LCOV - code coverage report
Current view: top level - libavcodec - vp8.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 1339 1583 84.6 %
Date: 2017-12-17 23:02:56 Functions: 74 83 89.2 %

          Line data    Source code
       1             : /*
       2             :  * VP7/VP8 compatible video decoder
       3             :  *
       4             :  * Copyright (C) 2010 David Conrad
       5             :  * Copyright (C) 2010 Ronald S. Bultje
       6             :  * Copyright (C) 2010 Fiona Glaser
       7             :  * Copyright (C) 2012 Daniel Kang
       8             :  * Copyright (C) 2014 Peter Ross
       9             :  *
      10             :  * This file is part of FFmpeg.
      11             :  *
      12             :  * FFmpeg is free software; you can redistribute it and/or
      13             :  * modify it under the terms of the GNU Lesser General Public
      14             :  * License as published by the Free Software Foundation; either
      15             :  * version 2.1 of the License, or (at your option) any later version.
      16             :  *
      17             :  * FFmpeg is distributed in the hope that it will be useful,
      18             :  * but WITHOUT ANY WARRANTY; without even the implied warranty of
      19             :  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
      20             :  * Lesser General Public License for more details.
      21             :  *
      22             :  * You should have received a copy of the GNU Lesser General Public
      23             :  * License along with FFmpeg; if not, write to the Free Software
      24             :  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
      25             :  */
      26             : 
      27             : #include "libavutil/imgutils.h"
      28             : 
      29             : #include "avcodec.h"
      30             : #include "hwaccel.h"
      31             : #include "internal.h"
      32             : #include "mathops.h"
      33             : #include "rectangle.h"
      34             : #include "thread.h"
      35             : #include "vp8.h"
      36             : #include "vp8data.h"
      37             : 
      38             : #if ARCH_ARM
      39             : #   include "arm/vp8.h"
      40             : #endif
      41             : 
      42             : #if CONFIG_VP7_DECODER && CONFIG_VP8_DECODER
      43             : #define VPX(vp7, f) (vp7 ? vp7_ ## f : vp8_ ## f)
      44             : #elif CONFIG_VP7_DECODER
      45             : #define VPX(vp7, f) vp7_ ## f
      46             : #else // CONFIG_VP8_DECODER
      47             : #define VPX(vp7, f) vp8_ ## f
      48             : #endif
      49             : 
      50          73 : static void free_buffers(VP8Context *s)
      51             : {
      52             :     int i;
      53          73 :     if (s->thread_data)
      54         333 :         for (i = 0; i < MAX_THREADS; i++) {
      55             : #if HAVE_THREADS
      56         296 :             pthread_cond_destroy(&s->thread_data[i].cond);
      57         296 :             pthread_mutex_destroy(&s->thread_data[i].lock);
      58             : #endif
      59         296 :             av_freep(&s->thread_data[i].filter_strength);
      60             :         }
      61          73 :     av_freep(&s->thread_data);
      62          73 :     av_freep(&s->macroblocks_base);
      63          73 :     av_freep(&s->intra4x4_pred_mode_top);
      64          73 :     av_freep(&s->top_nnz);
      65          73 :     av_freep(&s->top_border);
      66             : 
      67          73 :     s->macroblocks = NULL;
      68          73 : }
      69             : 
      70        1142 : static int vp8_alloc_frame(VP8Context *s, VP8Frame *f, int ref)
      71             : {
      72             :     int ret;
      73        1142 :     if ((ret = ff_thread_get_buffer(s->avctx, &f->tf,
      74             :                                     ref ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
      75           0 :         return ret;
      76        1142 :     if (!(f->seg_map = av_buffer_allocz(s->mb_width * s->mb_height)))
      77           0 :         goto fail;
      78        1142 :     if (s->avctx->hwaccel) {
      79           0 :         const AVHWAccel *hwaccel = s->avctx->hwaccel;
      80           0 :         if (hwaccel->frame_priv_data_size) {
      81           0 :             f->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
      82           0 :             if (!f->hwaccel_priv_buf)
      83           0 :                 goto fail;
      84           0 :             f->hwaccel_picture_private = f->hwaccel_priv_buf->data;
      85             :         }
      86             :     }
      87        1142 :     return 0;
      88             : 
      89           0 : fail:
      90           0 :     av_buffer_unref(&f->seg_map);
      91           0 :     ff_thread_release_buffer(s->avctx, &f->tf);
      92           0 :     return AVERROR(ENOMEM);
      93             : }
      94             : 
      95        1398 : static void vp8_release_frame(VP8Context *s, VP8Frame *f)
      96             : {
      97        1398 :     av_buffer_unref(&f->seg_map);
      98        1398 :     av_buffer_unref(&f->hwaccel_priv_buf);
      99        1398 :     f->hwaccel_picture_private = NULL;
     100        1398 :     ff_thread_release_buffer(s->avctx, &f->tf);
     101        1398 : }
     102             : 
     103             : #if CONFIG_VP8_DECODER
     104           0 : static int vp8_ref_frame(VP8Context *s, VP8Frame *dst, VP8Frame *src)
     105             : {
     106             :     int ret;
     107             : 
     108           0 :     vp8_release_frame(s, dst);
     109             : 
     110           0 :     if ((ret = ff_thread_ref_frame(&dst->tf, &src->tf)) < 0)
     111           0 :         return ret;
     112           0 :     if (src->seg_map &&
     113           0 :         !(dst->seg_map = av_buffer_ref(src->seg_map))) {
     114           0 :         vp8_release_frame(s, dst);
     115           0 :         return AVERROR(ENOMEM);
     116             :     }
     117           0 :     if (src->hwaccel_picture_private) {
     118           0 :         dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
     119           0 :         if (!dst->hwaccel_priv_buf)
     120           0 :             return AVERROR(ENOMEM);
     121           0 :         dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
     122             :     }
     123             : 
     124           0 :     return 0;
     125             : }
     126             : #endif /* CONFIG_VP8_DECODER */
     127             : 
     128          73 : static void vp8_decode_flush_impl(AVCodecContext *avctx, int free_mem)
     129             : {
     130          73 :     VP8Context *s = avctx->priv_data;
     131             :     int i;
     132             : 
     133         438 :     for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
     134         365 :         vp8_release_frame(s, &s->frames[i]);
     135          73 :     memset(s->framep, 0, sizeof(s->framep));
     136             : 
     137          73 :     if (free_mem)
     138          73 :         free_buffers(s);
     139          73 : }
     140             : 
     141           0 : static void vp8_decode_flush(AVCodecContext *avctx)
     142             : {
     143           0 :     vp8_decode_flush_impl(avctx, 0);
     144           0 : }
     145             : 
     146        1142 : static VP8Frame *vp8_find_free_buffer(VP8Context *s)
     147             : {
     148        1142 :     VP8Frame *frame = NULL;
     149             :     int i;
     150             : 
     151             :     // find a free buffer
     152        2793 :     for (i = 0; i < 5; i++)
     153        4996 :         if (&s->frames[i] != s->framep[VP56_FRAME_CURRENT]  &&
     154        4405 :             &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
     155        3802 :             &s->frames[i] != s->framep[VP56_FRAME_GOLDEN]   &&
     156        1600 :             &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2]) {
     157        1142 :             frame = &s->frames[i];
     158        1142 :             break;
     159             :         }
     160        1142 :     if (i == 5) {
     161           0 :         av_log(s->avctx, AV_LOG_FATAL, "Ran out of free frames!\n");
     162           0 :         abort();
     163             :     }
     164        1142 :     if (frame->tf.f->buf[0])
     165           0 :         vp8_release_frame(s, frame);
     166             : 
     167        1142 :     return frame;
     168             : }
     169             : 
     170          30 : static enum AVPixelFormat get_pixel_format(VP8Context *s)
     171             : {
     172          30 :     enum AVPixelFormat pix_fmts[] = {
     173             : #if CONFIG_VP8_VAAPI_HWACCEL
     174             :         AV_PIX_FMT_VAAPI,
     175             : #endif
     176             : #if CONFIG_VP8_NVDEC_HWACCEL
     177             :         AV_PIX_FMT_CUDA,
     178             : #endif
     179             :         AV_PIX_FMT_YUV420P,
     180             :         AV_PIX_FMT_NONE,
     181             :     };
     182             : 
     183          30 :     return ff_get_format(s->avctx, pix_fmts);
     184             : }
     185             : 
     186             : static av_always_inline
     187          37 : int update_dimensions(VP8Context *s, int width, int height, int is_vp7)
     188             : {
     189          37 :     AVCodecContext *avctx = s->avctx;
     190             :     int i, ret;
     191             : 
     192          61 :     if (width  != s->avctx->width || ((width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height) && s->macroblocks_base ||
     193          24 :         height != s->avctx->height) {
     194          13 :         vp8_decode_flush_impl(s->avctx, 1);
     195             : 
     196          13 :         ret = ff_set_dimensions(s->avctx, width, height);
     197          13 :         if (ret < 0)
     198           0 :             return ret;
     199             :     }
     200             : 
     201          37 :     if (!s->actually_webp && !is_vp7) {
     202          30 :         s->pix_fmt = get_pixel_format(s);
     203          30 :         if (s->pix_fmt < 0)
     204           0 :             return AVERROR(EINVAL);
     205          30 :         avctx->pix_fmt = s->pix_fmt;
     206             :     }
     207             : 
     208          37 :     s->mb_width  = (s->avctx->coded_width  + 15) / 16;
     209          37 :     s->mb_height = (s->avctx->coded_height + 15) / 16;
     210             : 
     211          37 :     s->mb_layout = is_vp7 || avctx->active_thread_type == FF_THREAD_SLICE &&
     212           0 :                    avctx->thread_count > 1;
     213          37 :     if (!s->mb_layout) { // Frame threading and one thread
     214          36 :         s->macroblocks_base       = av_mallocz((s->mb_width + s->mb_height * 2 + 1) *
     215             :                                                sizeof(*s->macroblocks));
     216          36 :         s->intra4x4_pred_mode_top = av_mallocz(s->mb_width * 4);
     217             :     } else // Sliced threading
     218           1 :         s->macroblocks_base = av_mallocz((s->mb_width + 2) * (s->mb_height + 2) *
     219             :                                          sizeof(*s->macroblocks));
     220          37 :     s->top_nnz     = av_mallocz(s->mb_width * sizeof(*s->top_nnz));
     221          37 :     s->top_border  = av_mallocz((s->mb_width + 1) * sizeof(*s->top_border));
     222          37 :     s->thread_data = av_mallocz(MAX_THREADS * sizeof(VP8ThreadData));
     223             : 
     224          74 :     if (!s->macroblocks_base || !s->top_nnz || !s->top_border ||
     225          74 :         !s->thread_data || (!s->intra4x4_pred_mode_top && !s->mb_layout)) {
     226           0 :         free_buffers(s);
     227           0 :         return AVERROR(ENOMEM);
     228             :     }
     229             : 
     230         333 :     for (i = 0; i < MAX_THREADS; i++) {
     231         592 :         s->thread_data[i].filter_strength =
     232         592 :             av_mallocz(s->mb_width * sizeof(*s->thread_data[0].filter_strength));
     233         296 :         if (!s->thread_data[i].filter_strength) {
     234           0 :             free_buffers(s);
     235           0 :             return AVERROR(ENOMEM);
     236             :         }
     237             : #if HAVE_THREADS
     238         296 :         pthread_mutex_init(&s->thread_data[i].lock, NULL);
     239         296 :         pthread_cond_init(&s->thread_data[i].cond, NULL);
     240             : #endif
     241             :     }
     242             : 
     243          37 :     s->macroblocks = s->macroblocks_base + 1;
     244             : 
     245          37 :     return 0;
     246             : }
     247             : 
     248           1 : static int vp7_update_dimensions(VP8Context *s, int width, int height)
     249             : {
     250           1 :     return update_dimensions(s, width, height, IS_VP7);
     251             : }
     252             : 
     253          36 : static int vp8_update_dimensions(VP8Context *s, int width, int height)
     254             : {
     255          36 :     return update_dimensions(s, width, height, IS_VP8);
     256             : }
     257             : 
     258             : 
     259         433 : static void parse_segment_info(VP8Context *s)
     260             : {
     261         433 :     VP56RangeCoder *c = &s->c;
     262             :     int i;
     263             : 
     264         433 :     s->segmentation.update_map = vp8_rac_get(c);
     265         433 :     s->segmentation.update_feature_data = vp8_rac_get(c);
     266             : 
     267         433 :     if (s->segmentation.update_feature_data) {
     268          68 :         s->segmentation.absolute_vals = vp8_rac_get(c);
     269             : 
     270         340 :         for (i = 0; i < 4; i++)
     271         272 :             s->segmentation.base_quant[i]   = vp8_rac_get_sint(c, 7);
     272             : 
     273         340 :         for (i = 0; i < 4; i++)
     274         272 :             s->segmentation.filter_level[i] = vp8_rac_get_sint(c, 6);
     275             :     }
     276         433 :     if (s->segmentation.update_map)
     277         272 :         for (i = 0; i < 3; i++)
     278         204 :             s->prob->segmentid[i] = vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255;
     279         433 : }
     280             : 
     281          42 : static void update_lf_deltas(VP8Context *s)
     282             : {
     283          42 :     VP56RangeCoder *c = &s->c;
     284             :     int i;
     285             : 
     286         210 :     for (i = 0; i < 4; i++) {
     287         168 :         if (vp8_rac_get(c)) {
     288         126 :             s->lf_delta.ref[i] = vp8_rac_get_uint(c, 6);
     289             : 
     290         126 :             if (vp8_rac_get(c))
     291          84 :                 s->lf_delta.ref[i] = -s->lf_delta.ref[i];
     292             :         }
     293             :     }
     294             : 
     295         210 :     for (i = MODE_I4x4; i <= VP8_MVMODE_SPLIT; i++) {
     296         168 :         if (vp8_rac_get(c)) {
     297         168 :             s->lf_delta.mode[i] = vp8_rac_get_uint(c, 6);
     298             : 
     299         168 :             if (vp8_rac_get(c))
     300          42 :                 s->lf_delta.mode[i] = -s->lf_delta.mode[i];
     301             :         }
     302             :     }
     303          42 : }
     304             : 
     305        1112 : static int setup_partitions(VP8Context *s, const uint8_t *buf, int buf_size)
     306             : {
     307        1112 :     const uint8_t *sizes = buf;
     308             :     int i;
     309             :     int ret;
     310             : 
     311        1112 :     s->num_coeff_partitions = 1 << vp8_rac_get_uint(&s->c, 2);
     312             : 
     313        1112 :     buf      += 3 * (s->num_coeff_partitions - 1);
     314        1112 :     buf_size -= 3 * (s->num_coeff_partitions - 1);
     315        1112 :     if (buf_size < 0)
     316           0 :         return -1;
     317             : 
     318        1451 :     for (i = 0; i < s->num_coeff_partitions - 1; i++) {
     319         339 :         int size = AV_RL24(sizes + 3 * i);
     320         339 :         if (buf_size - size < 0)
     321           0 :             return -1;
     322         339 :         s->coeff_partition_size[i] = size;
     323             : 
     324         339 :         ret = ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, size);
     325         339 :         if (ret < 0)
     326           0 :             return ret;
     327         339 :         buf      += size;
     328         339 :         buf_size -= size;
     329             :     }
     330             : 
     331        1112 :     s->coeff_partition_size[i] = buf_size;
     332        1112 :     ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size);
     333             : 
     334        1112 :     return 0;
     335             : }
     336             : 
     337          30 : static void vp7_get_quants(VP8Context *s)
     338             : {
     339          30 :     VP56RangeCoder *c = &s->c;
     340             : 
     341          30 :     int yac_qi  = vp8_rac_get_uint(c, 7);
     342          30 :     int ydc_qi  = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
     343          30 :     int y2dc_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
     344          30 :     int y2ac_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
     345          30 :     int uvdc_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
     346          30 :     int uvac_qi = vp8_rac_get(c) ? vp8_rac_get_uint(c, 7) : yac_qi;
     347             : 
     348          30 :     s->qmat[0].luma_qmul[0]    =       vp7_ydc_qlookup[ydc_qi];
     349          30 :     s->qmat[0].luma_qmul[1]    =       vp7_yac_qlookup[yac_qi];
     350          30 :     s->qmat[0].luma_dc_qmul[0] =       vp7_y2dc_qlookup[y2dc_qi];
     351          30 :     s->qmat[0].luma_dc_qmul[1] =       vp7_y2ac_qlookup[y2ac_qi];
     352          30 :     s->qmat[0].chroma_qmul[0]  = FFMIN(vp7_ydc_qlookup[uvdc_qi], 132);
     353          30 :     s->qmat[0].chroma_qmul[1]  =       vp7_yac_qlookup[uvac_qi];
     354          30 : }
     355             : 
     356        1112 : static void vp8_get_quants(VP8Context *s)
     357             : {
     358        1112 :     VP56RangeCoder *c = &s->c;
     359             :     int i, base_qi;
     360             : 
     361        1112 :     s->quant.yac_qi     = vp8_rac_get_uint(c, 7);
     362        1112 :     s->quant.ydc_delta  = vp8_rac_get_sint(c, 4);
     363        1112 :     s->quant.y2dc_delta = vp8_rac_get_sint(c, 4);
     364        1112 :     s->quant.y2ac_delta = vp8_rac_get_sint(c, 4);
     365        1112 :     s->quant.uvdc_delta = vp8_rac_get_sint(c, 4);
     366        1112 :     s->quant.uvac_delta = vp8_rac_get_sint(c, 4);
     367             : 
     368        5560 :     for (i = 0; i < 4; i++) {
     369        4448 :         if (s->segmentation.enabled) {
     370        1732 :             base_qi = s->segmentation.base_quant[i];
     371        1732 :             if (!s->segmentation.absolute_vals)
     372        1608 :                 base_qi += s->quant.yac_qi;
     373             :         } else
     374        2716 :             base_qi = s->quant.yac_qi;
     375             : 
     376        4448 :         s->qmat[i].luma_qmul[0]    = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.ydc_delta,  7)];
     377        4448 :         s->qmat[i].luma_qmul[1]    = vp8_ac_qlookup[av_clip_uintp2(base_qi,              7)];
     378        4448 :         s->qmat[i].luma_dc_qmul[0] = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.y2dc_delta, 7)] * 2;
     379             :         /* 101581>>16 is equivalent to 155/100 */
     380        4448 :         s->qmat[i].luma_dc_qmul[1] = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.y2ac_delta, 7)] * 101581 >> 16;
     381        4448 :         s->qmat[i].chroma_qmul[0]  = vp8_dc_qlookup[av_clip_uintp2(base_qi + s->quant.uvdc_delta, 7)];
     382        4448 :         s->qmat[i].chroma_qmul[1]  = vp8_ac_qlookup[av_clip_uintp2(base_qi + s->quant.uvac_delta, 7)];
     383             : 
     384        4448 :         s->qmat[i].luma_dc_qmul[1] = FFMAX(s->qmat[i].luma_dc_qmul[1], 8);
     385        4448 :         s->qmat[i].chroma_qmul[0]  = FFMIN(s->qmat[i].chroma_qmul[0], 132);
     386             :     }
     387        1112 : }
     388             : 
     389             : /**
     390             :  * Determine which buffers golden and altref should be updated with after this frame.
