]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/ffv1.c
iirfilter: fix biquad filter coefficients.
[ffmpeg] / libavcodec / ffv1.c
index e502e5099a4c10908453d022edb8605495477f96..53edbb3459b7024ea5db2d71531a8354d03007aa 100644 (file)
@@ -3,20 +3,20 @@
  *
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
@@ -218,6 +218,7 @@ typedef struct VlcState{
 
 typedef struct PlaneContext{
     int16_t quant_table[MAX_CONTEXT_INPUTS][256];
+    int quant_table_index;
     int context_count;
     uint8_t (*state)[CONTEXT_SIZE];
     VlcState *vlc_state;
@@ -232,6 +233,7 @@ typedef struct FFV1Context{
     GetBitContext gb;
     PutBitContext pb;
     uint64_t rc_stat[256][2];
+    uint64_t (*rc_stat2[MAX_QUANT_TABLES])[32][2];
     int version;
     int width, height;
     int chroma_h_shift, chroma_v_shift;
@@ -245,9 +247,11 @@ typedef struct FFV1Context{
     int16_t quant_tables[MAX_QUANT_TABLES][MAX_CONTEXT_INPUTS][256];
     int context_count[MAX_QUANT_TABLES];
     uint8_t state_transition[256];
+    uint8_t (*initial_states[MAX_QUANT_TABLES])[32];
     int run_index;
     int colorspace;
-    int_fast16_t *sample_buffer;
+    int16_t *sample_buffer;
+    int gob_count;
 
     int quant_table_count;
 
@@ -275,7 +279,8 @@ static av_always_inline int fold(int diff, int bits){
     return diff;
 }
 
-static inline int predict(int_fast16_t *src, int_fast16_t *last){
+static inline int predict(int16_t *src, int16_t *last)
+{
     const int LT= last[-1];
     const int  T= last[ 0];
     const int L =  src[-1];
@@ -283,7 +288,9 @@ static inline int predict(int_fast16_t *src, int_fast16_t *last){
     return mid_pred(L, L + T - LT, T);
 }
 
-static inline int get_context(PlaneContext *p, int_fast16_t *src, int_fast16_t *last, int_fast16_t *last2){
+static inline int get_context(PlaneContext *p, int16_t *src,
+                              int16_t *last, int16_t *last2)
+{
     const int LT= last[-1];
     const int  T= last[ 0];
     const int RT= last[ 1];
@@ -298,12 +305,57 @@ static inline int get_context(PlaneContext *p, int_fast16_t *src, int_fast16_t *
         return p->quant_table[0][(L-LT) & 0xFF] + p->quant_table[1][(LT-T) & 0xFF] + p->quant_table[2][(T-RT) & 0xFF];
 }
 
-static inline void put_symbol_inline(RangeCoder *c, uint8_t *state, int v, int is_signed, uint64_t rc_stat[256][2]){
+static void find_best_state(uint8_t best_state[256][256], const uint8_t one_state[256]){
+    int i,j,k,m;
+    double l2tab[256];
+
+    for(i=1; i<256; i++)
+        l2tab[i]= log2(i/256.0);
+
+    for(i=0; i<256; i++){
+        double best_len[256];
+        double p= i/256.0;
+
+        for(j=0; j<256; j++)
+            best_len[j]= 1<<30;
+
+        for(j=FFMAX(i-10,1); j<FFMIN(i+11,256); j++){
+            double occ[256]={0};
+            double len=0;
+            occ[j]=1.0;
+            for(k=0; k<256; k++){
+                double newocc[256]={0};
+                for(m=0; m<256; m++){
+                    if(occ[m]){
+                        len -=occ[m]*(     p *l2tab[    m]
+                                      + (1-p)*l2tab[256-m]);
+                    }
+                }
+                if(len < best_len[k]){
+                    best_len[k]= len;
+                    best_state[i][k]= j;
+                }
+                for(m=0; m<256; m++){
+                    if(occ[m]){
+                        newocc[    one_state[    m]] += occ[m]*   p ;
+                        newocc[256-one_state[256-m]] += occ[m]*(1-p);
+                    }
+                }
+                memcpy(occ, newocc, sizeof(occ));
+            }
+        }
+    }
+}
+
+static av_always_inline av_flatten void put_symbol_inline(RangeCoder *c, uint8_t *state, int v, int is_signed, uint64_t rc_stat[256][2], uint64_t rc_stat2[32][2]){
     int i;
 
