]> rtime.felk.cvut.cz Git - frescor/ffmpeg.git/blobdiff - libavcodec/vp6.c
frsh: Export information about the last RTP contract and VRES
[frescor/ffmpeg.git] / libavcodec / vp6.c
index bc0ce817b16b6bf23f509431565ea9bcf3d7fb8d..c1ca2273b142792af2ff7dcc6c73c9deafaddf09 100644 (file)
@@ -1,9 +1,13 @@
 /**
- * @file vp6.c
+ * @file libavcodec/vp6.c
  * VP6 compatible video decoder
  *
  * Copyright (C) 2006  Aurelien Jacobs <aurel@gnuage.org>
  *
+ * The VP6F decoder accepts an optional 1 byte extradata. It is composed of:
+ *  - upper 4bits: difference between encoded width and visible width
+ *  - lower 4bits: difference between encoded height and visible height
+ *
  * This file is part of FFmpeg.
  *
  * FFmpeg is free software; you can redistribute it and/or
  *
  * You should have received a copy of the GNU Lesser General Public
  * License along with FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
- *
- *
- * The VP6F decoder accept an optional 1 byte extradata. It is composed of:
- *  - upper 4bits: difference between encoded width and visible width
- *  - lower 4bits: difference between encoded height and visible height
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 #include <stdlib.h>
 
 #include "avcodec.h"
 #include "dsputil.h"
-#include "bitstream.h"
-#include "mpegvideo.h"
+#include "get_bits.h"
+#include "huffman.h"
 
 #include "vp56.h"
 #include "vp56data.h"
 #include "vp6data.h"
 
 
-static int vp6_parse_header(vp56_context_t *s, uint8_t *buf, int buf_size,
+static void vp6_parse_coeff(VP56Context *s);
+static void vp6_parse_coeff_huffman(VP56Context *s);
+
+static int vp6_parse_header(VP56Context *s, const uint8_t *buf, int buf_size,
                             int *golden_frame)
 {
-    vp56_range_coder_t *c = &s->c;
-    int parse_filter_info;
+    VP56RangeCoder *c = &s->c;
+    int parse_filter_info = 0;
+    int coeff_offset = 0;
+    int vrt_shift = 0;
+    int sub_version;
     int rows, cols;
     int res = 1;
+    int separated_coeff = buf[0] & 1;
 
-    if (buf[0] & 1)
-        return 0;
-
-    s->frames[VP56_FRAME_CURRENT].key_frame = !(buf[0] & 0x80);
+    s->framep[VP56_FRAME_CURRENT]->key_frame = !(buf[0] & 0x80);
     vp56_init_dequant(s, (buf[0] >> 1) & 0x3F);
 
-    if (s->frames[VP56_FRAME_CURRENT].key_frame) {
-        if ((buf[1] & 0xFE) != 0x46)  /* would be 0x36 for VP61 */
+    if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
+        sub_version = buf[1] >> 3;
+        if (sub_version > 8)
             return 0;
+        s->filter_header = buf[1] & 0x06;
         if (buf[1] & 1) {
             av_log(s->avctx, AV_LOG_ERROR, "interlacing not supported\n");
             return 0;
         }
+        if (separated_coeff || !s->filter_header) {
+            coeff_offset = AV_RB16(buf+2) - 2;
+            buf += 2;
+            buf_size -= 2;
+        }
 
         rows = buf[2];  /* number of stored macroblock rows */
         cols = buf[3];  /* number of stored macroblock cols */
         /* buf[4] is number of displayed macroblock rows */
         /* buf[5] is number of displayed macroblock cols */
 
-        if (16*cols != s->avctx->coded_width ||
+        if (!s->macroblocks || /* first frame */
+            16*cols != s->avctx->coded_width ||
             16*rows != s->avctx->coded_height) {
             avcodec_set_dimensions(s->avctx, 16*cols, 16*rows);
             if (s->avctx->extradata_size == 1) {
@@ -78,87 +89,153 @@ static int vp6_parse_header(vp56_context_t *s, uint8_t *buf, int buf_size,
         vp56_init_range_decoder(c, buf+6, buf_size-6);
         vp56_rac_gets(c, 2);
 