     391             :  * The spec isn't clear here, so I'm going by my understanding of what libvpx does
     392             :  *
     393             :  * Intra frames update all 3 references
     394             :  * Inter frames update VP56_FRAME_PREVIOUS if the update_last flag is set
     395             :  * If the update (golden|altref) flag is set, it's updated with the current frame
     396             :  *      if update_last is set, and VP56_FRAME_PREVIOUS otherwise.
     397             :  * If the flag is not set, the number read means:
     398             :  *      0: no update
     399             :  *      1: VP56_FRAME_PREVIOUS
     400             :  *      2: update golden with altref, or update altref with golden
     401             :  */
     402        2128 : static VP56Frame ref_to_update(VP8Context *s, int update, VP56Frame ref)
     403             : {
     404        2128 :     VP56RangeCoder *c = &s->c;
     405             : 
     406        2128 :     if (update)
     407          90 :         return VP56_FRAME_CURRENT;
     408             : 
     409        2038 :     switch (vp8_rac_get_uint(c, 2)) {
     410          31 :     case 1:
     411          31 :         return VP56_FRAME_PREVIOUS;
     412          84 :     case 2:
     413          84 :         return (ref == VP56_FRAME_GOLDEN) ? VP56_FRAME_GOLDEN2 : VP56_FRAME_GOLDEN;
     414             :     }
     415        1923 :     return VP56_FRAME_NONE;
     416             : }
     417             : 
     418          49 : static void vp78_reset_probability_tables(VP8Context *s)
     419             : {
     420             :     int i, j;
     421         245 :     for (i = 0; i < 4; i++)
     422        3332 :         for (j = 0; j < 16; j++)
     423        3136 :             memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
     424             :                    sizeof(s->prob->token[i][j]));
     425          49 : }
     426             : 
     427        1142 : static void vp78_update_probability_tables(VP8Context *s)
     428             : {
     429        1142 :     VP56RangeCoder *c = &s->c;
     430             :     int i, j, k, l, m;
     431             : 
     432        5710 :     for (i = 0; i < 4; i++)
     433       41112 :         for (j = 0; j < 8; j++)
     434      146176 :             for (k = 0; k < 3; k++)
     435     1315584 :                 for (l = 0; l < NUM_DCT_TOKENS-1; l++)
     436     1205952 :                     if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) {
     437        7203 :                         int prob = vp8_rac_get_uint(c, 8);
     438       23838 :                         for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
     439       16635 :                             s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
     440             :                     }
     441        1142 : }
     442             : 
     443             : #define VP7_MVC_SIZE 17
     444             : #define VP8_MVC_SIZE 19
     445             : 
     446        1093 : static void vp78_update_pred16x16_pred8x8_mvc_probabilities(VP8Context *s,
     447             :                                                             int mvc_size)
     448             : {
     449        1093 :     VP56RangeCoder *c = &s->c;
     450             :     int i, j;
     451             : 
     452        1093 :     if (vp8_rac_get(c))
     453          10 :         for (i = 0; i < 4; i++)
     454           8 :             s->prob->pred16x16[i] = vp8_rac_get_uint(c, 8);
     455        1093 :     if (vp8_rac_get(c))
     456           4 :         for (i = 0; i < 3; i++)
     457           3 :             s->prob->pred8x8c[i]  = vp8_rac_get_uint(c, 8);
     458             : 
     459             :     // 17.2 MV probability update
     460        3279 :     for (i = 0; i < 2; i++)
     461       43604 :         for (j = 0; j < mvc_size; j++)
     462       41418 :             if (vp56_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j]))
     463         306 :                 s->prob->mvc[i][j] = vp8_rac_get_nn(c);
     464        1093 : }
     465             : 
     466        1064 : static void update_refs(VP8Context *s)
     467             : {
     468        1064 :     VP56RangeCoder *c = &s->c;
     469             : 
     470        1064 :     int update_golden = vp8_rac_get(c);
     471        1064 :     int update_altref = vp8_rac_get(c);
     472             : 
     473        1064 :     s->update_golden = ref_to_update(s, update_golden, VP56_FRAME_GOLDEN);
     474        1064 :     s->update_altref = ref_to_update(s, update_altref, VP56_FRAME_GOLDEN2);
     475        1064 : }
     476             : 
     477           0 : static void copy_chroma(AVFrame *dst, AVFrame *src, int width, int height)
     478             : {
     479             :     int i, j;
     480             : 
     481           0 :     for (j = 1; j < 3; j++) {
     482           0 :         for (i = 0; i < height / 2; i++)
     483           0 :             memcpy(dst->data[j] + i * dst->linesize[j],
     484           0 :                    src->data[j] + i * src->linesize[j], width / 2);
     485             :     }
     486           0 : }
     487             : 
     488           0 : static void fade(uint8_t *dst, ptrdiff_t dst_linesize,
     489             :                  const uint8_t *src, ptrdiff_t src_linesize,
     490             :                  int width, int height,
     491             :                  int alpha, int beta)
     492             : {
     493             :     int i, j;
     494           0 :     for (j = 0; j < height; j++) {
     495           0 :         for (i = 0; i < width; i++) {
     496           0 :             uint8_t y = src[j * src_linesize + i];
     497           0 :             dst[j * dst_linesize + i] = av_clip_uint8(y + ((y * beta) >> 8) + alpha);
     498             :         }
     499             :     }
     500           0 : }
     501             : 
     502           0 : static int vp7_fade_frame(VP8Context *s, VP56RangeCoder *c)
     503             : {
     504           0 :     int alpha = (int8_t) vp8_rac_get_uint(c, 8);
     505           0 :     int beta  = (int8_t) vp8_rac_get_uint(c, 8);
     506             :     int ret;
     507             : 
     508           0 :     if (!s->keyframe && (alpha || beta)) {
     509           0 :         int width  = s->mb_width * 16;
     510           0 :         int height = s->mb_height * 16;
     511             :         AVFrame *src, *dst;
     512             : 
     513           0 :         if (!s->framep[VP56_FRAME_PREVIOUS] ||
     514           0 :             !s->framep[VP56_FRAME_GOLDEN]) {
     515           0 :             av_log(s->avctx, AV_LOG_WARNING, "Discarding interframe without a prior keyframe!\n");
     516           0 :             return AVERROR_INVALIDDATA;
     517             :         }
     518             : 
     519           0 :         dst =
     520           0 :         src = s->framep[VP56_FRAME_PREVIOUS]->tf.f;
     521             : 
     522             :         /* preserve the golden frame, write a new previous frame */
     523           0 :         if (s->framep[VP56_FRAME_GOLDEN] == s->framep[VP56_FRAME_PREVIOUS]) {
     524           0 :             s->framep[VP56_FRAME_PREVIOUS] = vp8_find_free_buffer(s);
     525           0 :             if ((ret = vp8_alloc_frame(s, s->framep[VP56_FRAME_PREVIOUS], 1)) < 0)
     526           0 :                 return ret;
     527             : 
     528           0 :             dst = s->framep[VP56_FRAME_PREVIOUS]->tf.f;
     529             : 
     530           0 :             copy_chroma(dst, src, width, height);
     531             :         }
     532             : 
     533           0 :         fade(dst->data[0], dst->linesize[0],
     534           0 :              src->data[0], src->linesize[0],
     535             :              width, height, alpha, beta);
     536             :     }
     537             : 
     538           0 :     return 0;
     539             : }
     540             : 
     541          30 : static int vp7_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
     542             : {
     543          30 :     VP56RangeCoder *c = &s->c;
     544             :     int part1_size, hscale, vscale, i, j, ret;
     545          30 :     int width  = s->avctx->width;
     546          30 :     int height = s->avctx->height;
     547             : 
     548          30 :     if (buf_size < 4) {
     549           0 :         return AVERROR_INVALIDDATA;
     550             :     }
     551             : 
     552          30 :     s->profile = (buf[0] >> 1) & 7;
     553          30 :     if (s->profile > 1) {
     554           0 :         avpriv_request_sample(s->avctx, "Unknown profile %d", s->profile);
     555           0 :         return AVERROR_INVALIDDATA;
     556             :     }
     557             : 
     558          30 :     s->keyframe  = !(buf[0] & 1);
     559          30 :     s->invisible = 0;
     560          30 :     part1_size   = AV_RL24(buf) >> 4;
     561             : 
     562          30 :     if (buf_size < 4 - s->profile + part1_size) {
     563           0 :         av_log(s->avctx, AV_LOG_ERROR, "Buffer size %d is too small, needed : %d\n", buf_size, 4 - s->profile + part1_size);
     564           0 :         return AVERROR_INVALIDDATA;
     565             :     }
     566             : 
     567          30 :     buf      += 4 - s->profile;
     568          30 :     buf_size -= 4 - s->profile;
     569             : 
     570          30 :     memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
     571             : 
     572          30 :     ret = ff_vp56_init_range_decoder(c, buf, part1_size);
     573          30 :     if (ret < 0)
     574           0 :         return ret;
     575          30 :     buf      += part1_size;
     576          30 :     buf_size -= part1_size;
     577             : 
     578             :     /* A. Dimension information (keyframes only) */
     579          30 :     if (s->keyframe) {
     580           1 :         width  = vp8_rac_get_uint(c, 12);
     581           1 :         height = vp8_rac_get_uint(c, 12);
     582           1 :         hscale = vp8_rac_get_uint(c, 2);
     583           1 :         vscale = vp8_rac_get_uint(c, 2);
     584           1 :         if (hscale || vscale)
     585           0 :             avpriv_request_sample(s->avctx, "Upscaling");
     586             : 
     587           1 :         s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
     588           1 :         vp78_reset_probability_tables(s);
     589           1 :         memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
     590             :                sizeof(s->prob->pred16x16));
     591           1 :         memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
     592             :                sizeof(s->prob->pred8x8c));
     593           3 :         for (i = 0; i < 2; i++)
     594           2 :             memcpy(s->prob->mvc[i], vp7_mv_default_prob[i],
     595             :                    sizeof(vp7_mv_default_prob[i]));
     596           1 :         memset(&s->segmentation, 0, sizeof(s->segmentation));
     597           1 :         memset(&s->lf_delta, 0, sizeof(s->lf_delta));
     598           1 :         memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
     599             :     }
     600             : 
     601          30 :     if (s->keyframe || s->profile > 0)
     602           1 :         memset(s->inter_dc_pred, 0 , sizeof(s->inter_dc_pred));
     603             : 
     604             :     /* B. Decoding information for all four macroblock-level features */
     605         150 :     for (i = 0; i < 4; i++) {
     606         120 :         s->feature_enabled[i] = vp8_rac_get(c);
     607         120 :         if (s->feature_enabled[i]) {
     608           0 :              s->feature_present_prob[i] = vp8_rac_get_uint(c, 8);
     609             : 
     610           0 :              for (j = 0; j < 3; j++)
     611           0 :                  s->feature_index_prob[i][j] =
     612           0 :                      vp8_rac_get(c) ? vp8_rac_get_uint(c, 8) : 255;
     613             : 
     614           0 :              if (vp7_feature_value_size[s->profile][i])
     615           0 :                  for (j = 0; j < 4; j++)
     616           0 :                      s->feature_value[i][j] =
     617           0 :                         vp8_rac_get(c) ? vp8_rac_get_uint(c, vp7_feature_value_size[s->profile][i]) : 0;
     618             :         }
     619             :     }
     620             : 
     621          30 :     s->segmentation.enabled    = 0;
     622          30 :     s->segmentation.update_map = 0;
     623          30 :     s->lf_delta.enabled        = 0;
     624             : 
     625          30 :     s->num_coeff_partitions = 1;
     626          30 :     ret = ff_vp56_init_range_decoder(&s->coeff_partition[0], buf, buf_size);
     627          30 :     if (ret < 0)
     628           0 :         return ret;
     629             : 
     630          59 :     if (!s->macroblocks_base || /* first frame */
     631          87 :         width != s->avctx->width || height != s->avctx->height ||
     632          58 :         (width + 15) / 16 != s->mb_width || (height + 15) / 16 != s->mb_height) {
     633           1 :         if ((ret = vp7_update_dimensions(s, width, height)) < 0)
     634           0 :             return ret;
     635             :     }
     636             : 
     637             :     /* C. Dequantization indices */
     638          30 :     vp7_get_quants(s);
     639             : 
     640             :     /* D. Golden frame update flag (a Flag) for interframes only */
     641          30 :     if (!s->keyframe) {
     642          29 :         s->update_golden = vp8_rac_get(c) ? VP56_FRAME_CURRENT : VP56_FRAME_NONE;
     643          29 :         s->sign_bias[VP56_FRAME_GOLDEN] = 0;
     644             :     }
     645             : 
     646          30 :     s->update_last          = 1;
     647          30 :     s->update_probabilities = 1;
     648          30 :     s->fade_present         = 1;
     649             : 
     650          30 :     if (s->profile > 0) {
     651           0 :         s->update_probabilities = vp8_rac_get(c);
     652           0 :         if (!s->update_probabilities)
     653           0 :             s->prob[1] = s->prob[0];
     654             : 
     655           0 :         if (!s->keyframe)
     656           0 :             s->fade_present = vp8_rac_get(c);
     657             :     }
     658             : 
     659             :     /* E. Fading information for previous frame */
     660          30 :     if (s->fade_present && vp8_rac_get(c)) {
     661           0 :         if ((ret = vp7_fade_frame(s ,c)) < 0)
     662           0 :             return ret;
     663             :     }
     664             : 
     665             :     /* F. Loop filter type */
     666          30 :     if (!s->profile)
     667          30 :         s->filter.simple = vp8_rac_get(c);
     668             : 
     669             :     /* G. DCT coefficient ordering specification */
     670          30 :     if (vp8_rac_get(c))
     671         208 :         for (i = 1; i < 16; i++)
     672         195 :             s->prob[0].scan[i] = ff_zigzag_scan[vp8_rac_get_uint(c, 4)];
     673             : 
     674             :     /* H. Loop filter levels  */
     675          30 :     if (s->profile > 0)
     676           0 :         s->filter.simple = vp8_rac_get(c);
     677          30 :     s->filter.level     = vp8_rac_get_uint(c, 6);
     678          30 :     s->filter.sharpness = vp8_rac_get_uint(c, 3);
     679             : 
     680             :     /* I. DCT coefficient probability update; 13.3 Token Probability Updates */
     681          30 :     vp78_update_probability_tables(s);
     682             : 
     683          30 :     s->mbskip_enabled = 0;
     684             : 
     685             :     /* J. The remaining frame header data occurs ONLY FOR INTERFRAMES */
     686          30 :     if (!s->keyframe) {
     687          29 :         s->prob->intra  = vp8_rac_get_uint(c, 8);
     688          29 :         s->prob->last   = vp8_rac_get_uint(c, 8);
     689          29 :         vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP7_MVC_SIZE);
     690             :     }
     691             : 
     692          30 :     return 0;
     693             : }
     694             : 
     695        1112 : static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
     696             : {
     697        1112 :     VP56RangeCoder *c = &s->c;
     698             :     int header_size, hscale, vscale, ret;
     699        1112 :     int width  = s->avctx->width;
     700        1112 :     int height = s->avctx->height;
     701             : 
     702        1112 :     if (buf_size < 3) {
     703           0 :         av_log(s->avctx, AV_LOG_ERROR, "Insufficent data (%d) for header\n", buf_size);
     704           0 :         return AVERROR_INVALIDDATA;
     705             :     }
     706             : 
     707        1112 :     s->keyframe  = !(buf[0] & 1);
     708        1112 :     s->profile   =  (buf[0]>>1) & 7;
     709        1112 :     s->invisible = !(buf[0] & 0x10);
     710        1112 :     header_size  = AV_RL24(buf) >> 5;
     711        1112 :     buf      += 3;
     712        1112 :     buf_size -= 3;
     713             : 
     714        1112 :     s->header_partition_size = header_size;
     715             : 
     716        1112 :     if (s->profile > 3)
     717           0 :         av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);
     718             : 
     719        1112 :     if (!s->profile)
     720         956 :         memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab,
     721             :                sizeof(s->put_pixels_tab));
     722             :     else    // profile 1-3 use bilinear, 4+ aren't defined so whatever
     723         156 :         memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab,
     724             :                sizeof(s->put_pixels_tab));
     725             : 
     726        1112 :     if (header_size > buf_size - 7 * s->keyframe) {
     727           0 :         av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
     728           0 :         return AVERROR_INVALIDDATA;
     729             :     }
     730             : 
     731        1112 :     if (s->keyframe) {
     732          48 :         if (AV_RL24(buf) != 0x2a019d) {
     733           0 :             av_log(s->avctx, AV_LOG_ERROR,
     734           0 :                    "Invalid start code 0x%x\n", AV_RL24(buf));
     735           0 :             return AVERROR_INVALIDDATA;
     736             :         }
     737          48 :         width     = AV_RL16(buf + 3) & 0x3fff;
     738          48 :         height    = AV_RL16(buf + 5) & 0x3fff;
     739          48 :         hscale    = buf[4] >> 6;
     740          48 :         vscale    = buf[6] >> 6;
     741          48 :         buf      += 7;
     742          48 :         buf_size -= 7;
     743             : 
     744          48 :         if (hscale || vscale)
     745           0 :             avpriv_request_sample(s->avctx, "Upscaling");
     746             : 
     747          48 :         s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
     748          48 :         vp78_reset_probability_tables(s);
     749          48 :         memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter,
     750             :                sizeof(s->prob->pred16x16));
     751          48 :         memcpy(s->prob->pred8x8c, vp8_pred8x8c_prob_inter,
     752             :                sizeof(s->prob->pred8x8c));
     753          48 :         memcpy(s->prob->mvc, vp8_mv_default_prob,
     754             :                sizeof(s->prob->mvc));
     755          48 :         memset(&s->segmentation, 0, sizeof(s->segmentation));
     756          48 :         memset(&s->lf_delta, 0, sizeof(s->lf_delta));
     757             :     }
     758             : 
     759        1112 :     ret = ff_vp56_init_range_decoder(c, buf, header_size);
     760        1112 :     if (ret < 0)
     761           0 :         return ret;
     762        1112 :     buf      += header_size;
     763        1112 :     buf_size -= header_size;
     764             : 
     765        1112 :     if (s->keyframe) {
     766          48 :         s->colorspace = vp8_rac_get(c);
     767          48 :         if (s->colorspace)
     768           0 :             av_log(s->avctx, AV_LOG_WARNING, "Unspecified colorspace\n");
     769          48 :         s->fullrange = vp8_rac_get(c);
     770             :     }
     771             : 
     772        1112 :     if ((s->segmentation.enabled = vp8_rac_get(c)))
     773         433 :         parse_segment_info(s);
     774             :     else
     775         679 :         s->segmentation.update_map = 0; // FIXME: move this to some init function?