 #define put_rac(C,S,B) \
 do{\
+    if(rc_stat){\
     rc_stat[*(S)][B]++;\
+        rc_stat2[(S)-state][B]++;\
+    }\
     put_rac(C,S,B);\
 }while(0)
 
@@ -343,8 +395,7 @@ do{\
 }
 
 static void av_noinline put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
-    uint64_t rc_stat[256][2]; //we dont bother counting header bits.
-    put_symbol_inline(c, state, v, is_signed, rc_stat);
+    put_symbol_inline(c, state, v, is_signed, NULL, NULL);
 }
 
 static inline av_flatten int get_symbol_inline(RangeCoder *c, uint8_t *state, int is_signed){
@@ -458,7 +509,10 @@ static inline int get_vlc_symbol(GetBitContext *gb, VlcState * const state, int
 }
 
 #if CONFIG_FFV1_ENCODER
-static inline int encode_line(FFV1Context *s, int w, int_fast16_t *sample[2], int plane_index, int bits){
+static av_always_inline int encode_line(FFV1Context *s, int w,
+                                        int16_t *sample[2],
+                                        int plane_index, int bits)
+{
     PlaneContext * const p= &s->plane[plane_index];
     RangeCoder * const c= &s->c;
     int x;
@@ -492,7 +546,11 @@ static inline int encode_line(FFV1Context *s, int w, int_fast16_t *sample[2], in
         diff= fold(diff, bits);
 
         if(s->ac){
-            put_symbol_inline(c, p->state[context], diff, 1, s->rc_stat);
+            if(s->flags & CODEC_FLAG_PASS1){
+                put_symbol_inline(c, p->state[context], diff, 1, s->rc_stat, s->rc_stat2[p->quant_table_index][context]);
+            }else{
+                put_symbol_inline(c, p->state[context], diff, 1, NULL, NULL);
+            }
         }else{
             if(context == 0) run_mode=1;
 
@@ -539,7 +597,7 @@ static inline int encode_line(FFV1Context *s, int w, int_fast16_t *sample[2], in
 static void encode_plane(FFV1Context *s, uint8_t *src, int w, int h, int stride, int plane_index){
     int x,y,i;
     const int ring_size= s->avctx->context_model ? 3 : 2;
-    int_fast16_t *sample[3];
+    int16_t *sample[3];
     s->run_index=0;
 
     memset(s->sample_buffer, 0, ring_size*(w+6)*sizeof(*s->sample_buffer));
@@ -569,7 +627,7 @@ static void encode_plane(FFV1Context *s, uint8_t *src, int w, int h, int stride,
 static void encode_rgb_frame(FFV1Context *s, uint32_t *src, int w, int h, int stride){
     int x, y, p, i;
     const int ring_size= s->avctx->context_model ? 3 : 2;
-    int_fast16_t *sample[3][3];
+    int16_t *sample[3][3];
     s->run_index=0;
 
