h264: stricter reference limit enforcement.
[ffmpeg.git] / libavcodec / h264.c
index 2a3357b..e0eb8e1 100644 (file)
@@ -2,20 +2,20 @@
  * H.26L/H.264/AVC/JVT/14496-10/... decoder
  * 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
  */
 
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
-#include "libavcore/imgutils.h"
+#include "libavutil/imgutils.h"
 #include "internal.h"
+#include "cabac.h"
+#include "cabac_functions.h"
 #include "dsputil.h"
 #include "avcodec.h"
 #include "mpegvideo.h"
 #include "golomb.h"
 #include "mathops.h"
 #include "rectangle.h"
+#include "thread.h"
 #include "vdpau_internal.h"
 #include "libavutil/avassert.h"
 
-#include "cabac.h"
-
 //#undef NDEBUG
 #include <assert.h>
 
-static const uint8_t rem6[52]={
-0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
+static const uint8_t rem6[QP_MAX_NUM+1]={
+0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
 };
 
-static const uint8_t div6[52]={
-0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
+static const uint8_t div6[QP_MAX_NUM+1]={
+0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
 };
 
 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
     PIX_FMT_DXVA2_VLD,
     PIX_FMT_VAAPI_VLD,
+    PIX_FMT_VDA_VLD,
     PIX_FMT_YUVJ420P,
     PIX_FMT_NONE
 };
 
-void ff_h264_write_back_intra_pred_mode(H264Context *h){
-    int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
-
-    AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
-    mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
-    mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
-    mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
-}
-
 /**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed and
+ * change the dc mode so it only uses the available blocks.
  */
 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
     MpegEncContext * const s = &h->s;
@@ -108,9 +102,10 @@ int ff_h264_check_intra4x4_pred_mode(H264Context *h){
 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 
 /**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed and
+ * change the dc mode so it only uses the available blocks.
  */
-int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
+int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma){
     MpegEncContext * const s = &h->s;
     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
@@ -130,7 +125,7 @@ int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 
     if((h->left_samples_available&0x8080) != 0x8080){
         mode= left[ mode ];
-        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
+        if(is_chroma && (h->left_samples_available&0x8080)){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
         }
         if(mode<0){
@@ -152,10 +147,6 @@ const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_l
     h->nal_unit_type= src[0]&0x1F;
 
     src++; length--;
-#if 0
-    for(i=0; i<length; i++)
-        printf("%2X ", src[i]);
-#endif
 
 #if HAVE_FAST_UNALIGNED
 # if HAVE_FAST_64BIT
@@ -249,6 +240,141 @@ static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
     return 0;
 }
 
+static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
+                                 int y_offset, int list){
+    int raw_my= h->mv_cache[list][ scan8[n] ][1];
+    int filter_height= (raw_my&3) ? 2 : 0;
+    int full_my= (raw_my>>2) + y_offset;
+    int top = full_my - filter_height, bottom = full_my + height + filter_height;
+
+    return FFMAX(abs(top), bottom);
+}
+
+static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
+                               int y_offset, int list0, int list1, int *nrefs){
+    MpegEncContext * const s = &h->s;
+    int my;
+
+    y_offset += 16*(s->mb_y >> MB_FIELD);
+
+    if(list0){
+        int ref_n = h->ref_cache[0][ scan8[n] ];
+        Picture *ref= &h->ref_list[0][ref_n];
+
+        // Error resilience puts the current picture in the ref list.
+        // Don't try to wait on these as it will cause a deadlock.
+        // Fields can wait on each other, though.
+        if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
+           (ref->f.reference & 3) != s->picture_structure) {
+            my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
+            if (refs[0][ref_n] < 0) nrefs[0] += 1;
+            refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
+        }
+    }
+
+    if(list1){
+        int ref_n = h->ref_cache[1][ scan8[n] ];
+        Picture *ref= &h->ref_list[1][ref_n];
+
+        if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
+           (ref->f.reference & 3) != s->picture_structure) {
+            my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
+            if (refs[1][ref_n] < 0) nrefs[1] += 1;
+            refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
+        }
+    }
+}
+
+/**
+ * Wait until all reference frames are available for MC operations.
+ *
+ * @param h the H264 context
+ */
+static void await_references(H264Context *h){
+    MpegEncContext * const s = &h->s;
+    const int mb_xy= h->mb_xy;
+    const int mb_type = s->current_picture.f.mb_type[mb_xy];
+    int refs[2][48];
+    int nrefs[2] = {0};
+    int ref, list;
+
+    memset(refs, -1, sizeof(refs));
+
+    if(IS_16X16(mb_type)){
+        get_lowest_part_y(h, refs, 0, 16, 0,
+                  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
+    }else if(IS_16X8(mb_type)){
+        get_lowest_part_y(h, refs, 0, 8, 0,
+                  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
+        get_lowest_part_y(h, refs, 8, 8, 8,
+                  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
+    }else if(IS_8X16(mb_type)){
+        get_lowest_part_y(h, refs, 0, 16, 0,
+                  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
+        get_lowest_part_y(h, refs, 4, 16, 0,
+                  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
+    }else{
+        int i;
+
+        assert(IS_8X8(mb_type));
+
+        for(i=0; i<4; i++){
+            const int sub_mb_type= h->sub_mb_type[i];
+            const int n= 4*i;
+            int y_offset= (i&2)<<2;
+
+            if(IS_SUB_8X8(sub_mb_type)){
+                get_lowest_part_y(h, refs, n  , 8, y_offset,
+                          IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+            }else if(IS_SUB_8X4(sub_mb_type)){
+                get_lowest_part_y(h, refs, n  , 4, y_offset,
+                          IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+                get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
+                          IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+            }else if(IS_SUB_4X8(sub_mb_type)){
+                get_lowest_part_y(h, refs, n  , 8, y_offset,
+                          IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+                get_lowest_part_y(h, refs, n+1, 8, y_offset,
+                          IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+            }else{
+                int j;
+                assert(IS_SUB_4X4(sub_mb_type));
+                for(j=0; j<4; j++){
+                    int sub_y_offset= y_offset + 2*(j&2);
+                    get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
+                              IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
+                }
+            }
+        }
+    }
+
+    for(list=h->list_count-1; list>=0; list--){
+        for(ref=0; ref<48 && nrefs[list]; ref++){
+            int row = refs[list][ref];
+            if(row >= 0){
+                Picture *ref_pic = &h->ref_list[list][ref];
+                int ref_field = ref_pic->f.reference - 1;
+                int ref_field_picture = ref_pic->field_picture;
+                int pic_height = 16*s->mb_height >> ref_field_picture;
+
+                row <<= MB_MBAFF;
+                nrefs[list]--;
+
+                if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
+                    ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
+                    ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1)           , pic_height-1), 0);
+                }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
+                    ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field    , pic_height-1), 0);
+                }else if(FIELD_PICTURE){
+                    ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
+                }else{
+                    ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
+                }
+            }
+        }
+    }
+}
+
 #if 0
 /**
  * DCT transforms the 16 dc values.
@@ -292,27 +418,6 @@ static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
 #undef xStride
 #undef stride
 