     776             : 
     777        1112 :     s->filter.simple    = vp8_rac_get(c);
     778        1112 :     s->filter.level     = vp8_rac_get_uint(c, 6);
     779        1112 :     s->filter.sharpness = vp8_rac_get_uint(c, 3);
     780             : 
     781        1112 :     if ((s->lf_delta.enabled = vp8_rac_get(c))) {
     782        1106 :         s->lf_delta.update = vp8_rac_get(c);
     783        1106 :         if (s->lf_delta.update)
     784          42 :             update_lf_deltas(s);
     785             :     }
     786             : 
     787        1112 :     if (setup_partitions(s, buf, buf_size)) {
     788           0 :         av_log(s->avctx, AV_LOG_ERROR, "Invalid partitions\n");
     789           0 :         return AVERROR_INVALIDDATA;
     790             :     }
     791             : 
     792        2197 :     if (!s->macroblocks_base || /* first frame */
     793        3241 :         width != s->avctx->width || height != s->avctx->height ||
     794        2152 :         (width+15)/16 != s->mb_width || (height+15)/16 != s->mb_height)
     795          36 :         if ((ret = vp8_update_dimensions(s, width, height)) < 0)
     796           0 :             return ret;
     797             : 
     798        1112 :     vp8_get_quants(s);
     799             : 
     800        1112 :     if (!s->keyframe) {
     801        1064 :         update_refs(s);
     802        1064 :         s->sign_bias[VP56_FRAME_GOLDEN]               = vp8_rac_get(c);
     803        1064 :         s->sign_bias[VP56_FRAME_GOLDEN2 /* altref */] = vp8_rac_get(c);
     804             :     }
     805             : 
     806             :     // if we aren't saving this frame's probabilities for future frames,
     807             :     // make a copy of the current probabilities
     808        1112 :     if (!(s->update_probabilities = vp8_rac_get(c)))
     809          64 :         s->prob[1] = s->prob[0];
     810             : 
     811        1112 :     s->update_last = s->keyframe || vp8_rac_get(c);
     812             : 
     813        1112 :     vp78_update_probability_tables(s);
     814             : 
     815        1112 :     if ((s->mbskip_enabled = vp8_rac_get(c)))
     816        1106 :         s->prob->mbskip = vp8_rac_get_uint(c, 8);
     817             : 
     818        1112 :     if (!s->keyframe) {
     819        1064 :         s->prob->intra  = vp8_rac_get_uint(c, 8);
     820        1064 :         s->prob->last   = vp8_rac_get_uint(c, 8);
     821        1064 :         s->prob->golden = vp8_rac_get_uint(c, 8);
     822        1064 :         vp78_update_pred16x16_pred8x8_mvc_probabilities(s, VP8_MVC_SIZE);
     823             :     }
     824             : 
     825             :     // Record the entropy coder state here so that hwaccels can use it.
     826        1112 :     s->c.code_word = vp56_rac_renorm(&s->c);
     827        1112 :     s->coder_state_at_header_end.input     = s->c.buffer - (-s->c.bits / 8);
     828        1112 :     s->coder_state_at_header_end.range     = s->c.high;
     829        1112 :     s->coder_state_at_header_end.value     = s->c.code_word >> 16;
     830        1112 :     s->coder_state_at_header_end.bit_count = -s->c.bits % 8;
     831             : 
     832        1112 :     return 0;
     833             : }
     834             : 
     835             : static av_always_inline
     836       57174 : void clamp_mv(VP8mvbounds *s, VP56mv *dst, const VP56mv *src)
     837             : {
     838       57174 :     dst->x = av_clip(src->x, av_clip(s->mv_min.x, INT16_MIN, INT16_MAX),
     839             :                              av_clip(s->mv_max.x, INT16_MIN, INT16_MAX));
     840       57174 :     dst->y = av_clip(src->y, av_clip(s->mv_min.y, INT16_MIN, INT16_MAX),
     841             :                              av_clip(s->mv_max.y, INT16_MIN, INT16_MAX));
     842       57174 : }
     843             : 
     844             : /**
     845             :  * Motion vector coding, 17.1.
     846             :  */
     847       84886 : static av_always_inline int read_mv_component(VP56RangeCoder *c, const uint8_t *p, int vp7)
     848             : {
     849       84886 :     int bit, x = 0;
     850             : 
     851       84886 :     if (vp56_rac_get_prob_branchy(c, p[0])) {
     852             :         int i;
     853             : 
     854       79804 :         for (i = 0; i < 3; i++)
     855       59853 :             x += vp56_rac_get_prob(c, p[9 + i]) << i;
     856      139657 :         for (i = (vp7 ? 7 : 9); i > 3; i--)
     857      119706 :             x += vp56_rac_get_prob(c, p[9 + i]) << i;
     858       19951 :         if (!(x & (vp7 ? 0xF0 : 0xFFF0)) || vp56_rac_get_prob(c, p[12]))
     859       13304 :             x += 8;
     860             :     } else {
     861             :         // small_mvtree
     862       64935 :         const uint8_t *ps = p + 2;
     863       64935 :         bit = vp56_rac_get_prob(c, *ps);
     864       64935 :         ps += 1 + 3 * bit;
     865       64935 :         x  += 4 * bit;
     866       64935 :         bit = vp56_rac_get_prob(c, *ps);
     867       64935 :         ps += 1 + bit;
     868       64935 :         x  += 2 * bit;
     869       64935 :         x  += vp56_rac_get_prob(c, *ps);
     870             :     }
     871             : 
     872       84886 :     return (x && vp56_rac_get_prob(c, p[1])) ? -x : x;
     873             : }
     874             : 
     875           0 : static int vp7_read_mv_component(VP56RangeCoder *c, const uint8_t *p)
     876             : {
     877           0 :     return read_mv_component(c, p, 1);
     878             : }
     879             : 
     880       28388 : static int vp8_read_mv_component(VP56RangeCoder *c, const uint8_t *p)
     881             : {
     882       28388 :     return read_mv_component(c, p, 0);
     883             : }
     884             : 
     885             : static av_always_inline
     886      136556 : const uint8_t *get_submv_prob(uint32_t left, uint32_t top, int is_vp7)
     887             : {
     888      136556 :     if (is_vp7)
     889           0 :         return vp7_submv_prob;
     890             : 
     891      136556 :     if (left == top)
     892       58677 :         return vp8_submv_prob[4 - !!left];
     893       77879 :     if (!top)
     894       18933 :         return vp8_submv_prob[2];
     895       58946 :     return vp8_submv_prob[1 - !!left];
     896             : }
     897             : 
     898             : /**
     899             :  * Split motion vector prediction, 16.4.
     900             :  * @returns the number of motion vectors parsed (2, 4 or 16)
     901             :  */
     902             : static av_always_inline
     903       18325 : int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
     904             :                     int layout, int is_vp7)
     905             : {
     906             :     int part_idx;
     907             :     int n, num;
     908             :     VP8Macroblock *top_mb;
     909       18325 :     VP8Macroblock *left_mb = &mb[-1];
     910       18325 :     const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning];
     911             :     const uint8_t *mbsplits_top, *mbsplits_cur, *firstidx;
     912             :     VP56mv *top_mv;
     913       18325 :     VP56mv *left_mv = left_mb->bmv;
     914       18325 :     VP56mv *cur_mv  = mb->bmv;
     915             : 
     916       18325 :     if (!layout) // layout is inlined, s->mb_layout is not
     917       18325 :         top_mb = &mb[2];
     918             :     else
     919           0 :         top_mb = &mb[-s->mb_width - 1];
     920       18325 :     mbsplits_top = vp8_mbsplits[top_mb->partitioning];
     921       18325 :     top_mv       = top_mb->bmv;
     922             : 
     923       18325 :     if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) {
     924       11713 :         if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1]))
     925        8044 :             part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]);
     926             :         else
     927        3669 :             part_idx = VP8_SPLITMVMODE_8x8;
     928             :     } else {
     929        6612 :         part_idx = VP8_SPLITMVMODE_4x4;
     930             :     }
     931             : 
     932       18325 :     num              = vp8_mbsplit_count[part_idx];
     933       18325 :     mbsplits_cur     = vp8_mbsplits[part_idx],
     934       18325 :     firstidx         = vp8_mbfirstidx[part_idx];
     935       18325 :     mb->partitioning = part_idx;
     936             : 
     937      154881 :     for (n = 0; n < num; n++) {
     938      136556 :         int k = firstidx[n];
     939             :         uint32_t left, above;
     940             :         const uint8_t *submv_prob;
     941             : 
     942      136556 :         if (!(k & 3))
     943       46739 :             left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
     944             :         else
     945       89817 :             left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]);
     946      136556 :         if (k <= 3)
     947       44965 :             above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
     948             :         else
     949       91591 :             above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]);
     950             : 
     951      136556 :         submv_prob = get_submv_prob(left, above, is_vp7);
     952             : 
     953      136556 :         if (vp56_rac_get_prob_branchy(c, submv_prob[0])) {
     954       50605 :             if (vp56_rac_get_prob_branchy(c, submv_prob[1])) {
     955       31737 :                 if (vp56_rac_get_prob_branchy(c, submv_prob[2])) {
     956       56498 :                     mb->bmv[n].y = mb->mv.y +
     957       28249 :                                    read_mv_component(c, s->prob->mvc[0], is_vp7);
     958       56498 :                     mb->bmv[n].x = mb->mv.x +
     959       28249 :                                    read_mv_component(c, s->prob->mvc[1], is_vp7);
     960             :                 } else {
     961        3488 :                     AV_ZERO32(&mb->bmv[n]);
     962             :                 }
     963             :             } else {
     964       18868 :                 AV_WN32A(&mb->bmv[n], above);
     965             :             }
     966             :         } else {
     967       85951 :             AV_WN32A(&mb->bmv[n], left);
     968             :         }
     969             :     }
     970             : 
     971       18325 :     return num;
     972             : }
     973             : 
     974             : /**
     975             :  * The vp7 reference decoder uses a padding macroblock column (added to right
     976             :  * edge of the frame) to guard against illegal macroblock offsets. The
     977             :  * algorithm has bugs that permit offsets to straddle the padding column.
     978             :  * This function replicates those bugs.
     979             :  *
     980             :  * @param[out] edge_x macroblock x address
     981             :  * @param[out] edge_y macroblock y address
     982             :  *
     983             :  * @return macroblock offset legal (boolean)
     984             :  */
     985       76560 : static int vp7_calculate_mb_offset(int mb_x, int mb_y, int mb_width,
     986             :                                    int xoffset, int yoffset, int boundary,
     987             :                                    int *edge_x, int *edge_y)
     988             : {
     989       76560 :     int vwidth = mb_width + 1;
     990       76560 :     int new = (mb_y + yoffset) * vwidth + mb_x + xoffset;
     991       76560 :     if (new < boundary || new % vwidth == vwidth - 1)
     992       11861 :         return 0;
     993       64699 :     *edge_y = new / vwidth;
     994       64699 :     *edge_x = new % vwidth;
     995       64699 :     return 1;
     996             : }
     997             : 
     998       64699 : static const VP56mv *get_bmv_ptr(const VP8Macroblock *mb, int subblock)
     999             : {
    1000       64699 :     return &mb->bmv[mb->mode == VP8_MVMODE_SPLIT ? vp8_mbsplits[mb->partitioning][subblock] : 0];
    1001             : }
    1002             : 
    1003             : static av_always_inline
    1004        6380 : void vp7_decode_mvs(VP8Context *s, VP8Macroblock *mb,
    1005             :                     int mb_x, int mb_y, int layout)
    1006             : {
    1007             :     VP8Macroblock *mb_edge[12];
    1008             :     enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR };
    1009             :     enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
    1010        6380 :     int idx = CNT_ZERO;
    1011             :     VP56mv near_mv[3];
    1012        6380 :     uint8_t cnt[3] = { 0 };
    1013        6380 :     VP56RangeCoder *c = &s->c;
    1014             :     int i;
    1015             : 
    1016        6380 :     AV_ZERO32(&near_mv[0]);
    1017        6380 :     AV_ZERO32(&near_mv[1]);
    1018        6380 :     AV_ZERO32(&near_mv[2]);
    1019             : 
    1020       82940 :     for (i = 0; i < VP7_MV_PRED_COUNT; i++) {
    1021       76560 :         const VP7MVPred * pred = &vp7_mv_pred[i];
    1022             :         int edge_x, edge_y;
    1023             : 
    1024      153120 :         if (vp7_calculate_mb_offset(mb_x, mb_y, s->mb_width, pred->xoffset,
    1025      153120 :                                     pred->yoffset, !s->profile, &edge_x, &edge_y)) {
    1026      129398 :             VP8Macroblock *edge = mb_edge[i] = (s->mb_layout == 1)
    1027      129398 :                                              ? s->macroblocks_base + 1 + edge_x +
    1028       64699 :                                                (s->mb_width + 1) * (edge_y + 1)
    1029      129398 :                                              : s->macroblocks + edge_x +
    1030           0 :                                                (s->mb_height - edge_y - 1) * 2;
    1031       64699 :             uint32_t mv = AV_RN32A(get_bmv_ptr(edge, vp7_mv_pred[i].subblock));
    1032       64699 :             if (mv) {
    1033           0 :                 if (AV_RN32A(&near_mv[CNT_NEAREST])) {
    1034           0 :                     if (mv == AV_RN32A(&near_mv[CNT_NEAREST])) {
    1035           0 :                         idx = CNT_NEAREST;
    1036           0 :                     } else if (AV_RN32A(&near_mv[CNT_NEAR])) {
    1037           0 :                         if (mv != AV_RN32A(&near_mv[CNT_NEAR]))
    1038           0 :                             continue;
    1039           0 :                         idx = CNT_NEAR;
    1040             :                     } else {
    1041           0 :                         AV_WN32A(&near_mv[CNT_NEAR], mv);
    1042           0 :                         idx = CNT_NEAR;
    1043             :                     }
    1044             :                 } else {
    1045           0 :                     AV_WN32A(&near_mv[CNT_NEAREST], mv);
    1046           0 :                     idx = CNT_NEAREST;
    1047             :                 }
    1048             :             } else {
    1049       64699 :                 idx = CNT_ZERO;
    1050             :             }
    1051             :         } else {
    1052       11861 :             idx = CNT_ZERO;
    1053             :         }
    1054       76560 :         cnt[idx] += vp7_mv_pred[i].score;
    1055             :     }
    1056             : 
    1057        6380 :     mb->partitioning = VP8_SPLITMVMODE_NONE;
    1058             : 
    1059        6380 :     if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_ZERO]][0])) {
    1060           0 :         mb->mode = VP8_MVMODE_MV;
    1061             : 
    1062           0 :         if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAREST]][1])) {
    1063             : 
    1064           0 :             if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][2])) {
    1065             : 
    1066           0 :                 if (cnt[CNT_NEAREST] > cnt[CNT_NEAR])
    1067           0 :                     AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAREST] ? 0 : AV_RN32A(&near_mv[CNT_NEAREST]));
    1068             :                 else
    1069           0 :                     AV_WN32A(&mb->mv, cnt[CNT_ZERO] > cnt[CNT_NEAR]    ? 0 : AV_RN32A(&near_mv[CNT_NEAR]));
    1070             : 
    1071           0 :                 if (vp56_rac_get_prob_branchy(c, vp7_mode_contexts[cnt[CNT_NEAR]][3])) {
    1072           0 :                     mb->mode = VP8_MVMODE_SPLIT;
    1073           0 :                     mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP7) - 1];
    1074             :                 } else {
    1075           0 :                     mb->mv.y += vp7_read_mv_component(c, s->prob->mvc[0]);
    1076           0 :                     mb->mv.x += vp7_read_mv_component(c, s->prob->mvc[1]);
    1077           0 :                     mb->bmv[0] = mb->mv;
    1078             :                 }
    1079             :             } else {
    1080           0 :                 mb->mv = near_mv[CNT_NEAR];
    1081           0 :                 mb->bmv[0] = mb->mv;
    1082             :             }
    1083             :         } else {
    1084           0 :             mb->mv = near_mv[CNT_NEAREST];
    1085           0 :             mb->bmv[0] = mb->mv;
    1086             :         }
    1087             :     } else {
    1088        6380 :         mb->mode = VP8_MVMODE_ZERO;
    1089        6380 :         AV_ZERO32(&mb->mv);
    1090        6380 :         mb->bmv[0] = mb->mv;
    1091             :     }
    1092        6380 : }
    1093             : 
    1094             : static av_always_inline
    1095      379497 : void vp8_decode_mvs(VP8Context *s, VP8mvbounds *mv_bounds, VP8Macroblock *mb,
    1096             :                     int mb_x, int mb_y, int layout)
    1097             : {
    1098      758994 :     VP8Macroblock *mb_edge[3] = { 0      /* top */,
    1099      379497 :                                   mb - 1 /* left */,
    1100             :                                   0      /* top-left */ };
    1101             :     enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
    1102             :     enum { VP8_EDGE_TOP, VP8_EDGE_LEFT, VP8_EDGE_TOPLEFT };
    1103      379497 :     int idx = CNT_ZERO;
    1104      379497 :     int cur_sign_bias = s->sign_bias[mb->ref_frame];
    1105      379497 :     int8_t *sign_bias = s->sign_bias;
    1106             :     VP56mv near_mv[4];
    1107      379497 :     uint8_t cnt[4] = { 0 };
    1108      379497 :     VP56RangeCoder *c = &s->c;
    1109             : 
    1110      379497 :     if (!layout) { // layout is inlined (s->mb_layout is not)
    1111      379497 :         mb_edge[0] = mb + 2;
    1112      379497 :         mb_edge[2] = mb + 1;
    1113             :     } else {
    1114           0 :         mb_edge[0] = mb - s->mb_width - 1;
    1115           0 :         mb_edge[2] = mb - s->mb_width - 2;
    1116             :     }
    1117             : 
    1118      379497 :     AV_ZERO32(&near_mv[0]);
    1119      379497 :     AV_ZERO32(&near_mv[1]);
    1120      379497 :     AV_ZERO32(&near_mv[2]);
    1121             : 
    1122             :     /* Process MB on top, left and top-left */
    1123             : #define MV_EDGE_CHECK(n)                                                      \
    1124             :     {                                                                         \
    1125             :         VP8Macroblock *edge = mb_edge[n];                                     \
    1126             :         int edge_ref = edge->ref_frame;                                       \
    1127             :         if (edge_ref != VP56_FRAME_CURRENT) {                                 \
    1128             :             uint32_t mv = AV_RN32A(&edge->mv);                                \
    1129             :             if (mv) {                                                         \
    1130             :                 if (cur_sign_bias != sign_bias[edge_ref]) {                   \
    1131             :                     /* SWAR negate of the values in mv. */                    \
    1132             :                     mv = ~mv;                                                 \
    1133             :                     mv = ((mv & 0x7fff7fff) +                                 \
    1134             :                           0x00010001) ^ (mv & 0x80008000);                    \
    1135             :                 }                                                             \
    1136             :                 if (!n || mv != AV_RN32A(&near_mv[idx]))                      \
    1137             :                     AV_WN32A(&near_mv[++idx], mv);                            \
    1138             :                 cnt[idx] += 1 + (n != 2);                                     \
    1139             :             } else                                                            \
    1140             :                 cnt[CNT_ZERO] += 1 + (n != 2);                                \
    1141             :         }                                                                     \
    1142             :     }
    1143             : 
    1144      379497 :     MV_EDGE_CHECK(0)
    1145      379497 :     MV_EDGE_CHECK(1)
    1146      379497 :     MV_EDGE_CHECK(2)
    1147             : 
    1148      379497 :     mb->partitioning = VP8_SPLITMVMODE_NONE;
    1149      379497 :     if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_ZERO]][0])) {
    1150       57174 :         mb->mode = VP8_MVMODE_MV;
    1151             : 
    1152             :         /* If we have three distinct MVs, merge first and last if they're the same */
    1153       69933 :         if (cnt[CNT_SPLITMV] &&
    1154       12759 :             AV_RN32A(&near_mv[1 + VP8_EDGE_TOP]) == AV_RN32A(&near_mv[1 + VP8_EDGE_TOPLEFT]))
    1155        5465 :             cnt[CNT_NEAREST] += 1;
    1156             : 
    1157             :         /* Swap near and nearest if necessary */
    1158       57174 :         if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
    1159        4482 :             FFSWAP(uint8_t,     cnt[CNT_NEAREST],     cnt[CNT_NEAR]);
    1160        4482 :             FFSWAP( VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
    1161             :         }
    1162             : 
    1163       57174 :         if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
