlc(&s->superblock_run_length_vlc);
01925     free_vlc(&s->fragment_run_length_vlc);
01926     free_vlc(&s->mode_code_vlc);
01927     free_vlc(&s->motion_vector_vlc);
01928 
01929     /* release all frames */
01930     if (s->golden_frame.data[0])
01931         avctx->release_buffer(avctx, &s->golden_frame);
01932     if (s->last_frame.data[0] && s->last_frame.type != FF_BUFFER_TYPE_COPY)
01933         avctx->release_buffer(avctx, &s->last_frame);
01934     /* no need to release the current_frame since it will always be pointing
01935      * to the same frame as either the golden or last frame */
01936 
01937     return 0;
01938 }
01939 
01940 static int read_huffman_tree(AVCodecContext *avctx, GetBitContext *gb)
01941 {
01942     Vp3DecodeContext *s = avctx->priv_data;
01943 
01944     if (get_bits1(gb)) {
01945         int token;
01946         if (s->entries >= 32) { /* overflow */
01947             av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
01948             return -1;
01949         }
01950         token = get_bits(gb, 5);
01951         //av_log(avctx, AV_LOG_DEBUG, "hti %d hbits %x token %d entry : %d size %d\n", s->hti, s->hbits, token, s->entries, s->huff_code_size);
01952         s->huffman_table[s->hti][token][0] = s->hbits;
01953         s->huffman_table[s->hti][token][1] = s->huff_code_size;
01954         s->entries++;
01955     }
01956     else {
01957         if (s->huff_code_size >= 32) {/* overflow */
01958             av_log(avctx, AV_LOG_ERROR, "huffman tree overflow\n");
01959             return -1;
01960         }
01961         s->huff_code_size++;
01962         s->hbits <<= 1;
01963         if (read_huffman_tree(avctx, gb))
01964             return -1;
01965         s->hbits |= 1;
01966         if (read_huffman_tree(avctx, gb))
01967             return -1;
01968         s->hbits >>= 1;
01969         s->huff_code_size--;
01970     }
01971     return 0;
01972 }
01973 
01974 #if CONFIG_THEORA_DECODER
01975 static const enum PixelFormat theora_pix_fmts[4] = {
01976     PIX_FMT_YUV420P, PIX_FMT_NONE, PIX_FMT_YUV422P, PIX_FMT_YUV444P
01977 };
01978 
01979 static int theora_decode_header(AVCodecContext *avctx, GetBitContext *gb)
01980 {
01981     Vp3DecodeContext *s = avctx->priv_data;
01982     int visible_width, visible_height, colorspace;
01983     int offset_x = 0, offset_y = 0;
01984     AVRational fps;
01985 
01986     s->theora = get_bits_long(gb, 24);
01987     av_log(avctx, AV_LOG_DEBUG, "Theora bitstream version %X\n", s->theora);
01988 
01989     /* 3.2.0 aka alpha3 has the same frame orientation as original vp3 */
01990     /* but previous versions have the image flipped relative to vp3 */
01991     if (s->theora < 0x030200)
01992     {
01993         s->flipped_image = 1;
01994         av_log(avctx, AV_LOG_DEBUG, "Old (<alpha3) Theora bitstream, flipped image\n");
01995     }
01996 
01997     visible_width  = s->width  = get_bits(gb, 16) << 4;
01998     visible_height = s->height = get_bits(gb, 16) << 4;
01999 
02000     if(avcodec_check_dimensions(avctx, s->width, s->height)){
02001         av_log(avctx, AV_LOG_ERROR, "Invalid dimensions (%dx%d)\n", s->width, s->height);
02002         s->width= s->height= 0;
02003         return -1;
02004     }
02005 
02006     if (s->theora >= 0x030200) {
02007         visible_width  = get_bits_long(gb, 24);
02008         visible_height = get_bits_long(gb, 24);
02009 
02010         offset_x = get_bits(gb, 8); /* offset x */
02011         offset_y = get_bits(gb, 8); /* offset y, from bottom */
02012     }
02013 
02014     fps.num = get_bits_long(gb, 32);
02015     fps.den = get_bits_long(gb, 32);
02016     if (fps.num && fps.den) {
02017         av_reduce(&avctx->time_base.num, &avctx->time_base.den,
02018                   fps.den, fps.num, 1<<30);
02019     }
02020 
02021     avctx->sample_aspect_ratio.num = get_bits_long(gb, 24);
02022     avctx->sample_aspect_ratio.den = get_bits_long(gb, 24);
02023 
02024     if (s->theora < 0x030200)
02025         skip_bits(gb, 5); /* keyframe frequency force */
