[FFmpeg-devel,4/4] VP4 video decoder

Submitted by Peter Ross on Jan. 6, 2019, 7:44 a.m.

Details

Message ID 38bb314c4a7feefd1fe561947b150bf1fb19c135.1546760371.git.pross@xvid.org
State New
Headers show

Commit Message

Peter Ross Jan. 6, 2019, 7:44 a.m.
---
 Changelog               |   1 +
 doc/general.texi        |   2 +
 libavcodec/codec_desc.c |   2 +-
 libavcodec/vp3.c        | 701 ++++++++++++++++++++++++++++++++++++++--
 libavformat/riff.c      |   1 +
 5 files changed, 671 insertions(+), 36 deletions(-)

Comments

Paul B Mahol Jan. 6, 2019, 9:11 a.m.
On 1/6/19, Peter Ross <pross@xvid.org> wrote:
> ---
>  Changelog               |   1 +
>  doc/general.texi        |   2 +
>  libavcodec/codec_desc.c |   2 +-
>  libavcodec/vp3.c        | 701 ++++++++++++++++++++++++++++++++++++++--
>  libavformat/riff.c      |   1 +
>  5 files changed, 671 insertions(+), 36 deletions(-)

Why is it not using own codec id? It is better that way IMHO.
Peter Ross Jan. 6, 2019, 11:50 a.m.
On Sun, Jan 06, 2019 at 10:11:14AM +0100, Paul B Mahol wrote:
> On 1/6/19, Peter Ross <pross@xvid.org> wrote:
> > ---
> >  Changelog               |   1 +
> >  doc/general.texi        |   2 +
> >  libavcodec/codec_desc.c |   2 +-
> >  libavcodec/vp3.c        | 701 ++++++++++++++++++++++++++++++++++++++--
> >  libavformat/riff.c      |   1 +
> >  5 files changed, 671 insertions(+), 36 deletions(-)
> 
> Why is it not using own codec id? It is better that way IMHO.

seperate ids would enable users to independently disable the decoder.
but other than that, i don't know what value this brings.

the codecs are very similar. even the reference implementation only uses the
version field to distinguish them. if i change a true VP3 file to use the
'VP40' fourcc, it decodes correctly. if i change a true VP4 file to use
the 'VP31' fourcc, it also decodes correctly.

-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
Carl Eugen Hoyos Jan. 6, 2019, 11:59 a.m.
2019-01-06 12:50 GMT+01:00, Peter Ross <pross@xvid.org>:
> On Sun, Jan 06, 2019 at 10:11:14AM +0100, Paul B Mahol wrote:
>> On 1/6/19, Peter Ross <pross@xvid.org> wrote:
>> > ---
>> >  Changelog               |   1 +
>> >  doc/general.texi        |   2 +
>> >  libavcodec/codec_desc.c |   2 +-
>> >  libavcodec/vp3.c        | 701 ++++++++++++++++++++++++++++++++++++++--
>> >  libavformat/riff.c      |   1 +
>> >  5 files changed, 671 insertions(+), 36 deletions(-)
>>
>> Why is it not using own codec id? It is better that way IMHO.
>
> seperate ids would enable users to independently disable the decoder.
> but other than that, i don't know what value this brings.

It would theoretically improve remuxing, if in doubt, please use a
separate id.

Carl Eugen

Patch hide | download patch | download mbox

diff --git a/Changelog b/Changelog
index f135fa56b6..f025d75ab9 100644
--- a/Changelog
+++ b/Changelog
@@ -13,6 +13,7 @@  version <next>:
 - GIF parser
 - vividas demuxer
 - hymt decoder
+- VP4 video decoder
 
 
 version 4.1:
diff --git a/doc/general.texi b/doc/general.texi
index 2bc33d180d..8e23c1d2c8 100644
--- a/doc/general.texi
+++ b/doc/general.texi
@@ -935,6 +935,8 @@  following image formats are supported:
     @tab Video encoding used in NuppelVideo files.
 @item On2 VP3                @tab     @tab  X
     @tab still experimental
+@item On2 VP4                @tab     @tab  X
+    @tab fourcc: VP40
 @item On2 VP5                @tab     @tab  X
     @tab fourcc: VP50
 @item On2 VP6                @tab     @tab  X
diff --git a/libavcodec/codec_desc.c b/libavcodec/codec_desc.c
index 2363a53283..2dfd24b9e0 100644
--- a/libavcodec/codec_desc.c
+++ b/libavcodec/codec_desc.c
@@ -237,7 +237,7 @@  static const AVCodecDescriptor codec_descriptors[] = {
         .id        = AV_CODEC_ID_VP3,
         .type      = AVMEDIA_TYPE_VIDEO,
         .name      = "vp3",
-        .long_name = NULL_IF_CONFIG_SMALL("On2 VP3"),
+        .long_name = NULL_IF_CONFIG_SMALL("On2 VP3 / On2 VP4"),
         .props     = AV_CODEC_PROP_LOSSY,
     },
     {
diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c
index a5d8c2ed0b..127189a22b 100644
--- a/libavcodec/vp3.c
+++ b/libavcodec/vp3.c
@@ -1,5 +1,6 @@ 
 /*
  * Copyright (C) 2003-2004 The FFmpeg project
+ * Copyright (C) 2019 Peter Ross
  *
  * This file is part of FFmpeg.
  *
@@ -20,7 +21,7 @@ 
 
 /**
  * @file
- * On2 VP3 Video Decoder
+ * On2 VP3/VP4 Video Decoder
  *
  * VP3 Video Decoder by Mike Melanson (mike at multimedia.cx)
  * For more information about the VP3 coding process, visit:
@@ -43,6 +44,7 @@ 
 #include "thread.h"
 #include "videodsp.h"
 #include "vp3data.h"
+#include "vp4data.h"
 #include "vp3dsp.h"
 #include "xiph.h"
 
@@ -127,6 +129,30 @@  static const uint8_t hilbert_offset[16][2] = {
     { 3, 1 }, { 2, 1 }, { 2, 0 }, { 3, 0 }
 };
 