    1164       37607 :             if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {
    1165             :                 /* Choose the best mv out of 0,0 and the nearest mv */
    1166       32519 :                 clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
    1167       97557 :                 cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode    == VP8_MVMODE_SPLIT) +
    1168       97557 :                                     (mb_edge[VP8_EDGE_TOP]->mode     == VP8_MVMODE_SPLIT)) * 2 +
    1169       32519 :                                     (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
    1170             : 
    1171       32519 :                 if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_SPLITMV]][3])) {
    1172       18325 :                     mb->mode = VP8_MVMODE_SPLIT;
    1173       18325 :                     mb->mv = mb->bmv[decode_splitmvs(s, c, mb, layout, IS_VP8) - 1];
    1174             :                 } else {
    1175       14194 :                     mb->mv.y  += vp8_read_mv_component(c, s->prob->mvc[0]);
    1176       14194 :                     mb->mv.x  += vp8_read_mv_component(c, s->prob->mvc[1]);
    1177       14194 :                     mb->bmv[0] = mb->mv;
    1178             :                 }
    1179             :             } else {
    1180        5088 :                 clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAR]);
    1181        5088 :                 mb->bmv[0] = mb->mv;
    1182             :             }
    1183             :         } else {
    1184       19567 :             clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAREST]);
    1185       19567 :             mb->bmv[0] = mb->mv;
    1186             :         }
    1187             :     } else {
    1188      322323 :         mb->mode = VP8_MVMODE_ZERO;
    1189      322323 :         AV_ZERO32(&mb->mv);
    1190      322323 :         mb->bmv[0] = mb->mv;
    1191             :     }
    1192      379497 : }
    1193             : 
    1194             : static av_always_inline
    1195       17675 : void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
    1196             :                            int mb_x, int keyframe, int layout)
    1197             : {
    1198       17675 :     uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
    1199             : 
    1200       17675 :     if (layout) {
    1201          33 :         VP8Macroblock *mb_top = mb - s->mb_width - 1;
    1202          33 :         memcpy(mb->intra4x4_pred_mode_top, mb_top->intra4x4_pred_mode_top, 4);
    1203             :     }
    1204       17675 :     if (keyframe) {
    1205             :         int x, y;
    1206             :         uint8_t *top;
    1207       10822 :         uint8_t *const left = s->intra4x4_pred_mode_left;
    1208       10822 :         if (layout)
    1209          33 :             top = mb->intra4x4_pred_mode_top;
    1210             :         else
    1211       10789 :             top = s->intra4x4_pred_mode_top + 4 * mb_x;
    1212       54110 :         for (y = 0; y < 4; y++) {
    1213      216440 :             for (x = 0; x < 4; x++) {
    1214             :                 const uint8_t *ctx;
    1215      173152 :                 ctx       = vp8_pred4x4_prob_intra[top[x]][left[y]];
    1216      173152 :                 *intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
    1217      173152 :                 left[y]   = top[x] = *intra4x4;
    1218      173152 :                 intra4x4++;
    1219             :             }
    1220             :         }
    1221             :     } else {
    1222             :         int i;
    1223      116501 :         for (i = 0; i < 16; i++)
    1224      109648 :             intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree,
    1225             :                                            vp8_pred4x4_prob_inter);
    1226             :     }
    1227       17675 : }
    1228             : 
    1229             : static av_always_inline
    1230      443209 : void decode_mb_mode(VP8Context *s, VP8mvbounds *mv_bounds,
    1231             :                     VP8Macroblock *mb, int mb_x, int mb_y,
    1232             :                     uint8_t *segment, uint8_t *ref, int layout, int is_vp7)
    1233             : {
    1234      443209 :     VP56RangeCoder *c = &s->c;
    1235             :     static const char * const vp7_feature_name[] = { "q-index",
    1236             :                                                      "lf-delta",
    1237             :                                                      "partial-golden-update",
    1238             :                                                      "blit-pitch" };
    1239      443209 :     if (is_vp7) {
    1240             :         int i;
    1241        6600 :         *segment = 0;
    1242       33000 :         for (i = 0; i < 4; i++) {
    1243       26400 :             if (s->feature_enabled[i]) {
    1244           0 :                 if (vp56_rac_get_prob_branchy(c, s->feature_present_prob[i])) {
    1245           0 :                       int index = vp8_rac_get_tree(c, vp7_feature_index_tree,
    1246           0 :                                                    s->feature_index_prob[i]);
    1247           0 :                       av_log(s->avctx, AV_LOG_WARNING,
    1248             :                              "Feature %s present in macroblock (value 0x%x)\n",
    1249           0 :                              vp7_feature_name[i], s->feature_value[i][index]);
    1250             :                 }
    1251             :            }
    1252             :         }
    1253      436609 :     } else if (s->segmentation.update_map) {
    1254        8289 :         int bit  = vp56_rac_get_prob(c, s->prob->segmentid[0]);
    1255        8289 :         *segment = vp56_rac_get_prob(c, s->prob->segmentid[1+bit]) + 2*bit;
    1256      428320 :     } else if (s->segmentation.enabled)
    1257       94425 :         *segment = ref ? *ref : *segment;
    1258      443209 :     mb->segment = *segment;
    1259             : 
    1260      443209 :     mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0;
    1261             : 
    1262      443209 :     if (s->keyframe) {
    1263       32872 :         mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra,
    1264             :                                     vp8_pred16x16_prob_intra);
    1265             : 
    1266       32872 :         if (mb->mode == MODE_I4x4) {
    1267       10822 :             decode_intra4x4_modes(s, c, mb, mb_x, 1, layout);
    1268             :         } else {
    1269       22050 :             const uint32_t modes = (is_vp7 ? vp7_pred4x4_mode
    1270       22050 :                                            : vp8_pred4x4_mode)[mb->mode] * 0x01010101u;
    1271       22050 :             if (s->mb_layout)
    1272         187 :                 AV_WN32A(mb->intra4x4_pred_mode_top, modes);
    1273             :             else
    1274       21863 :                 AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
    1275       22050 :             AV_WN32A(s->intra4x4_pred_mode_left, modes);
    1276             :         }
    1277             : 
    1278       32872 :         mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree,
    1279             :                                                 vp8_pred8x8c_prob_intra);
    1280       32872 :         mb->ref_frame        = VP56_FRAME_CURRENT;
    1281      410337 :     } else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) {
    1282             :         // inter MB, 16.2
    1283      385877 :         if (vp56_rac_get_prob_branchy(c, s->prob->last))
    1284       26184 :             mb->ref_frame =
    1285       17248 :                 (!is_vp7 && vp56_rac_get_prob(c, s->prob->golden)) ? VP56_FRAME_GOLDEN2 /* altref */
    1286             :                                                                    : VP56_FRAME_GOLDEN;
    1287             :         else
    1288      368597 :             mb->ref_frame = VP56_FRAME_PREVIOUS;
    1289      385877 :         s->ref_count[mb->ref_frame - 1]++;
    1290             : 
    1291             :         // motion vectors, 16.3
    1292      385877 :         if (is_vp7)
    1293        6380 :             vp7_decode_mvs(s, mb, mb_x, mb_y, layout);
    1294             :         else
    1295      379497 :             vp8_decode_mvs(s, mv_bounds, mb, mb_x, mb_y, layout);
    1296             :     } else {
    1297             :         // intra MB, 16.1
    1298       24460 :         mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
    1299             : 
    1300       24460 :         if (mb->mode == MODE_I4x4)
    1301        6853 :             decode_intra4x4_modes(s, c, mb, mb_x, 0, layout);
    1302             : 
    1303       24460 :         mb->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree,
    1304       24460 :                                                 s->prob->pred8x8c);
    1305       24460 :         mb->ref_frame        = VP56_FRAME_CURRENT;
    1306       24460 :         mb->partitioning     = VP8_SPLITMVMODE_NONE;
    1307       24460 :         AV_ZERO32(&mb->bmv[0]);
    1308             :     }
    1309      443209 : }
    1310             : 
    1311             : /**
    1312             :  * @param r     arithmetic bitstream reader context
    1313             :  * @param block destination for block coefficients
    1314             :  * @param probs probabilities to use when reading trees from the bitstream
    1315             :  * @param i     initial coeff index, 0 unless a separate DC block is coded
    1316             :  * @param qmul  array holding the dc/ac dequant factor at position 0/1
    1317             :  *
    1318             :  * @return 0 if no coeffs were decoded
    1319             :  *         otherwise, the index of the last coeff decoded plus one
    1320             :  */
    1321             : static av_always_inline
    1322      393067 : int decode_block_coeffs_internal(VP56RangeCoder *r, int16_t block[16],
    1323             :                                  uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
    1324             :                                  int i, uint8_t *token_prob, int16_t qmul[2],
    1325             :                                  const uint8_t scan[16], int vp7)
    1326             : {
    1327      393067 :     VP56RangeCoder c = *r;
    1328      393067 :     goto skip_eob;
    1329             :     do {
    1330             :         int coeff;
    1331     1101534 : restart:
    1332     1103653 :         if (!vp56_rac_get_prob_branchy(&c, token_prob[0]))   // DCT_EOB
    1333      373414 :             break;
    1334             : 
    1335      730239 : skip_eob:
    1336     1956826 :         if (!vp56_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0
    1337      835639 :             if (++i == 16)
    1338           0 :                 break; // invalid input; blocks should end with EOB
    1339      835639 :             token_prob = probs[i][0];
    1340      835639 :             if (vp7)
    1341        2119 :                 goto restart;
    1342      833520 :             goto skip_eob;
    1343             :         }
    1344             : 
    1345     1121187 :         if (!vp56_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1
    1346      719461 :             coeff = 1;
    1347      719461 :             token_prob = probs[i + 1][1];
    1348             :         } else {
    1349      401726 :             if (!vp56_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
    1350      274407 :                 coeff = vp56_rac_get_prob_branchy(&c, token_prob[4]);
    1351      274407 :                 if (coeff)
    1352      113420 :                     coeff += vp56_rac_get_prob(&c, token_prob[5]);
    1353      274407 :                 coeff += 2;
    1354             :             } else {
    1355             :                 // DCT_CAT*
    1356      127319 :                 if (!vp56_rac_get_prob_branchy(&c, token_prob[6])) {
    1357       83429 :                     if (!vp56_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1
    1358       45203 :                         coeff = 5 + vp56_rac_get_prob(&c, vp8_dct_cat1_prob[0]);
    1359             :                     } else {                                    // DCT_CAT2
    1360       38226 :                         coeff  = 7;
    1361       38226 :                         coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
    1362       38226 :                         coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[1]);
    1363             :                     }
    1364             :                 } else {    // DCT_CAT3 and up
    1365       43890 :                     int a   = vp56_rac_get_prob(&c, token_prob[8]);
    1366       43890 :                     int b   = vp56_rac_get_prob(&c, token_prob[9 + a]);
    1367       43890 :                     int cat = (a << 1) + b;
    1368       43890 :                     coeff  = 3 + (8 << cat);
    1369       43890 :                     coeff += vp8_rac_get_coeff(&c, ff_vp8_dct_cat_prob[cat]);
    1370             :                 }
    1371             :             }
    1372      401726 :             token_prob = probs[i + 1][2];
    1373             :         }
    1374     1121187 :         block[scan[i]] = (vp8_rac_get(&c) ? -coeff : coeff) * qmul[!!i];
    1375     1121187 :     } while (++i < 16);
    1376             : 
    1377      393067 :     *r = c;
    1378      393067 :     return i;
    1379             : }
    1380             : 
    1381             : static av_always_inline
    1382        6380 : int inter_predict_dc(int16_t block[16], int16_t pred[2])
    1383             : {
    1384        6380 :     int16_t dc = block[0];
    1385        6380 :     int ret = 0;
    1386             : 
    1387        6380 :     if (pred[1] > 3) {
    1388           0 :         dc += pred[0];
    1389           0 :         ret = 1;
    1390             :     }
    1391             : 
    1392        6380 :     if (!pred[0] | !dc | ((int32_t)pred[0] ^ (int32_t)dc) >> 31) {
    1393        6380 :         block[0] = pred[0] = dc;
    1394        6380 :         pred[1] = 0;
    1395             :     } else {
    1396           0 :         if (pred[0] == dc)
    1397           0 :             pred[1]++;
    1398           0 :         block[0] = pred[0] = dc;
    1399             :     }
    1400             : 
    1401        6380 :     return ret;
    1402             : }
    1403             : 
    1404         734 : static int vp7_decode_block_coeffs_internal(VP56RangeCoder *r,
    1405             :                                             int16_t block[16],
    1406             :                                             uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
    1407             :                                             int i, uint8_t *token_prob,
    1408             :                                             int16_t qmul[2],
    1409             :                                             const uint8_t scan[16])
    1410             : {
    1411         734 :     return decode_block_coeffs_internal(r, block, probs, i,
    1412             :                                         token_prob, qmul, scan, IS_VP7);
    1413             : }
    1414             : 
    1415             : #ifndef vp8_decode_block_coeffs_internal
    1416      392333 : static int vp8_decode_block_coeffs_internal(VP56RangeCoder *r,
    1417             :                                             int16_t block[16],
    1418             :                                             uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
    1419             :                                             int i, uint8_t *token_prob,
    1420             :                                             int16_t qmul[2])
    1421             : {
    1422      392333 :     return decode_block_coeffs_internal(r, block, probs, i,
    1423             :                                         token_prob, qmul, ff_zigzag_scan, IS_VP8);
    1424             : }
    1425             : #endif
    1426             : 
    1427             : /**
    1428             :  * @param c          arithmetic bitstream reader context
    1429             :  * @param block      destination for block coefficients
    1430             :  * @param probs      probabilities to use when reading trees from the bitstream
    1431             :  * @param i          initial coeff index, 0 unless a separate DC block is coded
    1432             :  * @param zero_nhood the initial prediction context for number of surrounding
    1433             :  *                   all-zero blocks (only left/top, so 0-2)
    1434             :  * @param qmul       array holding the dc/ac dequant factor at position 0/1
    1435             :  * @param scan       scan pattern (VP7 only)
    1436             :  *
    1437             :  * @return 0 if no coeffs were decoded
    1438             :  *         otherwise, the index of the last coeff decoded plus one
    1439             :  */
    1440             : static av_always_inline
    1441     2050151 : int decode_block_coeffs(VP56RangeCoder *c, int16_t block[16],
    1442             :                         uint8_t probs[16][3][NUM_DCT_TOKENS - 1],
    1443             :                         int i, int zero_nhood, int16_t qmul[2],
    1444             :                         const uint8_t scan[16], int vp7)
    1445             : {
    1446     2050151 :     uint8_t *token_prob = probs[i][zero_nhood];
    1447     2050151 :     if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
    1448     1657084 :         return 0;
    1449             :     return vp7 ? vp7_decode_block_coeffs_internal(c, block, probs, i,
    1450             :                                                   token_prob, qmul, scan)
    1451      393067 :                : vp8_decode_block_coeffs_internal(c, block, probs, i,
    1452             :                                                   token_prob, qmul);
    1453             : }
    1454             : 
    1455             : static av_always_inline
    1456       83297 : void decode_mb_coeffs(VP8Context *s, VP8ThreadData *td, VP56RangeCoder *c,
    1457             :                       VP8Macroblock *mb, uint8_t t_nnz[9], uint8_t l_nnz[9],
    1458             :                       int is_vp7)
    1459             : {
    1460       83297 :     int i, x, y, luma_start = 0, luma_ctx = 3;
    1461       83297 :     int nnz_pred, nnz, nnz_total = 0;
    1462       83297 :     int segment = mb->segment;
    1463       83297 :     int block_dc = 0;
    1464             : 
    1465       83297 :     if (mb->mode != MODE_I4x4 && (is_vp7 || mb->mode != VP8_MVMODE_SPLIT)) {
    1466       51023 :         nnz_pred = t_nnz[8] + l_nnz[8];
    1467             : 
    1468             :         // decode DC values and do hadamard
    1469       51023 :         nnz = decode_block_coeffs(c, td->block_dc, s->prob->token[1], 0,
    1470       51023 :                                   nnz_pred, s->qmat[segment].luma_dc_qmul,
    1471             :                                   ff_zigzag_scan, is_vp7);
    1472       51023 :         l_nnz[8] = t_nnz[8] = !!nnz;
    1473             : 
    1474       51023 :         if (is_vp7 && mb->mode > MODE_I4x4) {
    1475        6380 :             nnz |=  inter_predict_dc(td->block_dc,
    1476        6380 :                                      s->inter_dc_pred[mb->ref_frame - 1]);
    1477             :         }
    1478             : 
    1479       51023 :         if (nnz) {
    1480       34741 :             nnz_total += nnz;
    1481       34741 :             block_dc   = 1;
    1482       34741 :             if (nnz == 1)
    1483       10303 :                 s->vp8dsp.vp8_luma_dc_wht_dc(td->block, td->block_dc);
    1484             :             else
    1485       24438 :                 s->vp8dsp.vp8_luma_dc_wht(td->block, td->block_dc);
    1486             :         }
    1487       51023 :         luma_start = 1;
    1488       51023 :         luma_ctx   = 0;
    1489             :     }
    1490             : 
    1491             :     // luma blocks
    1492      416485 :     for (y = 0; y < 4; y++)
    1493     1665940 :         for (x = 0; x < 4; x++) {
    1494     1332752 :             nnz_pred = l_nnz[y] + t_nnz[x];
    1495     1332752 :             nnz = decode_block_coeffs(c, td->block[y][x],
    1496     1332752 :                                       s->prob->token[luma_ctx],
    1497             :                                       luma_start, nnz_pred,
    1498     1332752 :                                       s->qmat[segment].luma_qmul,
    1499     1332752 :                                       s->prob[0].scan, is_vp7);
    1500             :             /* nnz+block_dc may be one more than the actual last index,
    1501             :              * but we don't care */
    1502     1332752 :             td->non_zero_count_cache[y][x] = nnz + block_dc;
    1503     1332752 :             t_nnz[x] = l_nnz[y] = !!nnz;
    1504     1332752 :             nnz_total += nnz;
    1505             :         }
    1506             : 
    1507             :     // chroma blocks
    1508             :     // TODO: what to do about dimensions? 2nd dim for luma is x,
    1509             :     // but for chroma it's (y<<1)|x
    1510      249891 :     for (i = 4; i < 6; i++)
    1511      499782 :         for (y = 0; y < 2; y++)
    1512      999564 :             for (x = 0; x < 2; x++) {
    1513      666376 :                 nnz_pred = l_nnz[i + 2 * y] + t_nnz[i + 2 * x];
    1514      666376 :                 nnz = decode_block_coeffs(c, td->block[i][(y << 1) + x],
    1515      666376 :                                           s->prob->token[2], 0, nnz_pred,
    1516      666376 :                                           s->qmat[segment].chroma_qmul,
    1517      666376 :                                           s->prob[0].scan, is_vp7);
    1518      666376 :                 td->non_zero_count_cache[i][(y << 1) + x] = nnz;
    1519      666376 :                 t_nnz[i + 2 * x] = l_nnz[i + 2 * y] = !!nnz;
    1520      666376 :                 nnz_total += nnz;
    1521             :             }
    1522             : 
    1523             :     // if there were no coded coeffs despite the macroblock not being marked skip,
    1524             :     // we MUST not do the inner loop filter and should not do IDCT
    1525             :     // Since skip isn't used for bitstream prediction, just manually set it.