-        parse_filter_info = 1;
+        parse_filter_info = s->filter_header;
+        if (sub_version < 8)
+            vrt_shift = 5;
+        s->sub_version = sub_version;
     } else {
+        if (!s->sub_version)
+            return 0;
+
+        if (separated_coeff || !s->filter_header) {
+            coeff_offset = AV_RB16(buf+1) - 2;
+            buf += 2;
+            buf_size -= 2;
+        }
         vp56_init_range_decoder(c, buf+1, buf_size-1);
 
         *golden_frame = vp56_rac_get(c);
-        s->deblock_filtering = vp56_rac_get(c);
-        if (s->deblock_filtering)
-            vp56_rac_get(c);
-        parse_filter_info = vp56_rac_get(c);
+        if (s->filter_header) {
+            s->deblock_filtering = vp56_rac_get(c);
+            if (s->deblock_filtering)
+                vp56_rac_get(c);
+            if (s->sub_version > 7)
+                parse_filter_info = vp56_rac_get(c);
+        }
     }
 
     if (parse_filter_info) {
         if (vp56_rac_get(c)) {
             s->filter_mode = 2;
-            s->sample_variance_threshold = vp56_rac_gets(c, 5);
+            s->sample_variance_threshold = vp56_rac_gets(c, 5) << vrt_shift;
             s->max_vector_length = 2 << vp56_rac_gets(c, 3);
         } else if (vp56_rac_get(c)) {
             s->filter_mode = 1;
         } else {
             s->filter_mode = 0;
         }
-        s->filter_selection = vp56_rac_gets(c, 4);
+        if (s->sub_version > 7)
+            s->filter_selection = vp56_rac_gets(c, 4);
+        else
+            s->filter_selection = 16;
+    }
+
+    s->use_huffman = vp56_rac_get(c);
+
+    s->parse_coeff = vp6_parse_coeff;
+    if (coeff_offset) {
+        buf      += coeff_offset;
+        buf_size -= coeff_offset;
+        if (buf_size < 0)
+            return 0;
+        if (s->use_huffman) {
+            s->parse_coeff = vp6_parse_coeff_huffman;
+            init_get_bits(&s->gb, buf, buf_size<<3);
+        } else {
+            vp56_init_range_decoder(&s->cc, buf, buf_size);
+            s->ccp = &s->cc;
+        }
+    } else {
+        s->ccp = &s->c;
     }
 
-    vp56_rac_get(c);
     return res;
 }
 
-static void vp6_coeff_order_table_init(vp56_context_t *s)
+static void vp6_coeff_order_table_init(VP56Context *s)
 {
     int i, pos, idx = 1;
 
-    s->coeff_index_to_pos[0] = 0;
+    s->modelp->coeff_index_to_pos[0] = 0;
     for (i=0; i<16; i++)
         for (pos=1; pos<64; pos++)
-            if (s->coeff_reorder[pos] == i)
-                s->coeff_index_to_pos[idx++] = pos;
+            if (s->modelp->coeff_reorder[pos] == i)
+                s->modelp->coeff_index_to_pos[idx++] = pos;
 }
 
-static void vp6_default_models_init(vp56_context_t *s)
+static void vp6_default_models_init(VP56Context *s)
 {
-    s->vector_model_dct[0] = 0xA2;
-    s->vector_model_dct[1] = 0xA4;
-    s->vector_model_sig[0] = 0x80;
-    s->vector_model_sig[1] = 0x80;
+    VP56Model *model = s->modelp;
+
+    model->vector_dct[0] = 0xA2;
+    model->vector_dct[1] = 0xA4;
+    model->vector_sig[0] = 0x80;
+    model->vector_sig[1] = 0x80;
 