     memset(s->sample_buffer, 0, ring_size*3*(w+6)*sizeof(*s->sample_buffer));
@@ -658,8 +716,10 @@ static void write_header(FFV1Context *f){
             put_symbol(c, state, (fs->slice_y     +1)*f->num_v_slices / f->height  , 0);
             put_symbol(c, state, (fs->slice_width +1)*f->num_h_slices / f->width -1, 0);
             put_symbol(c, state, (fs->slice_height+1)*f->num_v_slices / f->height-1, 0);
-            for(j=0; j<f->plane_count; j++)
-            put_symbol(c, state, f->avctx->context_model, 0);
+            for(j=0; j<f->plane_count; j++){
+                put_symbol(c, state, f->plane[j].quant_table_index, 0);
+                av_assert0(f->plane[j].quant_table_index == f->avctx->context_model);
+            }
         }
     }
 }
@@ -731,27 +791,43 @@ static av_cold int init_slice_contexts(FFV1Context *f){
         int sye= f->avctx->height*(sy+1) / f->num_v_slices;
         f->slice_context[i]= fs;
         memcpy(fs, f, sizeof(*fs));
+        memset(fs->rc_stat2, 0, sizeof(fs->rc_stat2));
 
         fs->slice_width = sxe - sxs;
         fs->slice_height= sye - sys;
         fs->slice_x     = sxs;
         fs->slice_y     = sys;
 
-        fs->sample_buffer = av_malloc(6 * (fs->width+6) * sizeof(*fs->sample_buffer));
+        fs->sample_buffer = av_malloc(9 * (fs->width+6) * sizeof(*fs->sample_buffer));
         if (!fs->sample_buffer)
             return AVERROR(ENOMEM);
     }
     return 0;
 }
 
+static int allocate_initial_states(FFV1Context *f){
+    int i;
+
+    for(i=0; i<f->quant_table_count; i++){
+        f->initial_states[i]= av_malloc(f->context_count[i]*sizeof(*f->initial_states[i]));
+        if(!f->initial_states[i])
+            return AVERROR(ENOMEM);
+        memset(f->initial_states[i], 128, f->context_count[i]*sizeof(*f->initial_states[i]));
+    }
+    return 0;
+}
+
 #if CONFIG_FFV1_ENCODER
 static int write_extra_header(FFV1Context *f){
     RangeCoder * const c= &f->c;
     uint8_t state[CONTEXT_SIZE];
-    int i;
+    int i, j, k;
+    uint8_t state2[32][CONTEXT_SIZE];
+
+    memset(state2, 128, sizeof(state2));
     memset(state, 128, sizeof(state));
 
-    f->avctx->extradata= av_malloc(f->avctx->extradata_size= 10000);
+    f->avctx->extradata= av_malloc(f->avctx->extradata_size= 10000 + (11*11*5*5*5+11*11*11)*32);
     ff_init_range_encoder(c, f->avctx->extradata, f->avctx->extradata_size);
     ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 
@@ -775,15 +851,75 @@ static int write_extra_header(FFV1Context *f){
     for(i=0; i<f->quant_table_count; i++)
         write_quant_tables(c, f->quant_tables[i]);
 
+    for(i=0; i<f->quant_table_count; i++){
+        for(j=0; j<f->context_count[i]*CONTEXT_SIZE; j++)
+            if(f->initial_states[i] && f->initial_states[i][0][j] != 128)
+                break;
+        if(j<f->context_count[i]*CONTEXT_SIZE){
+            put_rac(c, state, 1);
+            for(j=0; j<f->context_count[i]; j++){
+                for(k=0; k<CONTEXT_SIZE; k++){
+                    int pred= j ? f->initial_states[i][j-1][k] : 128;
+                    put_symbol(c, state2[k], (int8_t)(f->initial_states[i][j][k]-pred), 1);
+                }
+            }
+        }else{
+            put_rac(c, state, 0);
+        }
+    }
+
     f->avctx->extradata_size= ff_rac_terminate(c);
 
