VP8: idct_mb optimizations
[ffmpeg.git] / libavcodec / huffyuv.c
index 6c4b3ce..2eb5864 100644 (file)
@@ -3,43 +3,48 @@
  *
  * Copyright (c) 2002-2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * This library is free software; you can redistribute it and/or
+ * see http://www.pcisys.net/~melanson/codecs/huffyuv.txt for a description of
+ * the algorithm used
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
  *
- * This library is distributed in the hope that it will be useful,
+ * FFmpeg 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 this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- *
- * see http://www.pcisys.net/~melanson/codecs/huffyuv.txt for a description of
- * the algorithm used 
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
+
 /**
- * @file huffyuv.c
+ * @file
  * huffyuv codec for libavcodec.
  */
 
-#include "common.h"
 #include "avcodec.h"
+#include "get_bits.h"
+#include "put_bits.h"
 #include "dsputil.h"
 
 #define VLC_BITS 11
 
-#ifdef WORDS_BIGENDIAN
+#if HAVE_BIGENDIAN
 #define B 3
 #define G 2
 #define R 1
+#define A 0
 #else
 #define B 0
 #define G 1
 #define R 2
+#define A 3
 #endif
 
 typedef enum Predictor{
@@ -47,7 +52,7 @@ typedef enum Predictor{
     PLANE,
     MEDIAN,
 } Predictor;
+
 typedef struct HYuvContext{
     AVCodecContext *avctx;
     Predictor predictor;
@@ -61,16 +66,19 @@ typedef struct HYuvContext{
     int bgr32;                              //use bgr32 instead of bgr24
     int width, height;
     int flags;
+    int context;
     int picture_number;
     int last_slice_end;
-    uint8_t __align8 temp[3][2560];
+    uint8_t *temp[3];
     uint64_t stats[3][256];
     uint8_t len[3][256];
     uint32_t bits[3][256];
-    VLC vlc[3];
+    uint32_t pix_bgr_map[1<<VLC_BITS];
+    VLC vlc[6];                             //Y,U,V,YY,YU,YV
     AVFrame picture;
-    uint8_t __align8 bitstream_buffer[1024*1024*3]; //FIXME dynamic alloc or some other solution
-    DSPContext dsp; 
+    uint8_t *bitstream_buffer;
+    unsigned int bitstream_buffer_size;
+    DSPContext dsp;
 }HYuvContext;
 
 static const unsigned char classic_shift_luma[] = {
@@ -123,64 +131,6 @@ static const unsigned char classic_add_chroma[256] = {
     6, 12,  8, 10,  7,  9,  6,  4,  6,  2,  2,  3,  3,  3,  3,  2,
 };
 
-static inline int add_left_prediction(uint8_t *dst, uint8_t *src, int w, int acc){
-    int i;
-
-    for(i=0; i<w-1; i++){
-        acc+= src[i];
-        dst[i]= acc;
-        i++;
-        acc+= src[i];
-        dst[i]= acc;
-    }
-
-    for(; i<w; i++){
-        acc+= src[i];
-        dst[i]= acc;
-    }
-
-    return acc;
-}
-
-static inline void add_median_prediction(uint8_t *dst, uint8_t *src1, uint8_t *diff, int w, int *left, int *left_top){
-    int i;
-    uint8_t l, lt;
-
-    l= *left;
-    lt= *left_top;
-
-    for(i=0; i<w; i++){
-        l= mid_pred(l, src1[i], (l + src1[i] - lt)&0xFF) + diff[i];
-        lt= src1[i];
-        dst[i]= l;
-    }    
-
-    *left= l;
-    *left_top= lt;
-}
-
-static inline void add_left_prediction_bgr32(uint8_t *dst, uint8_t *src, int w, int *red, int *green, int *blue){
-    int i;
-    int r,g,b;
-    r= *red;
-    g= *green;
-    b= *blue;
-
-    for(i=0; i<w; i++){
-        b+= src[4*i+B];
-        g+= src[4*i+G];
-        r+= src[4*i+R];
-        
-        dst[4*i+B]= b;
-        dst[4*i+G]= g;
-        dst[4*i+R]= r;
-    }
-
-    *red= r;
-    *green= g;
-    *blue= b;
-}
-
 static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst, uint8_t *src, int w, int left){
     int i;
     if(w<32){
@@ -201,21 +151,49 @@ static inline int sub_left_prediction(HYuvContext *s, uint8_t *dst, uint8_t *src
     }
 }
 
-static void read_len_table(uint8_t *dst, GetBitContext *gb){
+static inline void sub_left_prediction_bgr32(HYuvContext *s, uint8_t *dst, uint8_t *src, int w, int *red, int *green, int *blue){
+    int i;
+    int r,g,b;
+    r= *red;
+    g= *green;
+    b= *blue;
+    for(i=0; i<FFMIN(w,4); i++){
+        const int rt= src[i*4+R];
+        const int gt= src[i*4+G];
+        const int bt= src[i*4+B];
+        dst[i*4+R]= rt - r;
+        dst[i*4+G]= gt - g;
+        dst[i*4+B]= bt - b;
+        r = rt;
+        g = gt;
+        b = bt;
+    }
+    s->dsp.diff_bytes(dst+16, src+16, src+12, w*4-16);
+    *red=   src[(w-1)*4+R];
+    *green= src[(w-1)*4+G];
+    *blue=  src[(w-1)*4+B];
+}
+
+static int read_len_table(uint8_t *dst, GetBitContext *gb){
     int i, val, repeat;
-  
+
     for(i=0; i<256;){
         repeat= get_bits(gb, 3);
         val   = get_bits(gb, 5);
         if(repeat==0)
             repeat= get_bits(gb, 8);
 //printf("%d %d\n", val, repeat);
+        if(i+repeat > 256) {
+            av_log(NULL, AV_LOG_ERROR, "Error reading huffman table\n");
+            return -1;
+        }
         while (repeat--)
             dst[i++] = val;
     }
+    return 0;
 }
 
-static int generate_bits_table(uint32_t *dst, uint8_t *len_table){
+static int generate_bits_table(uint32_t *dst, const uint8_t *len_table){
     int len, index;
     uint32_t bits=0;
 
@@ -233,71 +211,143 @@ static int generate_bits_table(uint32_t *dst, uint8_t *len_table){
     return 0;
 }
 