-    memcpy(s->mb_types_stats, vp56_def_mb_types_stats, sizeof(s->mb_types_stats));
-    memcpy(s->vector_model_fdv, vp6_def_fdv_vector_model, sizeof(s->vector_model_fdv));
-    memcpy(s->vector_model_pdv, vp6_def_pdv_vector_model, sizeof(s->vector_model_pdv));
-    memcpy(s->coeff_model_runv, vp6_def_runv_coeff_model, sizeof(s->coeff_model_runv));
-    memcpy(s->coeff_reorder, vp6_def_coeff_reorder, sizeof(s->coeff_reorder));
+    memcpy(model->mb_types_stats, vp56_def_mb_types_stats, sizeof(model->mb_types_stats));
+    memcpy(model->vector_fdv, vp6_def_fdv_vector_model, sizeof(model->vector_fdv));
+    memcpy(model->vector_pdv, vp6_def_pdv_vector_model, sizeof(model->vector_pdv));
+    memcpy(model->coeff_runv, vp6_def_runv_coeff_model, sizeof(model->coeff_runv));
+    memcpy(model->coeff_reorder, vp6_def_coeff_reorder, sizeof(model->coeff_reorder));
 
     vp6_coeff_order_table_init(s);
 }
 
-static void vp6_parse_vector_models(vp56_context_t *s)
+static void vp6_parse_vector_models(VP56Context *s)
 {
-    vp56_range_coder_t *c = &s->c;
+    VP56RangeCoder *c = &s->c;
+    VP56Model *model = s->modelp;
     int comp, node;
 
     for (comp=0; comp<2; comp++) {
         if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][0]))
-            s->vector_model_dct[comp] = vp56_rac_gets_nn(c, 7);
+            model->vector_dct[comp] = vp56_rac_gets_nn(c, 7);
         if (vp56_rac_get_prob(c, vp6_sig_dct_pct[comp][1]))
-            s->vector_model_sig[comp] = vp56_rac_gets_nn(c, 7);
+            model->vector_sig[comp] = vp56_rac_gets_nn(c, 7);
     }
 
     for (comp=0; comp<2; comp++)
         for (node=0; node<7; node++)
             if (vp56_rac_get_prob(c, vp6_pdv_pct[comp][node]))
-                s->vector_model_pdv[comp][node] = vp56_rac_gets_nn(c, 7);
+                model->vector_pdv[comp][node] = vp56_rac_gets_nn(c, 7);
 
     for (comp=0; comp<2; comp++)
         for (node=0; node<8; node++)
             if (vp56_rac_get_prob(c, vp6_fdv_pct[comp][node]))
-                s->vector_model_fdv[comp][node] = vp56_rac_gets_nn(c, 7);
+                model->vector_fdv[comp][node] = vp56_rac_gets_nn(c, 7);
 }
 
-static void vp6_parse_coeff_models(vp56_context_t *s)
+/* nodes must ascend by count, but with descending symbol order */
+static int vp6_huff_cmp(const void *va, const void *vb)
 {
-    vp56_range_coder_t *c = &s->c;
+    const Node *a = va, *b = vb;
+    return (a->count - b->count)*16 + (b->sym - a->sym);
+}
+
+static void vp6_build_huff_tree(VP56Context *s, uint8_t coeff_model[],
+                                const uint8_t *map, unsigned size, VLC *vlc)
+{
+    Node nodes[2*size], *tmp = &nodes[size];
+    int a, b, i;
+
+    /* first compute probabilities from model */
+    tmp[0].count = 256;
+    for (i=0; i<size-1; i++) {
+        a = tmp[i].count *        coeff_model[i]  >> 8;
+        b = tmp[i].count * (255 - coeff_model[i]) >> 8;
+        nodes[map[2*i  ]].count = a + !a;
+        nodes[map[2*i+1]].count = b + !b;
+    }
+
+    /* then build the huffman tree accodring to probabilities */
+    ff_huff_build_tree(s->avctx, vlc, size, nodes, vp6_huff_cmp,
+                       FF_HUFFMAN_FLAG_HNODE_FIRST);
+}
+
+static void vp6_parse_coeff_models(VP56Context *s)
+{
+    VP56RangeCoder *c = &s->c;
+    VP56Model *model = s->modelp;
     int def_prob[11];
     int node, cg, ctx, pos;
     int ct;    /* code type */
@@ -170,22 +247,22 @@ static void vp6_parse_coeff_models(vp56_context_t *s)
         for (node=0; node<11; node++)
             if (vp56_rac_get_prob(c, vp6_dccv_pct[pt][node])) {
                 def_prob[node] = vp56_rac_gets_nn(c, 7);
-                s->coeff_model_dccv[pt][node] = def_prob[node];
-            } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
-                s->coeff_model_dccv[pt][node] = def_prob[node];
+                model->coeff_dccv[pt][node] = def_prob[node];
+            } else if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
+                model->coeff_dccv[pt][node] = def_prob[node];
             }
 