-static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
-    const int stride= 16*2;
-    const int xStride= 16;
-    int a,b,c,d,e;
-
-    a= block[stride*0 + xStride*0];
-    b= block[stride*0 + xStride*1];
-    c= block[stride*1 + xStride*0];
-    d= block[stride*1 + xStride*1];
-
-    e= a-b;
-    a= a+b;
-    b= c-d;
-    c= c+d;
-
-    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
-    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
-    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
-    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
-}
-
 #if 0
 static void chroma_dc_dct_c(DCTELEM *block){
     const int stride= 16*2;
@@ -336,15 +441,20 @@ static void chroma_dc_dct_c(DCTELEM *block){
 }
 #endif
 
-static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int src_x_offset, int src_y_offset,
-                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
+static av_always_inline void
+mc_dir_part(H264Context *h, Picture *pic, int n, int square,
+            int height, int delta, int list,
+            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+            int src_x_offset, int src_y_offset,
+            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
+            int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
     const int luma_xy= (mx&3) + ((my&3)<<2);
-    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
+    int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
+    uint8_t * src_y = pic->f.data[0] + offset;
     uint8_t * src_cb, * src_cr;
     int extra_width= h->emu_edge_width;
     int extra_height= h->emu_edge_height;
@@ -353,6 +463,7 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
     const int full_my= my>>2;
     const int pic_width  = 16*s->mb_width;
     const int pic_height = 16*s->mb_height >> MB_FIELD;
+    int ysh;
 
     if(mx&7) extra_width -= 3;
     if(my&7) extra_height -= 3;
@@ -361,8 +472,9 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
        || full_my < 0-extra_height
        || full_mx + 16/*FIXME*/ > pic_width + extra_width
        || full_my + 16/*FIXME*/ > pic_height + extra_height){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
-            src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
+                                16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+            src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
         emu=1;
     }
 
@@ -373,48 +485,91 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
 
     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 
-    if(MB_FIELD){
+    if(chroma_idc == 3 /* yuv444 */){
+        src_cb = pic->f.data[1] + offset;
+        if(emu){
+            s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
+                                    16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+            src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
+        }
+        qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
+        if(!square){
+            qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
+        }
+
+        src_cr = pic->f.data[2] + offset;
+        if(emu){
+            s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
+                                    16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+            src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
+        }
+        qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
+        if(!square){
+            qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
+        }
+        return;
+    }
+
+    ysh = 3 - (chroma_idc == 2 /* yuv422 */);
+    if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
         // chroma offset when predicting from a field of opposite parity
-        my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
+        my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
     }
-    src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
-    src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
+
+    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
+    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
+                                9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+                                pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
             src_cb= s->edge_emu_buffer;
     }
-    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
+              mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
+                                9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+                                pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
             src_cr= s->edge_emu_buffer;
     }
-    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
+              mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 }
 
-static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
-                           int list0, int list1){
+static av_always_inline void
+mc_part_std(H264Context *h, int n, int square, int height, int delta,
+            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+            int x_offset, int y_offset,
+            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
+            int list0, int list1, int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     qpel_mc_func *qpix_op=  qpix_put;
     h264_chroma_mc_func chroma_op= chroma_put;
 
-    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
+    dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    if (chroma_idc == 3 /* yuv444 */) {
+        dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+        dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    } else if (chroma_idc == 2 /* yuv422 */) {
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+    } else /* yuv420 */ {
+        dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+    }
     x_offset += 8*s->mb_x;
     y_offset += 8*(s->mb_y >> MB_FIELD);
 
     if(list0){
         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
+        mc_dir_part(h, ref, n, square, height, delta, 0,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift, chroma_idc);
 
         qpix_op=  qpix_avg;
         chroma_op= chroma_avg;
@@ -422,24 +577,39 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
 
     if(list1){
         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
+        mc_dir_part(h, ref, n, square, height, delta, 1,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift, chroma_idc);
     }
 }
 
-static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
-                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
-                           int list0, int list1){
+static av_always_inline void
+mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
+                 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+                 int x_offset, int y_offset,
+                 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+                 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
+                 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
+                 int list0, int list1, int pixel_shift, int chroma_idc){
     MpegEncContext * const s = &h->s;
-
-    dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;
-    dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;
-    dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;
+    int chroma_height;
+
+    dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    if (chroma_idc == 3 /* yuv444 */) {
+        chroma_height = height;
+        chroma_weight_avg = luma_weight_avg;
+        chroma_weight_op = luma_weight_op;
+        dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+        dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    } else if (chroma_idc == 2 /* yuv422 */) {
+        chroma_height = height;
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+    } else /* yuv420 */ {
+        chroma_height = height >> 1;
+        dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
+    }
     x_offset += 8*s->mb_x;
     y_offset += 8*(s->mb_y >> MB_FIELD);
 
@@ -447,32 +617,37 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         /* don't optimize for luma-only case, since B-frames usually
          * use implicit weights => chroma too. */
         uint8_t *tmp_cb = s->obmc_scratchpad;
-        uint8_t *tmp_cr = s->obmc_scratchpad + 8;
-        uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;
+        uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
+        uint8_t *tmp_y  = s->obmc_scratchpad + 16*h->mb_uvlinesize;
         int refn0 = h->ref_cache[0][ scan8[n] ];
         int refn1 = h->ref_cache[1][ scan8[n] ];
 
-        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
+        mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
                     dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put);
-        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
+                    x_offset, y_offset, qpix_put, chroma_put,
+                    pixel_shift, chroma_idc);
+        mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
                     tmp_y, tmp_cb, tmp_cr,
-                    x_offset, y_offset, qpix_put, chroma_put);
+                    x_offset, y_offset, qpix_put, chroma_put,
+                    pixel_shift, chroma_idc);
 
         if(h->use_weight == 2){
             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
             int weight1 = 64 - weight0;
-            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
+            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize,
+                              height,        5, weight0, weight1, 0);
+            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
+                              chroma_height, 5, weight0, weight1, 0);
+            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
+                              chroma_height, 5, weight0, weight1, 0);
         }else{
-            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
+            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
-            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
-            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
         }
@@ -480,40 +655,46 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         int list = list1 ? 1 : 0;
         int refn = h->ref_cache[list][ scan8[n] ];
         Picture *ref= &h->ref_list[list][refn];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
+        mc_dir_part(h, ref, n, square, height, delta, list,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                    qpix_put, chroma_put);
+                    qpix_put, chroma_put, pixel_shift, chroma_idc);
 