-static void generate_len_table(uint8_t *dst, uint64_t *stats, int size){
-    uint64_t counts[2*size];
-    int up[2*size];
+#if CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER
+typedef struct {
+    uint64_t val;
+    int name;
+} HeapElem;
+
+static void heap_sift(HeapElem *h, int root, int size)
+{
+    while(root*2+1 < size) {
+        int child = root*2+1;
+        if(child < size-1 && h[child].val > h[child+1].val)
+            child++;
+        if(h[root].val > h[child].val) {
+            FFSWAP(HeapElem, h[root], h[child]);
+            root = child;
+        } else
+            break;
+    }
+}
+
+static void generate_len_table(uint8_t *dst, const uint64_t *stats){
+    HeapElem h[256];
+    int up[2*256];
+    int len[2*256];
     int offset, i, next;
-    
+    int size = 256;
+
     for(offset=1; ; offset<<=1){
         for(i=0; i<size; i++){
-            counts[i]= stats[i] + offset - 1;
+            h[i].name = i;
+            h[i].val = (stats[i] << 8) + offset;
+        }
+        for(i=size/2-1; i>=0; i--)
+            heap_sift(h, i, size);
+
+        for(next=size; next<size*2-1; next++){
+            // merge the two smallest entries, and put it back in the heap
+            uint64_t min1v = h[0].val;
+            up[h[0].name] = next;
+            h[0].val = INT64_MAX;
+            heap_sift(h, 0, size);
+            up[h[0].name] = next;
+            h[0].name = next;
+            h[0].val += min1v;
+            heap_sift(h, 0, size);
+        }
+
+        len[2*size-2] = 0;
+        for(i=2*size-3; i>=size; i--)
+            len[i] = len[up[i]] + 1;
+        for(i=0; i<size; i++) {
+            dst[i] = len[up[i]] + 1;
+            if(dst[i] >= 32) break;
+        }
+        if(i==size) break;
+    }
+}
+#endif /* CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER */
+
+static void generate_joint_tables(HYuvContext *s){
+    uint16_t symbols[1<<VLC_BITS];
+    uint16_t bits[1<<VLC_BITS];
+    uint8_t len[1<<VLC_BITS];
+    if(s->bitstream_bpp < 24){
+        int p, i, y, u;
+        for(p=0; p<3; p++){
+            for(i=y=0; y<256; y++){
+                int len0 = s->len[0][y];
+                int limit = VLC_BITS - len0;
+                if(limit <= 0)
+                    continue;
+                for(u=0; u<256; u++){
+                    int len1 = s->len[p][u];
+                    if(len1 > limit)
+                        continue;
+                    len[i] = len0 + len1;
+                    bits[i] = (s->bits[0][y] << len1) + s->bits[p][u];
+                    symbols[i] = (y<<8) + u;
+                    if(symbols[i] != 0xffff) // reserved to mean "invalid"
+                        i++;
+                }
+            }
+            free_vlc(&s->vlc[3+p]);
+            init_vlc_sparse(&s->vlc[3+p], VLC_BITS, i, len, 1, 1, bits, 2, 2, symbols, 2, 2, 0);
         }
-        
-        for(next=size; next<size*2; next++){
-            uint64_t min1, min2;
-            int min1_i, min2_i;
-            
-            min1=min2= INT64_MAX;
-            min1_i= min2_i=-1;
-            
-            for(i=0; i<next; i++){
-                if(min2 > counts[i]){
-                    if(min1 > counts[i]){
-                        min2= min1;
-                        min2_i= min1_i;
-                        min1= counts[i];
-                        min1_i= i;
+    }else{
+        uint8_t (*map)[4] = (uint8_t(*)[4])s->pix_bgr_map;
+        int i, b, g, r, code;
+        int p0 = s->decorrelate;
+        int p1 = !s->decorrelate;
+        // restrict the range to +/-16 becaues that's pretty much guaranteed to
+        // cover all the combinations that fit in 11 bits total, and it doesn't
+        // matter if we miss a few rare codes.
+        for(i=0, g=-16; g<16; g++){
+            int len0 = s->len[p0][g&255];
+            int limit0 = VLC_BITS - len0;
+            if(limit0 < 2)
+                continue;
+            for(b=-16; b<16; b++){
+                int len1 = s->len[p1][b&255];
+                int limit1 = limit0 - len1;
+                if(limit1 < 1)
+                    continue;
+                code = (s->bits[p0][g&255] << len1) + s->bits[p1][b&255];
+                for(r=-16; r<16; r++){
+                    int len2 = s->len[2][r&255];
+                    if(len2 > limit1)
+                        continue;
+                    len[i] = len0 + len1 + len2;
+                    bits[i] = (code << len2) + s->bits[2][r&255];
+                    if(s->decorrelate){
+                        map[i][G] = g;
+                        map[i][B] = g+b;
+                        map[i][R] = g+r;
                     }else{
-                        min2= counts[i];
-                        min2_i= i;
+                        map[i][B] = g;
+                        map[i][G] = b;
+                        map[i][R] = r;
                     }
+                    i++;
                 }
             }
-            
-            if(min2==INT64_MAX) break;
-            
-            counts[next]= min1 + min2;
-            counts[min1_i]=
-            counts[min2_i]= INT64_MAX;
-            up[min1_i]=
-            up[min2_i]= next;
-            up[next]= -1;
         }
-        
-        for(i=0; i<size; i++){
-            int len;
-            int index=i;
-            
-            for(len=0; up[index] != -1; len++)
-                index= up[index];
-                
-            if(len > 32) break;
-            
-            dst[i]= len;
-        }
-        if(i==size) break;
+        free_vlc(&s->vlc[3]);
+        init_vlc(&s->vlc[3], VLC_BITS, i, len, 1, 1, bits, 2, 2, 0);
     }
 }
 
-static int read_huffman_tables(HYuvContext *s, uint8_t *src, int length){
+static int read_huffman_tables(HYuvContext *s, const uint8_t *src, int length){
     GetBitContext gb;
     int i;
-    
+
     init_get_bits(&gb, src, length*8);
-    
+
     for(i=0; i<3; i++){
-        read_len_table(s->len[i], &gb);
-        
+        if(read_len_table(s->len[i], &gb)<0)
+            return -1;
         if(generate_bits_table(s->bits[i], s->len[i])<0){
             return -1;
         }
@@ -306,10 +356,13 @@ for(j=0; j<256; j++){
 printf("%6X, %2d,  %3d\n", s->bits[i][j], s->len[i][j], j);
 }
 #endif
-        init_vlc(&s->vlc[i], VLC_BITS, 256, s->len[i], 1, 1, s->bits[i], 4, 4);
+        free_vlc(&s->vlc[i]);
+        init_vlc(&s->vlc[i], VLC_BITS, 256, s->len[i], 1, 1, s->bits[i], 4, 4, 0);
     }
-    
-    return 0;
+
+    generate_joint_tables(s);
+
+    return (get_bits_count(&gb)+7)/8;
 }
 
 static int read_old_huffman_tables(HYuvContext *s){
@@ -318,10 +371,12 @@ static int read_old_huffman_tables(HYuvContext *s){
     int i;
 
     init_get_bits(&gb, classic_shift_luma, sizeof(classic_shift_luma)*8);
-    read_len_table(s->len[0], &gb);
+    if(read_len_table(s->len[0], &gb)<0)
+        return -1;
     init_get_bits(&gb, classic_shift_chroma, sizeof(classic_shift_chroma)*8);
-    read_len_table(s->len[1], &gb);
-    
+    if(read_len_table(s->len[1], &gb)<0)
+        return -1;
+
     for(i=0; i<256; i++) s->bits[0][i] = classic_add_luma  [i];
     for(i=0; i<256; i++) s->bits[1][i] = classic_add_chroma[i];
 
@@ -331,57 +386,90 @@ static int read_old_huffman_tables(HYuvContext *s){
     }
     memcpy(s->bits[2], s->bits[1], 256*sizeof(uint32_t));
     memcpy(s->len[2] , s->len [1], 256*sizeof(uint8_t));
-    
-    for(i=0; i<3; i++)
-        init_vlc(&s->vlc[i], VLC_BITS, 256, s->len[i], 1, 1, s->bits[i], 4, 4);
-    
+
+    for(i=0; i<3; i++){
+        free_vlc(&s->vlc[i]);
+        init_vlc(&s->vlc[i], VLC_BITS, 256, s->len[i], 1, 1, s->bits[i], 4, 4, 0);
+    }
+
+    generate_joint_tables(s);
+
     return 0;
 #else
-    fprintf(stderr, "v1 huffyuv is not supported \n");
+    av_log(s->avctx, AV_LOG_DEBUG, "v1 huffyuv is not supported \n");
     return -1;
 #endif
 }
 
-static int decode_init(AVCodecContext *avctx)
-{
+static av_cold void alloc_temp(HYuvContext *s){
+    int i;
+
+    if(s->bitstream_bpp<24){
+        for(i=0; i<3; i++){
+            s->temp[i]= av_malloc(s->width + 16);
+        }
+    }else{
+        s->temp[0]= av_mallocz(4*s->width + 16);
+    }
+}
+
+static av_cold int common_init(AVCodecContext *avctx){
     HYuvContext *s = avctx->priv_data;
-    int width, height;
 
     s->avctx= avctx;
     s->flags= avctx->flags;
-        
+
     dsputil_init(&s->dsp, avctx);
-    
-    width= s->width= avctx->width;
-    height= s->height= avctx->height;
+
+    s->width= avctx->width;
+    s->height= avctx->height;
+    assert(s->width>0 && s->height>0);
+
+    return 0;
+}
+
+#if CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER
+static av_cold int decode_init(AVCodecContext *avctx)
+{
+    HYuvContext *s = avctx->priv_data;
+
+    common_init(avctx);
+    memset(s->vlc, 0, 3*sizeof(VLC));
+
     avctx->coded_frame= &s->picture;
+    s->interlaced= s->height > 288;
 
 s->bgr32=1;
-    assert(width && height);
 //if(avctx->extradata)
 //  printf("extradata:%X, extradata_size:%d\n", *(uint32_t*)avctx->extradata, avctx->extradata_size);
     if(avctx->extradata_size){
-        if((avctx->bits_per_sample&7) && avctx->bits_per_sample != 12)
+        if((avctx->bits_per_coded_sample&7) && avctx->bits_per_coded_sample != 12)
             s->version=1; // do such files exist at all?
         else
             s->version=2;
     }else
         s->version=0;
-    
+
     if(s->version==2){
-        int method;
+        int method, interlace;
+
+        if (avctx->extradata_size < 4)
+            return -1;
 