     return 0;
 }
 
+static int sort_stt(FFV1Context *s, uint8_t stt[256]){
+    int i,i2,changed,print=0;
+
+    do{
+        changed=0;
+        for(i=12; i<244; i++){
+            for(i2=i+1; i2<245 && i2<i+4; i2++){
+#define COST(old, new) \
+    s->rc_stat[old][0]*-log2((256-(new))/256.0)\
+   +s->rc_stat[old][1]*-log2(     (new) /256.0)
+
+#define COST2(old, new) \
+    COST(old, new)\
+   +COST(256-(old), 256-(new))
+
+                double size0= COST2(i, i ) + COST2(i2, i2);
+                double sizeX= COST2(i, i2) + COST2(i2, i );
+                if(sizeX < size0 && i!=128 && i2!=128){
+                    int j;
+                    FFSWAP(int, stt[    i], stt[    i2]);
+                    FFSWAP(int, s->rc_stat[i    ][0],s->rc_stat[    i2][0]);
+                    FFSWAP(int, s->rc_stat[i    ][1],s->rc_stat[    i2][1]);
+                    if(i != 256-i2){
+                        FFSWAP(int, stt[256-i], stt[256-i2]);
+                        FFSWAP(int, s->rc_stat[256-i][0],s->rc_stat[256-i2][0]);
+                        FFSWAP(int, s->rc_stat[256-i][1],s->rc_stat[256-i2][1]);
+                    }
+                    for(j=1; j<256; j++){
+                        if     (stt[j] == i ) stt[j] = i2;
+                        else if(stt[j] == i2) stt[j] = i ;
+                        if(i != 256-i2){
+                            if     (stt[256-j] == 256-i ) stt[256-j] = 256-i2;
+                            else if(stt[256-j] == 256-i2) stt[256-j] = 256-i ;
+                        }
+                    }
+                    print=changed=1;
+                }
+            }
+        }
+    }while(changed);
+    return print;
+}
+
 static av_cold int encode_init(AVCodecContext *avctx)
 {
     FFV1Context *s = avctx->priv_data;
-    int i, j, i2;
+    int i, j, k, m;
 
     common_init(avctx);
 
@@ -817,19 +953,21 @@ static av_cold int encode_init(AVCodecContext *avctx)
             s->quant_tables[1][4][i]= 5*5*11*11*quant5_10bit[i];
         }
     }
+    s->context_count[0]= (11*11*11+1)/2;
+    s->context_count[1]= (11*11*5*5*5+1)/2;
     memcpy(s->quant_table, s->quant_tables[avctx->context_model], sizeof(s->quant_table));
 
     for(i=0; i<s->plane_count; i++){
         PlaneContext * const p= &s->plane[i];
 
         memcpy(p->quant_table, s->quant_table, sizeof(p->quant_table));
-        if(avctx->context_model==0){
-            p->context_count= (11*11*11+1)/2;
-        }else{
-            p->context_count= (11*11*5*5*5+1)/2;
-        }
+        p->quant_table_index= avctx->context_model;
+        p->context_count= s->context_count[p->quant_table_index];
     }
 
+    if(allocate_initial_states(s) < 0)
+        return AVERROR(ENOMEM);
+
     avctx->coded_frame= &s->picture;
     switch(avctx->pix_fmt){
     case PIX_FMT_YUV444P16:
@@ -862,14 +1000,24 @@ static av_cold int encode_init(AVCodecContext *avctx)
 
     s->picture_number=0;
 
+    if(avctx->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)){
+        for(i=0; i<s->quant_table_count; i++){
+            s->rc_stat2[i]= av_mallocz(s->context_count[i]*sizeof(*s->rc_stat2[i]));
+            if(!s->rc_stat2[i])
+                return AVERROR(ENOMEM);
+        }
+    }
     if(avctx->stats_in){
         char *p= avctx->stats_in;
-        int changed;
+        uint8_t best_state[256][256];
+        int gob_count=0;
+        char *next;
+
+        av_assert0(s->version>=2);
 