02026     colorspace = get_bits(gb, 8);
02027     skip_bits(gb, 24); /* bitrate */
02028 
02029     skip_bits(gb, 6); /* quality hint */
02030 
02031     if (s->theora >= 0x030200)
02032     {
02033         skip_bits(gb, 5); /* keyframe frequency force */
02034         avctx->pix_fmt = theora_pix_fmts[get_bits(gb, 2)];
02035         skip_bits(gb, 3); /* reserved */
02036     }
02037 
02038 //    align_get_bits(gb);
02039 
02040     if (   visible_width  <= s->width  && visible_width  > s->width-16
02041         && visible_height <= s->height && visible_height > s->height-16
02042         && !offset_x && (offset_y == s->height - visible_height))
02043         avcodec_set_dimensions(avctx, visible_width, visible_height);
02044     else
02045         avcodec_set_dimensions(avctx, s->width, s->height);
02046 
02047     if (colorspace == 1) {
02048         avctx->color_primaries = AVCOL_PRI_BT470M;
02049     } else if (colorspace == 2) {
02050         avctx->color_primaries = AVCOL_PRI_BT470BG;
02051     }
02052     if (colorspace == 1 || colorspace == 2) {
02053         avctx->colorspace = AVCOL_SPC_BT470BG;
02054         avctx->color_trc  = AVCOL_TRC_BT709;
02055     }
02056 
02057     return 0;
02058 }
02059 
02060 static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
02061 {
02062     Vp3DecodeContext *s = avctx->priv_data;
02063     int i, n, matrices, inter, plane;
02064 
02065     if (s->theora >= 0x030200) {
02066         n = get_bits(gb, 3);
02067         /* loop filter limit values table */
02068         for (i = 0; i < 64; i++) {
02069             s->filter_limit_values[i] = get_bits(gb, n);
02070             if (s->filter_limit_values[i] > 127) {
02071                 av_log(avctx, AV_LOG_ERROR, "filter limit value too large (%i > 127), clamping\n", s->filter_limit_values[i]);
02072                 s->filter_limit_values[i] = 127;
02073             }
02074         }
02075     }
02076 
02077     if (s->theora >= 0x030200)
02078         n = get_bits(gb, 4) + 1;
02079     else
02080         n = 16;
02081     /* quality threshold table */
02082     for (i = 0; i < 64; i++)
02083         s->coded_ac_scale_factor[i] = get_bits(gb, n);
02084 
02085     if (s->theora >= 0x030200)
02086         n = get_bits(gb, 4) + 1;
02087     else
02088         n = 16;
02089     /* dc scale factor table */
02090     for (i = 0; i < 64; i++)
02091         s->coded_dc_scale_factor[i] = get_bits(gb, n);
02092 
02093     if (s->theora >= 0x030200)
02094         matrices = get_bits(gb, 9) + 1;
02095     else
02096         matrices = 3;
02097 
02098     if(matrices > 384){
02099         av_log(avctx, AV_LOG_ERROR, "invalid number of base matrixes\n");
02100         return -1;
02101     }
02102 
02103     for(n=0; n<matrices; n++){
02104         for (i = 0; i < 64; i++)
02105             s->base_matrix[n][i]= get_bits(gb, 8);
02106     }
02107 
02108     for (inter = 0; inter <= 1; inter++) {
02109         for (plane = 0; plane <= 2; plane++) {
02110             int newqr= 1;
02111             if (inter || plane > 0)
02112                 newqr = get_bits1(gb);
02113             if (!newqr) {
02114                 int qtj, plj;
02115                 if(inter && get_bits1(gb)){
02116                     qtj = 0;
02117                     plj = plane;
02118                 }else{
02119                     qtj= (3*inter + plane - 1) / 3;
02120                     plj= (plane + 2) % 3;
02121                 }
02122                 s->qr_count[inter][plane]= s->qr_count[qtj][plj];
02123                 memcpy(s->qr_size[inter][plane], s->qr_size[qtj][plj], sizeof(s->qr_size[0][0]));
02124                 memcpy(s->qr_base[inter][plane], s->qr_base[qtj][plj], sizeof(s->qr_base[0][0]));
02125             } else {
02126                 int qri= 0;
02127                 int qi = 0;
02128 
02129                 for(;;){
02130                     i= get_bits(gb, av_log2(matrices-1)+1);
02131                     if(i>= matrices){
02132                         av_log(avctx, AV_LOG_ERROR, "invalid base matrix index\n");