         method= ((uint8_t*)avctx->extradata)[0];
         s->decorrelate= method&64 ? 1 : 0;
         s->predictor= method&63;
         s->bitstream_bpp= ((uint8_t*)avctx->extradata)[1];
-        if(s->bitstream_bpp==0) 
-            s->bitstream_bpp= avctx->bits_per_sample&~7;
-            
-        if(read_huffman_tables(s, ((uint8_t*)avctx->extradata)+4, avctx->extradata_size) < 0)
+        if(s->bitstream_bpp==0)
+            s->bitstream_bpp= avctx->bits_per_coded_sample&~7;
+        interlace= (((uint8_t*)avctx->extradata)[2] & 0x30) >> 4;
+        s->interlaced= (interlace==1) ? 1 : (interlace==2) ? 0 : s->interlaced;
+        s->context= ((uint8_t*)avctx->extradata)[2] & 0x40 ? 1 : 0;
+
+        if(read_huffman_tables(s, ((uint8_t*)avctx->extradata)+4, avctx->extradata_size-4) < 0)
             return -1;
     }else{
-        switch(avctx->bits_per_sample&7){
+        switch(avctx->bits_per_coded_sample&7){
         case 1:
             s->predictor= LEFT;
             s->decorrelate= 0;
@@ -392,7 +480,7 @@ s->bgr32=1;
             break;
         case 3:
             s->predictor= PLANE;
-            s->decorrelate= avctx->bits_per_sample >= 24;
+            s->decorrelate= avctx->bits_per_coded_sample >= 24;
             break;
         case 4:
             s->predictor= MEDIAN;
@@ -403,21 +491,20 @@ s->bgr32=1;
             s->decorrelate= 0;
             break;
         }
-        s->bitstream_bpp= avctx->bits_per_sample & ~7;
-        
+        s->bitstream_bpp= avctx->bits_per_coded_sample & ~7;
+        s->context= 0;
+
         if(read_old_huffman_tables(s) < 0)
             return -1;
     }
-    
-    s->interlaced= height > 288;
-    
+
     switch(s->bitstream_bpp){
     case 12:
         avctx->pix_fmt = PIX_FMT_YUV420P;
         break;
     case 16:
         if(s->yuy2){
-            avctx->pix_fmt = PIX_FMT_YUV422;
+            avctx->pix_fmt = PIX_FMT_YUYV422;
         }else{
             avctx->pix_fmt = PIX_FMT_YUV422P;
         }
@@ -425,7 +512,7 @@ s->bgr32=1;
     case 24:
     case 32:
         if(s->bgr32){
-            avctx->pix_fmt = PIX_FMT_RGBA32;
+            avctx->pix_fmt = PIX_FMT_RGB32;
         }else{
             avctx->pix_fmt = PIX_FMT_BGR24;
         }
@@ -433,56 +520,52 @@ s->bgr32=1;
     default:
         assert(0);
     }
-    
-//    printf("pred:%d bpp:%d hbpp:%d il:%d\n", s->predictor, s->bitstream_bpp, avctx->bits_per_sample, s->interlaced);
-    
+
+    alloc_temp(s);
+
+//    av_log(NULL, AV_LOG_DEBUG, "pred:%d bpp:%d hbpp:%d il:%d\n", s->predictor, s->bitstream_bpp, avctx->bits_per_coded_sample, s->interlaced);
+
     return 0;
 }
+#endif /* CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER */
 
-static void store_table(HYuvContext *s, uint8_t *len){
+#if CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER
+static int store_table(HYuvContext *s, const uint8_t *len, uint8_t *buf){
     int i;
-    int index= s->avctx->extradata_size;
+    int index= 0;
 
     for(i=0; i<256;){
         int val= len[i];
         int repeat=0;
-        
+
         for(; i<256 && len[i]==val && repeat<255; i++)
             repeat++;
-        
+
         assert(val < 32 && val >0 && repeat<256 && repeat>0);
         if(repeat>7){
-            ((uint8_t*)s->avctx->extradata)[index++]= val;
-            ((uint8_t*)s->avctx->extradata)[index++]= repeat;
+            buf[index++]= val;
+            buf[index++]= repeat;
         }else{
-            ((uint8_t*)s->avctx->extradata)[index++]= val | (repeat<<5);
+            buf[index++]= val | (repeat<<5);
         }
     }
-    
-    s->avctx->extradata_size= index;
+
+    return index;
 }
 
-static int encode_init(AVCodecContext *avctx)
+static av_cold int encode_init(AVCodecContext *avctx)
 {
     HYuvContext *s = avctx->priv_data;
-    int i, j, width, height;
+    int i, j;
 
-    s->avctx= avctx;
-    s->flags= avctx->flags;
-        
-    dsputil_init(&s->dsp, avctx);
-    
-    width= s->width= avctx->width;
-    height= s->height= avctx->height;
-    
-    assert(width && height);
-    
-    avctx->extradata= av_mallocz(1024*30);
-    avctx->stats_out= av_mallocz(1024*30);
+    common_init(avctx);
+
+    avctx->extradata= av_mallocz(1024*30); // 256*3+4 == 772
+    avctx->stats_out= av_mallocz(1024*30); // 21*256*3(%llu ) + 3(\n) + 1(0) = 16132
     s->version=2;
-    
+
     avctx->coded_frame= &s->picture;
-    
+
     switch(avctx->pix_fmt){
     case PIX_FMT_YUV420P:
         s->bitstream_bpp= 12;
@@ -490,23 +573,54 @@ static int encode_init(AVCodecContext *avctx)
     case PIX_FMT_YUV422P:
         s->bitstream_bpp= 16;
         break;
+    case PIX_FMT_RGB32:
+        s->bitstream_bpp= 24;
+        break;
     default:
         av_log(avctx, AV_LOG_ERROR, "format not supported\n");
         return -1;
     }
-    avctx->bits_per_sample= s->bitstream_bpp;
+    avctx->bits_per_coded_sample= s->bitstream_bpp;
     s->decorrelate= s->bitstream_bpp >= 24;
     s->predictor= avctx->prediction_method;
-    
-    ((uint8_t*)avctx->extradata)[0]= s->predictor;
+    s->interlaced= avctx->flags&CODEC_FLAG_INTERLACED_ME ? 1 : 0;
+    if(avctx->context_model==1){
+        s->context= avctx->context_model;
+        if(s->flags & (CODEC_FLAG_PASS1|CODEC_FLAG_PASS2)){
+            av_log(avctx, AV_LOG_ERROR, "context=1 is not compatible with 2 pass huffyuv encoding\n");
+            return -1;
+        }
+    }else s->context= 0;
+
+    if(avctx->codec->id==CODEC_ID_HUFFYUV){
+        if(avctx->pix_fmt==PIX_FMT_YUV420P){
+            av_log(avctx, AV_LOG_ERROR, "Error: YV12 is not supported by huffyuv; use vcodec=ffvhuff or format=422p\n");
+            return -1;
+        }
+        if(avctx->context_model){
+            av_log(avctx, AV_LOG_ERROR, "Error: per-frame huffman tables are not supported by huffyuv; use vcodec=ffvhuff\n");
+            return -1;
+        }
+        if(s->interlaced != ( s->height > 288 ))
+            av_log(avctx, AV_LOG_INFO, "using huffyuv 2.2.0 or newer interlacing flag\n");
+    }
+
+    if(s->bitstream_bpp>=24 && s->predictor==MEDIAN){
+        av_log(avctx, AV_LOG_ERROR, "Error: RGB is incompatible with median predictor\n");
+        return -1;
+    }
+
+    ((uint8_t*)avctx->extradata)[0]= s->predictor | (s->decorrelate << 6);
     ((uint8_t*)avctx->extradata)[1]= s->bitstream_bpp;
-    ((uint8_t*)avctx->extradata)[2]=
+    ((uint8_t*)avctx->extradata)[2]= s->interlaced ? 0x10 : 0x20;
+    if(s->context)
+        ((uint8_t*)avctx->extradata)[2]|= 0x40;
     ((uint8_t*)avctx->extradata)[3]= 0;
     s->avctx->extradata_size= 4;
-    
+
     if(avctx->stats_in){
         char *p= avctx->stats_in;
-    
+
         for(i=0; i<3; i++)
             for(j=0; j<256; j++)
                 s->stats[i][j]= 1;
@@ -519,7 +633,7 @@ static int encode_init(AVCodecContext *avctx)
                     s->stats[i][j]+= strtol(p, &next, 0);
                     if(next==p) return -1;
                     p=next;
-                }        
+                }
             }
             if(p[0]==0 || p[1]==0 || p[2]==0) break;
         }
@@ -527,142 +641,277 @@ static int encode_init(AVCodecContext *avctx)
         for(i=0; i<3; i++)
             for(j=0; j<256; j++){
                 int d= FFMIN(j, 256-j);
-                
+
                 s->stats[i][j]= 100000000/(d+1);
             }
     }
-    
+
     for(i=0; i<3; i++){
-        generate_len_table(s->len[i], s->stats[i], 256);
+        generate_len_table(s->len[i], s->stats[i]);
 
         if(generate_bits_table(s->bits[i], s->len[i])<0){
             return -1;
         }
-        
-        store_table(s, s->len[i]);
+
+        s->avctx->extradata_size+=
+        store_table(s, s->len[i], &((uint8_t*)s->avctx->extradata)[s->avctx->extradata_size]);
     }
 