         for(;;){
             for(j=0; j<256; j++){
                 for(i=0; i<2; i++){
-                    char *next;
                     s->rc_stat[j][i]= strtol(p, &next, 0);
                     if(next==p){
                         av_log(avctx, AV_LOG_ERROR, "2Pass file invalid at %d %d [%s]\n", j,i,p);
@@ -878,43 +1026,44 @@ static av_cold int encode_init(AVCodecContext *avctx)
                     p=next;
                 }
             }
+            for(i=0; i<s->quant_table_count; i++){
+                for(j=0; j<s->context_count[i]; j++){
+                    for(k=0; k<32; k++){
+                        for(m=0; m<2; m++){
+                            s->rc_stat2[i][j][k][m]= strtol(p, &next, 0);
+                            if(next==p){
+                                av_log(avctx, AV_LOG_ERROR, "2Pass file invalid at %d %d %d %d [%s]\n", i,j,k,m,p);
+                                return -1;
+                            }
+                            p=next;
+                        }
+                    }
+                }
+            }
+            gob_count= strtol(p, &next, 0);
+            if(next==p || gob_count <0){
+                av_log(avctx, AV_LOG_ERROR, "2Pass file invalid\n");
+                return -1;
+            }
+            p=next;
             while(*p=='\n' || *p==' ') p++;
             if(p[0]==0) break;
         }
+        sort_stt(s, s->state_transition);
 
-        do{
-            changed=0;
-            for(i=12; i<244; i++){
-                for(i2=i+1; i2<245; i2++){
-#define COST(old, new) \
-    s->rc_stat[old][0]*-log2((256-(new))/256.0)\
-   +s->rc_stat[old][1]*-log2(     (new) /256.0)
-
-#define COST2(old, new) \
-    COST(old, new)\
-   +COST(256-(old), 256-(new))
+        find_best_state(best_state, s->state_transition);
 
-                    double size0= COST2(i, i ) + COST2(i2, i2);
-                    double sizeX= COST2(i, i2) + COST2(i2, i );
-                    if(sizeX < size0 && i!=128 && i2!=128){ //128 is special we cant swap it around FIXME 127<->129 swap
-                        int j;
-                        FFSWAP(int, s->state_transition[    i], s->state_transition[    i2]);
-                        FFSWAP(int, s->state_transition[256-i], s->state_transition[256-i2]);
-                        FFSWAP(int, s->rc_stat[i    ][0],s->rc_stat[    i2][0]);
-                        FFSWAP(int, s->rc_stat[i    ][1],s->rc_stat[    i2][1]);
-                        FFSWAP(int, s->rc_stat[256-i][0],s->rc_stat[256-i2][0]);
-                        FFSWAP(int, s->rc_stat[256-i][1],s->rc_stat[256-i2][1]);
-                        for(j=1; j<256; j++){
-                            if     (s->state_transition[j] == i ) s->state_transition[j] = i2;
-                            else if(s->state_transition[j] == i2) s->state_transition[j] = i ;
-                            if     (s->state_transition[256-j] == 256-i ) s->state_transition[256-j] = 256-i2;
-                            else if(s->state_transition[256-j] == 256-i2) s->state_transition[256-j] = 256-i ;
-                        }
-                        changed=1;
+        for(i=0; i<s->quant_table_count; i++){
+            for(j=0; j<s->context_count[i]; j++){
+                for(k=0; k<32; k++){
+                    double p= 128;
+                    if(s->rc_stat2[i][j][k][0]+s->rc_stat2[i][j][k][1]){
+                        p=256.0*s->rc_stat2[i][j][k][1] / (s->rc_stat2[i][j][k][0]+s->rc_stat2[i][j][k][1]);
                     }
+                    s->initial_states[i][j][k]= best_state[av_clip(round(p), 1, 255)][av_clip((s->rc_stat2[i][j][k][0]+s->rc_stat2[i][j][k][1])/gob_count, 0, 255)];
                 }
             }
-        }while(changed);
+        }
     }
 
     if(s->version>1){
@@ -928,7 +1077,19 @@ static av_cold int encode_init(AVCodecContext *avctx)
     if(init_slice_state(s) < 0)
         return -1;
 