+enum {
+    VP4_DC_INTRA  = 0,
+    VP4_DC_INTER  = 1,
+    VP4_DC_GOLDEN = 2,
+    NB_VP4_DC_TYPES,
+    VP4_DC_UNDEFINED = NB_VP4_DC_TYPES
+};
+
+static const uint8_t vp4_pred_block_type_map[8] = {
+    [MODE_INTER_NO_MV]      = VP4_DC_INTER,
+    [MODE_INTRA]            = VP4_DC_INTRA,
+    [MODE_INTER_PLUS_MV]    = VP4_DC_INTER,
+    [MODE_INTER_LAST_MV]    = VP4_DC_INTER,
+    [MODE_INTER_PRIOR_LAST] = VP4_DC_INTER,
+    [MODE_USING_GOLDEN]     = VP4_DC_GOLDEN,
+    [MODE_GOLDEN_MV]        = VP4_DC_GOLDEN,
+    [MODE_INTER_FOURMV]     = VP4_DC_INTER,
+};
+
+typedef struct {
+    int dc;
+    int type;
+} VP4Predictor;
+
 #define MIN_DEQUANT_VAL 2
 
 typedef struct Vp3DecodeContext {
@@ -164,9 +190,13 @@  typedef struct Vp3DecodeContext {
     int v_superblock_start;
     unsigned char *superblock_coding;
 
-    int macroblock_count;
+    int macroblock_count; /* y macroblock count */
     int macroblock_width;
     int macroblock_height;
+    int c_macroblock_count;
+    int c_macroblock_width;
+    int c_macroblock_height;
+    int yuv_macroblock_count; /* y+u+v macroblock count */
 
     int fragment_count;
     int fragment_width[2];
@@ -182,7 +212,7 @@  typedef struct Vp3DecodeContext {
     int8_t (*motion_val[2])[2];
 
     /* tables */
-    uint16_t coded_dc_scale_factor[64];
+    uint16_t coded_dc_scale_factor[2][64];
     uint32_t coded_ac_scale_factor[64];
     uint8_t base_matrix[384][64];
     uint8_t qr_count[2][3];
@@ -233,10 +263,12 @@  typedef struct Vp3DecodeContext {
     VLC ac_vlc_3[16];
     VLC ac_vlc_4[16];
 
-    VLC superblock_run_length_vlc;
-    VLC fragment_run_length_vlc;
+    VLC superblock_run_length_vlc; /* version < 2 */
+    VLC fragment_run_length_vlc; /* version < 2 */
+    VLC block_pattern_vlc[2]; /* version >= 2*/
     VLC mode_code_vlc;
-    VLC motion_vector_vlc;
+    VLC motion_vector_vlc; /* version < 2 */
+    VLC vp4_mv_vlc[2][7]; /* version >=2 */
 
     /* these arrays need to be on 16-byte boundaries since SSE2 operations
      * index into them */
@@ -263,6 +295,9 @@  typedef struct Vp3DecodeContext {
 
     uint8_t filter_limit_values[64];
     DECLARE_ALIGNED(8, int, bounding_values_array)[256 + 2];
+
+    VP4Predictor * dc_pred_row; /* dc_pred_row[y_superblock_width * 4] */
+
 } Vp3DecodeContext;
 
 /************************************************************************
@@ -280,6 +315,7 @@  static av_cold void free_tables(AVCodecContext *avctx)
     av_freep(&s->dct_tokens_base);
     av_freep(&s->superblock_fragments);
     av_freep(&s->macroblock_coding);
+    av_freep(&s->dc_pred_row);
     av_freep(&s->motion_val[0]);
     av_freep(&s->motion_val[1]);
 }
@@ -299,7 +335,7 @@  static void vp3_decode_flush(AVCodecContext *avctx)
 static av_cold int vp3_decode_end(AVCodecContext *avctx)
 {
     Vp3DecodeContext *s = avctx->priv_data;
-    int i;
+    int i, j;
 
     free_tables(avctx);
     av_freep(&s->edge_emu_buffer);
@@ -328,6 +364,12 @@  static av_cold int vp3_decode_end(AVCodecContext *avctx)
     ff_free_vlc(&s->mode_code_vlc);
     ff_free_vlc(&s->motion_vector_vlc);
 
+    for (j = 0; j < 2; j++)
+        for (i = 0; i < 7; i++)
+            ff_free_vlc(&s->vp4_mv_vlc[j][i]);
+
+    for (i = 0; i < 2; i++)
+        ff_free_vlc(&s->block_pattern_vlc[i]);
     return 0;
 }
 
@@ -375,11 +417,11 @@  static int init_block_mapping(Vp3DecodeContext *s)
 static void init_dequantizer(Vp3DecodeContext *s, int qpi)
 {
     int ac_scale_factor = s->coded_ac_scale_factor[s->qps[qpi]];
-    int dc_scale_factor = s->coded_dc_scale_factor[s->qps[qpi]];
     int i, plane, inter, qri, bmi, bmj, qistart;
 
     for (inter = 0; inter < 2; inter++) {
         for (plane = 0; plane < 3; plane++) {
+            int dc_scale_factor = s->coded_dc_scale_factor[!!plane][s->qps[qpi]];
             int sum = 0;
             for (qri = 0; qri < s->qr_count[inter][plane]; qri++) {
                 sum += s->qr_size[inter][plane][qri];
@@ -397,9 +439,10 @@  static void init_dequantizer(Vp3DecodeContext *s, int qpi)
 
                 int qmin   = 8 << (inter + !i);
                 int qscale = i ? ac_scale_factor : dc_scale_factor;
-
+                int qbias = (1 + inter) * 3;
                 s->qmat[qpi][inter][plane][s->idct_permutation[i]] =
-                    av_clip((qscale * coeff) / 100 * 4, qmin, 4096);
+                    (i == 0 || s->version < 2) ? av_clip((qscale * coeff) / 100 * 4, qmin, 4096)
+                                               : (qscale * (coeff - qbias) / 100 + qbias) * 4;
             }
             /* all DC coefficients use the same quant so as not to interfere
              * with DC prediction */
@@ -618,6 +661,131 @@  static int unpack_superblocks(Vp3DecodeContext *s, GetBitContext *gb)
     return 0;
 }
 