-    for(i=0; i<3; i++)
-        for(j=0; j<256; j++)
-            s->stats[i][j]= 0;
-    
-    s->interlaced= height > 288;
+    if(s->context){
+        for(i=0; i<3; i++){
+            int pels = s->width*s->height / (i?40:10);
+            for(j=0; j<256; j++){
+                int d= FFMIN(j, 256-j);
+                s->stats[i][j]= pels/(d+1);
+            }
+        }
+    }else{
+        for(i=0; i<3; i++)
+            for(j=0; j<256; j++)
+                s->stats[i][j]= 0;
+    }
+
+//    printf("pred:%d bpp:%d hbpp:%d il:%d\n", s->predictor, s->bitstream_bpp, avctx->bits_per_coded_sample, s->interlaced);
 
-//    printf("pred:%d bpp:%d hbpp:%d il:%d\n", s->predictor, s->bitstream_bpp, avctx->bits_per_sample, s->interlaced);
+    alloc_temp(s);
 
     s->picture_number=0;
 
     return 0;
 }
+#endif /* CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER */
+
+/* TODO instead of restarting the read when the code isn't in the first level
+ * of the joint table, jump into the 2nd level of the individual table. */
+#define READ_2PIX(dst0, dst1, plane1){\
+    uint16_t code = get_vlc2(&s->gb, s->vlc[3+plane1].table, VLC_BITS, 1);\
+    if(code != 0xffff){\
+        dst0 = code>>8;\
+        dst1 = code;\
+    }else{\
+        dst0 = get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3);\
+        dst1 = get_vlc2(&s->gb, s->vlc[plane1].table, VLC_BITS, 3);\
+    }\
+}
 
 static void decode_422_bitstream(HYuvContext *s, int count){
     int i;
 
     count/=2;
-    
-    for(i=0; i<count; i++){
-        s->temp[0][2*i  ]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3); 
-        s->temp[1][  i  ]= get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3); 
-        s->temp[0][2*i+1]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3); 
-        s->temp[2][  i  ]= get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3); 
+
+    if(count >= (get_bits_left(&s->gb))/(31*4)){
+        for(i=0; i<count && get_bits_count(&s->gb) < s->gb.size_in_bits; i++){
+            READ_2PIX(s->temp[0][2*i  ], s->temp[1][i], 1);
+            READ_2PIX(s->temp[0][2*i+1], s->temp[2][i], 2);
+        }
+    }else{
+        for(i=0; i<count; i++){
+            READ_2PIX(s->temp[0][2*i  ], s->temp[1][i], 1);
+            READ_2PIX(s->temp[0][2*i+1], s->temp[2][i], 2);
+        }
     }
 }
 
 static void decode_gray_bitstream(HYuvContext *s, int count){
     int i;
-    
+
     count/=2;
-    
-    for(i=0; i<count; i++){
-        s->temp[0][2*i  ]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3); 
-        s->temp[0][2*i+1]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3); 
+
+    if(count >= (get_bits_left(&s->gb))/(31*2)){
+        for(i=0; i<count && get_bits_count(&s->gb) < s->gb.size_in_bits; i++){
+            READ_2PIX(s->temp[0][2*i  ], s->temp[0][2*i+1], 0);
+        }
+    }else{
+        for(i=0; i<count; i++){
+            READ_2PIX(s->temp[0][2*i  ], s->temp[0][2*i+1], 0);
+        }
     }
 }
 
-static void encode_422_bitstream(HYuvContext *s, int count){
+#if CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER
+static int encode_422_bitstream(HYuvContext *s, int offset, int count){
     int i;
-    
+    const uint8_t *y = s->temp[0] + offset;
+    const uint8_t *u = s->temp[1] + offset/2;
+    const uint8_t *v = s->temp[2] + offset/2;
+
+    if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 2*4*count){
+        av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
+        return -1;
+    }
+
+#define LOAD4\
+            int y0 = y[2*i];\
+            int y1 = y[2*i+1];\
+            int u0 = u[i];\
+            int v0 = v[i];
+
     count/=2;
     if(s->flags&CODEC_FLAG_PASS1){
         for(i=0; i<count; i++){
-            s->stats[0][ s->temp[0][2*i  ] ]++;
-            s->stats[1][ s->temp[1][  i  ] ]++;
-            s->stats[0][ s->temp[0][2*i+1] ]++;
-            s->stats[2][ s->temp[2][  i  ] ]++;
+            LOAD4;
+            s->stats[0][y0]++;
+            s->stats[1][u0]++;
+            s->stats[0][y1]++;
+            s->stats[2][v0]++;
+        }
+    }
+    if(s->avctx->flags2&CODEC_FLAG2_NO_OUTPUT)
+        return 0;
+    if(s->context){
+        for(i=0; i<count; i++){
+            LOAD4;
+            s->stats[0][y0]++;
+            put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
+            s->stats[1][u0]++;
+            put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
+            s->stats[0][y1]++;
+            put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
+            s->stats[2][v0]++;
+            put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
         }
     }else{
         for(i=0; i<count; i++){
-            put_bits(&s->pb, s->len[0][ s->temp[0][2*i  ] ], s->bits[0][ s->temp[0][2*i  ] ]);
-            put_bits(&s->pb, s->len[1][ s->temp[1][  i  ] ], s->bits[1][ s->temp[1][  i  ] ]);
-            put_bits(&s->pb, s->len[0][ s->temp[0][2*i+1] ], s->bits[0][ s->temp[0][2*i+1] ]);
-            put_bits(&s->pb, s->len[2][ s->temp[2][  i  ] ], s->bits[2][ s->temp[2][  i  ] ]);
+            LOAD4;
+            put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);
+            put_bits(&s->pb, s->len[1][u0], s->bits[1][u0]);
+            put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
+            put_bits(&s->pb, s->len[2][v0], s->bits[2][v0]);
         }
     }
+    return 0;
 }
 
-static void encode_gray_bitstream(HYuvContext *s, int count){
+static int encode_gray_bitstream(HYuvContext *s, int count){
     int i;
-    
+
+    if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 4*count){
+        av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
+        return -1;
+    }
+
+#define LOAD2\
+            int y0 = s->temp[0][2*i];\
+            int y1 = s->temp[0][2*i+1];
+#define STAT2\
+            s->stats[0][y0]++;\
+            s->stats[0][y1]++;
+#define WRITE2\
+            put_bits(&s->pb, s->len[0][y0], s->bits[0][y0]);\
+            put_bits(&s->pb, s->len[0][y1], s->bits[0][y1]);
+
     count/=2;
     if(s->flags&CODEC_FLAG_PASS1){
         for(i=0; i<count; i++){
-            s->stats[0][ s->temp[0][2*i  ] ]++;
-            s->stats[0][ s->temp[0][2*i+1] ]++;
+            LOAD2;
+            STAT2;
+        }
+    }
+    if(s->avctx->flags2&CODEC_FLAG2_NO_OUTPUT)
+        return 0;
+
+    if(s->context){
+        for(i=0; i<count; i++){
+            LOAD2;
+            STAT2;
+            WRITE2;
         }
     }else{
         for(i=0; i<count; i++){
-            put_bits(&s->pb, s->len[0][ s->temp[0][2*i  ] ], s->bits[0][ s->temp[0][2*i  ] ]);
-            put_bits(&s->pb, s->len[0][ s->temp[0][2*i+1] ], s->bits[0][ s->temp[0][2*i+1] ]);
+            LOAD2;
+            WRITE2;
         }
     }
+    return 0;
 }
+#endif /* CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER */
 