     if (vp56_rac_get(c)) {
         for (pos=1; pos<64; pos++)
             if (vp56_rac_get_prob(c, vp6_coeff_reorder_pct[pos]))
-                s->coeff_reorder[pos] = vp56_rac_gets(c, 4);
+                model->coeff_reorder[pos] = vp56_rac_gets(c, 4);
         vp6_coeff_order_table_init(s);
     }
 
     for (cg=0; cg<2; cg++)
         for (node=0; node<14; node++)
             if (vp56_rac_get_prob(c, vp6_runv_pct[cg][node]))
-                s->coeff_model_runv[cg][node] = vp56_rac_gets_nn(c, 7);
+                model->coeff_runv[cg][node] = vp56_rac_gets_nn(c, 7);
 
     for (ct=0; ct<3; ct++)
         for (pt=0; pt<2; pt++)
@@ -193,46 +270,62 @@ static void vp6_parse_coeff_models(vp56_context_t *s)
                 for (node=0; node<11; node++)
                     if (vp56_rac_get_prob(c, vp6_ract_pct[ct][pt][cg][node])) {
                         def_prob[node] = vp56_rac_gets_nn(c, 7);
-                        s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
-                    } else if (s->frames[VP56_FRAME_CURRENT].key_frame) {
-                        s->coeff_model_ract[pt][ct][cg][node] = def_prob[node];
+                        model->coeff_ract[pt][ct][cg][node] = def_prob[node];
+                    } else if (s->framep[VP56_FRAME_CURRENT]->key_frame) {
+                        model->coeff_ract[pt][ct][cg][node] = def_prob[node];
                     }
 
-    /* coeff_model_dcct is a linear combination of coeff_model_dccv */
+    if (s->use_huffman) {
+        for (pt=0; pt<2; pt++) {
+            vp6_build_huff_tree(s, model->coeff_dccv[pt],
+                                vp6_huff_coeff_map, 12, &s->dccv_vlc[pt]);
+            vp6_build_huff_tree(s, model->coeff_runv[pt],
+                                vp6_huff_run_map, 9, &s->runv_vlc[pt]);
+            for (ct=0; ct<3; ct++)
+                for (cg = 0; cg < 6; cg++)
+                    vp6_build_huff_tree(s, model->coeff_ract[pt][ct][cg],
+                                        vp6_huff_coeff_map, 12,
+                                        &s->ract_vlc[pt][ct][cg]);
+        }
+        memset(s->nb_null, 0, sizeof(s->nb_null));
+    } else {
+    /* coeff_dcct is a linear combination of coeff_dccv */
     for (pt=0; pt<2; pt++)
         for (ctx=0; ctx<3; ctx++)
             for (node=0; node<5; node++)
-                s->coeff_model_dcct[pt][ctx][node] = clip(((s->coeff_model_dccv[pt][node] * vp6_dccv_lc[ctx][node][0] + 128) >> 8) + vp6_dccv_lc[ctx][node][1], 1, 255);
+                model->coeff_dcct[pt][ctx][node] = av_clip(((model->coeff_dccv[pt][node] * vp6_dccv_lc[ctx][node][0] + 128) >> 8) + vp6_dccv_lc[ctx][node][1], 1, 255);
+    }
 }
 
-static void vp6_parse_vector_adjustment(vp56_context_t *s, vp56_mv_t *vect)
+static void vp6_parse_vector_adjustment(VP56Context *s, VP56mv *vect)
 {
-    vp56_range_coder_t *c = &s->c;
+    VP56RangeCoder *c = &s->c;
+    VP56Model *model = s->modelp;
     int comp;
 
-    *vect = (vp56_mv_t) {0,0};
+    *vect = (VP56mv) {0,0};
     if (s->vector_candidate_pos < 2)
         *vect = s->vector_candidate[0];
 
     for (comp=0; comp<2; comp++) {
         int i, delta = 0;
 