-    avctx->stats_out= av_mallocz(1024*30);
+#define STATS_OUT_SIZE 1024*1024*6
+    if(avctx->flags & CODEC_FLAG_PASS1){
+        avctx->stats_out= av_mallocz(STATS_OUT_SIZE);
+        for(i=0; i<s->quant_table_count; i++){
+            for(j=0; j<s->slice_count; j++){
+                FFV1Context *sf= s->slice_context[j];
+                av_assert0(!sf->rc_stat2[i]);
+                sf->rc_stat2[i]= av_mallocz(s->context_count[i]*sizeof(*sf->rc_stat2[i]));
+                if(!sf->rc_stat2[i])
+                    return AVERROR(ENOMEM);
+            }
+        }
+    }
 
     return 0;
 }
@@ -946,15 +1107,18 @@ static void clear_state(FFV1Context *f){
             p->interlace_bit_state[0]= 128;
             p->interlace_bit_state[1]= 128;
 
+            if(fs->ac){
+                if(f->initial_states[p->quant_table_index]){
+                    memcpy(p->state, f->initial_states[p->quant_table_index], CONTEXT_SIZE*p->context_count);
+                }else
+                memset(p->state, 128, CONTEXT_SIZE*p->context_count);
+            }else{
             for(j=0; j<p->context_count; j++){
-                if(fs->ac){
-                    memset(p->state[j], 128, sizeof(uint8_t)*CONTEXT_SIZE);
-                }else{
                     p->vlc_state[j].drift= 0;
                     p->vlc_state[j].error_sum= 4; //FFMAX((RANGE + 32)/64, 2);
                     p->vlc_state[j].bias= 0;
                     p->vlc_state[j].count= 1;
-                }
+            }
             }
         }
     }
@@ -1002,11 +1166,12 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
     ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 
     *p = *pict;
-    p->pict_type= FF_I_TYPE;
+    p->pict_type= AV_PICTURE_TYPE_I;
 
     if(avctx->gop_size==0 || f->picture_number % avctx->gop_size == 0){
         put_rac(c, &keystate, 1);
         p->key_frame= 1;
+        f->gob_count++;
         write_header(f);
         clear_state(f);
     }else{
@@ -1064,17 +1229,28 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
     }
 
     if((avctx->flags&CODEC_FLAG_PASS1) && (f->picture_number&31)==0){
-        int j;
+        int j, k, m;
         char *p= avctx->stats_out;
-        char *end= p + 1024*30;
+        char *end= p + STATS_OUT_SIZE;
 
         memset(f->rc_stat, 0, sizeof(f->rc_stat));
+        for(i=0; i<f->quant_table_count; i++)
+            memset(f->rc_stat2[i], 0, f->context_count[i]*sizeof(*f->rc_stat2[i]));
+
         for(j=0; j<f->slice_count; j++){
             FFV1Context *fs= f->slice_context[j];
             for(i=0; i<256; i++){
                 f->rc_stat[i][0] += fs->rc_stat[i][0];
                 f->rc_stat[i][1] += fs->rc_stat[i][1];
             }
+            for(i=0; i<f->quant_table_count; i++){
+                for(k=0; k<f->context_count[i]; k++){
+                    for(m=0; m<32; m++){
+                        f->rc_stat2[i][k][m][0] += fs->rc_stat2[i][k][m][0];
+                        f->rc_stat2[i][k][m][1] += fs->rc_stat2[i][k][m][1];
+                    }
+                }
+            }
         }
 