-static void decode_bgr_bitstream(HYuvContext *s, int count){
+static av_always_inline void decode_bgr_1(HYuvContext *s, int count, int decorrelate, int alpha){
     int i;
+    for(i=0; i<count; i++){
+        int code = get_vlc2(&s->gb, s->vlc[3].table, VLC_BITS, 1);
+        if(code != -1){
+            *(uint32_t*)&s->temp[0][4*i] = s->pix_bgr_map[code];
+        }else if(decorrelate){
+            s->temp[0][4*i+G] = get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3);
+            s->temp[0][4*i+B] = get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3) + s->temp[0][4*i+G];
+            s->temp[0][4*i+R] = get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3) + s->temp[0][4*i+G];
+        }else{
+            s->temp[0][4*i+B] = get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3);
+            s->temp[0][4*i+G] = get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3);
+            s->temp[0][4*i+R] = get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3);
+        }
+        if(alpha)
+            s->temp[0][4*i+A] = get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3);
+    }
+}
 
+static void decode_bgr_bitstream(HYuvContext *s, int count){
     if(s->decorrelate){
-        if(s->bitstream_bpp==24){
-            for(i=0; i<count; i++){
-                s->temp[0][4*i+G]= get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3); 
-                s->temp[0][4*i+B]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3) + s->temp[0][4*i+G];
-                s->temp[0][4*i+R]= get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3) + s->temp[0][4*i+G];
-            }
-        }else{
-            for(i=0; i<count; i++){
-                s->temp[0][4*i+G]= get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3); 
-                s->temp[0][4*i+B]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3) + s->temp[0][4*i+G];
-                s->temp[0][4*i+R]= get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3) + s->temp[0][4*i+G]; 
-                                   get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3); //?!
-            }
+        if(s->bitstream_bpp==24)
+            decode_bgr_1(s, count, 1, 0);
+        else
+            decode_bgr_1(s, count, 1, 1);
+    }else{
+        if(s->bitstream_bpp==24)
+            decode_bgr_1(s, count, 0, 0);
+        else
+            decode_bgr_1(s, count, 0, 1);
+    }
+}
+
+static int encode_bgr_bitstream(HYuvContext *s, int count){
+    int i;
+
+    if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < 3*4*count){
+        av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
+        return -1;
+    }
+
+#define LOAD3\
+            int g= s->temp[0][4*i+G];\
+            int b= (s->temp[0][4*i+B] - g) & 0xff;\
+            int r= (s->temp[0][4*i+R] - g) & 0xff;
+#define STAT3\
+            s->stats[0][b]++;\
+            s->stats[1][g]++;\
+            s->stats[2][r]++;
+#define WRITE3\
+            put_bits(&s->pb, s->len[1][g], s->bits[1][g]);\
+            put_bits(&s->pb, s->len[0][b], s->bits[0][b]);\
+            put_bits(&s->pb, s->len[2][r], s->bits[2][r]);
+
+    if((s->flags&CODEC_FLAG_PASS1) && (s->avctx->flags2&CODEC_FLAG2_NO_OUTPUT)){
+        for(i=0; i<count; i++){
+            LOAD3;
+            STAT3;
+        }
+    }else if(s->context || (s->flags&CODEC_FLAG_PASS1)){
+        for(i=0; i<count; i++){
+            LOAD3;
+            STAT3;
+            WRITE3;
         }
     }else{
-        if(s->bitstream_bpp==24){
-            for(i=0; i<count; i++){
-                s->temp[0][4*i+B]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3);
-                s->temp[0][4*i+G]= get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3); 
-                s->temp[0][4*i+R]= get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3); 
-            }
-        }else{
-            for(i=0; i<count; i++){
-                s->temp[0][4*i+B]= get_vlc2(&s->gb, s->vlc[0].table, VLC_BITS, 3);
-                s->temp[0][4*i+G]= get_vlc2(&s->gb, s->vlc[1].table, VLC_BITS, 3); 
-                s->temp[0][4*i+R]= get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3); 
-                                   get_vlc2(&s->gb, s->vlc[2].table, VLC_BITS, 3); //?!
-            }
+        for(i=0; i<count; i++){
+            LOAD3;
+            WRITE3;
         }
     }
+    return 0;
 }
 
+#if CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER
 static void draw_slice(HYuvContext *s, int y){
     int h, cy;
     int offset[4];
-    
-    if(s->avctx->draw_horiz_band==NULL) 
+
+    if(s->avctx->draw_horiz_band==NULL)
         return;
-        
+
     h= y - s->last_slice_end;
     y -= h;
-    
+
     if(s->bitstream_bpp==12){
         cy= y>>1;
     }else{
@@ -676,27 +925,29 @@ static void draw_slice(HYuvContext *s, int y){
     emms_c();
 
     s->avctx->draw_horiz_band(s->avctx, &s->picture, offset, y, 3, h);
-    
+
     s->last_slice_end= y + h;
 }
 
-static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){
+static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt){
+    const uint8_t *buf = avpkt->data;
+    int buf_size = avpkt->size;
     HYuvContext *s = avctx->priv_data;
     const int width= s->width;
     const int width2= s->width>>1;
     const int height= s->height;
     int fake_ystride, fake_ustride, fake_vstride;
     AVFrame * const p= &s->picture;
+    int table_size= 0;
 
     AVFrame *picture = data;
 
-    /* no supplementary picture */
-    if (buf_size == 0)
-        return 0;
+    av_fast_malloc(&s->bitstream_buffer, &s->bitstream_buffer_size, buf_size + FF_INPUT_BUFFER_PADDING_SIZE);
+    if (!s->bitstream_buffer)
+        return AVERROR(ENOMEM);
 
-    s->dsp.bswap_buf((uint32_t*)s->bitstream_buffer, (uint32_t*)buf, buf_size/4);
-    
-    init_get_bits(&s->gb, s->bitstream_buffer, buf_size*8);
+    memset(s->bitstream_buffer + buf_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
+    s->dsp.bswap_buf((uint32_t*)s->bitstream_buffer, (const uint32_t*)buf, buf_size/4);
 
     if(p->data[0])
         avctx->release_buffer(avctx, p);
@@ -707,51 +958,62 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
         return -1;
     }
 
+    if(s->context){
+        table_size = read_huffman_tables(s, s->bitstream_buffer, buf_size);
+        if(table_size < 0)
+            return -1;
+    }
+
+    if((unsigned)(buf_size-table_size) >= INT_MAX/8)
+        return -1;
+
+    init_get_bits(&s->gb, s->bitstream_buffer+table_size, (buf_size-table_size)*8);
+
     fake_ystride= s->interlaced ? p->linesize[0]*2  : p->linesize[0];
     fake_ustride= s->interlaced ? p->linesize[1]*2  : p->linesize[1];
     fake_vstride= s->interlaced ? p->linesize[2]*2  : p->linesize[2];
-    
+
     s->last_slice_end= 0;
-        
+
     if(s->bitstream_bpp<24){
         int y, cy;
         int lefty, leftu, leftv;
         int lefttopy, lefttopu, lefttopv;
-        
+
         if(s->yuy2){
             p->data[0][3]= get_bits(&s->gb, 8);
             p->data[0][2]= get_bits(&s->gb, 8);
             p->data[0][1]= get_bits(&s->gb, 8);
             p->data[0][0]= get_bits(&s->gb, 8);
-            
-            av_log(avctx, AV_LOG_ERROR, "YUY2 output isnt implemenetd yet\n");
+
+            av_log(avctx, AV_LOG_ERROR, "YUY2 output is not implemented yet\n");
             return -1;
         }else{
-        
+
             leftv= p->data[2][0]= get_bits(&s->gb, 8);
             lefty= p->data[0][1]= get_bits(&s->gb, 8);
             leftu= p->data[1][0]= get_bits(&s->gb, 8);
                    p->data[0][0]= get_bits(&s->gb, 8);
-        
+
             switch(s->predictor){
             case LEFT:
             case PLANE:
                 decode_422_bitstream(s, width-2);
-                lefty= add_left_prediction(p->data[0] + 2, s->temp[0], width-2, lefty);
+                lefty= s->dsp.add_hfyu_left_prediction(p->data[0] + 2, s->temp[0], width-2, lefty);
                 if(!(s->flags&CODEC_FLAG_GRAY)){
-                    leftu= add_left_prediction(p->data[1] + 1, s->temp[1], width2-1, leftu);
-                    leftv= add_left_prediction(p->data[2] + 1, s->temp[2], width2-1, leftv);
+                    leftu= s->dsp.add_hfyu_left_prediction(p->data[1] + 1, s->temp[1], width2-1, leftu);
+                    leftv= s->dsp.add_hfyu_left_prediction(p->data[2] + 1, s->temp[2], width2-1, leftv);
                 }
 
                 for(cy=y=1; y<s->height; y++,cy++){
                     uint8_t *ydst, *udst, *vdst;
-                    
+
                     if(s->bitstream_bpp==12){
                         decode_gray_bitstream(s, width);
-                    
+
                         ydst= p->data[0] + p->linesize[0]*y;
 