+static int read_mb_value(GetBitContext *gb)
+{
+    int v = 1;
+    int size;
+
+    do {
+        size = 0;
+        if (!get_bits1(gb))
+            break;
+        v++;
+        do {
+            if (!get_bits1(gb))
+                break;
+            v += 1 << size++;
+        } while(size < 8);
+    } while (size == 8);
+
+    if (size)
+        v += get_bits(gb, size);
+
+    return v;
+}
+
+static int vp4_get_block_pattern(Vp3DecodeContext *s, GetBitContext *gb, int *next_block_pattern_table)
+{
+    int v = get_vlc2(gb, s->block_pattern_vlc[*next_block_pattern_table].table, 3, 2);
+    if (v == -1) {
+        av_log(s->avctx, AV_LOG_ERROR, "Invalid block pattern\n");
+        *next_block_pattern_table = 0;
+        return 0;
+    }
+    *next_block_pattern_table = vp4_block_pattern_table_selector[v];
+    return v + 1;
+}
+
+static int vp4_unpack_macroblocks(Vp3DecodeContext *s, GetBitContext *gb)
+{
+    int plane, i, j, k, fragment;
+    int next_block_pattern_table;
+    int bit, current_run, has_partial;
+
+    memset(s->macroblock_coding, MODE_COPY, s->macroblock_count);
+
+    if (s->keyframe)
+        return 0;
+
+    has_partial = 0;
+    bit         = get_bits1(gb);
+    current_run = read_mb_value(gb);
+
+    for (i = 0; i < s->yuv_macroblock_count; i++) {
+        if (!current_run) {
+            bit ^= 1;
+            current_run = read_mb_value(gb);
+        }
+        s->superblock_coding[i] = 2 * bit;
+        has_partial |= bit == 0;
+        current_run--;
+    }
+
+    if (has_partial) {
+        bit  = get_bits1(gb);
+        current_run = read_mb_value(gb);
+        for (i = 0; i < s->yuv_macroblock_count; i++) {
+            if (!s->superblock_coding[i]) {
+                if (!current_run) {
+                    bit ^= 1;
+                    current_run = read_mb_value(gb);
+                }
+                s->superblock_coding[i] = bit;
+                current_run--;
+            }
+        }
+    }
+
+    next_block_pattern_table = 0;
+    i = 0;
+    for (plane = 0; plane < 3; plane++) {
+        int sb_x, sb_y;
+        int sb_width = plane ? s->c_superblock_width : s->y_superblock_width;
+        int sb_height = plane ? s->c_superblock_height : s->y_superblock_height;
+        int mb_width = plane ? s->c_macroblock_width : s->macroblock_width;
+        int mb_height = plane ? s->c_macroblock_height : s->macroblock_height;
+        int fragment_width = s->fragment_width[!!plane];
+        int fragment_height = s->fragment_height[!!plane];
+
+        for (sb_y = 0; sb_y < sb_height; sb_y++) {
+            for (sb_x = 0; sb_x < sb_width; sb_x++) {
+                for (j = 0; j < 4; j++) {
+                    int mb_x = 2 * sb_x + (j >> 1);
+                    int mb_y = 2 * sb_y + (((j >> 1) + j) & 1);
+                    int mb_coded, pattern, coded;
+
+                    if (mb_x >= mb_width || mb_y >= mb_height)
+                        continue;
+
+                    mb_coded = s->superblock_coding[i++];
+
+                    if (mb_coded == SB_FULLY_CODED)
+                        pattern = 0xF;
+                    else if (mb_coded == SB_PARTIALLY_CODED)
+                        pattern = vp4_get_block_pattern(s, gb, &next_block_pattern_table);
+                    else
+                        pattern = 0;
+
+#define BLOCK_X (2 * mb_x + (k & 1))
+#define BLOCK_Y (2 * mb_y + (k >> 1))
+                    for (k = 0; k < 4; k++) {
+                        if (BLOCK_X >= fragment_width || BLOCK_Y >= fragment_height)
+                            continue;
+                        fragment = s->fragment_start[plane] + BLOCK_Y * fragment_width + BLOCK_X;
+
+                        coded = pattern & (1 << (3 - k));
+
+                        /* MODE_INTER_NO_MV is the default for coded fragments.
+                           the actual method is decoded in the next phase. */
+                        s->all_fragments[fragment].coding_method = coded ? MODE_INTER_NO_MV : MODE_COPY;
+                    }
+                }
+            }
+        }
+    }
+    return 0;
+}
+
 /*
  * This function unpacks all the coding mode data for individual macroblocks
  * from the bitstream.
@@ -666,8 +834,6 @@  static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
                         mb_y >= s->macroblock_height)
                         continue;
 
-#define BLOCK_X (2 * mb_x + (k & 1))
-#define BLOCK_Y (2 * mb_y + (k >> 1))
                     /* coding modes are only stored if the macroblock has
                      * at least one luma block coded, otherwise it must be
                      * INTER_NO_MV */
@@ -727,6 +893,15 @@  static int unpack_modes(Vp3DecodeContext *s, GetBitContext *gb)
     return 0;
 }
 
+static int vp4_get_mv(Vp3DecodeContext *s, GetBitContext *gb, int axis, int last_motion)
+{
+    int v = get_vlc2(gb, s->vp4_mv_vlc[axis][vp4_mv_table_selector[FFABS(last_motion)]].table, 6, 2);
+    v -= 31;
+    if (last_motion < 0)
+        v = -v;
+    return v;
+}
+
 /*
  * This function unpacks all the motion vectors for the individual
  * macroblocks from the bitstream.
@@ -741,6 +916,8 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
     int last_motion_y = 0;
     int prior_last_motion_x = 0;
     int prior_last_motion_y = 0;
+    int last_gold_motion_x = 0;
+    int last_gold_motion_y = 0;
     int current_macroblock;
     int current_fragment;
     int frag;
@@ -748,8 +925,8 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
     if (s->keyframe)
         return 0;
 
-    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme */
-    coding_mode = get_bits1(gb);
+    /* coding mode 0 is the VLC scheme; 1 is the fixed code scheme; 2 is VP4 code scheme */
+    coding_mode = s->version < 2 ? get_bits1(gb) : 2;
 