02133                         return -1;
02134                     }
02135                     s->qr_base[inter][plane][qri]= i;
02136                     if(qi >= 63)
02137                         break;
02138                     i = get_bits(gb, av_log2(63-qi)+1) + 1;
02139                     s->qr_size[inter][plane][qri++]= i;
02140                     qi += i;
02141                 }
02142 
02143                 if (qi > 63) {
02144                     av_log(avctx, AV_LOG_ERROR, "invalid qi %d > 63\n", qi);
02145                     return -1;
02146                 }
02147                 s->qr_count[inter][plane]= qri;
02148             }
02149         }
02150     }
02151 
02152     /* Huffman tables */
02153     for (s->hti = 0; s->hti < 80; s->hti++) {
02154         s->entries = 0;
02155         s->huff_code_size = 1;
02156         if (!get_bits1(gb)) {
02157             s->hbits = 0;
02158             if(read_huffman_tree(avctx, gb))
02159                 return -1;
02160             s->hbits = 1;
02161             if(read_huffman_tree(avctx, gb))
02162                 return -1;
02163         }
02164     }
02165 
02166     s->theora_tables = 1;
02167 
02168     return 0;
02169 }
02170 
02171 static av_cold int theora_decode_init(AVCodecContext *avctx)
02172 {
02173     Vp3DecodeContext *s = avctx->priv_data;
02174     GetBitContext gb;
02175     int ptype;
02176     uint8_t *header_start[3];
02177     int header_len[3];
02178     int i;
02179 
02180     s->theora = 1;
02181 
02182     if (!avctx->extradata_size)
02183     {
02184         av_log(avctx, AV_LOG_ERROR, "Missing extradata!\n");
02185         return -1;
02186     }
02187 
02188     if (ff_split_xiph_headers(avctx->extradata, avctx->extradata_size,
02189                               42, header_start, header_len) < 0) {
02190         av_log(avctx, AV_LOG_ERROR, "Corrupt extradata\n");
02191         return -1;
02192     }
02193 
02194   for(i=0;i<3;i++) {
02195     init_get_bits(&gb, header_start[i], header_len[i] * 8);
02196 
02197     ptype = get_bits(&gb, 8);
02198 
02199      if (!(ptype & 0x80))
02200      {
02201         av_log(avctx, AV_LOG_ERROR, "Invalid extradata!\n");
02202 //        return -1;
02203      }
02204 
02205     // FIXME: Check for this as well.
02206     skip_bits_long(&gb, 6*8); /* "theora" */
02207 
02208     switch(ptype)
02209     {
02210         case 0x80:
02211             theora_decode_header(avctx, &gb);
02212                 break;
02213         case 0x81:
02214 // FIXME: is this needed? it breaks sometimes
02215 //            theora_decode_comments(avctx, gb);
02216             break;
02217         case 0x82:
02218             if (theora_decode_tables(avctx, &gb))
02219                 return -1;
02220             break;
02221         default:
02222             av_log(avctx, AV_LOG_ERROR, "Unknown Theora config packet: %d\n", ptype&~0x80);
02223             break;
02224     }
02225     if(ptype != 0x81 && 8*header_len[i] != get_bits_count(&gb))
02226         av_log(avctx, AV_LOG_WARNING, "%d bits left in packet %X\n", 8*header_len[i] - get_bits_count(&gb), ptype);
02227     if (s->theora < 0x030200)
02228         break;
02229   }
02230 
02231     return vp3_decode_init(avctx);
02232 }
02233 
02234 AVCodec theora_decoder = {
02235     "theora",
02236     AVMEDIA_TYPE_VIDEO,
02237     CODEC_ID_THEORA,
02238     sizeof(Vp3DecodeContext),
02239     theora_decode_init,
02240     NULL,
02241     vp3_decode_end,
02242     vp3_decode_frame,
02243     CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
02244     NULL,
02245     .long_name = NULL_IF_CONFIG_SMALL("Theora"),
02246 };
02247 #endif
02248 
02249 AVCodec vp3_decoder = {
02250     "vp3",
02251     AVMEDIA_TYPE_VIDEO,
02252     CODEC_ID_VP3,
02253     sizeof(Vp3DecodeContext),
02254     vp3_decode_init,
02255     NULL,
02256     vp3_decode_end,
02257     vp3_decode_frame,
02258     CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
02259     NULL,
02260     .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
02261 };