-        if (vp56_rac_get_prob(c, s->vector_model_dct[comp])) {
+        if (vp56_rac_get_prob(c, model->vector_dct[comp])) {
             static const uint8_t prob_order[] = {0, 1, 2, 7, 6, 5, 4};
             for (i=0; i<sizeof(prob_order); i++) {
                 int j = prob_order[i];
-                delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][j])<<j;
+                delta |= vp56_rac_get_prob(c, model->vector_fdv[comp][j])<<j;
             }
             if (delta & 0xF0)
-                delta |= vp56_rac_get_prob(c, s->vector_model_fdv[comp][3])<<3;
+                delta |= vp56_rac_get_prob(c, model->vector_fdv[comp][3])<<3;
             else
                 delta |= 8;
         } else {
             delta = vp56_rac_get_tree(c, vp56_pva_tree,
-                                      s->vector_model_pdv[comp]);
+                                      model->vector_pdv[comp]);
         }
 
-        if (delta && vp56_rac_get_prob(c, s->vector_model_sig[comp]))
+        if (delta && vp56_rac_get_prob(c, model->vector_sig[comp]))
             delta = -delta;
 
         if (!comp)
@@ -242,11 +335,83 @@ static void vp6_parse_vector_adjustment(vp56_context_t *s, vp56_mv_t *vect)
     }
 }
 
-static void vp6_parse_coeff(vp56_context_t *s)
+/**
+ * Read number of consecutive blocks with null DC or AC.
+ * This value is < 74.
+ */
+static unsigned vp6_get_nb_null(VP56Context *s)
+{
+    unsigned val = get_bits(&s->gb, 2);
+    if (val == 2)
+        val += get_bits(&s->gb, 2);
+    else if (val == 3) {
+        val = get_bits1(&s->gb) << 2;
+        val = 6+val + get_bits(&s->gb, 2+val);
+    }
+    return val;
+}
+
+static void vp6_parse_coeff_huffman(VP56Context *s)
+{
+    VP56Model *model = s->modelp;
+    uint8_t *permute = s->scantable.permutated;
+    VLC *vlc_coeff;
+    int coeff, sign, coeff_idx;
+    int b, cg, idx;
+    int pt = 0;    /* plane type (0 for Y, 1 for U or V) */
+
+    for (b=0; b<6; b++) {
+        int ct = 0;    /* code type */
+        if (b > 3) pt = 1;
+        vlc_coeff = &s->dccv_vlc[pt];
+
+        for (coeff_idx=0; coeff_idx<64; ) {
+            int run = 1;
+            if (coeff_idx<2 && s->nb_null[coeff_idx][pt]) {
+                s->nb_null[coeff_idx][pt]--;
+                if (coeff_idx)
+                    break;
+            } else {
+                coeff = get_vlc2(&s->gb, vlc_coeff->table, 9, 3);
+                if (coeff == 0) {
+                    if (coeff_idx) {
+                        int pt = (coeff_idx >= 6);
+                        run += get_vlc2(&s->gb, s->runv_vlc[pt].table, 9, 3);
+                        if (run >= 9)
+                            run += get_bits(&s->gb, 6);
+                    } else
+                        s->nb_null[0][pt] = vp6_get_nb_null(s);
+                    ct = 0;
+                } else if (coeff == 11) {  /* end of block */
+                    if (coeff_idx == 1)    /* first AC coeff ? */
+                        s->nb_null[1][pt] = vp6_get_nb_null(s);
+                    break;
+                } else {
+                    int coeff2 = vp56_coeff_bias[coeff];
+                    if (coeff > 4)
+                        coeff2 += get_bits(&s->gb, coeff <= 9 ? coeff - 4 : 11);
+                    ct = 1 + (coeff2 > 1);
+                    sign = get_bits1(&s->gb);
+                    coeff2 = (coeff2 ^ -sign) + sign;
+                    if (coeff_idx)
+                        coeff2 *= s->dequant_ac;
+                    idx = model->coeff_index_to_pos[coeff_idx];
+                    s->block_coeff[b][permute[idx]] = coeff2;
+                }
+            }
+            coeff_idx+=run;
+            cg = FFMIN(vp6_coeff_groups[coeff_idx], 3);
+            vlc_coeff = &s->ract_vlc[pt][ct][cg];
+        }
+    }
+}
+
+static void vp6_parse_coeff(VP56Context *s)
 {
-    vp56_range_coder_t *c = &s->c;
+    VP56RangeCoder *c = s->ccp;
+    VP56Model *model = s->modelp;
     uint8_t *permute = s->scantable.permutated;
-    uint8_t *model, *model2, *model3;
+    uint8_t *model1, *model2, *model3;
     int coeff, sign, coeff_idx;
     int b, i, cg, idx, ctx;
     int pt = 0;    /* plane type (0 for Y, 1 for U or V) */
@@ -259,26 +424,21 @@ static void vp6_parse_coeff(vp56_context_t *s)
 