     /* iterate through all of the macroblocks that contain 1 or more
      * coded fragments */
@@ -769,15 +946,25 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
                     continue;
 
                 switch (s->macroblock_coding[current_macroblock]) {
-                case MODE_INTER_PLUS_MV:
                 case MODE_GOLDEN_MV:
+                    if (coding_mode == 2) { /* VP4 */
+                        motion_x[0] = vp4_get_mv(s, gb, 0, last_gold_motion_x);
+                        motion_y[0] = vp4_get_mv(s, gb, 1, last_gold_motion_y);
+                        last_gold_motion_x = motion_x[0];
+                        last_gold_motion_y = motion_y[0];
+                        break;
+                    } /* otherwise fall through */
+                case MODE_INTER_PLUS_MV:
                     /* all 6 fragments use the same motion vector */
                     if (coding_mode == 0) {
                         motion_x[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
                         motion_y[0] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
-                    } else {
+                    } else if (coding_mode == 1) {
                         motion_x[0] = fixed_motion_vector_table[get_bits(gb, 6)];
                         motion_y[0] = fixed_motion_vector_table[get_bits(gb, 6)];
+                    } else { /* VP4 */
+                        motion_x[0] = vp4_get_mv(s, gb, 0, last_motion_x);
+                        motion_y[0] = vp4_get_mv(s, gb, 1, last_motion_y);
                     }
 
                     /* vector maintenance, only on MODE_INTER_PLUS_MV */
@@ -802,9 +989,12 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
                             if (coding_mode == 0) {
                                 motion_x[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
                                 motion_y[k] = motion_vector_table[get_vlc2(gb, s->motion_vector_vlc.table, 6, 2)];
-                            } else {
+                            } else if (coding_mode == 1) {
                                 motion_x[k] = fixed_motion_vector_table[get_bits(gb, 6)];
                                 motion_y[k] = fixed_motion_vector_table[get_bits(gb, 6)];
+                            } else { /* VP4 */
+                                motion_x[k] = vp4_get_mv(s, gb, 0, prior_last_motion_x);
+                                motion_y[k] = vp4_get_mv(s, gb, 1, prior_last_motion_y);
                             }
                             last_motion_x = motion_x[k];
                             last_motion_y = motion_y[k];
@@ -866,8 +1056,10 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
                         motion_y[0] = RSHIFT(motion_y[0] + motion_y[1] +
                                              motion_y[2] + motion_y[3], 2);
                     }
-                    motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
-                    motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
+                    if (s->version <= 2) {
+                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
+                        motion_y[0] = (motion_y[0] >> 1) | (motion_y[0] & 1);
+                    }
                     frag = mb_y * s->fragment_width[1] + mb_x;
                     s->motion_val[1][frag][0] = motion_x[0];
                     s->motion_val[1][frag][1] = motion_y[0];
@@ -881,9 +1073,10 @@  static int unpack_vectors(Vp3DecodeContext *s, GetBitContext *gb)
                         motion_x[1] = motion_x[0];
                         motion_y[1] = motion_y[0];
                     }
-                    motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
-                    motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
-
+                    if (s->version <= 2) {
+                        motion_x[0] = (motion_x[0] >> 1) | (motion_x[0] & 1);
+                        motion_x[1] = (motion_x[1] >> 1) | (motion_x[1] & 1);
+                    }
                     frag = 2 * mb_y * s->fragment_width[1] + mb_x;
                     for (k = 0; k < 2; k++) {
                         s->motion_val[1][frag][0] = motion_x[k];
@@ -1006,6 +1199,7 @@  static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
     while (coeff_i < num_coeffs && get_bits_left(gb) > 0) {
         /* decode a VLC into a token */
         token = get_vlc2(gb, vlc_table, 11, 3);
+
         /* use the token to get a zero run, a coefficient, and an eob run */
         if ((unsigned) token <= 6U) {
             eob_run = eob_run_base[token];
@@ -1032,6 +1226,7 @@  static int unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
             bits_to_get = coeff_get_bits[token];
             if (bits_to_get)
                 bits_to_get = get_bits(gb, bits_to_get);
+
             coeff = coeff_tables[token][bits_to_get];
 
             zero_run = zero_run_base[token];
@@ -1188,6 +1383,242 @@  static int unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
     return 0;
 }
 