-                        lefty= add_left_prediction(ydst, s->temp[0], width, lefty);
+                        lefty= s->dsp.add_hfyu_left_prediction(ydst, s->temp[0], width, lefty);
                         if(s->predictor == PLANE){
                             if(y>s->interlaced)
                                 s->dsp.add_bytes(ydst, ydst - fake_ystride, width);
@@ -759,18 +1021,18 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
                         y++;
                         if(y>=s->height) break;
                     }
-                    
+
                     draw_slice(s, y);
-                    
+
                     ydst= p->data[0] + p->linesize[0]*y;
                     udst= p->data[1] + p->linesize[1]*cy;
                     vdst= p->data[2] + p->linesize[2]*cy;
-                    
+
                     decode_422_bitstream(s, width);
-                    lefty= add_left_prediction(ydst, s->temp[0], width, lefty);
+                    lefty= s->dsp.add_hfyu_left_prediction(ydst, s->temp[0], width, lefty);
                     if(!(s->flags&CODEC_FLAG_GRAY)){
-                        leftu= add_left_prediction(udst, s->temp[1], width2, leftu);
-                        leftv= add_left_prediction(vdst, s->temp[2], width2, leftv);
+                        leftu= s->dsp.add_hfyu_left_prediction(udst, s->temp[1], width2, leftu);
+                        leftv= s->dsp.add_hfyu_left_prediction(vdst, s->temp[2], width2, leftv);
                     }
                     if(s->predictor == PLANE){
                         if(cy>s->interlaced){
@@ -783,50 +1045,50 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
                     }
                 }
                 draw_slice(s, height);
-                
+
                 break;
             case MEDIAN:
                 /* first line except first 2 pixels is left predicted */
                 decode_422_bitstream(s, width-2);
-                lefty= add_left_prediction(p->data[0] + 2, s->temp[0], width-2, lefty);
+                lefty= s->dsp.add_hfyu_left_prediction(p->data[0] + 2, s->temp[0], width-2, lefty);
                 if(!(s->flags&CODEC_FLAG_GRAY)){
-                    leftu= add_left_prediction(p->data[1] + 1, s->temp[1], width2-1, leftu);
-                    leftv= add_left_prediction(p->data[2] + 1, s->temp[2], width2-1, leftv);
+                    leftu= s->dsp.add_hfyu_left_prediction(p->data[1] + 1, s->temp[1], width2-1, leftu);
+                    leftv= s->dsp.add_hfyu_left_prediction(p->data[2] + 1, s->temp[2], width2-1, leftv);
                 }
-                
+
                 cy=y=1;
-                
+
                 /* second line is left predicted for interlaced case */
                 if(s->interlaced){
                     decode_422_bitstream(s, width);
-                    lefty= add_left_prediction(p->data[0] + p->linesize[0], s->temp[0], width, lefty);
+                    lefty= s->dsp.add_hfyu_left_prediction(p->data[0] + p->linesize[0], s->temp[0], width, lefty);
                     if(!(s->flags&CODEC_FLAG_GRAY)){
-                        leftu= add_left_prediction(p->data[1] + p->linesize[2], s->temp[1], width2, leftu);
-                        leftv= add_left_prediction(p->data[2] + p->linesize[1], s->temp[2], width2, leftv);
+                        leftu= s->dsp.add_hfyu_left_prediction(p->data[1] + p->linesize[2], s->temp[1], width2, leftu);
+                        leftv= s->dsp.add_hfyu_left_prediction(p->data[2] + p->linesize[1], s->temp[2], width2, leftv);
                     }
                     y++; cy++;
                 }
 
                 /* next 4 pixels are left predicted too */
                 decode_422_bitstream(s, 4);
-                lefty= add_left_prediction(p->data[0] + fake_ystride, s->temp[0], 4, lefty);
+                lefty= s->dsp.add_hfyu_left_prediction(p->data[0] + fake_ystride, s->temp[0], 4, lefty);
                 if(!(s->flags&CODEC_FLAG_GRAY)){
-                    leftu= add_left_prediction(p->data[1] + fake_ustride, s->temp[1], 2, leftu);
-                    leftv= add_left_prediction(p->data[2] + fake_vstride, s->temp[2], 2, leftv);
+                    leftu= s->dsp.add_hfyu_left_prediction(p->data[1] + fake_ustride, s->temp[1], 2, leftu);
+                    leftv= s->dsp.add_hfyu_left_prediction(p->data[2] + fake_vstride, s->temp[2], 2, leftv);
                 }
 
                 /* next line except the first 4 pixels is median predicted */
                 lefttopy= p->data[0][3];
                 decode_422_bitstream(s, width-4);
-                add_median_prediction(p->data[0] + fake_ystride+4, p->data[0]+4, s->temp[0], width-4, &lefty, &lefttopy);
+                s->dsp.add_hfyu_median_prediction(p->data[0] + fake_ystride+4, p->data[0]+4, s->temp[0], width-4, &lefty, &lefttopy);
                 if(!(s->flags&CODEC_FLAG_GRAY)){
                     lefttopu= p->data[1][1];
                     lefttopv= p->data[2][1];
-                    add_median_prediction(p->data[1] + fake_ustride+2, p->data[1]+2, s->temp[1], width2-2, &leftu, &lefttopu);
-                    add_median_prediction(p->data[2] + fake_vstride+2, p->data[2]+2, s->temp[2], width2-2, &leftv, &lefttopv);
+                    s->dsp.add_hfyu_median_prediction(p->data[1] + fake_ustride+2, p->data[1]+2, s->temp[1], width2-2, &leftu, &lefttopu);
+                    s->dsp.add_hfyu_median_prediction(p->data[2] + fake_vstride+2, p->data[2]+2, s->temp[2], width2-2, &leftv, &lefttopv);
                 }
                 y++; cy++;
-                
+
                 for(; y<height; y++,cy++){
                     uint8_t *ydst, *udst, *vdst;
 
@@ -834,7 +1096,7 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
                         while(2*cy > y){
                             decode_gray_bitstream(s, width);
                             ydst= p->data[0] + p->linesize[0]*y;
-                            add_median_prediction(ydst, ydst - fake_ystride, s->temp[0], width, &lefty, &lefttopy);
+                            s->dsp.add_hfyu_median_prediction(ydst, ydst - fake_ystride, s->temp[0], width, &lefty, &lefttopy);
                             y++;
                         }
                         if(y>=height) break;
@@ -847,10 +1109,10 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
                     udst= p->data[1] + p->linesize[1]*cy;
                     vdst= p->data[2] + p->linesize[2]*cy;
 
-                    add_median_prediction(ydst, ydst - fake_ystride, s->temp[0], width, &lefty, &lefttopy);
+                    s->dsp.add_hfyu_median_prediction(ydst, ydst - fake_ystride, s->temp[0], width, &lefty, &lefttopy);
                     if(!(s->flags&CODEC_FLAG_GRAY)){
-                        add_median_prediction(udst, udst - fake_ustride, s->temp[1], width2, &leftu, &lefttopu);
-                        add_median_prediction(vdst, vdst - fake_vstride, s->temp[2], width2, &leftv, &lefttopv);
+                        s->dsp.add_hfyu_median_prediction(udst, udst - fake_ustride, s->temp[1], width2, &leftu, &lefttopu);
+                        s->dsp.add_hfyu_median_prediction(vdst, vdst - fake_vstride, s->temp[2], width2, &leftv, &lefttopv);
                     }
                 }
 