         ctx = s->left_block[vp56_b6to4[b]].not_null_dc
               + s->above_blocks[s->above_block_idx[b]].not_null_dc;
-        model = s->coeff_model_dccv[pt];
-        model2 = s->coeff_model_dcct[pt][ctx];
+        model1 = model->coeff_dccv[pt];
+        model2 = model->coeff_dcct[pt][ctx];
 
         for (coeff_idx=0; coeff_idx<64; ) {
             if ((coeff_idx>1 && ct==0) || vp56_rac_get_prob(c, model2[0])) {
                 /* parse a coeff */
-                if (coeff_idx == 0) {
-                    s->left_block[vp56_b6to4[b]].not_null_dc = 1;
-                    s->above_blocks[s->above_block_idx[b]].not_null_dc = 1;
-                }
-
                 if (vp56_rac_get_prob(c, model2[2])) {
                     if (vp56_rac_get_prob(c, model2[3])) {
-                        idx = vp56_rac_get_tree(c, vp56_pc_tree, model);
-                        coeff = vp56_coeff_bias[idx];
+                        idx = vp56_rac_get_tree(c, vp56_pc_tree, model1);
+                        coeff = vp56_coeff_bias[idx+5];
                         for (i=vp56_coeff_bit_length[idx]; i>=0; i--)
                             coeff += vp56_rac_get_prob(c, vp56_coeff_parse_table[idx][i]) << i;
                     } else {
                         if (vp56_rac_get_prob(c, model2[4]))
-                            coeff = 3 + vp56_rac_get_prob(c, model[5]);
+                            coeff = 3 + vp56_rac_get_prob(c, model1[5]);
                         else
                             coeff = 2;
                     }
@@ -291,20 +451,17 @@ static void vp6_parse_coeff(vp56_context_t *s)
                 coeff = (coeff ^ -sign) + sign;
                 if (coeff_idx)
                     coeff *= s->dequant_ac;
-                idx = s->coeff_index_to_pos[coeff_idx];
+                idx = model->coeff_index_to_pos[coeff_idx];
                 s->block_coeff[b][permute[idx]] = coeff;
                 run = 1;
             } else {
                 /* parse a run */
                 ct = 0;
-                if (coeff_idx == 0) {
-                    s->left_block[vp56_b6to4[b]].not_null_dc = 0;
-                    s->above_blocks[s->above_block_idx[b]].not_null_dc = 0;
-                } else {
+                if (coeff_idx > 0) {
                     if (!vp56_rac_get_prob(c, model2[1]))
                         break;
 
-                    model3 = s->coeff_model_runv[coeff_idx >= 6];
+                    model3 = model->coeff_runv[coeff_idx >= 6];
                     run = vp56_rac_get_tree(c, vp6_pcr_tree, model3);
                     if (!run)
                         for (run=9, i=0; i<6; i++)
@@ -313,8 +470,11 @@ static void vp6_parse_coeff(vp56_context_t *s)
             }
 
             cg = vp6_coeff_groups[coeff_idx+=run];
-            model = model2 = s->coeff_model_ract[pt][ct][cg];
+            model1 = model2 = model->coeff_ract[pt][ct][cg];
         }
+
+        s->left_block[vp56_b6to4[b]].not_null_dc =
+        s->above_blocks[s->above_block_idx[b]].not_null_dc = !!s->block_coeff[b][0];
     }
 }
 
@@ -346,14 +506,6 @@ static int vp6_block_variance(uint8_t *src, int stride)
     return (16*square_sum - sum*sum) >> 8;
 }
 