+/**
+ * eob_tracker[] is instead of TOKEN_EOB(value)
+ * a dummy TOKEN_EOB(0) value is used to make vp3_dequant work
+ *
+ * @return < 0 on error
+ */
+static int vp4_unpack_vlcs(Vp3DecodeContext *s, GetBitContext *gb,
+                       VLC *vlc_tables[64],
+                       int plane, int eob_tracker[64], int fragment)
+{
+    int token;
+    int zero_run  = 0;
+    int16_t coeff = 0;
+    int bits_to_get;
+    int coeff_i = 0;
+    int eob_run;
+
+    while(!eob_tracker[coeff_i]) {
+
+        token = get_vlc2(gb, vlc_tables[coeff_i]->table, 11, 3);
+
+        /* use the token to get a zero run, a coefficient, and an eob run */
+        if ((unsigned) token <= 6U) {
+            eob_run = eob_run_base[token];
+            if (eob_run_get_bits[token])
+                eob_run += get_bits(gb, eob_run_get_bits[token]);
+            *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
+            eob_tracker[coeff_i] = eob_run - 1;
+            return 0;
+        } else if (token >= 0) {
+            bits_to_get = coeff_get_bits[token];
+
+            if (bits_to_get)
+                bits_to_get = get_bits(gb, bits_to_get);
+            coeff = coeff_tables[token][bits_to_get];
+
+            zero_run = zero_run_base[token];
+            if (zero_run_get_bits[token])
+                zero_run += get_bits(gb, zero_run_get_bits[token]);
+
+            if (zero_run) {
+                if (coeff_i + zero_run > 64) {
+                    av_log(s->avctx, AV_LOG_DEBUG,
+                        "Invalid zero run of %d with %d coeffs left\n",
+                        zero_run, 64 - coeff_i);
+                    zero_run = 64 - coeff_i;
+                }
+                *s->dct_tokens[plane][coeff_i]++ = TOKEN_ZERO_RUN(coeff, zero_run);
+                coeff_i += zero_run;
+            } else {
+                if (!coeff_i)
+                    s->all_fragments[fragment].dc = coeff;
+
+                *s->dct_tokens[plane][coeff_i]++ = TOKEN_COEFF(coeff);
+            }
+            coeff_i++;
+            if (coeff_i >= 64) /* > 64 occurs when there is a zero_run overflow */
+                return 0; /* stop */
+        } else {
+            av_log(s->avctx, AV_LOG_ERROR, "Invalid token %d\n", token);
+            return -1;
+        }
+    }
+    *s->dct_tokens[plane][coeff_i]++ = TOKEN_EOB(0);
+    eob_tracker[coeff_i]--;
+    return 0;
+}
+
+static void vp4_dc_predictor_reset(VP4Predictor *p)
+{
+    p->dc = 0;
+    p->type = VP4_DC_UNDEFINED;
+}
+
+static void vp4_dc_pred_before(const Vp3DecodeContext *s, VP4Predictor * dc_pred, int sb_x)
+{
+    int i, j;
+
+    for (i = 0; i < 4; i++)
+        dc_pred[i + 1] = s->dc_pred_row[sb_x * 4 + i];
+
+    for (j = 0; j < 4; j++)
+        for (i = 0; i < 4; i++)
+            vp4_dc_predictor_reset(&dc_pred[j * 6 + i + 7]);
+}
+
+static void vp4_dc_pred_after(Vp3DecodeContext *s, VP4Predictor * dc_pred, int sb_x)
+{
+    int i;
+
+    for (i = 0; i < 4; i++)
+        s->dc_pred_row[sb_x * 4 + i] = dc_pred[25 + i];
+
+    for (i = 1; i < 5; i++)
+        dc_pred[6 * i] = dc_pred[6 * i + 4];
+}
+
+static int vp4_dc_pred(const Vp3DecodeContext *s, const VP4Predictor * dc_pred, const int * last_dc, int idx, int type, int plane)
+{
+    int count = 0;
+    int dc = 0;
+
+    if (dc_pred[idx - 6].type == type) {
+        dc += dc_pred[idx - 6].dc;
+        count++;
+    }
+
+    if (dc_pred[idx + 6].type == type) {
+        dc += dc_pred[idx + 6].dc;
+        count++;
+    }
+
+    if (count != 2 && dc_pred[idx - 1].type == type) {
+        dc += dc_pred[idx - 1].dc;
+        count++;
+    }
+
+    if (count != 2 && dc_pred[idx + 1].type == type) {
+        dc += dc_pred[idx + 1].dc;
+        count++;
+    }
+
+    return count == 2 ? dc / count : last_dc[type];
+}
+
+static void vp4_set_tokens_base(Vp3DecodeContext *s)
+{
+    int plane, i;
+    int16_t *base = s->dct_tokens_base;
+    for (plane = 0; plane < 3; plane++) {
+        for (i = 0; i < 64; i++) {
+            s->dct_tokens[plane][i] = base;
+            base += s->fragment_width[!!plane] * s->fragment_height[!!plane];
+        }
+    }
+}
+
+static int vp4_unpack_dct_coeffs(Vp3DecodeContext *s, GetBitContext *gb)
+{
+    int i, j;
+    int dc_y_table;
+    int dc_c_table;
+    int ac_y_table;
+    int ac_c_table;
+    VLC *tables[2][64];
+    int plane, sb_y, sb_x;
+    int eob_tracker[64];
+    VP4Predictor dc_pred[6 * 6];
+    int last_dc[NB_VP4_DC_TYPES];
+
+    if (get_bits_left(gb) < 16)
+        return AVERROR_INVALIDDATA;
+
+    /* fetch the DC table indexes */
+    dc_y_table = get_bits(gb, 4);
+    dc_c_table = get_bits(gb, 4);
+
+    ac_y_table = get_bits(gb, 4);
+    ac_c_table = get_bits(gb, 4);
+
+    /* build tables of DC/AC VLC tables */
+
+    tables[0][0] = &s->dc_vlc[dc_y_table];
+    tables[1][0] = &s->dc_vlc[dc_c_table];
+    for (i = 1; i <= 5; i++) {
+        tables[0][i] = &s->ac_vlc_1[ac_y_table];
+        tables[1][i] = &s->ac_vlc_1[ac_c_table];
+    }
+    for (i = 6; i <= 14; i++) {
+        tables[0][i] = &s->ac_vlc_2[ac_y_table];
+        tables[1][i] = &s->ac_vlc_2[ac_c_table];
+    }
+    for (i = 15; i <= 27; i++) {
+        tables[0][i] = &s->ac_vlc_3[ac_y_table];
+        tables[1][i] = &s->ac_vlc_3[ac_c_table];
+    }
+    for (i = 28; i <= 63; i++) {
+        tables[0][i] = &s->ac_vlc_4[ac_y_table];
+        tables[1][i] = &s->ac_vlc_4[ac_c_table];
+    }
+
+    vp4_set_tokens_base(s);
+
+    memset(last_dc, 0, sizeof(last_dc));
+
+    for (plane = 0; plane < ((s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 1 : 3); plane++) {
+        memset(eob_tracker, 0, sizeof(eob_tracker));
+
+        /* initialise dc prediction */
+        for (i = 0; i < s->fragment_width[!!plane]; i++)
+            vp4_dc_predictor_reset(&s->dc_pred_row[i]);
+
+        for (i = 1; i < 35; i++)
+            vp4_dc_predictor_reset(&dc_pred[i]);
+
+        for (sb_y = 0; sb_y * 4 < s->fragment_height[!!plane]; sb_y++) {
+            for (sb_x = 0; sb_x *4 < s->fragment_width[!!plane]; sb_x++) {
+                vp4_dc_pred_before(s, dc_pred, sb_x);
+                for (j = 0; j < 16; j++) {
+                        int hx = hilbert_offset[j][0];
+                        int hy = hilbert_offset[j][1];
+                        int x  = 4 * sb_x + hilbert_offset[j][0];
+                        int y  = 4 * sb_y + hilbert_offset[j][1];
+                        int fragment, dc_block_idx, dc_block_type;
+
+                        if (x >= s->fragment_width[!!plane] || y >= s->fragment_height[!!plane])
+                            continue;
+
+                        fragment = s->fragment_start[plane] + y * s->fragment_width[!!plane] + x;
+
+                        if (s->all_fragments[fragment].coding_method == MODE_COPY)
+                            continue;
+
+                        if (vp4_unpack_vlcs(s, gb, tables[!!plane], plane, eob_tracker, fragment) < 0)
+                            return -1;
+
+                        dc_block_idx = 6 * (hy + 1) + (hx + 1);
+                        dc_block_type = vp4_pred_block_type_map[s->all_fragments[fragment].coding_method];
+
+                        s->all_fragments[fragment].dc +=
+                            vp4_dc_pred(s, dc_pred, last_dc, dc_block_idx, dc_block_type, plane);
+
+                        dc_pred[dc_block_idx].type = dc_block_type;
+                        dc_pred[dc_block_idx].dc   =
+                        last_dc[dc_block_type]     = s->all_fragments[fragment].dc;
+                }
+                vp4_dc_pred_after(s, dc_pred, sb_x);
+            }
+        }
+    }
+
+    vp4_set_tokens_base(s);
+
+    return 0;
+}
+
 /*
  * This function reverses the DC prediction for each coded fragment in
  * the frame. Much of this function is adapted directly from the original
@@ -1520,6 +1951,98 @@  static void await_reference_row(Vp3DecodeContext *s, Vp3Fragment *fragment,
     ff_thread_await_progress(ref_frame, ref_row, 0);
 }
 