-        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
+        luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
         if(h->use_weight_chroma){
-            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
-            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
         }
     }
 }
 
-static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
-                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
-                           int list0, int list1){
+static av_always_inline void
+mc_part(H264Context *h, int n, int square, int height, int delta,
+        uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+        int x_offset, int y_offset,
+        qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+        qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
+        h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+        int list0, int list1, int pixel_shift, int chroma_idc)
+{
     if((h->use_weight==2 && list0 && list1
         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
        || h->use_weight==1)
-        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
+        mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
                          x_offset, y_offset, qpix_put, chroma_put,
-                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
+                         weight_op[0], weight_op[1], weight_avg[0],
+                         weight_avg[1], list0, list1, pixel_shift, chroma_idc);
     else
-        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
+        mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
+                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
+                    chroma_avg, list0, list1, pixel_shift, chroma_idc);
 }
 
-static inline void prefetch_motion(H264Context *h, int list){
+static av_always_inline void
+prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
+{
     /* fetch pixels for estimated mv 4 macroblocks ahead
      * optimized for 64byte cache lines */
     MpegEncContext * const s = &h->s;
@@ -521,49 +702,63 @@ static inline void prefetch_motion(H264Context *h, int list){
     if(refn >= 0){
         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
-        uint8_t **src= h->ref_list[list][refn].data;
-        int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
+        uint8_t **src = h->ref_list[list][refn].f.data;
+        int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
         s->dsp.prefetch(src[0]+off, s->linesize, 4);
-        off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
-        s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
+        if (chroma_idc == 3 /* yuv444 */) {
+            s->dsp.prefetch(src[1]+off, s->linesize, 4);
+            s->dsp.prefetch(src[2]+off, s->linesize, 4);
+        }else{
+            off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
+            s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
+        }
     }
 }
 
-static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
-                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
+                      h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+                      int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
-    const int mb_type= s->current_picture.mb_type[mb_xy];
+    const int mb_type = s->current_picture.f.mb_type[mb_xy];
 
     assert(IS_INTER(mb_type));
 
-    prefetch_motion(h, 0);
+    if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
+        await_references(h);
+    prefetch_motion(h, 0, pixel_shift, chroma_idc);
 
     if(IS_16X16(mb_type)){
-        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
                 weight_op, weight_avg,
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift, chroma_idc);
     }else if(IS_16X8(mb_type)){
-        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
-                &weight_op[1], &weight_avg[1],
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
-        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
+                weight_op, weight_avg,
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift, chroma_idc);
+        mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
-                &weight_op[1], &weight_avg[1],
-                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
+                weight_op, weight_avg,
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift, chroma_idc);
     }else if(IS_8X16(mb_type)){
-        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                &weight_op[2], &weight_avg[2],
-                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
-        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
+                &weight_op[1], &weight_avg[1],
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift, chroma_idc);
+        mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                &weight_op[2], &weight_avg[2],
-                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
+                &weight_op[1], &weight_avg[1],
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift, chroma_idc);
     }else{
         int i;
 
@@ -576,50 +771,78 @@ static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t
             int y_offset= (i&2)<<1;
 
             if(IS_SUB_8X8(sub_mb_type)){
-                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                    &weight_op[3], &weight_avg[3],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    &weight_op[1], &weight_avg[1],
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift, chroma_idc);
             }else if(IS_SUB_8X4(sub_mb_type)){
-                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n  , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
-                    &weight_op[4], &weight_avg[4],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
-                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
+                    &weight_op[1], &weight_avg[1],
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift, chroma_idc);
+                mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
-                    &weight_op[4], &weight_avg[4],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    &weight_op[1], &weight_avg[1],
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift, chroma_idc);
             }else if(IS_SUB_4X8(sub_mb_type)){
-                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n  , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                    &weight_op[5], &weight_avg[5],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
-                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
+                    &weight_op[2], &weight_avg[2],
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift, chroma_idc);
+                mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                    &weight_op[5], &weight_avg[5],
-                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                    &weight_op[2], &weight_avg[2],
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift, chroma_idc);
             }else{
                 int j;
                 assert(IS_SUB_4X4(sub_mb_type));
                 for(j=0; j<4; j++){
                     int sub_x_offset= x_offset + 2*(j&1);
                     int sub_y_offset= y_offset +   (j&2);
-                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
+                    mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                        &weight_op[6], &weight_avg[6],
-                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
+                        &weight_op[2], &weight_avg[2],
+                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                        pixel_shift, chroma_idc);
                 }
             }
         }
     }
 
-    prefetch_motion(h, 1);
+    prefetch_motion(h, 1, pixel_shift, chroma_idc);
 }
 
+static av_always_inline void
+hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+              qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
+              qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
+              h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+              int pixel_shift)
+{
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
+              qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
+}
+
+static av_always_inline void
+hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+              qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
+              qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
+              h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+              int pixel_shift)
+{
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
+              qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
+}
 
 static void free_tables(H264Context *h, int free_rbsp){
     int i;
     H264Context *hx;
+
     av_freep(&h->intra4x4_pred_mode);
     av_freep(&h->chroma_pred_mode_table);
     av_freep(&h->cbp_table);
@@ -651,17 +874,21 @@ static void free_tables(H264Context *h, int free_rbsp){
 }
 
 static void init_dequant8_coeff_table(H264Context *h){
-    int i,q,x;
-    h->dequant8_coeff[0] = h->dequant8_buffer[0];
-    h->dequant8_coeff[1] = h->dequant8_buffer[1];
+    int i,j,q,x;
+    const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
 
-    for(i=0; i<2; i++ ){
-        if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
-            h->dequant8_coeff[1] = h->dequant8_buffer[0];
-            break;
+    for(i=0; i<6; i++ ){
+        h->dequant8_coeff[i] = h->dequant8_buffer[i];
+        for(j=0; j<i; j++){
+            if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
+                h->dequant8_coeff[i] = h->dequant8_buffer[j];
+                break;
+            }
         }
+        if(j<i)
+            continue;
 