-static void vp6_filter_hv2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
-                           int stride, int delta, int16_t weight)
-{
-    s->dsp.put_pixels_tab[1][0](dst, src, stride, 8);
-    s->dsp.biweight_h264_pixels_tab[3](dst, src+delta, stride, 2,
-                                       8-weight, weight, 0);
-}
-
 static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
                            int delta, const int16_t *weights)
 {
@@ -361,7 +513,7 @@ static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
 
     for (y=0; y<8; y++) {
         for (x=0; x<8; x++) {
-            dst[x] = clip_uint8((  src[x-delta  ] * weights[0]
+            dst[x] = av_clip_uint8((  src[x-delta  ] * weights[0]
                                  + src[x        ] * weights[1]
                                  + src[x+delta  ] * weights[2]
                                  + src[x+2*delta] * weights[3] + 64) >> 7);
@@ -371,60 +523,17 @@ static void vp6_filter_hv4(uint8_t *dst, uint8_t *src, int stride,
     }
 }
 
-static void vp6_filter_diag2(vp56_context_t *s, uint8_t *dst, uint8_t *src,
+static void vp6_filter_diag2(VP56Context *s, uint8_t *dst, uint8_t *src,
                              int stride, int h_weight, int v_weight)
 {
     uint8_t *tmp = s->edge_emu_buffer+16;
-    int x, xmax;
-
-    s->dsp.put_pixels_tab[1][0](tmp, src, stride, 8);
-    s->dsp.biweight_h264_pixels_tab[3](tmp, src+1, stride, 2,
-                                       8-h_weight, h_weight, 0);
-    /* we need a 8x9 block to do vertical filter, so compute one more line */
-    for (x=8*stride, xmax=x+8; x<xmax; x++)
-        tmp[x] = (src[x]*(8-h_weight) + src[x+1]*h_weight + 4) >> 3;
-
-    s->dsp.put_pixels_tab[1][0](dst, tmp, stride, 8);
-    s->dsp.biweight_h264_pixels_tab[3](dst, tmp+stride, stride, 2,
-                                       8-v_weight, v_weight, 0);
-}
-
-static void vp6_filter_diag4(uint8_t *dst, uint8_t *src, int stride,
-                             const int16_t *h_weights,const int16_t *v_weights)
-{
-    int x, y;
-    int tmp[8*11];
-    int *t = tmp;
-
-    src -= stride;
-
-    for (y=0; y<11; y++) {
-        for (x=0; x<8; x++) {
-            t[x] = clip_uint8((  src[x-1] * h_weights[0]
-                               + src[x  ] * h_weights[1]
-                               + src[x+1] * h_weights[2]
-                               + src[x+2] * h_weights[3] + 64) >> 7);
-        }
-        src += stride;
-        t += 8;
-    }
-
-    t = tmp + 8;
-    for (y=0; y<8; y++) {
-        for (x=0; x<8; x++) {
-            dst[x] = clip_uint8((  t[x-8 ] * v_weights[0]
-                                 + t[x   ] * v_weights[1]
-                                 + t[x+8 ] * v_weights[2]
-                                 + t[x+16] * v_weights[3] + 64) >> 7);
-        }
-        dst += stride;
-        t += 8;
-    }
+    s->dsp.put_h264_chroma_pixels_tab[0](tmp, src, stride, 9, h_weight, 0);
+    s->dsp.put_h264_chroma_pixels_tab[0](dst, tmp, stride, 8, 0, v_weight);
 }
 