@@ -860,11 +1122,11 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
         }
     }else{
         int y;
-        int leftr, leftg, leftb;
+        int leftr, leftg, leftb, lefta;
         const int last_line= (height-1)*p->linesize[0];
-        
+
         if(s->bitstream_bpp==32){
-            skip_bits(&s->gb, 8);
+            lefta= p->data[0][last_line+A]= get_bits(&s->gb, 8);
             leftr= p->data[0][last_line+R]= get_bits(&s->gb, 8);
             leftg= p->data[0][last_line+G]= get_bits(&s->gb, 8);
             leftb= p->data[0][last_line+B]= get_bits(&s->gb, 8);
@@ -872,58 +1134,79 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8
             leftr= p->data[0][last_line+R]= get_bits(&s->gb, 8);
             leftg= p->data[0][last_line+G]= get_bits(&s->gb, 8);
             leftb= p->data[0][last_line+B]= get_bits(&s->gb, 8);
+            lefta= p->data[0][last_line+A]= 255;
             skip_bits(&s->gb, 8);
         }
-        
+
         if(s->bgr32){
             switch(s->predictor){
             case LEFT:
             case PLANE:
                 decode_bgr_bitstream(s, width-1);
-                add_left_prediction_bgr32(p->data[0] + last_line+4, s->temp[0], width-1, &leftr, &leftg, &leftb);
+                s->dsp.add_hfyu_left_prediction_bgr32(p->data[0] + last_line+4, s->temp[0], width-1, &leftr, &leftg, &leftb, &lefta);
 
-                for(y=s->height-2; y>=0; y--){ //yes its stored upside down
+                for(y=s->height-2; y>=0; y--){ //Yes it is stored upside down.
                     decode_bgr_bitstream(s, width);
-                    
-                    add_left_prediction_bgr32(p->data[0] + p->linesize[0]*y, s->temp[0], width, &leftr, &leftg, &leftb);
+
+                    s->dsp.add_hfyu_left_prediction_bgr32(p->data[0] + p->linesize[0]*y, s->temp[0], width, &leftr, &leftg, &leftb, &lefta);
                     if(s->predictor == PLANE){
-                        if((y&s->interlaced)==0 && y<s->height-2){
-                            s->dsp.add_bytes(p->data[0] + p->linesize[0]*y, 
+                        if(s->bitstream_bpp!=32) lefta=0;
+                        if((y&s->interlaced)==0 && y<s->height-1-s->interlaced){
+                            s->dsp.add_bytes(p->data[0] + p->linesize[0]*y,
                                              p->data[0] + p->linesize[0]*y + fake_ystride, fake_ystride);
                         }
                     }
                 }
-                draw_slice(s, height); // just 1 large slice as this isnt possible in reverse order
+                draw_slice(s, height); // just 1 large slice as this is not possible in reverse order
                 break;
             default:
                 av_log(avctx, AV_LOG_ERROR, "prediction type not supported!\n");
             }
         }else{
 
-            av_log(avctx, AV_LOG_ERROR, "BGR24 output isnt implemenetd yet\n");
+            av_log(avctx, AV_LOG_ERROR, "BGR24 output is not implemented yet\n");
             return -1;
         }
     }
     emms_c();
-    
+
     *picture= *p;
     *data_size = sizeof(AVFrame);
-    
-    return (get_bits_count(&s->gb)+31)/32*4;
+
+    return (get_bits_count(&s->gb)+31)/32*4 + table_size;
 }
+#endif /* CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER */
+
+static int common_end(HYuvContext *s){
+    int i;
 
-static int decode_end(AVCodecContext *avctx)
+    for(i=0; i<3; i++){
+        av_freep(&s->temp[i]);
+    }
+    return 0;
+}
+
+#if CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER
+static av_cold int decode_end(AVCodecContext *avctx)
 {
     HYuvContext *s = avctx->priv_data;
     int i;
-    
-    for(i=0; i<3; i++){
+
+    if (s->picture.data[0])
+        avctx->release_buffer(avctx, &s->picture);
+
+    common_end(s);
+    av_freep(&s->bitstream_buffer);
+
+    for(i=0; i<6; i++){
         free_vlc(&s->vlc[i]);
     }
 
     return 0;
 }
+#endif /* CONFIG_HUFFYUV_DECODER || CONFIG_FFVHUFF_DECODER */
 
+#if CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER
 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
     HYuvContext *s = avctx->priv_data;
     AVFrame *pict = data;
@@ -934,14 +1217,27 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
     const int fake_ustride= s->interlaced ? pict->linesize[1]*2  : pict->linesize[1];
     const int fake_vstride= s->interlaced ? pict->linesize[2]*2  : pict->linesize[2];
     AVFrame * const p= &s->picture;
-    int i, size;
+    int i, j, size=0;
 
-    init_put_bits(&s->pb, buf, buf_size);
-    
     *p = *pict;
     p->pict_type= FF_I_TYPE;
     p->key_frame= 1;
-    
+
+    if(s->context){
+        for(i=0; i<3; i++){
+            generate_len_table(s->len[i], s->stats[i]);
+            if(generate_bits_table(s->bits[i], s->len[i])<0)
+                return -1;
+            size+= store_table(s, s->len[i], &buf[size]);
+        }
+
+        for(i=0; i<3; i++)
+            for(j=0; j<256; j++)
+                s->stats[i][j] >>= 1;
+    }
+
+    init_put_bits(&s->pb, buf+size, buf_size-size);
+
     if(avctx->pix_fmt == PIX_FMT_YUV422P || avctx->pix_fmt == PIX_FMT_YUV420P){
         int lefty, leftu, leftv, y, cy;
 
@@ -949,13 +1245,13 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
         put_bits(&s->pb, 8, lefty= p->data[0][1]);
         put_bits(&s->pb, 8, leftu= p->data[1][0]);
         put_bits(&s->pb, 8,        p->data[0][0]);
-        
-        lefty= sub_left_prediction(s, s->temp[0], p->data[0]+2, width-2 , lefty);
-        leftu= sub_left_prediction(s, s->temp[1], p->data[1]+1, width2-1, leftu);
-        leftv= sub_left_prediction(s, s->temp[2], p->data[2]+1, width2-1, leftv);
-        
-        encode_422_bitstream(s, width-2);
-        
+
+        lefty= sub_left_prediction(s, s->temp[0], p->data[0], width , 0);
+        leftu= sub_left_prediction(s, s->temp[1], p->data[1], width2, 0);
+        leftv= sub_left_prediction(s, s->temp[2], p->data[2], width2, 0);
+
+        encode_422_bitstream(s, 2, width-2);
+
         if(s->predictor==MEDIAN){
             int lefttopy, lefttopu, lefttopv;
             cy=y=1;
@@ -963,16 +1259,16 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
                 lefty= sub_left_prediction(s, s->temp[0], p->data[0]+p->linesize[0], width , lefty);
                 leftu= sub_left_prediction(s, s->temp[1], p->data[1]+p->linesize[1], width2, leftu);
                 leftv= sub_left_prediction(s, s->temp[2], p->data[2]+p->linesize[2], width2, leftv);
-        
-                encode_422_bitstream(s, width);
+
+                encode_422_bitstream(s, 0, width);
                 y++; cy++;
             }
-            
+
             lefty= sub_left_prediction(s, s->temp[0], p->data[0]+fake_ystride, 4, lefty);
-            leftu= sub_left_prediction(s, s->temp[1], p->data[1]+fake_ystride, 2, leftu);
-            leftv= sub_left_prediction(s, s->temp[2], p->data[2]+fake_ystride, 2, leftv);
-        
-            encode_422_bitstream(s, 4);
+            leftu= sub_left_prediction(s, s->temp[1], p->data[1]+fake_ustride, 2, leftu);
+            leftv= sub_left_prediction(s, s->temp[2], p->data[2]+fake_vstride, 2, leftv);
+
+            encode_422_bitstream(s, 0, 4);
 
             lefttopy= p->data[0][3];
             lefttopu= p->data[1][1];
@@ -980,12 +1276,12 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
             s->dsp.sub_hfyu_median_prediction(s->temp[0], p->data[0]+4, p->data[0] + fake_ystride+4, width-4 , &lefty, &lefttopy);
             s->dsp.sub_hfyu_median_prediction(s->temp[1], p->data[1]+2, p->data[1] + fake_ustride+2, width2-2, &leftu, &lefttopu);
             s->dsp.sub_hfyu_median_prediction(s->temp[2], p->data[2]+2, p->data[2] + fake_vstride+2, width2-2, &leftv, &lefttopv);
-            encode_422_bitstream(s, width-4);
+            encode_422_bitstream(s, 0, width-4);
             y++; cy++;
 
             for(; y<height; y++,cy++){
                 uint8_t *ydst, *udst, *vdst;
-                    
+
                 if(s->bitstream_bpp==12){
                     while(2*cy > y){
                         ydst= p->data[0] + p->linesize[0]*y;
@@ -1003,12 +1299,12 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
                 s->dsp.sub_hfyu_median_prediction(s->temp[1], udst - fake_ustride, udst, width2, &leftu, &lefttopu);
                 s->dsp.sub_hfyu_median_prediction(s->temp[2], vdst - fake_vstride, vdst, width2, &leftv, &lefttopv);
 