-        for(q=0; q<52; q++){
+        for(q=0; q<max_qp+1; q++){
             int shift = div6[q];
             int idx = rem6[q];
             for(x=0; x<64; x++)
@@ -674,6 +901,7 @@ static void init_dequant8_coeff_table(H264Context *h){
 
 static void init_dequant4_coeff_table(H264Context *h){
     int i,j,q,x;
+    const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
     for(i=0; i<6; i++ ){
         h->dequant4_coeff[i] = h->dequant4_buffer[i];
         for(j=0; j<i; j++){
@@ -685,7 +913,7 @@ static void init_dequant4_coeff_table(H264Context *h){
         if(j<i)
             continue;
 
-        for(q=0; q<52; q++){
+        for(q=0; q<max_qp+1; q++){
             int shift = div6[q] + 2;
             int idx = rem6[q];
             for(x=0; x<16; x++)
@@ -706,7 +934,7 @@ static void init_dequant_tables(H264Context *h){
             for(x=0; x<16; x++)
                 h->dequant4_coeff[i][0][x] = 1<<6;
         if(h->pps.transform_8x8_mode)
-            for(i=0; i<2; i++)
+            for(i=0; i<6; i++)
                 for(x=0; x<64; x++)
                     h->dequant8_coeff[i][0][x] = 1<<6;
     }
@@ -721,7 +949,7 @@ int ff_h264_alloc_tables(H264Context *h){
 
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8  * sizeof(uint8_t), fail)
 
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 32 * sizeof(uint8_t), fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count    , big_mb_num * 48 * sizeof(uint8_t), fail)
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
 
@@ -775,7 +1003,7 @@ static void clone_tables(H264Context *dst, H264Context *src, int i){
     dst->list_counts              = src->list_counts;
 
     dst->s.obmc_scratchpad = NULL;
-    ff_h264_pred_init(&dst->hpc, src->s.codec_id);
+    ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
 }
 
 /**
@@ -783,8 +1011,8 @@ static void clone_tables(H264Context *dst, H264Context *src, int i){
  * Allocate buffers which are not shared amongst multiple threads.
  */
 static int context_init(H264Context *h){
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
-    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * sizeof(uint8_t), fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
 
     h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
     h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
@@ -803,12 +1031,11 @@ static av_cold void common_init(H264Context *h){
     s->height = s->avctx->height;
     s->codec_id= s->avctx->codec->id;
 
-    ff_h264dsp_init(&h->h264dsp);
-    ff_h264_pred_init(&h->hpc, s->codec_id);
+    ff_h264dsp_init(&h->h264dsp, 8, 1);
+    ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
 
     h->dequant_coeff_pps= -1;
     s->unrestricted_mv=1;
-    s->decode=1; //FIXME
 
     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 
@@ -820,7 +1047,7 @@ int ff_h264_decode_extradata(H264Context *h)
 {
     AVCodecContext *avctx = h->s.avctx;
 
-    if(*(char *)avctx->extradata == 1){
+    if(avctx->extradata[0] == 1){
         int i, cnt, nalsize;
         unsigned char *p = avctx->extradata;
 
@@ -838,6 +1065,8 @@ int ff_h264_decode_extradata(H264Context *h)
         p += 6;
         for (i = 0; i < cnt; i++) {
             nalsize = AV_RB16(p) + 2;
+            if (p - avctx->extradata + nalsize > avctx->extradata_size)
+                return -1;
             if(decode_nal_units(h, p, nalsize) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
                 return -1;
@@ -848,14 +1077,16 @@ int ff_h264_decode_extradata(H264Context *h)
         cnt = *(p++); // Number of pps
         for (i = 0; i < cnt; i++) {
             nalsize = AV_RB16(p) + 2;
-            if(decode_nal_units(h, p, nalsize)  != nalsize) {
+            if (p - avctx->extradata + nalsize > avctx->extradata_size)
+                return -1;
+            if (decode_nal_units(h, p, nalsize) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
                 return -1;
             }
             p += nalsize;
         }
         // Now store right nal length size, that will be use to parse all other nals
-        h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
+        h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
     } else {
         h->is_avc = 0;
         if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
@@ -867,6 +1098,7 @@ int ff_h264_decode_extradata(H264Context *h)
 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
     H264Context *h= avctx->priv_data;
     MpegEncContext * const s = &h->s;
+    int i;
 
     MPV_decode_defaults(s);
 
@@ -886,8 +1118,13 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 
     ff_h264_decode_init_vlc();
 
+    h->pixel_shift = 0;
+    h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
+
     h->thread_context[0] = h;
-    h->outputed_poc = INT_MIN;
+    h->outputed_poc = h->next_outputed_poc = INT_MIN;
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+        h->last_pocs[i] = INT_MIN;
     h->prev_poc_msb= 1<<16;
     h->x264_build = -1;
     ff_h264_reset_sei(h);
@@ -910,9 +1147,139 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
     return 0;
 }
 