         for(j=0; j<256; j++){
@@ -1082,7 +1258,17 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
             p+= strlen(p);
         }
         snprintf(p, end-p, "\n");
-    } else
+
+        for(i=0; i<f->quant_table_count; i++){
+            for(j=0; j<f->context_count[i]; j++){
+                for(m=0; m<32; m++){
+                    snprintf(p, end-p, "%"PRIu64" %"PRIu64" ", f->rc_stat2[i][j][m][0], f->rc_stat2[i][j][m][1]);
+                    p+= strlen(p);
+                }
+            }
+        }
+        snprintf(p, end-p, "%d\n", f->gob_count);
+    } else if(avctx->flags&CODEC_FLAG_PASS1)
         avctx->stats_out[0] = '\0';
 
     f->picture_number++;
@@ -1094,6 +1280,9 @@ static av_cold int common_end(AVCodecContext *avctx){
     FFV1Context *s = avctx->priv_data;
     int i, j;
 
+    if (avctx->codec->decode && s->picture.data[0])
+        avctx->release_buffer(avctx, &s->picture);
+
     for(j=0; j<s->slice_count; j++){
         FFV1Context *fs= s->slice_context[j];
         for(i=0; i<s->plane_count; i++){
@@ -1106,11 +1295,26 @@ static av_cold int common_end(AVCodecContext *avctx){
     }
 
     av_freep(&avctx->stats_out);
+    for(j=0; j<s->quant_table_count; j++){
+        av_freep(&s->initial_states[j]);
+        for(i=0; i<s->slice_count; i++){
+            FFV1Context *sf= s->slice_context[i];
+            av_freep(&sf->rc_stat2[j]);
+        }
+        av_freep(&s->rc_stat2[j]);
+    }
+
+    for(i=0; i<s->slice_count; i++){
+        av_freep(&s->slice_context[i]);
+    }
 
     return 0;
 }
 
-static av_always_inline void decode_line(FFV1Context *s, int w, int_fast16_t *sample[2], int plane_index, int bits){
+static av_always_inline void decode_line(FFV1Context *s, int w,
+                                         int16_t *sample[2],
+                                         int plane_index, int bits)
+{
     PlaneContext * const p= &s->plane[plane_index];
     RangeCoder * const c= &s->c;
     int x;
@@ -1170,7 +1374,7 @@ static av_always_inline void decode_line(FFV1Context *s, int w, int_fast16_t *sa
 
 static void decode_plane(FFV1Context *s, uint8_t *src, int w, int h, int stride, int plane_index){
     int x, y;
-    int_fast16_t *sample[2];
+    int16_t *sample[2];
     sample[0]=s->sample_buffer    +3;
     sample[1]=s->sample_buffer+w+6+3;
 
@@ -1179,7 +1383,7 @@ static void decode_plane(FFV1Context *s, uint8_t *src, int w, int h, int stride,
     memset(s->sample_buffer, 0, 2*(w+6)*sizeof(*s->sample_buffer));
 
     for(y=0; y<h; y++){
-        int_fast16_t *temp= sample[0]; //FIXME try a normal buffer
+        int16_t *temp = sample[0]; //FIXME try a normal buffer
 
         sample[0]= sample[1];
         sample[1]= temp;
@@ -1205,7 +1409,7 @@ static void decode_plane(FFV1Context *s, uint8_t *src, int w, int h, int stride,
 
 static void decode_rgb_frame(FFV1Context *s, uint32_t *src, int w, int h, int stride){
     int x, y, p;
-    int_fast16_t *sample[3][2];
+    int16_t *sample[3][2];
     for(x=0; x<3; x++){
         sample[x][0] = s->sample_buffer +  x*2   *(w+6) + 3;
         sample[x][1] = s->sample_buffer + (x*2+1)*(w+6) + 3;
@@ -1217,7 +1421,7 @@ static void decode_rgb_frame(FFV1Context *s, uint32_t *src, int w, int h, int st
 
     for(y=0; y<h; y++){
         for(p=0; p<3; p++){
-            int_fast16_t *temp= sample[p][0]; //FIXME try a normal buffer
+            int16_t *temp = sample[p][0]; //FIXME try a normal buffer
 
             sample[p][0]= sample[p][1];
             sample[p][1]= temp;
@@ -1317,8 +1521,10 @@ static int read_quant_tables(RangeCoder *c, int16_t quant_table[MAX_CONTEXT_INPU
 static int read_extra_header(FFV1Context *f){
     RangeCoder * const c= &f->c;
     uint8_t state[CONTEXT_SIZE];
-    int i;
+    int i, j, k;
+    uint8_t state2[32][CONTEXT_SIZE];
 