+/**
+ * @return non-zero if temp (edge_emu_buffer) was populated
+ */
+static int vp4_mc_loop_filter(Vp3DecodeContext *s, int plane, int motion_x, int motion_y, int bx, int by,
+       uint8_t * motion_source, int stride, int src_x, int src_y, uint8_t *temp)
+{
+    static const int motion_shift[2] = { 1, 2 };
+    static const int subpel_mask[2]  = { 1, 3 };
+    int *bounding_values = s->bounding_values_array + 127;
+
+    int i;
+    int x, y;
+    int x2, y2;
+    int x_subpel, y_subpel;
+    int x_offset, y_offset;
+
+    int block_width = plane ? 8 : 16;
+    int plane_width  = s->width  >> (plane && s->chroma_x_shift);
+    int plane_height = s->height >> (plane && s->chroma_y_shift);
+
+#define loop_stride 12
+    uint8_t loop[12 * loop_stride];
+
+#define SHIFT(v, shift) ((v) >> (shift))
+#define ABS_SHIFT(v, shift) ((v) > 0 ? SHIFT(v, shift) : -SHIFT(-v, shift))
+    x = 8 * bx + ABS_SHIFT(motion_x, motion_shift[!!plane]);
+    y = 8 * by + ABS_SHIFT(motion_y, motion_shift[!!plane]);
+
+    x_subpel = motion_x & subpel_mask[!!plane];
+    y_subpel = motion_y & subpel_mask[!!plane];
+
+    if (x_subpel || y_subpel) {
+        x--;
+        y--;
+
+        if (x_subpel)
+            x = FFMIN(x, x + FFSIGN(motion_x));
+
+        if (y_subpel)
+            y = FFMIN(y, y + FFSIGN(motion_y));
+
+        x2 = x + block_width;
+        y2 = y + block_width;
+
+        if (x2 < 0 || x2 >= plane_width || y2 < 0 || y2 >= plane_height)
+            return 0;
+
+        x_offset = (-(x + 2) & 7) + 2;
+        y_offset = (-(y + 2) & 7) + 2;
+
+        if (x_offset > 8 + x_subpel && y_offset > 8 + y_subpel)
+            return 0;
+
+        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
+             loop_stride, stride,
+             12, 12, src_x - 1, src_y - 1,
+             plane_width,
+             plane_height);
+
+        if (x_offset <= 8 + x_subpel)
+            ff_vp4_h_loop_filter_12_c(loop + x_offset, loop_stride, bounding_values);
+
+        if (y_offset <= 8 + y_subpel)
+            ff_vp4_v_loop_filter_12_c(loop + y_offset*loop_stride, loop_stride, bounding_values);
+
+    } else {
+
+        x_offset = -x & 7;
+        y_offset = -y & 7;
+
+        if (!x_offset && !y_offset)
+            return 0;
+
+        s->vdsp.emulated_edge_mc(loop, motion_source - stride - 1,
+             loop_stride, stride,
+             12, 12, src_x - 1, src_y - 1,
+             plane_width,
+             plane_height);
+
+        if (x_offset)
+            s->vp3dsp.h_loop_filter(loop + loop_stride + x_offset + 1, loop_stride, bounding_values);
+
+        if (y_offset)
+            s->vp3dsp.v_loop_filter(loop + (y_offset + 1)*loop_stride + 1, loop_stride, bounding_values);
+    }
+
+    for (i = 0; i < 9; i++)
+        memcpy(temp + i*stride, loop + (i + 1) * loop_stride + 1, 9);
+
+    return 1;
+}
+
 /*
  * Perform the final rendering for a particular slice of data.
  * The slice number ranges from 0..(c_superblock_height - 1).
@@ -1605,8 +2128,13 @@  static void render_slice(Vp3DecodeContext *s, int slice)
                         if ((s->all_fragments[i].coding_method > MODE_INTRA) &&
                             (s->all_fragments[i].coding_method != MODE_USING_GOLDEN)) {
                             int src_x, src_y;
+                            int standard_mc = 1;
                             motion_x = motion_val[fragment][0];
                             motion_y = motion_val[fragment][1];
+                            if (plane && s->version >= 2) {
+                                motion_x = (motion_x >> 1) | (motion_x & 1);
+                                motion_y = (motion_y >> 1) | (motion_y & 1);
+                            }
 
                             src_x = (motion_x >> 1) + 8 * x;
                             src_y = (motion_y >> 1) + 8 * y;
@@ -1617,9 +2145,20 @@  static void render_slice(Vp3DecodeContext *s, int slice)
                             motion_halfpel_index |= (motion_y & 0x01) << 1;
                             motion_source        += ((motion_y >> 1) * stride);
 
-                            if (src_x < 0 || src_y < 0 ||
+                            if (s->version >= 2) {
+                                uint8_t *temp = s->edge_emu_buffer;
+                                if (stride < 0)
+                                    temp -= 8 * stride;
+                                if (vp4_mc_loop_filter(s, plane, motion_val[fragment][0], motion_val[fragment][1], x, y, motion_source, stride, src_x, src_y, temp)) {
+                                    motion_source = temp;
+                                    standard_mc = 0;
+                                }
+                            }
+
+                            if (standard_mc && (
+                                src_x < 0 || src_y < 0 ||
                                 src_x + 9 >= plane_width ||
-                                src_y + 9 >= plane_height) {
+                                src_y + 9 >= plane_height)) {
                                 uint8_t *temp = s->edge_emu_buffer;
                                 if (stride < 0)
                                     temp -= 8 * stride;
@@ -1685,7 +2224,7 @@  static void render_slice(Vp3DecodeContext *s, int slice)
             }
 