    1526       83297 :     if (!nnz_total)
    1527        6457 :         mb->skip = 1;
    1528       83297 : }
    1529             : 
    1530             : static av_always_inline
    1531      424229 : void backup_mb_border(uint8_t *top_border, uint8_t *src_y,
    1532             :                       uint8_t *src_cb, uint8_t *src_cr,
    1533             :                       ptrdiff_t linesize, ptrdiff_t uvlinesize, int simple)
    1534             : {
    1535      424229 :     AV_COPY128(top_border, src_y + 15 * linesize);
    1536      424229 :     if (!simple) {
    1537      416507 :         AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
    1538      416507 :         AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
    1539             :     }
    1540      424229 : }
    1541             : 
    1542             : static av_always_inline
    1543       84130 : void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb,
    1544             :                     uint8_t *src_cr, ptrdiff_t linesize, ptrdiff_t uvlinesize, int mb_x,
    1545             :                     int mb_y, int mb_width, int simple, int xchg)
    1546             : {
    1547       84130 :     uint8_t *top_border_m1 = top_border - 32;     // for TL prediction
    1548       84130 :     src_y  -= linesize;
    1549       84130 :     src_cb -= uvlinesize;
    1550       84130 :     src_cr -= uvlinesize;
    1551             : 
    1552             : #define XCHG(a, b, xchg)                                                      \
    1553             :     do {                                                                      \
    1554             :         if (xchg)                                                             \
    1555             :             AV_SWAP64(b, a);                                                  \
    1556             :         else                                                                  \
    1557             :             AV_COPY64(b, a);                                                  \
    1558             :     } while (0)
    1559             : 
    1560       84130 :     XCHG(top_border_m1 + 8, src_y - 8, xchg);
    1561       84130 :     XCHG(top_border, src_y, xchg);
    1562       84130 :     XCHG(top_border + 8, src_y + 8, 1);
    1563       84130 :     if (mb_x < mb_width - 1)
    1564       80706 :         XCHG(top_border + 32, src_y + 16, 1);
    1565             : 
    1566             :     // only copy chroma for normal loop filter
    1567             :     // or to initialize the top row to 127
    1568       84130 :     if (!simple || !mb_y) {
    1569       81048 :         XCHG(top_border_m1 + 16, src_cb - 8, xchg);
    1570       81048 :         XCHG(top_border_m1 + 24, src_cr - 8, xchg);
    1571       81048 :         XCHG(top_border + 16, src_cb, 1);
    1572       81048 :         XCHG(top_border + 24, src_cr, 1);
    1573             :     }
    1574       84130 : }
    1575             : 
    1576             : static av_always_inline
    1577       72888 : int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
    1578             : {
    1579       72888 :     if (!mb_x)
    1580        3029 :         return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
    1581             :     else
    1582       69859 :         return mb_y ? mode : LEFT_DC_PRED8x8;
    1583             : }
    1584             : 
    1585             : static av_always_inline
    1586        1809 : int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y, int vp7)
    1587             : {
    1588        1809 :     if (!mb_x)
    1589           1 :         return mb_y ? VERT_PRED8x8 : (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8);
    1590             :     else
    1591        1808 :         return mb_y ? mode : HOR_PRED8x8;
    1592             : }
    1593             : 
    1594             : static av_always_inline
    1595       96989 : int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y, int vp7)
    1596             : {
    1597       96989 :     switch (mode) {
    1598       72888 :     case DC_PRED8x8:
    1599       72888 :         return check_dc_pred8x8_mode(mode, mb_x, mb_y);
    1600        5757 :     case VERT_PRED8x8:
    1601        5757 :         return !mb_y ? (vp7 ? DC_128_PRED8x8 : DC_127_PRED8x8) : mode;
    1602       16535 :     case HOR_PRED8x8:
    1603       16535 :         return !mb_x ? (vp7 ? DC_128_PRED8x8 : DC_129_PRED8x8) : mode;
    1604        1809 :     case PLANE_PRED8x8: /* TM */
    1605        1809 :         return check_tm_pred8x8_mode(mode, mb_x, mb_y, vp7);
    1606             :     }
    1607           0 :     return mode;
    1608             : }
    1609             : 
    1610             : static av_always_inline
    1611       49009 : int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y, int vp7)
    1612             : {
    1613       49009 :     if (!mb_x) {
    1614        1362 :         return mb_y ? VERT_VP8_PRED : (vp7 ? DC_128_PRED : DC_129_PRED);
    1615             :     } else {
    1616       47647 :         return mb_y ? mode : HOR_VP8_PRED;
    1617             :     }
    1618             : }
    1619             : 
    1620             : static av_always_inline
    1621      282800 : int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y,
    1622             :                                      int *copy_buf, int vp7)
    1623             : {
    1624      282800 :     switch (mode) {
    1625       13608 :     case VERT_PRED:
    1626       13608 :         if (!mb_x && mb_y) {
    1627          95 :             *copy_buf = 1;
    1628          95 :             return mode;
    1629             :         }
    1630             :         /* fall-through */
    1631             :     case DIAG_DOWN_LEFT_PRED:
    1632             :     case VERT_LEFT_PRED:
    1633       34820 :         return !mb_y ? (vp7 ? DC_128_PRED : DC_127_PRED) : mode;
    1634       19464 :     case HOR_PRED:
    1635       19464 :         if (!mb_y) {
    1636         128 :             *copy_buf = 1;
    1637         128 :             return mode;
    1638             :         }
    1639             :         /* fall-through */
    1640             :     case HOR_UP_PRED:
    1641       31686 :         return !mb_x ? (vp7 ? DC_128_PRED : DC_129_PRED) : mode;
    1642       49009 :     case TM_VP8_PRED:
    1643       49009 :         return check_tm_pred4x4_mode(mode, mb_x, mb_y, vp7);
    1644      167062 :     case DC_PRED: /* 4x4 DC doesn't use the same "H.264-style" exceptions
    1645             :                    * as 16x16/8x8 DC */
    1646             :     case DIAG_DOWN_RIGHT_PRED:
    1647             :     case VERT_RIGHT_PRED:
    1648             :     case HOR_DOWN_PRED:
    1649      167062 :         if (!mb_y || !mb_x)
    1650        1140 :             *copy_buf = 1;
    1651      167062 :         return mode;
    1652             :     }
    1653           0 :     return mode;
    1654             : }
    1655             : 
    1656             : static av_always_inline
    1657       57332 : void intra_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
    1658             :                    VP8Macroblock *mb, int mb_x, int mb_y, int is_vp7)
    1659             : {
    1660             :     int x, y, mode, nnz;
    1661             :     uint32_t tr;
    1662             : 
    1663             :     /* for the first row, we need to run xchg_mb_border to init the top edge
    1664             :      * to 127 otherwise, skip it if we aren't going to deblock */
    1665       57332 :     if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
    1666       84130 :         xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
    1667       42065 :                        s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
    1668       42065 :                        s->filter.simple, 1);
    1669             : 
    1670       57332 :     if (mb->mode < MODE_I4x4) {
    1671       39657 :         mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y, is_vp7);
    1672       39657 :         s->hpc.pred16x16[mode](dst[0], s->linesize);
    1673             :     } else {
    1674       17675 :         uint8_t *ptr = dst[0];
    1675       17675 :         uint8_t *intra4x4 = mb->intra4x4_pred_mode_mb;
    1676       17675 :         const uint8_t lo = is_vp7 ? 128 : 127;
    1677       17675 :         const uint8_t hi = is_vp7 ? 128 : 129;
    1678       17675 :         uint8_t tr_top[4] = { lo, lo, lo, lo };
    1679             : 
    1680             :         // all blocks on the right edge of the macroblock use bottom edge
    1681             :         // the top macroblock for their topright edge
    1682       17675 :         uint8_t *tr_right = ptr - s->linesize + 16;
    1683             : 
    1684             :         // if we're on the right edge of the frame, said edge is extended
    1685             :         // from the top macroblock
    1686       17675 :         if (mb_y && mb_x == s->mb_width - 1) {
    1687         635 :             tr       = tr_right[-1] * 0x01010101u;
    1688         635 :             tr_right = (uint8_t *) &tr;
    1689             :         }
    1690             : 
    1691       17675 :         if (mb->skip)
    1692         460 :             AV_ZERO128(td->non_zero_count_cache);
    1693             : 
    1694       88375 :         for (y = 0; y < 4; y++) {
    1695       70700 :             uint8_t *topright = ptr + 4 - s->linesize;
    1696      353500 :             for (x = 0; x < 4; x++) {
    1697      282800 :                 int copy = 0;
    1698      282800 :                 ptrdiff_t linesize = s->linesize;
    1699      282800 :                 uint8_t *dst = ptr + 4 * x;
    1700      282800 :                 LOCAL_ALIGNED(4, uint8_t, copy_dst, [5 * 8]);
    1701             : 
    1702      282800 :                 if ((y == 0 || x == 3) && mb_y == 0) {
    1703        4417 :                     topright = tr_top;
    1704      278383 :                 } else if (x == 3)
    1705       68176 :                     topright = tr_right;
    1706             : 
    1707      282800 :                 mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x,
    1708             :                                                         mb_y + y, &copy, is_vp7);
    1709      282800 :                 if (copy) {
    1710        1363 :                     dst      = copy_dst + 12;
    1711        1363 :                     linesize = 8;
    1712        1363 :                     if (!(mb_y + y)) {
    1713         702 :                         copy_dst[3] = lo;
    1714         702 :                         AV_WN32A(copy_dst + 4, lo * 0x01010101U);
    1715             :                     } else {
    1716         661 :                         AV_COPY32(copy_dst + 4, ptr + 4 * x - s->linesize);
    1717         661 :                         if (!(mb_x + x)) {
    1718         661 :                             copy_dst[3] = hi;
    1719             :                         } else {
    1720           0 :                             copy_dst[3] = ptr[4 * x - s->linesize - 1];
    1721             :                         }
    1722             :                     }
    1723        1363 :                     if (!(mb_x + x)) {
    1724        1404 :                         copy_dst[11] =
    1725        2106 :                         copy_dst[19] =
    1726        2106 :                         copy_dst[27] =
    1727        1404 :                         copy_dst[35] = hi;
    1728             :                     } else {
    1729         661 :                         copy_dst[11] = ptr[4 * x                   - 1];
    1730         661 :                         copy_dst[19] = ptr[4 * x + s->linesize     - 1];
    1731         661 :                         copy_dst[27] = ptr[4 * x + s->linesize * 2 - 1];
    1732         661 :                         copy_dst[35] = ptr[4 * x + s->linesize * 3 - 1];
    1733             :                     }
    1734             :                 }
    1735      282800 :                 s->hpc.pred4x4[mode](dst, topright, linesize);
    1736      282800 :                 if (copy) {
    1737        1363 :                     AV_COPY32(ptr + 4 * x,                   copy_dst + 12);
    1738        1363 :                     AV_COPY32(ptr + 4 * x + s->linesize,     copy_dst + 20);
    1739        1363 :                     AV_COPY32(ptr + 4 * x + s->linesize * 2, copy_dst + 28);
    1740        1363 :                     AV_COPY32(ptr + 4 * x + s->linesize * 3, copy_dst + 36);
    1741             :                 }
    1742             : 
    1743      282800 :                 nnz = td->non_zero_count_cache[y][x];
    1744      282800 :                 if (nnz) {
    1745      122340 :                     if (nnz == 1)
    1746       70262 :                         s->vp8dsp.vp8_idct_dc_add(ptr + 4 * x,
    1747       35131 :                                                   td->block[y][x], s->linesize);
    1748             :                     else
    1749      174418 :                         s->vp8dsp.vp8_idct_add(ptr + 4 * x,
    1750       87209 :                                                td->block[y][x], s->linesize);
    1751             :                 }
    1752      282800 :                 topright += 4;
    1753             :             }
    1754             : 
    1755       70700 :             ptr      += 4 * s->linesize;
    1756       70700 :             intra4x4 += 4;
    1757             :         }
    1758             :     }
    1759             : 
    1760       57332 :     mode = check_intra_pred8x8_mode_emuedge(mb->chroma_pred_mode,
    1761             :                                             mb_x, mb_y, is_vp7);
    1762       57332 :     s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
    1763       57332 :     s->hpc.pred8x8[mode](dst[2], s->uvlinesize);
    1764             : 
    1765       57332 :     if (mb_y && (s->deblock_filter || !mb_y) && td->thread_nr == 0)
    1766       84130 :         xchg_mb_border(s->top_border[mb_x + 1], dst[0], dst[1], dst[2],
    1767       42065 :                        s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
    1768       42065 :                        s->filter.simple, 0);
    1769       57332 : }
    1770             : 
    1771             : static const uint8_t subpel_idx[3][8] = {
    1772             :     { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
    1773             :                                 // also function pointer index
    1774             :     { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
    1775             :     { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
    1776             : };
    1777             : 
    1778             : /**
    1779             :  * luma MC function
    1780             :  *
    1781             :  * @param s        VP8 decoding context
    1782             :  * @param dst      target buffer for block data at block position
    1783             :  * @param ref      reference picture buffer at origin (0, 0)
    1784             :  * @param mv       motion vector (relative to block position) to get pixel data from
    1785             :  * @param x_off    horizontal position of block from origin (0, 0)
    1786             :  * @param y_off    vertical position of block from origin (0, 0)
    1787             :  * @param block_w  width of block (16, 8 or 4)
    1788             :  * @param block_h  height of block (always same as block_w)
    1789             :  * @param width    width of src/dst plane data
    1790             :  * @param height   height of src/dst plane data
    1791             :  * @param linesize size of a single line of plane data, including padding
    1792             :  * @param mc_func  motion compensation function pointers (bilinear or sixtap MC)
    1793             :  */
    1794             : static av_always_inline
    1795      504108 : void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst,
    1796             :                  ThreadFrame *ref, const VP56mv *mv,
    1797             :                  int x_off, int y_off, int block_w, int block_h,
    1798             :                  int width, int height, ptrdiff_t linesize,
    1799             :                  vp8_mc_func mc_func[3][3])
    1800             : {
    1801      504108 :     uint8_t *src = ref->f->data[0];
    1802             : 
    1803      504108 :     if (AV_RN32A(mv)) {
    1804      124886 :         ptrdiff_t src_linesize = linesize;
    1805             : 
    1806      124886 :         int mx = (mv->x * 2) & 7, mx_idx = subpel_idx[0][mx];
    1807      124886 :         int my = (mv->y * 2) & 7, my_idx = subpel_idx[0][my];
    1808             : 
    1809      124886 :         x_off += mv->x >> 2;
    1810      124886 :         y_off += mv->y >> 2;
    1811             : 
    1812             :         // edge emulation
    1813      124886 :         ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 4, 0);
    1814      124886 :         src += y_off * linesize + x_off;
    1815      124886 :         if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
    1816      118693 :             y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
    1817       36944 :             s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
    1818        9236 :                                      src - my_idx * linesize - mx_idx,
    1819             :                                      EDGE_EMU_LINESIZE, linesize,
    1820        9236 :                                      block_w + subpel_idx[1][mx],
    1821        9236 :                                      block_h + subpel_idx[1][my],
    1822             :                                      x_off - mx_idx, y_off - my_idx,
    1823             :                                      width, height);
    1824        9236 :             src = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
    1825        9236 :             src_linesize = EDGE_EMU_LINESIZE;
    1826             :         }
    1827      124886 :         mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my);
    1828             :     } else {
    1829      379222 :         ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0);
    1830      379222 :         mc_func[0][0](dst, linesize, src + y_off * linesize + x_off,
    1831             :                       linesize, block_h, 0, 0);
    1832             :     }
    1833      504108 : }
    1834             : 
    1835             : /**
    1836             :  * chroma MC function
    1837             :  *
    1838             :  * @param s        VP8 decoding context
    1839             :  * @param dst1     target buffer for block data at block position (U plane)
    1840             :  * @param dst2     target buffer for block data at block position (V plane)
    1841             :  * @param ref      reference picture buffer at origin (0, 0)
    1842             :  * @param mv       motion vector (relative to block position) to get pixel data from
    1843             :  * @param x_off    horizontal position of block from origin (0, 0)
    1844             :  * @param y_off    vertical position of block from origin (0, 0)
    1845             :  * @param block_w  width of block (16, 8 or 4)
    1846             :  * @param block_h  height of block (always same as block_w)
    1847             :  * @param width    width of src/dst plane data
    1848             :  * @param height   height of src/dst plane data
    1849             :  * @param linesize size of a single line of plane data, including padding
    1850             :  * @param mc_func  motion compensation function pointers (bilinear or sixtap MC)
    1851             :  */
    1852             : static av_always_inline
    1853      424764 : void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1,
    1854             :                    uint8_t *dst2, ThreadFrame *ref, const VP56mv *mv,
    1855             :                    int x_off, int y_off, int block_w, int block_h,
    1856             :                    int width, int height, ptrdiff_t linesize,
    1857             :                    vp8_mc_func mc_func[3][3])
    1858             : {
    1859      424764 :     uint8_t *src1 = ref->f->data[1], *src2 = ref->f->data[2];
    1860             : 
    1861      424764 :     if (AV_RN32A(mv)) {
    1862       79185 :         int mx = mv->x & 7, mx_idx = subpel_idx[0][mx];
    1863       79185 :         int my = mv->y & 7, my_idx = subpel_idx[0][my];
    1864             : 
    1865       79185 :         x_off += mv->x >> 3;
    1866       79185 :         y_off += mv->y >> 3;
    1867             : 
    1868             :         // edge emulation
    1869       79185 :         src1 += y_off * linesize + x_off;
    1870       79185 :         src2 += y_off * linesize + x_off;
    1871       79185 :         ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0);
    1872       79185 :         if (x_off < mx_idx || x_off >= width  - block_w - subpel_idx[2][mx] ||
    1873       73045 :             y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) {
    1874       36628 :             s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
    1875        9157 :                                      src1 - my_idx * linesize - mx_idx,
    1876             :                                      EDGE_EMU_LINESIZE, linesize,
    1877        9157 :                                      block_w + subpel_idx[1][mx],
    1878        9157 :                                      block_h + subpel_idx[1][my],
    1879             :                                      x_off - mx_idx, y_off - my_idx, width, height);
    1880        9157 :             src1 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
    1881        9157 :             mc_func[my_idx][mx_idx](dst1, linesize, src1, EDGE_EMU_LINESIZE, block_h, mx, my);
    1882             : 
    1883       36628 :             s->vdsp.emulated_edge_mc(td->edge_emu_buffer,
    1884        9157 :                                      src2 - my_idx * linesize - mx_idx,
    1885             :                                      EDGE_EMU_LINESIZE, linesize,
    1886        9157 :                                      block_w + subpel_idx[1][mx],
    1887        9157 :                                      block_h + subpel_idx[1][my],
    1888             :                                      x_off - mx_idx, y_off - my_idx, width, height);
    1889        9157 :             src2 = td->edge_emu_buffer + mx_idx + EDGE_EMU_LINESIZE * my_idx;
    1890        9157 :             mc_func[my_idx][mx_idx](dst2, linesize, src2, EDGE_EMU_LINESIZE, block_h, mx, my);
    1891             :         } else {
    1892       70028 :             mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my);
    1893       70028 :             mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my);
    1894             :         }
    1895             :     } else {
    1896      345579 :         ff_thread_await_progress(ref, (3 + y_off + block_h) >> 3, 0);
    1897      345579 :         mc_func[0][0](dst1, linesize, src1 + y_off * linesize + x_off, linesize, block_h, 0, 0);
    1898      345579 :         mc_func[0][0](dst2, linesize, src2 + y_off * linesize + x_off, linesize, block_h, 0, 0);
    1899             :     }
    1900      424764 : }
    1901             : 
    1902             : static av_always_inline
    1903      398316 : void vp8_mc_part(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
    1904             :                  ThreadFrame *ref_frame, int x_off, int y_off,
    1905             :                  int bx_off, int by_off, int block_w, int block_h,
    1906             :                  int width, int height, VP56mv *mv)
    1907             : {
    1908      398316 :     VP56mv uvmv = *mv;
    1909             : 
    1910             :     /* Y */
    1911      398316 :     vp8_mc_luma(s, td, dst[0] + by_off * s->linesize + bx_off,
    1912             :                 ref_frame, mv, x_off + bx_off, y_off + by_off,
    1913             :                 block_w, block_h, width, height, s->linesize,
    1914      398316 :                 s->put_pixels_tab[block_w == 8]);
    1915             : 
    1916             :     /* U/V */
    1917      398316 :     if (s->profile == 3) {
    1918             :         /* this block only applies VP8; it is safe to check
    1919             :          * only the profile, as VP7 profile <= 1 */
    1920        4775 :         uvmv.x &= ~7;
    1921        4775 :         uvmv.y &= ~7;
    1922             :     }
    1923      398316 :     x_off   >>= 1;
    1924      398316 :     y_off   >>= 1;
    1925      398316 :     bx_off  >>= 1;
    1926      398316 :     by_off  >>= 1;
    1927      398316 :     width   >>= 1;
    1928      398316 :     height  >>= 1;
    1929      398316 :     block_w >>= 1;
    1930      398316 :     block_h >>= 1;
    1931      796632 :     vp8_mc_chroma(s, td, dst[1] + by_off * s->uvlinesize + bx_off,
    1932      398316 :                   dst[2] + by_off * s->uvlinesize + bx_off, ref_frame,
    1933             :                   &uvmv, x_off + bx_off, y_off + by_off,
    1934             :                   block_w, block_h, width, height, s->uvlinesize,
    1935      398316 :                   s->put_pixels_tab[1 + (block_w == 4)]);
    1936      398316 : }
    1937             : 
    1938             : /* Fetch pixels for estimated mv 4 macroblocks ahead.