+    memset(state2, 128, sizeof(state2));
     memset(state, 128, sizeof(state));
 
     ff_init_range_decoder(c, f->avctx->extradata, f->avctx->extradata_size);
@@ -1355,6 +1561,20 @@ static int read_extra_header(FFV1Context *f){
         }
     }
 
+    if(allocate_initial_states(f) < 0)
+        return AVERROR(ENOMEM);
+
+    for(i=0; i<f->quant_table_count; i++){
+        if(get_rac(c, state)){
+            for(j=0; j<f->context_count[i]; j++){
+                for(k=0; k<CONTEXT_SIZE; k++){
+                    int pred= j ? f->initial_states[i][j-1][k] : 128;
+                    f->initial_states[i][j][k]= (pred+get_symbol(c, state2[k], 1))&0xFF;
+                }
+            }
+        }
+    }
+
     return 0;
 }
 
@@ -1459,6 +1679,7 @@ static int read_header(FFV1Context *f){
                     av_log(f->avctx, AV_LOG_ERROR, "quant_table_index out of range\n");
                     return -1;
                 }
+                p->quant_table_index= idx;
                 memcpy(p->quant_table, f->quant_tables[idx], sizeof(p->quant_table));
                 context_count= f->context_count[idx];
             }else{
@@ -1503,11 +1724,15 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
 
     AVFrame *picture = data;
 
+    /* release previously stored data */
+    if (p->data[0])
+        avctx->release_buffer(avctx, p);
+
     ff_init_range_decoder(c, buf, buf_size);
     ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 
 
-    p->pict_type= FF_I_TYPE; //FIXME I vs. P
+    p->pict_type= AV_PICTURE_TYPE_I; //FIXME I vs. P
     if(get_rac(c, &keystate)){
         p->key_frame= 1;
         if(read_header(f) < 0)
@@ -1565,15 +1790,12 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
     f->picture_number++;
 
     *picture= *p;
-
-    avctx->release_buffer(avctx, p); //FIXME
-
     *data_size = sizeof(AVFrame);
 
     return buf_size;
 }
 
-AVCodec ffv1_decoder = {
+AVCodec ff_ffv1_decoder = {
     "ffv1",
     AVMEDIA_TYPE_VIDEO,
     CODEC_ID_FFV1,
@@ -1582,13 +1804,13 @@ AVCodec ffv1_decoder = {
     NULL,
     common_end,
     decode_frame,
-    CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
+    CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/ | CODEC_CAP_SLICE_THREADS,
     NULL,
     .long_name= NULL_IF_CONFIG_SMALL("FFmpeg video codec #1"),
 };
 
 #if CONFIG_FFV1_ENCODER
-AVCodec ffv1_encoder = {
+AVCodec ff_ffv1_encoder = {
     "ffv1",
     AVMEDIA_TYPE_VIDEO,
     CODEC_ID_FFV1,
@@ -1596,6 +1818,7 @@ AVCodec ffv1_encoder = {
     encode_init,
     encode_frame,
     common_end,
+    .capabilities = CODEC_CAP_SLICE_THREADS,
     .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV444P, PIX_FMT_YUV422P, PIX_FMT_YUV411P, PIX_FMT_YUV410P, PIX_FMT_RGB32, PIX_FMT_YUV420P16, PIX_FMT_YUV422P16, PIX_FMT_YUV444P16, PIX_FMT_NONE},
     .long_name= NULL_IF_CONFIG_SMALL("FFmpeg video codec #1"),
 };