+#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
+static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
+{
+    int i;
+
+    for (i=0; i<count; i++){
+        assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
+                IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
+                !from[i]));
+        to[i] = REBASE_PICTURE(from[i], new_base, old_base);
+    }
+}
+
+static void copy_parameter_set(void **to, void **from, int count, int size)
+{
+    int i;
+
+    for (i=0; i<count; i++){
+        if (to[i] && !from[i]) av_freep(&to[i]);
+        else if (from[i] && !to[i]) to[i] = av_malloc(size);
+
+        if (from[i]) memcpy(to[i], from[i], size);
+    }
+}
+
+static int decode_init_thread_copy(AVCodecContext *avctx){
+    H264Context *h= avctx->priv_data;
+
+    if (!avctx->internal->is_copy)
+        return 0;
+    memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
+    memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
+
+    return 0;
+}
+
+#define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
+static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src){
+    H264Context *h= dst->priv_data, *h1= src->priv_data;
+    MpegEncContext * const s = &h->s, * const s1 = &h1->s;
+    int inited = s->context_initialized, err;
+    int i;
+
+    if(dst == src || !s1->context_initialized) return 0;
+
+    err = ff_mpeg_update_thread_context(dst, src);
+    if(err) return err;
+
+    //FIXME handle width/height changing
+    if(!inited){
+        for(i = 0; i < MAX_SPS_COUNT; i++)
+            av_freep(h->sps_buffers + i);
+
+        for(i = 0; i < MAX_PPS_COUNT; i++)
+            av_freep(h->pps_buffers + i);
+
+        memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
+        memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
+        memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
+        if (ff_h264_alloc_tables(h) < 0) {
+            av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
+            return AVERROR(ENOMEM);
+        }
+        context_init(h);
+
+        for(i=0; i<2; i++){
+            h->rbsp_buffer[i] = NULL;
+            h->rbsp_buffer_size[i] = 0;
+        }
+
+        h->thread_context[0] = h;
+
+        // frame_start may not be called for the next thread (if it's decoding a bottom field)
+        // so this has to be allocated here
+        h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
+
+        s->dsp.clear_blocks(h->mb);
+        s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
+    }
+
+    //extradata/NAL handling
+    h->is_avc          = h1->is_avc;
+
+    //SPS/PPS
+    copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
+    h->sps             = h1->sps;
+    copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
+    h->pps             = h1->pps;
+
+    //Dequantization matrices
+    //FIXME these are big - can they be only copied when PPS changes?
+    copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
+
+    for(i=0; i<6; i++)
+        h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
+
+    for(i=0; i<6; i++)
+        h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
+
+    h->dequant_coeff_pps = h1->dequant_coeff_pps;
+
+    //POC timing
+    copy_fields(h, h1, poc_lsb, redundant_pic_count);
+
+    //reference lists
+    copy_fields(h, h1, ref_count, list_count);
+    copy_fields(h, h1, ref_list,  intra_gb);
+    copy_fields(h, h1, short_ref, cabac_init_idc);
+
+    copy_picture_range(h->short_ref,   h1->short_ref,   32, s, s1);
+    copy_picture_range(h->long_ref,    h1->long_ref,    32, s, s1);
+    copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
+
+    h->last_slice_type = h1->last_slice_type;
+
+    if(!s->current_picture_ptr) return 0;
+
+    if(!s->dropable) {
+        err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+        h->prev_poc_msb     = h->poc_msb;
+        h->prev_poc_lsb     = h->poc_lsb;
+    }
+    h->prev_frame_num_offset= h->frame_num_offset;
+    h->prev_frame_num       = h->frame_num;
+    h->outputed_poc         = h->next_outputed_poc;
+
+    return err;
+}
+
 int ff_h264_frame_start(H264Context *h){
     MpegEncContext * const s = &h->s;
     int i;
+    const int pixel_shift = h->pixel_shift;
 
     if(MPV_frame_start(s, s->avctx) < 0)
         return -1;
@@ -923,32 +1290,32 @@ int ff_h264_frame_start(H264Context *h){
      * Zero here; IDR markings per slice in frame or fields are ORed in later.
      * See decode_nal_units().
      */
-    s->current_picture_ptr->key_frame= 0;
+    s->current_picture_ptr->f.key_frame = 0;
     s->current_picture_ptr->mmco_reset= 0;
 
     assert(s->linesize && s->uvlinesize);
 
     for(i=0; i<16; i++){
-        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
-        h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
     }
-    for(i=0; i<4; i++){
+    for(i=0; i<16; i++){
         h->block_offset[16+i]=
-        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
-        h->block_offset[24+16+i]=
-        h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
+        h->block_offset[48+16+i]=
+        h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
     }
 
     /* can't be in alloc_tables because linesize isn't known there.
      * FIXME: redo bipred weight to not require extra buffer? */
-    for(i = 0; i < s->avctx->thread_count; i++)
+    for(i = 0; i < s->slice_context_count; i++)
         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
-            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
+            h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
 
     /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
     memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
 
-//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
+//    s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
 
     // We mark the current picture as non-reference after allocating it, so
     // that if we break out due to an error it can be released automatically
@@ -957,19 +1324,236 @@ int ff_h264_frame_start(H264Context *h){
     // get released even with set reference, besides SVQ3 and others do not
     // mark frames as reference later "naturally".
     if(s->codec_id != CODEC_ID_SVQ3)
-        s->current_picture_ptr->reference= 0;
+        s->current_picture_ptr->f.reference = 0;
 
     s->current_picture_ptr->field_poc[0]=
     s->current_picture_ptr->field_poc[1]= INT_MAX;
+
+    h->next_output_pic = NULL;
+
     assert(s->current_picture_ptr->long_ref==0);
 