    1939             :  * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */
    1940             : static av_always_inline
    1941     1329627 : void prefetch_motion(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
    1942             :                      int mb_xy, int ref)
    1943             : {
    1944             :     /* Don't prefetch refs that haven't been used very often this frame. */
    1945     1329627 :     if (s->ref_count[ref - 1] > (mb_xy >> 5)) {
    1946      478745 :         int x_off = mb_x << 4, y_off = mb_y << 4;
    1947      478745 :         int mx = (mb->mv.x >> 2) + x_off + 8;
    1948      478745 :         int my = (mb->mv.y >> 2) + y_off;
    1949      478745 :         uint8_t **src = s->framep[ref]->tf.f->data;
    1950      478745 :         int off = mx + (my + (mb_x & 3) * 4) * s->linesize + 64;
    1951             :         /* For threading, a ff_thread_await_progress here might be useful, but
    1952             :          * it actually slows down the decoder. Since a bad prefetch doesn't
    1953             :          * generate bad decoder output, we don't run it here. */
    1954      478745 :         s->vdsp.prefetch(src[0] + off, s->linesize, 4);
    1955      478745 :         off = (mx >> 1) + ((my >> 1) + (mb_x & 7)) * s->uvlinesize + 64;
    1956      478745 :         s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
    1957             :     }
    1958     1329627 : }
    1959             : 
    1960             : /**
    1961             :  * Apply motion vectors to prediction buffer, chapter 18.
    1962             :  */
    1963             : static av_always_inline
    1964      385877 : void inter_predict(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3],
    1965             :                    VP8Macroblock *mb, int mb_x, int mb_y)
    1966             : {
    1967      385877 :     int x_off = mb_x << 4, y_off = mb_y << 4;
    1968      385877 :     int width = 16 * s->mb_width, height = 16 * s->mb_height;
    1969      385877 :     ThreadFrame *ref = &s->framep[mb->ref_frame]->tf;
    1970      385877 :     VP56mv *bmv = mb->bmv;
    1971             : 
    1972      385877 :     switch (mb->partitioning) {
    1973      367552 :     case VP8_SPLITMVMODE_NONE:
    1974      367552 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    1975             :                     0, 0, 16, 16, width, height, &mb->mv);
    1976      367552 :         break;
    1977        6612 :     case VP8_SPLITMVMODE_4x4: {
    1978             :         int x, y;
    1979             :         VP56mv uvmv;
    1980             : 
    1981             :         /* Y */
    1982       33060 :         for (y = 0; y < 4; y++) {
    1983      132240 :             for (x = 0; x < 4; x++) {
    1984      423168 :                 vp8_mc_luma(s, td, dst[0] + 4 * y * s->linesize + x * 4,
    1985      105792 :                             ref, &bmv[4 * y + x],
    1986      211584 :                             4 * x + x_off, 4 * y + y_off, 4, 4,
    1987             :                             width, height, s->linesize,
    1988      105792 :                             s->put_pixels_tab[2]);
    1989             :             }
    1990             :         }
    1991             : 
    1992             :         /* U/V */
    1993        6612 :         x_off  >>= 1;
    1994        6612 :         y_off  >>= 1;
    1995        6612 :         width  >>= 1;
    1996        6612 :         height >>= 1;
    1997       19836 :         for (y = 0; y < 2; y++) {
    1998       39672 :             for (x = 0; x < 2; x++) {
    1999       79344 :                 uvmv.x = mb->bmv[2 * y       * 4 + 2 * x    ].x +
    2000       52896 :                          mb->bmv[2 * y       * 4 + 2 * x + 1].x +
    2001       52896 :                          mb->bmv[(2 * y + 1) * 4 + 2 * x    ].x +
    2002       26448 :                          mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].x;
    2003       79344 :                 uvmv.y = mb->bmv[2 * y       * 4 + 2 * x    ].y +
    2004       52896 :                          mb->bmv[2 * y       * 4 + 2 * x + 1].y +
    2005       52896 :                          mb->bmv[(2 * y + 1) * 4 + 2 * x    ].y +
    2006       26448 :                          mb->bmv[(2 * y + 1) * 4 + 2 * x + 1].y;
    2007       26448 :                 uvmv.x = (uvmv.x + 2 + FF_SIGNBIT(uvmv.x)) >> 2;
    2008       26448 :                 uvmv.y = (uvmv.y + 2 + FF_SIGNBIT(uvmv.y)) >> 2;
    2009       26448 :                 if (s->profile == 3) {
    2010         492 :                     uvmv.x &= ~7;
    2011         492 :                     uvmv.y &= ~7;
    2012             :                 }
    2013      105792 :                 vp8_mc_chroma(s, td, dst[1] + 4 * y * s->uvlinesize + x * 4,
    2014       26448 :                               dst[2] + 4 * y * s->uvlinesize + x * 4, ref,
    2015       52896 :                               &uvmv, 4 * x + x_off, 4 * y + y_off, 4, 4,
    2016             :                               width, height, s->uvlinesize,
    2017       26448 :                               s->put_pixels_tab[2]);
    2018             :             }
    2019             :         }
    2020        6612 :         break;
    2021             :     }
    2022        4909 :     case VP8_SPLITMVMODE_16x8:
    2023        4909 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2024             :                     0, 0, 16, 8, width, height, &bmv[0]);
    2025        4909 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2026             :                     0, 8, 16, 8, width, height, &bmv[1]);
    2027        4909 :         break;
    2028        3135 :     case VP8_SPLITMVMODE_8x16:
    2029        3135 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2030             :                     0, 0, 8, 16, width, height, &bmv[0]);
    2031        3135 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2032             :                     8, 0, 8, 16, width, height, &bmv[1]);
    2033        3135 :         break;
    2034        3669 :     case VP8_SPLITMVMODE_8x8:
    2035        3669 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2036             :                     0, 0, 8, 8, width, height, &bmv[0]);
    2037        3669 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2038             :                     8, 0, 8, 8, width, height, &bmv[1]);
    2039        3669 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2040             :                     0, 8, 8, 8, width, height, &bmv[2]);
    2041        3669 :         vp8_mc_part(s, td, dst, ref, x_off, y_off,
    2042             :                     8, 8, 8, 8, width, height, &bmv[3]);
    2043        3669 :         break;
    2044             :     }
    2045      385877 : }
    2046             : 
    2047             : static av_always_inline
    2048       76840 : void idct_mb(VP8Context *s, VP8ThreadData *td, uint8_t *dst[3], VP8Macroblock *mb)
    2049             : {
    2050             :     int x, y, ch;
    2051             : 
    2052       76840 :     if (mb->mode != MODE_I4x4) {
    2053       59625 :         uint8_t *y_dst = dst[0];
    2054      298125 :         for (y = 0; y < 4; y++) {
    2055      238500 :             uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[y]);
    2056      238500 :             if (nnz4) {
    2057      175833 :                 if (nnz4 & ~0x01010101) {
    2058      155935 :                     for (x = 0; x < 4; x++) {
    2059      155935 :                         if ((uint8_t) nnz4 == 1)
    2060      108102 :                             s->vp8dsp.vp8_idct_dc_add(y_dst + 4 * x,
    2061       54051 :                                                       td->block[y][x],
    2062             :                                                       s->linesize);
    2063      101884 :                         else if ((uint8_t) nnz4 > 1)
    2064      148856 :                             s->vp8dsp.vp8_idct_add(y_dst + 4 * x,
    2065       74428 :                                                    td->block[y][x],
    2066             :                                                    s->linesize);
    2067      155935 :                         nnz4 >>= 8;
    2068      155935 :                         if (!nnz4)
    2069       44887 :                             break;
    2070             :                     }
    2071             :                 } else {
    2072      130946 :                     s->vp8dsp.vp8_idct_dc_add4y(y_dst, td->block[y], s->linesize);
    2073             :                 }
    2074             :             }
    2075      238500 :             y_dst += 4 * s->linesize;
    2076             :         }
    2077             :     }
    2078             : 
    2079      230520 :     for (ch = 0; ch < 2; ch++) {
    2080      153680 :         uint32_t nnz4 = AV_RL32(td->non_zero_count_cache[4 + ch]);
    2081      153680 :         if (nnz4) {
    2082       60025 :             uint8_t *ch_dst = dst[1 + ch];
    2083       60025 :             if (nnz4 & ~0x01010101) {
    2084       55530 :                 for (y = 0; y < 2; y++) {
    2085      131913 :                     for (x = 0; x < 2; x++) {
    2086      106337 :                         if ((uint8_t) nnz4 == 1)
    2087       32108 :                             s->vp8dsp.vp8_idct_dc_add(ch_dst + 4 * x,
    2088       16054 :                                                       td->block[4 + ch][(y << 1) + x],
    2089             :                                                       s->uvlinesize);
    2090       90283 :                         else if ((uint8_t) nnz4 > 1)
    2091      133844 :                             s->vp8dsp.vp8_idct_add(ch_dst + 4 * x,
    2092       66922 :                                                    td->block[4 + ch][(y << 1) + x],
    2093             :                                                    s->uvlinesize);
    2094      106337 :                         nnz4 >>= 8;
    2095      106337 :                         if (!nnz4)
    2096       29954 :                             goto chroma_idct_end;
    2097             :                     }
    2098       25576 :                     ch_dst += 4 * s->uvlinesize;
    2099             :                 }
    2100             :             } else {
    2101       30071 :                 s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, td->block[4 + ch], s->uvlinesize);
    2102             :             }
    2103             :         }
    2104      247335 : chroma_idct_end:
    2105             :         ;
    2106             :     }
    2107       76840 : }
    2108             : 
    2109             : static av_always_inline
    2110      424229 : void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb,
    2111             :                          VP8FilterStrength *f, int is_vp7)
    2112             : {
    2113             :     int interior_limit, filter_level;
    2114             : 
    2115      424229 :     if (s->segmentation.enabled) {
    2116      102219 :         filter_level = s->segmentation.filter_level[mb->segment];
    2117      102219 :         if (!s->segmentation.absolute_vals)
    2118       98895 :             filter_level += s->filter.level;
    2119             :     } else
    2120      322010 :         filter_level = s->filter.level;
    2121             : 
    2122      424229 :     if (s->lf_delta.enabled) {
    2123      416875 :         filter_level += s->lf_delta.ref[mb->ref_frame];
    2124      416875 :         filter_level += s->lf_delta.mode[mb->mode];
    2125             :     }
    2126             : 
    2127      424229 :     filter_level = av_clip_uintp2(filter_level, 6);
    2128             : 
    2129      424229 :     interior_limit = filter_level;
    2130      424229 :     if (s->filter.sharpness) {
    2131        2772 :         interior_limit >>= (s->filter.sharpness + 3) >> 2;
    2132        2772 :         interior_limit = FFMIN(interior_limit, 9 - s->filter.sharpness);
    2133             :     }
    2134      424229 :     interior_limit = FFMAX(interior_limit, 1);
    2135             : 
    2136      424229 :     f->filter_level = filter_level;
    2137      424229 :     f->inner_limit = interior_limit;
    2138      777260 :     f->inner_filter = is_vp7 || !mb->skip || mb->mode == MODE_I4x4 ||
    2139      353031 :                       mb->mode == VP8_MVMODE_SPLIT;
    2140      424229 : }
    2141             : 
    2142             : static av_always_inline
    2143      416507 : void filter_mb(VP8Context *s, uint8_t *dst[3], VP8FilterStrength *f,
    2144             :                int mb_x, int mb_y, int is_vp7)
    2145             : {
    2146             :     int mbedge_lim, bedge_lim_y, bedge_lim_uv, hev_thresh;
    2147      416507 :     int filter_level = f->filter_level;
    2148      416507 :     int inner_limit = f->inner_limit;
    2149      416507 :     int inner_filter = f->inner_filter;
    2150      416507 :     ptrdiff_t linesize   = s->linesize;
    2151      416507 :     ptrdiff_t uvlinesize = s->uvlinesize;
    2152             :     static const uint8_t hev_thresh_lut[2][64] = {
    2153             :         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
    2154             :           2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
    2155             :           3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
    2156             :           3, 3, 3, 3 },
    2157             :         { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
    2158             :           1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    2159             :           2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
    2160             :           2, 2, 2, 2 }
    2161             :     };
    2162             : 
    2163      416507 :     if (!filter_level)
    2164        5858 :         return;
    2165             : 
    2166      410649 :     if (is_vp7) {
    2167        6600 :         bedge_lim_y  = filter_level;
    2168        6600 :         bedge_lim_uv = filter_level * 2;
    2169        6600 :         mbedge_lim   = filter_level + 2;
    2170             :     } else {
    2171      404049 :         bedge_lim_y  =
    2172      404049 :         bedge_lim_uv = filter_level * 2 + inner_limit;
    2173      404049 :         mbedge_lim   = bedge_lim_y + 4;
    2174             :     }
    2175             : 
    2176      410649 :     hev_thresh = hev_thresh_lut[s->keyframe][filter_level];
    2177             : 
    2178      410649 :     if (mb_x) {
    2179      396544 :         s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize,
    2180             :                                        mbedge_lim, inner_limit, hev_thresh);
    2181      396544 :         s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize,
    2182             :                                        mbedge_lim, inner_limit, hev_thresh);
    2183             :     }
    2184             : 
    2185             : #define H_LOOP_FILTER_16Y_INNER(cond)                                         \
    2186             :     if (cond && inner_filter) {                                               \
    2187             :         s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] +  4, linesize,           \
    2188             :                                              bedge_lim_y, inner_limit,        \
    2189             :                                              hev_thresh);                     \
    2190             :         s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] +  8, linesize,           \
    2191             :                                              bedge_lim_y, inner_limit,        \
    2192             :                                              hev_thresh);                     \
    2193             :         s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0] + 12, linesize,           \
    2194             :                                              bedge_lim_y, inner_limit,        \
    2195             :                                              hev_thresh);                     \
    2196             :         s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] +  4, dst[2] + 4,         \
    2197             :                                              uvlinesize,  bedge_lim_uv,       \
    2198             :                                              inner_limit, hev_thresh);        \
    2199             :     }
    2200             : 
    2201      410649 :     H_LOOP_FILTER_16Y_INNER(!is_vp7)
    2202             : 
    2203      410649 :     if (mb_y) {
    2204      389635 :         s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize,
    2205             :                                        mbedge_lim, inner_limit, hev_thresh);
    2206      389635 :         s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize,
    2207             :                                        mbedge_lim, inner_limit, hev_thresh);
    2208             :     }
    2209             : 
    2210      410649 :     if (inner_filter) {
    2211       69204 :         s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] +  4 * linesize,
    2212             :                                              linesize, bedge_lim_y,
    2213             :                                              inner_limit, hev_thresh);
    2214       69204 :         s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] +  8 * linesize,
    2215             :                                              linesize, bedge_lim_y,
    2216             :                                              inner_limit, hev_thresh);
    2217       69204 :         s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0] + 12 * linesize,
    2218             :                                              linesize, bedge_lim_y,
    2219             :                                              inner_limit, hev_thresh);
    2220      138408 :         s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] +  4 * uvlinesize,
    2221       69204 :                                              dst[2] +  4 * uvlinesize,
    2222             :                                              uvlinesize, bedge_lim_uv,
    2223             :                                              inner_limit, hev_thresh);
    2224             :     }
    2225             : 
    2226      410649 :     H_LOOP_FILTER_16Y_INNER(is_vp7)
    2227             : }
    2228             : 
    2229             : static av_always_inline
    2230        7722 : void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8FilterStrength *f,
    2231             :                       int mb_x, int mb_y)
    2232             : {
    2233             :     int mbedge_lim, bedge_lim;
    2234        7722 :     int filter_level = f->filter_level;
    2235        7722 :     int inner_limit  = f->inner_limit;
    2236        7722 :     int inner_filter = f->inner_filter;
    2237        7722 :     ptrdiff_t linesize = s->linesize;
    2238             : 
    2239        7722 :     if (!filter_level)
    2240         332 :         return;
    2241             : 
    2242        7390 :     bedge_lim  = 2 * filter_level + inner_limit;
    2243        7390 :     mbedge_lim = bedge_lim + 4;
    2244             : 
    2245        7390 :     if (mb_x)
    2246        6713 :         s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim);
    2247        7390 :     if (inner_filter) {
    2248        3523 :         s->vp8dsp.vp8_h_loop_filter_simple(dst +  4, linesize, bedge_lim);
    2249        3523 :         s->vp8dsp.vp8_h_loop_filter_simple(dst +  8, linesize, bedge_lim);
    2250        3523 :         s->vp8dsp.vp8_h_loop_filter_simple(dst + 12, linesize, bedge_lim);
    2251             :     }
    2252             : 
    2253        7390 :     if (mb_y)
    2254        6609 :         s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim);
    2255        7390 :     if (inner_filter) {
    2256        3523 :         s->vp8dsp.vp8_v_loop_filter_simple(dst +  4 * linesize, linesize, bedge_lim);
    2257        3523 :         s->vp8dsp.vp8_v_loop_filter_simple(dst +  8 * linesize, linesize, bedge_lim);
    2258        3523 :         s->vp8dsp.vp8_v_loop_filter_simple(dst + 12 * linesize, linesize, bedge_lim);
    2259             :     }
    2260             : }
    2261             : 
    2262             : #define MARGIN (16 << 2)
    2263             : static av_always_inline
    2264          30 : void vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe,
    2265             :                                     VP8Frame *prev_frame, int is_vp7)
    2266             : {
    2267          30 :     VP8Context *s = avctx->priv_data;
    2268             :     int mb_x, mb_y;
    2269             : 
    2270          30 :     s->mv_bounds.mv_min.y = -MARGIN;
    2271          30 :     s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
    2272         360 :     for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
    2273         660 :         VP8Macroblock *mb = s->macroblocks_base +
    2274         330 :                             ((s->mb_width + 1) * (mb_y + 1) + 1);
    2275         330 :         int mb_xy = mb_y * s->mb_width;
    2276             : 
    2277         330 :         AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
    2278             : 
    2279         330 :         s->mv_bounds.mv_min.x = -MARGIN;
    2280         330 :         s->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
    2281        6930 :         for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
    2282        6600 :             if (mb_y == 0)
    2283         600 :                 AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,
    2284             :                          DC_PRED * 0x01010101);
    2285       12980 :             decode_mb_mode(s, &s->mv_bounds, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
    2286        6380 :                            prev_frame && prev_frame->seg_map ?