             // Filter up to the last row in the superblock row
-            if (!s->skip_loop_filter)
+            if (s->version < 2 && !s->skip_loop_filter)
                 apply_loop_filter(s, plane, 4 * sb_y - !!sb_y,
                                   FFMIN(4 * sb_y + 3, fragment_height - 1));
         }
@@ -1714,7 +2253,8 @@  static av_cold int allocate_tables(AVCodecContext *avctx)
     y_fragment_count = s->fragment_width[0] * s->fragment_height[0];
     c_fragment_count = s->fragment_width[1] * s->fragment_height[1];
 
-    s->superblock_coding = av_mallocz(s->superblock_count);
+    /* superblock_coding is used by unpack_superblocks (VP3/Theora) and vp4_unpack_macroblocks (VP4) */
+    s->superblock_coding = av_mallocz(FFMAX(s->superblock_count, s->yuv_macroblock_count));
     s->all_fragments     = av_mallocz_array(s->fragment_count, sizeof(Vp3Fragment));
 
     s-> kf_coded_fragment_list = av_mallocz_array(s->fragment_count, sizeof(int));
@@ -1730,10 +2270,13 @@  static av_cold int allocate_tables(AVCodecContext *avctx)
     s->superblock_fragments = av_mallocz_array(s->superblock_count, 16 * sizeof(int));
     s->macroblock_coding    = av_mallocz(s->macroblock_count + 1);
 
+    s->dc_pred_row = av_malloc_array(s->y_superblock_width * 4, sizeof(*s->dc_pred_row));
+
     if (!s->superblock_coding    || !s->all_fragments          ||
         !s->dct_tokens_base      || !s->kf_coded_fragment_list ||
         !s->nkf_coded_fragment_list ||
         !s->superblock_fragments || !s->macroblock_coding      ||
+        !s->dc_pred_row ||
         !s->motion_val[0]        || !s->motion_val[1]) {
         vp3_decode_end(avctx);
         return -1;
@@ -1763,7 +2306,7 @@  static av_cold int init_frames(Vp3DecodeContext *s)
 static av_cold int vp3_decode_init(AVCodecContext *avctx)
 {
     Vp3DecodeContext *s = avctx->priv_data;
-    int i, inter, plane, ret;
+    int i, j, inter, plane, ret;
     int c_width;
     int c_height;
     int y_fragment_count, c_fragment_count;
@@ -1774,7 +2317,9 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
 
     avctx->internal->allocate_progress = 1;
 
-    if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
+    if (avctx->codec_tag == MKTAG('V', 'P', '4', '0'))
+        s->version = 3;
+    else if (avctx->codec_tag == MKTAG('V', 'P', '3', '0'))
         s->version = 0;
     else
         s->version = 1;
@@ -1823,6 +2368,10 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
     s->macroblock_width  = (s->width  + 15) / 16;
     s->macroblock_height = (s->height + 15) / 16;
     s->macroblock_count  = s->macroblock_width * s->macroblock_height;
+    s->c_macroblock_width  = (c_width  + 15) / 16;
+    s->c_macroblock_height = (c_height + 15) / 16;
+    s->c_macroblock_count  = s->c_macroblock_width * s->c_macroblock_height;
+    s->yuv_macroblock_count = s->macroblock_count + 2 * s->c_macroblock_count;
 
     s->fragment_width[0]  = s->width / FRAGMENT_PIXELS;
     s->fragment_height[0] = s->height / FRAGMENT_PIXELS;
@@ -1838,12 +2387,13 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
 
     if (!s->theora_tables) {
         for (i = 0; i < 64; i++) {
-            s->coded_dc_scale_factor[i] = vp31_dc_scale_factor[i];
-            s->coded_ac_scale_factor[i] = vp31_ac_scale_factor[i];
-            s->base_matrix[0][i]        = vp31_intra_y_dequant[i];
-            s->base_matrix[1][i]        = vp31_intra_c_dequant[i];
-            s->base_matrix[2][i]        = vp31_inter_dequant[i];
-            s->filter_limit_values[i]   = vp31_filter_limit_values[i];
+            s->coded_dc_scale_factor[0][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_y_dc_scale_factor[i];
+            s->coded_dc_scale_factor[1][i] = s->version < 2 ? vp31_dc_scale_factor[i] : vp4_uv_dc_scale_factor[i];
+            s->coded_ac_scale_factor[i] = s->version < 2 ? vp31_ac_scale_factor[i] : vp4_ac_scale_factor[i];
+            s->base_matrix[0][i]        = s->version < 2 ? vp31_intra_y_dequant[i] : vp4_generic_dequant[i];
+            s->base_matrix[1][i]        = s->version < 2 ? vp31_intra_c_dequant[i] : vp4_generic_dequant[i];
+            s->base_matrix[2][i]        = s->version < 2 ? vp31_inter_dequant[i]   : vp4_generic_dequant[i];
+            s->filter_limit_values[i]   = s->version < 2 ? vp31_filter_limit_values[i] : vp4_filter_limit_values[i];
         }
 
         for (inter = 0; inter < 2; inter++) {
@@ -1856,6 +2406,7 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
         }
 
         /* init VLC tables */
+        if (s->version < 2) {
         for (i = 0; i < 16; i++) {
             /* DC histograms */
             init_vlc(&s->dc_vlc[i], 11, 32,
@@ -1882,6 +2433,34 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
                      &ac_bias_3[i][0][1], 4, 2,
                      &ac_bias_3[i][0][0], 4, 2, 0);
         }
+        } else { /* version >= 2 */
+            for (i = 0; i < 16; i++) {
+                /* DC histograms */
+                init_vlc(&s->dc_vlc[i], 11, 32,
+                         &vp4_dc_bias[i][0][1], 4, 2,
+                         &vp4_dc_bias[i][0][0], 4, 2, 0);
+
+                /* group 1 AC histograms */
+                init_vlc(&s->ac_vlc_1[i], 11, 32,
+                         &vp4_ac_bias_0[i][0][1], 4, 2,
+                         &vp4_ac_bias_0[i][0][0], 4, 2, 0);
+
+                /* group 2 AC histograms */
+                init_vlc(&s->ac_vlc_2[i], 11, 32,
+                         &vp4_ac_bias_1[i][0][1], 4, 2,
+                         &vp4_ac_bias_1[i][0][0], 4, 2, 0);
+
+                /* group 3 AC histograms */
+                init_vlc(&s->ac_vlc_3[i], 11, 32,
+                         &vp4_ac_bias_2[i][0][1], 4, 2,
+                         &vp4_ac_bias_2[i][0][0], 4, 2, 0);
+
+                /* group 4 AC histograms */
+                init_vlc(&s->ac_vlc_4[i], 11, 32,
+                         &vp4_ac_bias_3[i][0][1], 4, 2,
+                         &vp4_ac_bias_3[i][0][0], 4, 2, 0);
+            }
+        }
     } else {
         for (i = 0; i < 16; i++) {
             /* DC histograms */
@@ -1932,6 +2511,19 @@  static av_cold int vp3_decode_init(AVCodecContext *avctx)
              &motion_vector_vlc_table[0][1], 2, 1,
              &motion_vector_vlc_table[0][0], 2, 1, 0);
 