-static void vp6_filter(vp56_context_t *s, uint8_t *dst, uint8_t *src,
+static void vp6_filter(VP56Context *s, uint8_t *dst, uint8_t *src,
                        int offset1, int offset2, int stride,
-                       vp56_mv_t mv, int mask, int select, int luma)
+                       VP56mv mv, int mask, int select, int luma)
 {
     int filter4 = 0;
     int x8 = mv.x & mask;
@@ -439,8 +548,8 @@ static void vp6_filter(vp56_context_t *s, uint8_t *dst, uint8_t *src,
                 (FFABS(mv.x) > s->max_vector_length ||
                  FFABS(mv.y) > s->max_vector_length)) {
                 filter4 = 0;
-            } else if (!s->sample_variance_threshold
-                       || (vp6_block_variance(src+offset1, stride)
+            } else if (s->sample_variance_threshold
+                       && (vp6_block_variance(src+offset1, stride)
                            < s->sample_variance_threshold)) {
                 filter4 = 0;
             }
@@ -458,38 +567,30 @@ static void vp6_filter(vp56_context_t *s, uint8_t *dst, uint8_t *src,
         } else if (!x8) {               /* above or below combine */
             vp6_filter_hv4(dst, src+offset1, stride, stride,
                            vp6_block_copy_filter[select][y8]);
-        } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
-            vp6_filter_diag4(dst, src+offset1-1, stride,
-                             vp6_block_copy_filter[select][x8],
-                             vp6_block_copy_filter[select][y8]);
-        } else {                        /* lower-right or upper-left combine */
-            vp6_filter_diag4(dst, src+offset1, stride,
+        } else {
+            s->dsp.vp6_filter_diag4(dst, src+offset1+((mv.x^mv.y)>>31), stride,
                              vp6_block_copy_filter[select][x8],
                              vp6_block_copy_filter[select][y8]);
         }
     } else {
-        if (!y8) {                      /* left or right combine */
-            vp6_filter_hv2(s, dst, src+offset1, stride, 1, x8);
-        } else if (!x8) {               /* above or below combine */
-            vp6_filter_hv2(s, dst, src+offset1, stride, stride, y8);
-        } else if ((mv.x^mv.y) >> 31) { /* lower-left or upper-right combine */
-            vp6_filter_diag2(s, dst, src+offset1-1, stride, x8, y8);
-        } else {                        /* lower-right or upper-left combine */
-            vp6_filter_diag2(s, dst, src+offset1, stride, x8, y8);
+        if (!x8 || !y8) {
+            s->dsp.put_h264_chroma_pixels_tab[0](dst, src+offset1, stride, 8, x8, y8);
+        } else {
+            vp6_filter_diag2(s, dst, src+offset1 + ((mv.x^mv.y)>>31), stride, x8, y8);
         }
     }
 }
 
-static int vp6_decode_init(AVCodecContext *avctx)
+static av_cold int vp6_decode_init(AVCodecContext *avctx)
 {
-    vp56_context_t *s = avctx->priv_data;
+    VP56Context *s = avctx->priv_data;
 
-    vp56_init(s, avctx, avctx->codec->id == CODEC_ID_VP6);
+    vp56_init(avctx, avctx->codec->id == CODEC_ID_VP6,
+                     avctx->codec->id == CODEC_ID_VP6A);
     s->vp56_coord_div = vp6_coord_div;
     s->parse_vector_adjustment = vp6_parse_vector_adjustment;
     s->adjust = vp6_adjust;
     s->filter = vp6_filter;
-    s->parse_coeff = vp6_parse_coeff;
     s->default_models_init = vp6_default_models_init;
     s->parse_vector_models = vp6_parse_vector_models;
     s->parse_coeff_models = vp6_parse_coeff_models;
@@ -502,11 +603,13 @@ AVCodec vp6_decoder = {
     "vp6",
     CODEC_TYPE_VIDEO,
     CODEC_ID_VP6,
-    sizeof(vp56_context_t),
+    sizeof(VP56Context),
     vp6_decode_init,
     NULL,
     vp56_free,
     vp56_decode_frame,
+    CODEC_CAP_DR1,
+    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6"),
 };
 
 /* flash version, not flipped upside-down */
@@ -514,9 +617,25 @@ AVCodec vp6f_decoder = {
     "vp6f",
     CODEC_TYPE_VIDEO,
     CODEC_ID_VP6F,
-    sizeof(vp56_context_t),
+    sizeof(VP56Context),
+    vp6_decode_init,
+    NULL,
+    vp56_free,
+    vp56_decode_frame,
+    CODEC_CAP_DR1,
+    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6 (Flash version)"),
+};
+
+/* flash version, not flipped upside-down, with alpha channel */
+AVCodec vp6a_decoder = {
+    "vp6a",
+    CODEC_TYPE_VIDEO,
+    CODEC_ID_VP6A,
+    sizeof(VP56Context),
     vp6_decode_init,
     NULL,
     vp56_free,
     vp56_decode_frame,
+    CODEC_CAP_DR1,
+    .long_name = NULL_IF_CONFIG_SMALL("On2 VP6 (Flash version, with alpha channel)"),
 };