    2287        6380 :                            prev_frame->seg_map->data + mb_xy : NULL, 1, is_vp7);
    2288        6600 :             s->mv_bounds.mv_min.x -= 64;
    2289        6600 :             s->mv_bounds.mv_max.x -= 64;
    2290             :         }
    2291         330 :         s->mv_bounds.mv_min.y -= 64;
    2292         330 :         s->mv_bounds.mv_max.y -= 64;
    2293             :     }
    2294          30 : }
    2295             : 
    2296          30 : static void vp7_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame,
    2297             :                                    VP8Frame *prev_frame)
    2298             : {
    2299          30 :     vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP7);
    2300          30 : }
    2301             : 
    2302           0 : static void vp8_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *cur_frame,
    2303             :                                    VP8Frame *prev_frame)
    2304             : {
    2305           0 :     vp78_decode_mv_mb_modes(avctx, cur_frame, prev_frame, IS_VP8);
    2306           0 : }
    2307             : 
    2308             : #if HAVE_THREADS
    2309             : #define check_thread_pos(td, otd, mb_x_check, mb_y_check)                     \
    2310             :     do {                                                                      \
    2311             :         int tmp = (mb_y_check << 16) | (mb_x_check & 0xFFFF);                 \
    2312             :         if (atomic_load(&otd->thread_mb_pos) < tmp) {                         \
    2313             :             pthread_mutex_lock(&otd->lock);                                   \
    2314             :             atomic_store(&td->wait_mb_pos, tmp);                              \
    2315             :             do {                                                              \
    2316             :                 if (atomic_load(&otd->thread_mb_pos) >= tmp)                  \
    2317             :                     break;                                                    \
    2318             :                 pthread_cond_wait(&otd->cond, &otd->lock);                    \
    2319             :             } while (1);                                                      \
    2320             :             atomic_store(&td->wait_mb_pos, INT_MAX);                          \
    2321             :             pthread_mutex_unlock(&otd->lock);                                 \
    2322             :         }                                                                     \
    2323             :     } while (0)
    2324             : 
    2325             : #define update_pos(td, mb_y, mb_x)                                            \
    2326             :     do {                                                                      \
    2327             :         int pos              = (mb_y << 16) | (mb_x & 0xFFFF);                \
    2328             :         int sliced_threading = (avctx->active_thread_type == FF_THREAD_SLICE) && \
    2329             :                                (num_jobs > 1);                                \
    2330             :         int is_null          = !next_td || !prev_td;                          \
    2331             :         int pos_check        = (is_null) ? 1 :                                \
    2332             :             (next_td != td && pos >= atomic_load(&next_td->wait_mb_pos)) ||   \
    2333             :             (prev_td != td && pos >= atomic_load(&prev_td->wait_mb_pos));     \
    2334             :         atomic_store(&td->thread_mb_pos, pos);                                \
    2335             :         if (sliced_threading && pos_check) {                                  \
    2336             :             pthread_mutex_lock(&td->lock);                                    \
    2337             :             pthread_cond_broadcast(&td->cond);                                \
    2338             :             pthread_mutex_unlock(&td->lock);                                  \
    2339             :         }                                                                     \
    2340             :     } while (0)
    2341             : #else
    2342             : #define check_thread_pos(td, otd, mb_x_check, mb_y_check) while(0)
    2343             : #define update_pos(td, mb_y, mb_x) while(0)
    2344             : #endif
    2345             : 
    2346       16181 : static av_always_inline int decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
    2347             :                                         int jobnr, int threadnr, int is_vp7)
    2348             : {
    2349       16181 :     VP8Context *s = avctx->priv_data;
    2350       16181 :     VP8ThreadData *prev_td, *next_td, *td = &s->thread_data[threadnr];
    2351       16181 :     int mb_y = atomic_load(&td->thread_mb_pos) >> 16;
    2352       16181 :     int mb_x, mb_xy = mb_y * s->mb_width;
    2353       16181 :     int num_jobs = s->num_jobs;
    2354       16181 :     VP8Frame *curframe = s->curframe, *prev_frame = s->prev_frame;
    2355       16181 :     VP56RangeCoder *c  = &s->coeff_partition[mb_y & (s->num_coeff_partitions - 1)];
    2356             :     VP8Macroblock *mb;
    2357       48543 :     uint8_t *dst[3] = {
    2358       16181 :         curframe->tf.f->data[0] + 16 * mb_y * s->linesize,
    2359       16181 :         curframe->tf.f->data[1] +  8 * mb_y * s->uvlinesize,
    2360       16181 :         curframe->tf.f->data[2] +  8 * mb_y * s->uvlinesize
    2361             :     };
    2362             : 
    2363       16181 :     if (c->end <= c->buffer && c->bits >= 0)
    2364           0 :          return AVERROR_INVALIDDATA;
    2365             : 
    2366       16181 :     if (mb_y == 0)
    2367        1142 :         prev_td = td;
    2368             :     else
    2369       15039 :         prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
    2370       16181 :     if (mb_y == s->mb_height - 1)
    2371        1142 :         next_td = td;
    2372             :     else
    2373       15039 :         next_td = &s->thread_data[(jobnr + 1) % num_jobs];
    2374       16181 :     if (s->mb_layout == 1)
    2375         330 :         mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
    2376             :     else {
    2377             :         // Make sure the previous frame has read its segmentation map,
    2378             :         // if we re-use the same map.
    2379       21449 :         if (prev_frame && s->segmentation.enabled &&
    2380        5598 :             !s->segmentation.update_map)
    2381        5025 :             ff_thread_await_progress(&prev_frame->tf, mb_y, 0);
    2382       15851 :         mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
    2383       15851 :         memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
    2384       15851 :         AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
    2385             :     }
    2386             : 
    2387       16181 :     if (!is_vp7 || mb_y == 0)
    2388       15881 :         memset(td->left_nnz, 0, sizeof(td->left_nnz));
    2389             : 
    2390       16181 :     td->mv_bounds.mv_min.x = -MARGIN;
    2391       16181 :     td->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
    2392             : 
    2393      459390 :     for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
    2394      443209 :         if (c->end <= c->buffer && c->bits >= 0)
    2395           0 :             return AVERROR_INVALIDDATA;
    2396             :         // Wait for previous thread to read mb_x+2, and reach mb_y-1.
    2397      443209 :         if (prev_td != td) {
    2398           0 :             if (threadnr != 0) {
    2399           0 :                 check_thread_pos(td, prev_td,
    2400             :                                  mb_x + (is_vp7 ? 2 : 1),
    2401             :                                  mb_y - (is_vp7 ? 2 : 1));
    2402             :             } else {
    2403           0 :                 check_thread_pos(td, prev_td,
    2404             :                                  mb_x + (is_vp7 ? 2 : 1) + s->mb_width + 3,
    2405             :                                  mb_y - (is_vp7 ? 2 : 1));
    2406             :             }
    2407             :         }
    2408             : 
    2409      443209 :         s->vdsp.prefetch(dst[0] + (mb_x & 3) * 4 * s->linesize + 64,
    2410             :                          s->linesize, 4);
    2411     1329627 :         s->vdsp.prefetch(dst[1] + (mb_x & 7) * s->uvlinesize + 64,
    2412      886418 :                          dst[2] - dst[1], 2);
    2413             : 
    2414      443209 :         if (!s->mb_layout)
    2415      842558 :             decode_mb_mode(s, &td->mv_bounds, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
    2416      405949 :                            prev_frame && prev_frame->seg_map ?
    2417      405949 :                            prev_frame->seg_map->data + mb_xy : NULL, 0, is_vp7);
    2418             : 
    2419      443209 :         prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
    2420             : 
    2421      443209 :         if (!mb->skip)
    2422       83297 :             decode_mb_coeffs(s, td, c, mb, s->top_nnz[mb_x], td->left_nnz, is_vp7);
    2423             : 
    2424      443209 :         if (mb->mode <= MODE_I4x4)
    2425       57332 :             intra_predict(s, td, dst, mb, mb_x, mb_y, is_vp7);
    2426             :         else
    2427      385877 :             inter_predict(s, td, dst, mb, mb_x, mb_y);
    2428             : 
    2429      443209 :         prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);
    2430             : 
    2431      443209 :         if (!mb->skip) {
    2432       76840 :             idct_mb(s, td, dst, mb);
    2433             :         } else {
    2434      366369 :             AV_ZERO64(td->left_nnz);
    2435      366369 :             AV_WN64(s->top_nnz[mb_x], 0);   // array of 9, so unaligned
    2436             : 
    2437             :             /* Reset DC block predictors if they would exist
    2438             :              * if the mb had coefficients */
    2439      366369 :             if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
    2440      362642 :                 td->left_nnz[8]     = 0;
    2441      362642 :                 s->top_nnz[mb_x][8] = 0;
    2442             :             }
    2443             :         }
    2444             : 
    2445      443209 :         if (s->deblock_filter)
    2446      424229 :             filter_level_for_mb(s, mb, &td->filter_strength[mb_x], is_vp7);
    2447             : 
    2448      443209 :         if (s->deblock_filter && num_jobs != 1 && threadnr == num_jobs - 1) {
    2449           0 :             if (s->filter.simple)
    2450           0 :                 backup_mb_border(s->top_border[mb_x + 1], dst[0],
    2451             :                                  NULL, NULL, s->linesize, 0, 1);
    2452             :             else
    2453           0 :                 backup_mb_border(s->top_border[mb_x + 1], dst[0],
    2454             :                                  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
    2455             :         }
    2456             : 
    2457      443209 :         prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2);
    2458             : 
    2459      443209 :         dst[0]      += 16;
    2460      443209 :         dst[1]      += 8;
    2461      443209 :         dst[2]      += 8;
    2462      443209 :         td->mv_bounds.mv_min.x -= 64;
    2463      443209 :         td->mv_bounds.mv_max.x -= 64;
    2464             : 
    2465      443209 :         if (mb_x == s->mb_width + 1) {
    2466           0 :             update_pos(td, mb_y, s->mb_width + 3);
    2467             :         } else {
    2468      443209 :             update_pos(td, mb_y, mb_x);
    2469             :         }
    2470             :     }
    2471       16181 :     return 0;
    2472             : }
    2473             : 
    2474         330 : static int vp7_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
    2475             :                                         int jobnr, int threadnr)
    2476             : {
    2477         330 :     return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 1);
    2478             : }
    2479             : 
    2480       15851 : static int vp8_decode_mb_row_no_filter(AVCodecContext *avctx, void *tdata,
    2481             :                                         int jobnr, int threadnr)
    2482             : {
    2483       15851 :     return decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr, 0);
    2484             : }
    2485             : 
    2486       15115 : static av_always_inline void filter_mb_row(AVCodecContext *avctx, void *tdata,
    2487             :                               int jobnr, int threadnr, int is_vp7)
    2488             : {
    2489       15115 :     VP8Context *s = avctx->priv_data;
    2490       15115 :     VP8ThreadData *td = &s->thread_data[threadnr];
    2491       15115 :     int mb_x, mb_y = atomic_load(&td->thread_mb_pos) >> 16, num_jobs = s->num_jobs;
    2492       15115 :     AVFrame *curframe = s->curframe->tf.f;
    2493             :     VP8Macroblock *mb;
    2494             :     VP8ThreadData *prev_td, *next_td;
    2495       45345 :     uint8_t *dst[3] = {
    2496       15115 :         curframe->data[0] + 16 * mb_y * s->linesize,
    2497       15115 :         curframe->data[1] +  8 * mb_y * s->uvlinesize,
    2498       15115 :         curframe->data[2] +  8 * mb_y * s->uvlinesize
    2499             :     };
    2500             : 
    2501       15115 :     if (s->mb_layout == 1)
    2502         330 :         mb = s->macroblocks_base + ((s->mb_width + 1) * (mb_y + 1) + 1);
    2503             :     else
    2504       14785 :         mb = s->macroblocks + (s->mb_height - mb_y - 1) * 2;
    2505             : 
    2506       15115 :     if (mb_y == 0)
    2507        1034 :         prev_td = td;
    2508             :     else
    2509       14081 :         prev_td = &s->thread_data[(jobnr + num_jobs - 1) % num_jobs];
    2510       15115 :     if (mb_y == s->mb_height - 1)
    2511        1034 :         next_td = td;
    2512             :     else
    2513       14081 :         next_td = &s->thread_data[(jobnr + 1) % num_jobs];
    2514             : 
    2515      439344 :     for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb++) {
    2516      424229 :         VP8FilterStrength *f = &td->filter_strength[mb_x];
    2517      424229 :         if (prev_td != td)
    2518           0 :             check_thread_pos(td, prev_td,
    2519             :                              (mb_x + 1) + (s->mb_width + 3), mb_y - 1);
    2520      424229 :         if (next_td != td)
    2521           0 :             if (next_td != &s->thread_data[0])
    2522           0 :                 check_thread_pos(td, next_td, mb_x + 1, mb_y + 1);
    2523             : 
    2524      424229 :         if (num_jobs == 1) {
    2525      424229 :             if (s->filter.simple)
    2526        7722 :                 backup_mb_border(s->top_border[mb_x + 1], dst[0],
    2527             :                                  NULL, NULL, s->linesize, 0, 1);
    2528             :             else
    2529      416507 :                 backup_mb_border(s->top_border[mb_x + 1], dst[0],
    2530             :                                  dst[1], dst[2], s->linesize, s->uvlinesize, 0);
    2531             :         }
    2532             : 
    2533      424229 :         if (s->filter.simple)
    2534        7722 :             filter_mb_simple(s, dst[0], f, mb_x, mb_y);
    2535             :         else
    2536      416507 :             filter_mb(s, dst, f, mb_x, mb_y, is_vp7);
    2537      424229 :         dst[0] += 16;
    2538      424229 :         dst[1] += 8;
    2539      424229 :         dst[2] += 8;
    2540             : 
    2541      424229 :         update_pos(td, mb_y, (s->mb_width + 3) + mb_x);
    2542             :     }
    2543       15115 : }
    2544             : 
    2545         330 : static void vp7_filter_mb_row(AVCodecContext *avctx, void *tdata,
    2546             :                               int jobnr, int threadnr)
    2547             : {
    2548         330 :     filter_mb_row(avctx, tdata, jobnr, threadnr, 1);
    2549         330 : }
    2550             : 
    2551       14785 : static void vp8_filter_mb_row(AVCodecContext *avctx, void *tdata,
    2552             :                               int jobnr, int threadnr)
    2553             : {
    2554       14785 :     filter_mb_row(avctx, tdata, jobnr, threadnr, 0);
    2555       14785 : }
    2556             : 
    2557             : static av_always_inline
    2558        1142 : int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
    2559             :                               int threadnr, int is_vp7)
    2560             : {
    2561        1142 :     VP8Context *s = avctx->priv_data;
    2562        1142 :     VP8ThreadData *td = &s->thread_data[jobnr];
    2563        1142 :     VP8ThreadData *next_td = NULL, *prev_td = NULL;
    2564        1142 :     VP8Frame *curframe = s->curframe;
    2565        1142 :     int mb_y, num_jobs = s->num_jobs;
    2566             :     int ret;
    2567             : 
    2568        1142 :     td->thread_nr = threadnr;
    2569        1142 :     td->mv_bounds.mv_min.y   = -MARGIN - 64 * threadnr;
    2570        1142 :     td->mv_bounds.mv_max.y   = ((s->mb_height - 1) << 6) + MARGIN - 64 * threadnr;
    2571       17323 :     for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) {
    2572       16181 :         atomic_store(&td->thread_mb_pos, mb_y << 16);
    2573       16181 :         ret = s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr);
    2574       16181 :         if (ret < 0) {
    2575           0 :             update_pos(td, s->mb_height, INT_MAX & 0xFFFF);
    2576           0 :             return ret;
    2577             :         }
    2578       16181 :         if (s->deblock_filter)
    2579       15115 :             s->filter_mb_row(avctx, tdata, jobnr, threadnr);
    2580       16181 :         update_pos(td, mb_y, INT_MAX & 0xFFFF);
    2581             : 
    2582       16181 :         td->mv_bounds.mv_min.y -= 64 * num_jobs;
    2583       16181 :         td->mv_bounds.mv_max.y -= 64 * num_jobs;
    2584             : 
    2585       16181 :         if (avctx->active_thread_type == FF_THREAD_FRAME)
    2586           0 :             ff_thread_report_progress(&curframe->tf, mb_y, 0);
    2587             :     }
    2588             : 
    2589        1142 :     return 0;
    2590             : }
    2591             : 
    2592          30 : static int vp7_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
    2593             :                                     int jobnr, int threadnr)
    2594             : {
    2595          30 :     return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP7);
    2596             : }
    2597             : 
    2598        1112 : static int vp8_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata,
    2599             :                                     int jobnr, int threadnr)
    2600             : {
    2601        1112 :     return vp78_decode_mb_row_sliced(avctx, tdata, jobnr, threadnr, IS_VP8);
    2602             : }
    2603             : 
    2604             : static av_always_inline
    2605        1142 : int vp78_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
    2606             :                       AVPacket *avpkt, int is_vp7)
    2607             : {
    2608        1142 :     VP8Context *s = avctx->priv_data;
    2609             :     int ret, i, referenced, num_jobs;
    2610             :     enum AVDiscard skip_thresh;
    2611        1142 :     VP8Frame *av_uninit(curframe), *prev_frame;
    2612             : 
    2613        1142 :     if (is_vp7)
    2614          30 :         ret = vp7_decode_frame_header(s, avpkt->data, avpkt->size);
    2615             :     else
    2616        1112 :         ret = vp8_decode_frame_header(s, avpkt->data, avpkt->size);
    2617             : 
    2618        1142 :     if (ret < 0)
    2619           0 :         goto err;
    2620             : 
    2621        1142 :     if (s->actually_webp) {
    2622             :         // avctx->pix_fmt already set in caller.