+    for (j = 0; j < 2; j++)
+        for (i = 0; i < 7; i++)
+            init_vlc(&s->vp4_mv_vlc[j][i], 6, 63,
+                 &vp4_mv_vlc[j][i][0][1], 4, 2,
+                 &vp4_mv_vlc[j][i][0][0], 4, 2, 0);
+
+    /* version >= 2 */
+    for (i = 0; i < 2; i++)
+        init_vlc(&s->block_pattern_vlc[i], 3, 14,
+             &vp4_block_pattern_vlc[i][0][1], 2, 1,
+             &vp4_block_pattern_vlc[i][0][0], 2, 1, 0);
+
+
     return allocate_tables(avctx);
 
 vlc_fail:
@@ -2155,6 +2747,27 @@  static int vp3_decode_frame(AVCodecContext *avctx,
                 av_log(s->avctx, AV_LOG_ERROR,
                        "Warning, unsupported keyframe coding type?!\n");
             skip_bits(&gb, 2); /* reserved? */
+
+            if (s->version >= 2) {
+                int mb_height, mb_width;
+                int mb_width_mul, mb_width_div, mb_height_mul, mb_height_div;
+
+                mb_height = get_bits(&gb, 8);
+                mb_width  = get_bits(&gb, 8);
+                if (mb_height != s->macroblock_height ||
+                    mb_width != s->macroblock_width)
+                    av_log(s->avctx, AV_LOG_WARNING, "VP4 header: Warning, macroblock dimension mismatch");
+
+                mb_width_mul = get_bits(&gb, 5);
+                mb_width_div = get_bits(&gb, 3);
+                mb_height_mul = get_bits(&gb, 5);
+                mb_height_div = get_bits(&gb, 3);
+                if (mb_width_mul != 1 || mb_width_div != 1 || mb_height_mul != 1 || mb_height_div != 1)
+                      av_log(s->avctx, AV_LOG_WARNING, "VP4 header: Warning, unexpected macroblock dimension multipler/divider");
+
+                if (get_bits(&gb, 2))
+                    av_log(s->avctx, AV_LOG_WARNING, "VP4 header: Warning, unknown bits set");
+            }
         }
     } else {
         if (!s->golden_frame.f->data[0]) {
@@ -2176,10 +2789,17 @@  static int vp3_decode_frame(AVCodecContext *avctx,
     memset(s->all_fragments, 0, s->fragment_count * sizeof(Vp3Fragment));
     ff_thread_finish_setup(avctx);
 
+    if (s->version < 2) {
     if (unpack_superblocks(s, &gb)) {
         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_superblocks\n");
         goto error;
     }
+    } else {
+        if (vp4_unpack_macroblocks(s, &gb)) {
+            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_macroblocks\n");
+            goto error;
+    }
+    }
     if (unpack_modes(s, &gb)) {
         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_modes\n");
         goto error;
@@ -2192,10 +2812,18 @@  static int vp3_decode_frame(AVCodecContext *avctx,
         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_block_qpis\n");
         goto error;
     }
+
+    if (s->version < 2) {
     if (unpack_dct_coeffs(s, &gb)) {
         av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
         goto error;
     }
+    } else {
+        if (vp4_unpack_dct_coeffs(s, &gb)) {
+            av_log(s->avctx, AV_LOG_ERROR, "error in vp4_unpack_dct_coeffs\n");
+            goto error;
+        }
+    }
 
     for (i = 0; i < 3; i++) {
         int height = s->height >> (i && s->chroma_y_shift);
@@ -2210,6 +2838,7 @@  static int vp3_decode_frame(AVCodecContext *avctx,
         render_slice(s, i);
 
     // filter the last row
+    if (s->version < 2)
     for (i = 0; i < 3; i++) {
         int row = (s->height >> (3 + (i && s->chroma_y_shift))) - 1;
         apply_loop_filter(s, i, row, row + 1);
@@ -2294,6 +2923,7 @@  static int vp3_init_thread_copy(AVCodecContext *avctx)
     s->motion_val[0]          = NULL;
     s->motion_val[1]          = NULL;
     s->edge_emu_buffer        = NULL;
+    s->dc_pred_row            = NULL;
 
     return init_frames(s);
 }
@@ -2442,7 +3072,8 @@  static int theora_decode_tables(AVCodecContext *avctx, GetBitContext *gb)
         n = 16;
     /* dc scale factor table */
     for (i = 0; i < 64; i++)
-        s->coded_dc_scale_factor[i] = get_bits(gb, n);
+        s->coded_dc_scale_factor[0][i] =
+        s->coded_dc_scale_factor[1][i] = get_bits(gb, n);
 
     if (s->theora >= 0x030200)
         matrices = get_bits(gb, 9) + 1;
diff --git a/libavformat/riff.c b/libavformat/riff.c
index d5a509c8b9..6d36293c50 100644
--- a/libavformat/riff.c
+++ b/libavformat/riff.c
@@ -307,6 +307,7 @@  const AVCodecTag ff_codec_bmp_tags[] = {
     { AV_CODEC_ID_INDEO5,       MKTAG('I', 'V', '5', '0') },
     { AV_CODEC_ID_VP3,          MKTAG('V', 'P', '3', '1') },
     { AV_CODEC_ID_VP3,          MKTAG('V', 'P', '3', '0') },
+    { AV_CODEC_ID_VP3,          MKTAG('V', 'P', '4', '0') },
     { AV_CODEC_ID_VP5,          MKTAG('V', 'P', '5', '0') },
     { AV_CODEC_ID_VP6,          MKTAG('V', 'P', '6', '0') },
     { AV_CODEC_ID_VP6,          MKTAG('V', 'P', '6', '1') },