     return 0;
 }
 
-static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int simple){
+/**
+  * Run setup operations that must be run after slice header decoding.
+  * This includes finding the next displayed frame.
+  *
+  * @param h h264 master context
+  * @param setup_finished enough NALs have been read that we can call
+  * ff_thread_finish_setup()
+  */
+static void decode_postinit(H264Context *h, int setup_finished){
+    MpegEncContext * const s = &h->s;
+    Picture *out = s->current_picture_ptr;
+    Picture *cur = s->current_picture_ptr;
+    int i, pics, out_of_order, out_idx;
+    int invalid = 0, cnt = 0;
+
+    s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
+    s->current_picture_ptr->f.pict_type   = s->pict_type;
+
+    if (h->next_output_pic) return;
+
+    if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
+        //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
+        //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
+        //to find this yet, so we assume the worst for now.
+        //if (setup_finished)
+        //    ff_thread_finish_setup(s->avctx);
+        return;
+    }
+
+    cur->f.interlaced_frame = 0;
+    cur->f.repeat_pict      = 0;
+
+    /* Signal interlacing information externally. */
+    /* Prioritize picture timing SEI information over used decoding process if it exists. */
+
+    if(h->sps.pic_struct_present_flag){
+        switch (h->sei_pic_struct)
+        {
+        case SEI_PIC_STRUCT_FRAME:
+            break;
+        case SEI_PIC_STRUCT_TOP_FIELD:
+        case SEI_PIC_STRUCT_BOTTOM_FIELD:
+            cur->f.interlaced_frame = 1;
+            break;
+        case SEI_PIC_STRUCT_TOP_BOTTOM:
+        case SEI_PIC_STRUCT_BOTTOM_TOP:
+            if (FIELD_OR_MBAFF_PICTURE)
+                cur->f.interlaced_frame = 1;
+            else
+                // try to flag soft telecine progressive
+                cur->f.interlaced_frame = h->prev_interlaced_frame;
+            break;
+        case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
+        case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
+            // Signal the possibility of telecined film externally (pic_struct 5,6)
+            // From these hints, let the applications decide if they apply deinterlacing.
+            cur->f.repeat_pict = 1;
+            break;
+        case SEI_PIC_STRUCT_FRAME_DOUBLING:
+            // Force progressive here, as doubling interlaced frame is a bad idea.
+            cur->f.repeat_pict = 2;
+            break;
+        case SEI_PIC_STRUCT_FRAME_TRIPLING:
+            cur->f.repeat_pict = 4;
+            break;
+        }
+
+        if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
+            cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
+    }else{
+        /* Derive interlacing flag from used decoding process. */
+        cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
+    }
+    h->prev_interlaced_frame = cur->f.interlaced_frame;
+
+    if (cur->field_poc[0] != cur->field_poc[1]){
+        /* Derive top_field_first from field pocs. */
+        cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
+    }else{
+        if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
+            /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
+            if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
+              || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
+                cur->f.top_field_first = 1;
+            else
+                cur->f.top_field_first = 0;
+        }else{
+            /* Most likely progressive */
+            cur->f.top_field_first = 0;
+        }
+    }
+
+    //FIXME do something with unavailable reference frames
+
+    /* Sort B-frames into display order */
+
+    if(h->sps.bitstream_restriction_flag
+       && s->avctx->has_b_frames < h->sps.num_reorder_frames){
+        s->avctx->has_b_frames = h->sps.num_reorder_frames;
+        s->low_delay = 0;
+    }
+
+    if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
+       && !h->sps.bitstream_restriction_flag){
+        s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
+        s->low_delay= 0;
+    }
+
+    pics = 0;
+    while(h->delayed_pic[pics]) pics++;
+
+    assert(pics <= MAX_DELAYED_PIC_COUNT);
+
+    h->delayed_pic[pics++] = cur;
+    if (cur->f.reference == 0)
+        cur->f.reference = DELAYED_PIC_REF;
+
+    /* Frame reordering. This code takes pictures from coding order and sorts
+     * them by their incremental POC value into display order. It supports POC
+     * gaps, MMCO reset codes and random resets.
+     * A "display group" can start either with a IDR frame (f.key_frame = 1),
+     * and/or can be closed down with a MMCO reset code. In sequences where
+     * there is no delay, we can't detect that (since the frame was already
+     * output to the user), so we also set h->mmco_reset to detect the MMCO
+     * reset code.
+     * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
+     * we increase the delay between input and output. All frames affected by
+     * the lag (e.g. those that should have been output before another frame
+     * that we already returned to the user) will be dropped. This is a bug
+     * that we will fix later. */
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
+        cnt     += out->poc < h->last_pocs[i];
+        invalid += out->poc == INT_MIN;
+    }
+    if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
+        h->mmco_reset = 2;
+        if (pics > 1)
+            h->delayed_pic[pics - 2]->mmco_reset = 2;
+    }
+    if (h->mmco_reset || cur->f.key_frame) {
+        for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+            h->last_pocs[i] = INT_MIN;
+        cnt     = 0;
+        invalid = MAX_DELAYED_PIC_COUNT;
+    }
+    out = h->delayed_pic[0];
+    out_idx = 0;
+    for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
+         !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
+    {
+        if(h->delayed_pic[i]->poc < out->poc){
+            out = h->delayed_pic[i];
+            out_idx = i;
+        }
+    }
+    if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
+        h->next_outputed_poc = INT_MIN;
+    out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
+
+    if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
+        { }
+    else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
+             s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
+        if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
+            s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
+        }
+        s->low_delay = 0;
+    } else if (s->low_delay &&
+               ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
+                cur->f.pict_type == AV_PICTURE_TYPE_B)) {
+        s->low_delay = 0;
+        s->avctx->has_b_frames++;
+    }
+
+    if(pics > s->avctx->has_b_frames){
+        out->f.reference &= ~DELAYED_PIC_REF;
+        out->owner2 = s; // for frame threading, the owner must be the second field's thread
+                         // or else the first thread can release the picture and reuse it unsafely
+        for(i=out_idx; h->delayed_pic[i]; i++)
+            h->delayed_pic[i] = h->delayed_pic[i+1];
+    }
+    memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
+    h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
+    if(!out_of_order && pics > s->avctx->has_b_frames){
+        h->next_output_pic = out;
+        if (out->mmco_reset) {
+            if (out_idx > 0) {
+                h->next_outputed_poc = out->poc;
+                h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
+            } else {
+                h->next_outputed_poc = INT_MIN;
+            }
+        } else {
+            if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
+                h->next_outputed_poc = INT_MIN;
+            } else {
+                h->next_outputed_poc = out->poc;
+            }
+        }
+        h->mmco_reset = 0;
+    }else{
+        av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
+    }
+
+    if (setup_finished)
+        ff_thread_finish_setup(s->avctx);
+}
+
+static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
+                                              uint8_t *src_cb, uint8_t *src_cr,
+                                              int linesize, int uvlinesize, int simple)
+{
     MpegEncContext * const s = &h->s;
     uint8_t *top_border;
     int top_idx = 1;
+    const int pixel_shift = h->pixel_shift;
+    int chroma444 = CHROMA444;
+    int chroma422 = CHROMA422;
 
     src_y  -=   linesize;
     src_cb -= uvlinesize;
@@ -980,9 +1564,36 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
             if(!MB_MBAFF){
                 top_border = h->top_borders[0][s->mb_x];
                 AV_COPY128(top_border, src_y + 15*linesize);
+                if (pixel_shift)
+                    AV_COPY128(top_border+16, src_y+15*linesize+16);
                 if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                    AV_COPY64(top_border+16, src_cb+7*uvlinesize);
-                    AV_COPY64(top_border+24, src_cr+7*uvlinesize);
+                    if(chroma444){
+                        if (pixel_shift){
+                            AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
+                            AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
+                            AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
+                            AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
+                        } else {
+                            AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
+                            AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
+                        }
+                    } else if(chroma422) {
+                        if (pixel_shift) {
+                            AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
+                            AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
+                        } else {
+                            AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
+                            AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
+                        }
+                    } else {
+                        if (pixel_shift) {
+                            AV_COPY128(top_border+32, src_cb+7*uvlinesize);
+                            AV_COPY128(top_border+48, src_cr+7*uvlinesize);
+                        } else {
+                            AV_COPY64(top_border+16, src_cb+7*uvlinesize);
+                            AV_COPY64(top_border+24, src_cr+7*uvlinesize);
+                        }
+                    }
                 }
             }
         }else if(MB_MBAFF){
@@ -995,16 +1606,47 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
     // There are two lines saved, the line above the the top macroblock of a pair,
     // and the line above the bottom macroblock
     AV_COPY128(top_border, src_y + 16*linesize);
+    if (pixel_shift)
+        AV_COPY128(top_border+16, src_y+16*linesize+16);
 