    2623        1136 :     } else if (!is_vp7 && s->pix_fmt == AV_PIX_FMT_NONE) {
    2624           0 :         s->pix_fmt = get_pixel_format(s);
    2625           0 :         if (s->pix_fmt < 0) {
    2626           0 :             ret = AVERROR(EINVAL);
    2627           0 :             goto err;
    2628             :         }
    2629           0 :         avctx->pix_fmt = s->pix_fmt;
    2630             :     }
    2631             : 
    2632        1142 :     prev_frame = s->framep[VP56_FRAME_CURRENT];
    2633             : 
    2634        1153 :     referenced = s->update_last || s->update_golden == VP56_FRAME_CURRENT ||
    2635          11 :                  s->update_altref == VP56_FRAME_CURRENT;
    2636             : 
    2637        1142 :     skip_thresh = !referenced ? AVDISCARD_NONREF
    2638        1142 :                               : !s->keyframe ? AVDISCARD_NONKEY
    2639             :                                              : AVDISCARD_ALL;
    2640             : 
    2641        1142 :     if (avctx->skip_frame >= skip_thresh) {
    2642           0 :         s->invisible = 1;
    2643           0 :         memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
    2644           0 :         goto skip_decode;
    2645             :     }
    2646        1142 :     s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;
    2647             : 
    2648             :     // release no longer referenced frames
    2649        6852 :     for (i = 0; i < 5; i++)
    2650        9536 :         if (s->frames[i].tf.f->buf[0] &&
    2651        6547 :             &s->frames[i] != prev_frame &&
    2652        5431 :             &s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
    2653        4451 :             &s->frames[i] != s->framep[VP56_FRAME_GOLDEN]   &&
    2654        1741 :             &s->frames[i] != s->framep[VP56_FRAME_GOLDEN2])
    2655        1033 :             vp8_release_frame(s, &s->frames[i]);
    2656             : 
    2657        1142 :     curframe = s->framep[VP56_FRAME_CURRENT] = vp8_find_free_buffer(s);
    2658             : 
    2659        1142 :     if (!s->colorspace)
    2660        1142 :         avctx->colorspace = AVCOL_SPC_BT470BG;
    2661        1142 :     if (s->fullrange)
    2662           0 :         avctx->color_range = AVCOL_RANGE_JPEG;
    2663             :     else
    2664        1142 :         avctx->color_range = AVCOL_RANGE_MPEG;
    2665             : 
    2666             :     /* Given that arithmetic probabilities are updated every frame, it's quite
    2667             :      * likely that the values we have on a random interframe are complete
    2668             :      * junk if we didn't start decode on a keyframe. So just don't display
    2669             :      * anything rather than junk. */
    2670        2235 :     if (!s->keyframe && (!s->framep[VP56_FRAME_PREVIOUS] ||
    2671        2186 :                          !s->framep[VP56_FRAME_GOLDEN]   ||
    2672        1093 :                          !s->framep[VP56_FRAME_GOLDEN2])) {
    2673           0 :         av_log(avctx, AV_LOG_WARNING,
    2674             :                "Discarding interframe without a prior keyframe!\n");
    2675           0 :         ret = AVERROR_INVALIDDATA;
    2676           0 :         goto err;
    2677             :     }
    2678             : 
    2679        1142 :     curframe->tf.f->key_frame = s->keyframe;
    2680        2284 :     curframe->tf.f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
    2681        1142 :                                             : AV_PICTURE_TYPE_P;
    2682        1142 :     if ((ret = vp8_alloc_frame(s, curframe, referenced)) < 0)
    2683           0 :         goto err;
    2684             : 
    2685             :     // check if golden and altref are swapped
    2686        1142 :     if (s->update_altref != VP56_FRAME_NONE)
    2687         196 :         s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[s->update_altref];
    2688             :     else
    2689         946 :         s->next_framep[VP56_FRAME_GOLDEN2] = s->framep[VP56_FRAME_GOLDEN2];
    2690             : 
    2691        1142 :     if (s->update_golden != VP56_FRAME_NONE)
    2692         139 :         s->next_framep[VP56_FRAME_GOLDEN] = s->framep[s->update_golden];
    2693             :     else
    2694        1003 :         s->next_framep[VP56_FRAME_GOLDEN] = s->framep[VP56_FRAME_GOLDEN];
    2695             : 
    2696        1142 :     if (s->update_last)
    2697        1131 :         s->next_framep[VP56_FRAME_PREVIOUS] = curframe;
    2698             :     else
    2699          11 :         s->next_framep[VP56_FRAME_PREVIOUS] = s->framep[VP56_FRAME_PREVIOUS];
    2700             : 
    2701        1142 :     s->next_framep[VP56_FRAME_CURRENT] = curframe;
    2702             : 
    2703        1142 :     ff_thread_finish_setup(avctx);
    2704             : 
    2705        1142 :     if (avctx->hwaccel) {
    2706           0 :         ret = avctx->hwaccel->start_frame(avctx, avpkt->data, avpkt->size);
    2707           0 :         if (ret < 0)
    2708           0 :             goto err;
    2709             : 
    2710           0 :         ret = avctx->hwaccel->decode_slice(avctx, avpkt->data, avpkt->size);
    2711           0 :         if (ret < 0)
    2712           0 :             goto err;
    2713             : 
    2714           0 :         ret = avctx->hwaccel->end_frame(avctx);
    2715           0 :         if (ret < 0)
    2716           0 :             goto err;
    2717             : 
    2718             :     } else {
    2719        1142 :         s->linesize   = curframe->tf.f->linesize[0];
    2720        1142 :         s->uvlinesize = curframe->tf.f->linesize[1];
    2721             : 
    2722        1142 :         memset(s->top_nnz, 0, s->mb_width * sizeof(*s->top_nnz));
    2723             :         /* Zero macroblock structures for top/top-left prediction
    2724             :          * from outside the frame. */
    2725        1142 :         if (!s->mb_layout)
    2726        1112 :             memset(s->macroblocks + s->mb_height * 2 - 1, 0,
    2727        1112 :                    (s->mb_width + 1) * sizeof(*s->macroblocks));
    2728        1142 :         if (!s->mb_layout && s->keyframe)
    2729          48 :             memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width * 4);
    2730             : 
    2731        1142 :         memset(s->ref_count, 0, sizeof(s->ref_count));
    2732             : 
    2733        1142 :         if (s->mb_layout == 1) {
    2734             :             // Make sure the previous frame has read its segmentation map,
    2735             :             // if we re-use the same map.
    2736          30 :             if (prev_frame && s->segmentation.enabled &&
    2737           0 :                 !s->segmentation.update_map)
    2738           0 :                 ff_thread_await_progress(&prev_frame->tf, 1, 0);
    2739          30 :             if (is_vp7)
    2740          30 :                 vp7_decode_mv_mb_modes(avctx, curframe, prev_frame);
    2741             :             else
    2742           0 :                 vp8_decode_mv_mb_modes(avctx, curframe, prev_frame);
    2743             :         }
    2744             : 
    2745        1142 :         if (avctx->active_thread_type == FF_THREAD_FRAME)
    2746           0 :             num_jobs = 1;
    2747             :         else
    2748        1142 :             num_jobs = FFMIN(s->num_coeff_partitions, avctx->thread_count);
    2749        1142 :         s->num_jobs   = num_jobs;
    2750        1142 :         s->curframe   = curframe;
    2751        1142 :         s->prev_frame = prev_frame;
    2752        1142 :         s->mv_bounds.mv_min.y   = -MARGIN;
    2753        1142 :         s->mv_bounds.mv_max.y   = ((s->mb_height - 1) << 6) + MARGIN;
    2754       10278 :         for (i = 0; i < MAX_THREADS; i++) {
    2755        9136 :             VP8ThreadData *td = &s->thread_data[i];
    2756        9136 :             atomic_init(&td->thread_mb_pos, 0);
    2757        9136 :             atomic_init(&td->wait_mb_pos, INT_MAX);
    2758             :         }
    2759        1142 :         if (is_vp7)
    2760          30 :             avctx->execute2(avctx, vp7_decode_mb_row_sliced, s->thread_data, NULL,
    2761             :                             num_jobs);
    2762             :         else
    2763        1112 :             avctx->execute2(avctx, vp8_decode_mb_row_sliced, s->thread_data, NULL,
    2764             :                             num_jobs);
    2765             :     }
    2766             : 
    2767        1142 :     ff_thread_report_progress(&curframe->tf, INT_MAX, 0);
    2768        1142 :     memcpy(&s->framep[0], &s->next_framep[0], sizeof(s->framep[0]) * 4);
    2769             : 
    2770        1142 : skip_decode:
    2771             :     // if future frames don't use the updated probabilities,
    2772             :     // reset them to the values we saved
    2773        1142 :     if (!s->update_probabilities)
    2774          64 :         s->prob[0] = s->prob[1];
    2775             : 
    2776        1142 :     if (!s->invisible) {
    2777        1133 :         if ((ret = av_frame_ref(data, curframe->tf.f)) < 0)
    2778           0 :             return ret;
    2779        1133 :         *got_frame = 1;
    2780             :     }
    2781             : 
    2782        1142 :     return avpkt->size;
    2783           0 : err:
    2784           0 :     memcpy(&s->next_framep[0], &s->framep[0], sizeof(s->framep[0]) * 4);
    2785           0 :     return ret;
    2786             : }
    2787             : 
    2788        1112 : int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
    2789             :                         AVPacket *avpkt)
    2790             : {
    2791        1112 :     return vp78_decode_frame(avctx, data, got_frame, avpkt, IS_VP8);
    2792             : }
    2793             : 
    2794             : #if CONFIG_VP7_DECODER
    2795          30 : static int vp7_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
    2796             :                             AVPacket *avpkt)
    2797             : {
    2798          30 :     return vp78_decode_frame(avctx, data, got_frame, avpkt, IS_VP7);
    2799             : }
    2800             : #endif /* CONFIG_VP7_DECODER */
    2801             : 
    2802          60 : av_cold int ff_vp8_decode_free(AVCodecContext *avctx)
    2803             : {
    2804          60 :     VP8Context *s = avctx->priv_data;
    2805             :     int i;
    2806             : 
    2807          60 :     if (!s)
    2808           0 :         return 0;
    2809             : 
    2810          60 :     vp8_decode_flush_impl(avctx, 1);
    2811         360 :     for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
    2812         300 :         av_frame_free(&s->frames[i].tf.f);
    2813             : 
    2814          60 :     return 0;
    2815             : }
    2816             : 
    2817          60 : static av_cold int vp8_init_frames(VP8Context *s)
    2818             : {
    2819             :     int i;
    2820         360 :     for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) {
    2821         300 :         s->frames[i].tf.f = av_frame_alloc();
    2822         300 :         if (!s->frames[i].tf.f)
    2823           0 :             return AVERROR(ENOMEM);
    2824             :     }
    2825          60 :     return 0;
    2826             : }
    2827             : 
    2828             : static av_always_inline
    2829          60 : int vp78_decode_init(AVCodecContext *avctx, int is_vp7)
    2830             : {
    2831          60 :     VP8Context *s = avctx->priv_data;
    2832             :     int ret;
    2833             : 
    2834          60 :     s->avctx = avctx;
    2835          60 :     s->vp7   = avctx->codec->id == AV_CODEC_ID_VP7;
    2836          60 :     s->pix_fmt = AV_PIX_FMT_NONE;
    2837          60 :     avctx->pix_fmt = AV_PIX_FMT_YUV420P;
    2838          60 :     avctx->internal->allocate_progress = 1;
    2839             : 
    2840          60 :     ff_videodsp_init(&s->vdsp, 8);
    2841             : 
    2842          60 :     ff_vp78dsp_init(&s->vp8dsp);
    2843          60 :     if (CONFIG_VP7_DECODER && is_vp7) {
    2844           3 :         ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP7, 8, 1);
    2845           3 :         ff_vp7dsp_init(&s->vp8dsp);
    2846           3 :         s->decode_mb_row_no_filter = vp7_decode_mb_row_no_filter;
    2847           3 :         s->filter_mb_row           = vp7_filter_mb_row;
    2848          57 :     } else if (CONFIG_VP8_DECODER && !is_vp7) {
    2849          57 :         ff_h264_pred_init(&s->hpc, AV_CODEC_ID_VP8, 8, 1);
    2850          57 :         ff_vp8dsp_init(&s->vp8dsp);
    2851          57 :         s->decode_mb_row_no_filter = vp8_decode_mb_row_no_filter;
    2852          57 :         s->filter_mb_row           = vp8_filter_mb_row;
    2853             :     }
    2854             : 
    2855             :     /* does not change for VP8 */
    2856          60 :     memcpy(s->prob[0].scan, ff_zigzag_scan, sizeof(s->prob[0].scan));
    2857             : 
    2858          60 :     if ((ret = vp8_init_frames(s)) < 0) {
    2859           0 :         ff_vp8_decode_free(avctx);
    2860           0 :         return ret;
    2861             :     }
    2862             : 
    2863          60 :     return 0;
    2864             : }
    2865             : 
    2866             : #if CONFIG_VP7_DECODER
    2867           3 : static int vp7_decode_init(AVCodecContext *avctx)
    2868             : {
    2869           3 :     return vp78_decode_init(avctx, IS_VP7);
    2870             : }
    2871             : #endif /* CONFIG_VP7_DECODER */
    2872             : 
    2873          57 : av_cold int ff_vp8_decode_init(AVCodecContext *avctx)
    2874             : {
    2875          57 :     return vp78_decode_init(avctx, IS_VP8);
    2876             : }
    2877             : 
    2878             : #if CONFIG_VP8_DECODER
    2879             : #if HAVE_THREADS
    2880           0 : static av_cold int vp8_decode_init_thread_copy(AVCodecContext *avctx)
    2881             : {
    2882           0 :     VP8Context *s = avctx->priv_data;
    2883             :     int ret;
    2884             : 
    2885           0 :     s->avctx = avctx;
    2886             : 
    2887           0 :     if ((ret = vp8_init_frames(s)) < 0) {
    2888           0 :         ff_vp8_decode_free(avctx);
    2889           0 :         return ret;
    2890             :     }
    2891             : 
    2892           0 :     return 0;
    2893             : }
    2894             : 
    2895             : #define REBASE(pic) ((pic) ? (pic) - &s_src->frames[0] + &s->frames[0] : NULL)
    2896             : 
    2897           0 : static int vp8_decode_update_thread_context(AVCodecContext *dst,
    2898             :                                             const AVCodecContext *src)
    2899             : {
    2900           0 :     VP8Context *s = dst->priv_data, *s_src = src->priv_data;
    2901             :     int i;
    2902             : 
    2903           0 :     if (s->macroblocks_base &&
    2904           0 :         (s_src->mb_width != s->mb_width || s_src->mb_height != s->mb_height)) {
    2905           0 :         free_buffers(s);
    2906           0 :         s->mb_width  = s_src->mb_width;
    2907           0 :         s->mb_height = s_src->mb_height;
    2908             :     }
    2909             : 
    2910           0 :     s->pix_fmt      = s_src->pix_fmt;
    2911           0 :     s->prob[0]      = s_src->prob[!s_src->update_probabilities];
    2912           0 :     s->segmentation = s_src->segmentation;
    2913           0 :     s->lf_delta     = s_src->lf_delta;
    2914           0 :     memcpy(s->sign_bias, s_src->sign_bias, sizeof(s->sign_bias));
    2915             : 
    2916           0 :     for (i = 0; i < FF_ARRAY_ELEMS(s_src->frames); i++) {
    2917           0 :         if (s_src->frames[i].tf.f->buf[0]) {
    2918           0 :             int ret = vp8_ref_frame(s, &s->frames[i], &s_src->frames[i]);
    2919           0 :             if (ret < 0)
    2920           0 :                 return ret;
    2921             :         }
    2922             :     }
    2923             : 
    2924           0 :     s->framep[0] = REBASE(s_src->next_framep[0]);
    2925           0 :     s->framep[1] = REBASE(s_src->next_framep[1]);
    2926           0 :     s->framep[2] = REBASE(s_src->next_framep[2]);
    2927           0 :     s->framep[3] = REBASE(s_src->next_framep[3]);
    2928             : 
    2929           0 :     return 0;
    2930             : }
    2931             : #endif /* HAVE_THREADS */
    2932             : #endif /* CONFIG_VP8_DECODER */
    2933             : 
    2934             : #if CONFIG_VP7_DECODER
    2935             : AVCodec ff_vp7_decoder = {
    2936             :     .name                  = "vp7",
    2937             :     .long_name             = NULL_IF_CONFIG_SMALL("On2 VP7"),
    2938             :     .type                  = AVMEDIA_TYPE_VIDEO,
    2939             :     .id                    = AV_CODEC_ID_VP7,
    2940             :     .priv_data_size        = sizeof(VP8Context),
    2941             :     .init                  = vp7_decode_init,
    2942             :     .close                 = ff_vp8_decode_free,
    2943             :     .decode                = vp7_decode_frame,
    2944             :     .capabilities          = AV_CODEC_CAP_DR1,
    2945             :     .flush                 = vp8_decode_flush,
    2946             : };
    2947             : #endif /* CONFIG_VP7_DECODER */
    2948             : 
    2949             : #if CONFIG_VP8_DECODER
    2950             : AVCodec ff_vp8_decoder = {
    2951             :     .name                  = "vp8",
    2952             :     .long_name             = NULL_IF_CONFIG_SMALL("On2 VP8"),
    2953             :     .type                  = AVMEDIA_TYPE_VIDEO,
    2954             :     .id                    = AV_CODEC_ID_VP8,
    2955             :     .priv_data_size        = sizeof(VP8Context),
    2956             :     .init                  = ff_vp8_decode_init,
    2957             :     .close                 = ff_vp8_decode_free,
    2958             :     .decode                = ff_vp8_decode_frame,
    2959             :     .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS |
    2960             :                              AV_CODEC_CAP_SLICE_THREADS,
    2961             :     .flush                 = vp8_decode_flush,
    2962             :     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(vp8_decode_init_thread_copy),
    2963             :     .update_thread_context = ONLY_IF_THREADS_ENABLED(vp8_decode_update_thread_context),
    2964             :     .hw_configs            = (const AVCodecHWConfigInternal*[]) {
    2965             : #if CONFIG_VP8_VAAPI_HWACCEL
    2966             :                                HWACCEL_VAAPI(vp8),
    2967             : #endif
    2968             : #if CONFIG_VP8_NVDEC_HWACCEL
    2969             :                                HWACCEL_NVDEC(vp8),
    2970             : #endif
    2971             :                                NULL
    2972             :                            },
    2973             : };
    2974             : #endif /* CONFIG_VP7_DECODER */

Generated by: LCOV version 1.13