-                encode_422_bitstream(s, width);
+                encode_422_bitstream(s, 0, width);
             }
         }else{
             for(cy=y=1; y<height; y++,cy++){
                 uint8_t *ydst, *udst, *vdst;
-                
+
                 /* encode a luma only line & y++ */
                 if(s->bitstream_bpp==12){
                     ydst= p->data[0] + p->linesize[0]*y;
@@ -1024,7 +1320,7 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
                     y++;
                     if(y>=height) break;
                 }
-                
+
                 ydst= p->data[0] + p->linesize[0]*y;
                 udst= p->data[1] + p->linesize[1]*cy;
                 vdst= p->data[2] + p->linesize[2]*cy;
@@ -1032,68 +1328,97 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size,
                 if(s->predictor == PLANE && s->interlaced < cy){
                     s->dsp.diff_bytes(s->temp[1], ydst, ydst - fake_ystride, width);
                     s->dsp.diff_bytes(s->temp[2], udst, udst - fake_ustride, width2);
-                    s->dsp.diff_bytes(s->temp[2] + 1250, vdst, vdst - fake_vstride, width2);
+                    s->dsp.diff_bytes(s->temp[2] + width2, vdst, vdst - fake_vstride, width2);
 
                     lefty= sub_left_prediction(s, s->temp[0], s->temp[1], width , lefty);
                     leftu= sub_left_prediction(s, s->temp[1], s->temp[2], width2, leftu);
-                    leftv= sub_left_prediction(s, s->temp[2], s->temp[2] + 1250, width2, leftv);
+                    leftv= sub_left_prediction(s, s->temp[2], s->temp[2] + width2, width2, leftv);
                 }else{
                     lefty= sub_left_prediction(s, s->temp[0], ydst, width , lefty);
                     leftu= sub_left_prediction(s, s->temp[1], udst, width2, leftu);
                     leftv= sub_left_prediction(s, s->temp[2], vdst, width2, leftv);
                 }
 
-                encode_422_bitstream(s, width);
+                encode_422_bitstream(s, 0, width);
+            }
+        }
+    }else if(avctx->pix_fmt == PIX_FMT_RGB32){
+        uint8_t *data = p->data[0] + (height-1)*p->linesize[0];
+        const int stride = -p->linesize[0];
+        const int fake_stride = -fake_ystride;
+        int y;
+        int leftr, leftg, leftb;
+
+        put_bits(&s->pb, 8, leftr= data[R]);
+        put_bits(&s->pb, 8, leftg= data[G]);
+        put_bits(&s->pb, 8, leftb= data[B]);
+        put_bits(&s->pb, 8, 0);
+
+        sub_left_prediction_bgr32(s, s->temp[0], data+4, width-1, &leftr, &leftg, &leftb);
+        encode_bgr_bitstream(s, width-1);
+
+        for(y=1; y<s->height; y++){
+            uint8_t *dst = data + y*stride;
+            if(s->predictor == PLANE && s->interlaced < y){
+                s->dsp.diff_bytes(s->temp[1], dst, dst - fake_stride, width*4);
+                sub_left_prediction_bgr32(s, s->temp[0], s->temp[1], width, &leftr, &leftg, &leftb);
+            }else{
+                sub_left_prediction_bgr32(s, s->temp[0], dst, width, &leftr, &leftg, &leftb);
             }
-        }        
+            encode_bgr_bitstream(s, width);
+        }
     }else{
         av_log(avctx, AV_LOG_ERROR, "Format not supported!\n");
     }
     emms_c();
-    
-    size= (put_bits_count(&s->pb)+31)/32;
-    
+
+    size+= (put_bits_count(&s->pb)+31)/8;
+    put_bits(&s->pb, 16, 0);
+    put_bits(&s->pb, 15, 0);
+    size/= 4;
+
     if((s->flags&CODEC_FLAG_PASS1) && (s->picture_number&31)==0){
         int j;
         char *p= avctx->stats_out;
+        char *end= p + 1024*30;
         for(i=0; i<3; i++){
             for(j=0; j<256; j++){
-                sprintf(p, "%llu ", s->stats[i][j]);
+                snprintf(p, end-p, "%"PRIu64" ", s->stats[i][j]);
                 p+= strlen(p);
                 s->stats[i][j]= 0;
             }
-            sprintf(p, "\n");
+            snprintf(p, end-p, "\n");
             p++;
         }
-    }else{
+    } else
+        avctx->stats_out[0] = '\0';
+    if(!(s->avctx->flags2 & CODEC_FLAG2_NO_OUTPUT)){
         flush_put_bits(&s->pb);
         s->dsp.bswap_buf((uint32_t*)buf, (uint32_t*)buf, size);
     }
-    
+
     s->picture_number++;
 
     return size*4;
 }
 
-static int encode_end(AVCodecContext *avctx)
+static av_cold int encode_end(AVCodecContext *avctx)
 {
-//    HYuvContext *s = avctx->priv_data;
+    HYuvContext *s = avctx->priv_data;
+
+    common_end(s);
 
     av_freep(&avctx->extradata);
     av_freep(&avctx->stats_out);
-    
+
     return 0;
 }
+#endif /* CONFIG_HUFFYUV_ENCODER || CONFIG_FFVHUFF_ENCODER */
 
-static const AVOption huffyuv_options[] =
-{
-    AVOPTION_CODEC_INT("prediction_method", "prediction_method", prediction_method, 0, 2, 0),
-    AVOPTION_END()
-};
-
-AVCodec huffyuv_decoder = {
+#if CONFIG_HUFFYUV_DECODER
+AVCodec ff_huffyuv_decoder = {
     "huffyuv",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_HUFFYUV,
     sizeof(HYuvContext),
     decode_init,
@@ -1101,20 +1426,51 @@ AVCodec huffyuv_decoder = {
     decode_end,
     decode_frame,
     CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
-    NULL
+    NULL,
+    .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
 };
+#endif
 
-#ifdef CONFIG_ENCODERS
+#if CONFIG_FFVHUFF_DECODER
+AVCodec ff_ffvhuff_decoder = {
+    "ffvhuff",
+    AVMEDIA_TYPE_VIDEO,
+    CODEC_ID_FFVHUFF,
+    sizeof(HYuvContext),
+    decode_init,
+    NULL,
+    decode_end,
+    decode_frame,
+    CODEC_CAP_DR1 | CODEC_CAP_DRAW_HORIZ_BAND,
+    NULL,
+    .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),
+};
+#endif
 
-AVCodec huffyuv_encoder = {
+#if CONFIG_HUFFYUV_ENCODER
+AVCodec ff_huffyuv_encoder = {
     "huffyuv",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_HUFFYUV,
     sizeof(HYuvContext),
     encode_init,
     encode_frame,
     encode_end,
-    .options = huffyuv_options,
+    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV422P, PIX_FMT_RGB32, PIX_FMT_NONE},
+    .long_name = NULL_IF_CONFIG_SMALL("Huffyuv / HuffYUV"),
 };
+#endif
 
-#endif //CONFIG_ENCODERS
+#if CONFIG_FFVHUFF_ENCODER
+AVCodec ff_ffvhuff_encoder = {
+    "ffvhuff",
+    AVMEDIA_TYPE_VIDEO,
+    CODEC_ID_FFVHUFF,
+    sizeof(HYuvContext),
+    encode_init,
+    encode_frame,
+    encode_end,
+    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_RGB32, PIX_FMT_NONE},
+    .long_name = NULL_IF_CONFIG_SMALL("Huffyuv FFmpeg variant"),
+};
+#endif