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-        AV_COPY64(top_border+16, src_cb+8*uvlinesize);
-        AV_COPY64(top_border+24, src_cr+8*uvlinesize);
+        if(chroma444){
+            if (pixel_shift){
+                AV_COPY128(top_border+32, src_cb + 16*linesize);
+                AV_COPY128(top_border+48, src_cb + 16*linesize+16);
+                AV_COPY128(top_border+64, src_cr + 16*linesize);
+                AV_COPY128(top_border+80, src_cr + 16*linesize+16);
+            } else {
+                AV_COPY128(top_border+16, src_cb + 16*linesize);
+                AV_COPY128(top_border+32, src_cr + 16*linesize);
+            }
+        } else if(chroma422) {
+            if (pixel_shift) {
+                AV_COPY128(top_border+32, src_cb+16*uvlinesize);
+                AV_COPY128(top_border+48, src_cr+16*uvlinesize);
+            } else {
+                AV_COPY64(top_border+16, src_cb+16*uvlinesize);
+                AV_COPY64(top_border+24, src_cr+16*uvlinesize);
+            }
+        } else {
+            if (pixel_shift) {
+                AV_COPY128(top_border+32, src_cb+8*uvlinesize);
+                AV_COPY128(top_border+48, src_cr+8*uvlinesize);
+            } else {
+                AV_COPY64(top_border+16, src_cb+8*uvlinesize);
+                AV_COPY64(top_border+24, src_cr+8*uvlinesize);
+            }
+        }
     }
 }
 
-static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg, int simple){
+static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
+                                  uint8_t *src_cb, uint8_t *src_cr,
+                                  int linesize, int uvlinesize,
+                                  int xchg, int chroma444,
+                                  int simple, int pixel_shift){
     MpegEncContext * const s = &h->s;
-    int deblock_left;
+    int deblock_topleft;
     int deblock_top;
     int top_idx = 1;
     uint8_t *top_border_m1;
@@ -1020,80 +1662,271 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_c
     }
 
     if(h->deblocking_filter == 2) {
-        deblock_left = h->left_type[0];
-        deblock_top  = h->top_type;
+        deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
+        deblock_top     = h->top_type;
     } else {
-        deblock_left = (s->mb_x > 0);
-        deblock_top  (s->mb_y > !!MB_FIELD);
+        deblock_topleft = (s->mb_x > 0);
+        deblock_top     = (s->mb_y > !!MB_FIELD);
     }
 
-    src_y  -=   linesize + 1;
-    src_cb -= uvlinesize + 1;
-    src_cr -= uvlinesize + 1;
+    src_y  -=   linesize + 1 + pixel_shift;
+    src_cb -= uvlinesize + 1 + pixel_shift;
+    src_cr -= uvlinesize + 1 + pixel_shift;
 
     top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
     top_border    = h->top_borders[top_idx][s->mb_x];
 
 #define XCHG(a,b,xchg)\
+    if (pixel_shift) {\
+        if (xchg) {\
+            AV_SWAP64(b+0,a+0);\
+            AV_SWAP64(b+8,a+8);\
+        } else {\
+            AV_COPY128(b,a); \
+        }\
+    } else \
 if (xchg) AV_SWAP64(b,a);\
 else      AV_COPY64(b,a);
 
     if(deblock_top){
-        if(deblock_left){
-            XCHG(top_border_m1+8, src_y -7, 1);
+        if(deblock_topleft){
+            XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
         }
-        XCHG(top_border+0, src_y +1, xchg);
-        XCHG(top_border+8, src_y +9, 1);
+        XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
+        XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
         if(s->mb_x+1 < s->mb_width){
-            XCHG(h->top_borders[top_idx][s->mb_x+1], src_y +17, 1);
+            XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
         }
     }
-
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-        if(deblock_top){
-            if(deblock_left){
-                XCHG(top_border_m1+16, src_cb -7, 1);
-                XCHG(top_border_m1+24, src_cr -7, 1);
+        if(chroma444){
+            if(deblock_topleft){
+                XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
+                XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
+            }
+            XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
+            XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
+            XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
+            XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
+            if(s->mb_x+1 < s->mb_width){
+                XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
+                XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
+            }
+        } else {
+            if(deblock_top){
+                if(deblock_topleft){
+                    XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
+                    XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
+                }
+                XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
+                XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
             }
-            XCHG(top_border+16, src_cb+1, 1);
-            XCHG(top_border+24, src_cr+1, 1);
         }
     }
 }
 
-static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
+static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
+    if (high_bit_depth) {
+        return AV_RN32A(((int32_t*)mb) + index);
+    } else
+        return AV_RN16A(mb + index);
+}
+
+static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
+    if (high_bit_depth) {
+        AV_WN32A(((int32_t*)mb) + index, value);
+    } else
+        AV_WN16A(mb + index, value);
+}
+
+static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
+                                                       int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
+{
+    MpegEncContext * const s = &h->s;
+    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
+    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
+    int i;
+    int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
+    block_offset += 16*p;
+    if(IS_INTRA4x4(mb_type)){
+        if(simple || !s->encoding){
+            if(IS_8x8DCT(mb_type)){
+                if(transform_bypass){
+                    idct_dc_add =
+                    idct_add    = s->dsp.add_pixels8;
+                }else{
+                    idct_dc_add = h->h264dsp.h264_idct8_dc_add;
+                    idct_add    = h->h264dsp.h264_idct8_add;
+                }
+                for(i=0; i<16; i+=4){
+                    uint8_t * const ptr= dest_y + block_offset[i];
+                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
+                    if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
+                        h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                    }else{
+                        const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
+                        h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
+                                                    (h->topright_samples_available<<i)&0x4000, linesize);
+                        if(nnz){
+                            if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
+                                idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                            else
+                                idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                        }
+                    }
+                }
+            }else{
+                if(transform_bypass){
+                    idct_dc_add =
+                    idct_add    = s->dsp.add_pixels4;
+                }else{
+                    idct_dc_add = h->h264dsp.h264_idct_dc_add;
+                    idct_add    = h->h264dsp.h264_idct_add;
+                }
+                for(i=0; i<16; i++){
+                    uint8_t * const ptr= dest_y + block_offset[i];
+                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
+
+                    if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
+                        h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                    }else{
+                        uint8_t *topright;
+                        int nnz, tr;
+                        uint64_t tr_high;
+                        if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
+                            const int topright_avail= (h->topright_samples_available<<i)&0x8000;
+                            assert(s->mb_y || linesize <= block_offset[i]);
+                            if(!topright_avail){
+                                if (pixel_shift) {
+                                    tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
+                                    topright= (uint8_t*) &tr_high;
+                                } else {
+                                    tr= ptr[3 - linesize]*0x01010101u;
+                                    topright= (uint8_t*) &tr;
+                                }
+                            }else
+                                topright= ptr + (4 << pixel_shift) - linesize;
+                        }else
+                            topright= NULL;
+
+                        h->hpc.pred4x4[ dir ](ptr, topright, linesize);
+                        nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
+                        if(nnz){
+                            if(is_h264){
+                                if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
+                                    idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                                else
+                                    idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
+                            } else if (CONFIG_SVQ3_DECODER)
+                                ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
+                        }
+                    }
+                }
+            }
+        }
+    }else{
+        h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
+        if(is_h264){
+            if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX+p] ]){
+                if(!transform_bypass)
+                    h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
+                else{
+                    static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
+                                                            8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
+                    for(i = 0; i < 16; i++)
+                        dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
+                }
+            }
+        } else if (CONFIG_SVQ3_DECODER)
+            ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
+    }
+}
+
+static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
+                                                    int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
+{
+    MpegEncContext * const s = &h->s;
+    void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
+    int i;
+    block_offset += 16*p;
+    if(!IS_INTRA4x4(mb_type)){
+        if(is_h264){
+            if(IS_INTRA16x16(mb_type)){
+                if(transform_bypass){
+                    if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
+                        h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
+                    }else{
+                        for(i=0; i<16; i++){
+                            if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
+                                s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
+                        }
+                    }
+                }else{
+                    h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
+                }
+            }else if(h->cbp&15){
+                if(transform_bypass){
+                    const int di = IS_8x8DCT(mb_type) ? 4 : 1;
+                    idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
+                    for(i=0; i<16; i+=di){
+                        if(h->non_zero_count_cache[ scan8[i+p*16] ]){
+                            idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
+                        }
+                    }
+                }else{
+                    if(IS_8x8DCT(mb_type)){
+                        h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
+                    }else{
+                        h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
+                    }
+                }
+            }
+        } else if (CONFIG_SVQ3_DECODER) {
+            for(i=0; i<16; i++){
+                if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
+                    uint8_t * const ptr= dest_y + block_offset[i];
+                    ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
+                }
+            }
+        }
+    }
+}
+
+static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
+{
     MpegEncContext * const s = &h->s;
     const int mb_x= s->mb_x;
     const int mb_y= s->mb_y;
     const int mb_xy= h->mb_xy;
-    const int mb_type= s->current_picture.mb_type[mb_xy];
+    const int mb_type = s->current_picture.f.mb_type[mb_xy];
     uint8_t  *dest_y, *dest_cb, *dest_cr;
     int linesize, uvlinesize /*dct_offset*/;
-    int i;
+    int i, j;
     int *block_offset = &h->block_offset[0];
     const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
     /* is_h264 should always be true if SVQ3 is disabled. */
     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
-    void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
+    const int block_h = 16 >> s->chroma_y_shift;
+    const int chroma422 = CHROMA422;
 
-    dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
-    dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
-    dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
+    dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+    dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
+    dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
 
-    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4);
-    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2);
+    s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
+    s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
 
     h->list_counts[mb_xy]= h->list_count;
 
     if (!simple && MB_FIELD) {
         linesize   = h->mb_linesize   = s->linesize * 2;
         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
-        block_offset = &h->block_offset[24];
+        block_offset = &h->block_offset[48];
         if(mb_y&1){ //FIXME move out of this function?
             dest_y -= s->linesize*15;
-            dest_cb-= s->uvlinesize*7;
-            dest_cr-= s->uvlinesize*7;
+            dest_cb-= s->uvlinesize * (block_h - 1);
+            dest_cr-= s->uvlinesize * (block_h - 1);
         }
         if(FRAME_MBAFF) {
             int list;
@@ -1119,190 +1952,143 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     }
 
     if (!simple && IS_INTRA_PCM(mb_type)) {
-        for (i=0; i<16; i++) {
-            memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
-        }
-        for (i=0; i<8; i++) {
-            memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
-            memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
+        if (pixel_shift) {
+            const int bit_depth = h->sps.bit_depth_luma;
+            int j;
+            GetBitContext gb;
+            init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
+
+            for (i = 0; i < 16; i++) {
+                uint16_t *tmp_y  = (uint16_t*)(dest_y  + i*linesize);
+                for (j = 0; j < 16; j++)
+                    tmp_y[j] = get_bits(&gb, bit_depth);
+            }
+            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
+                if (!h->sps.chroma_format_idc) {
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
+                        for (j = 0; j < 8; j++) {
+                            tmp_cb[j] = 1 << (bit_depth - 1);
+                        }
+                    }
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
+                        for (j = 0; j < 8; j++) {
+                            tmp_cr[j] = 1 << (bit_depth - 1);
+                        }
+                    }
+                } else {
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
+                        for (j = 0; j < 8; j++)
+                            tmp_cb[j] = get_bits(&gb, bit_depth);
+                    }
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
+                        for (j = 0; j < 8; j++)
+                            tmp_cr[j] = get_bits(&gb, bit_depth);
+                    }
+                }
+            }
+        } else {
+            for (i=0; i<16; i++) {
+                memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
+            }
+            if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
+                if (!h->sps.chroma_format_idc) {
+                    for (i = 0; i < block_h; i++) {
+                        memset(dest_cb + i*uvlinesize, 128, 8);
+                        memset(dest_cr + i*uvlinesize, 128, 8);
+                    }
+                } else {
+                    for (i = 0; i < block_h; i++) {
+                        memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
+                        memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
+                    }
+                }
+            }
         }
     } else {
         if(IS_INTRA(mb_type)){
             if(h->deblocking_filter)
-                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, simple);
+                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
 
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
             }
 
-            if(IS_INTRA4x4(mb_type)){
-                if(simple || !s->encoding){
-                    if(IS_8x8DCT(mb_type)){
-                        if(transform_bypass){
-                            idct_dc_add =
-                            idct_add    = s->dsp.add_pixels8;
-                        }else{
-                            idct_dc_add = h->h264dsp.h264_idct8_dc_add;
-                            idct_add    = h->h264dsp.h264_idct8_add;
-                        }
-                        for(i=0; i<16; i+=4){
-                            uint8_t * const ptr= dest_y + block_offset[i];
-                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
-                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
-                                h->hpc.pred8x8l_add[dir](ptr, h->mb + i*16, linesize);
-                            }else{
-                                const int nnz = h->non_zero_count_cache[ scan8[i] ];
-                                h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
-                                                            (h->topright_samples_available<<i)&0x4000, linesize);
-                                if(nnz){
-                                    if(nnz == 1 && h->mb[i*16])
-                                        idct_dc_add(ptr, h->mb + i*16, linesize);
-                                    else
-                                        idct_add   (ptr, h->mb + i*16, linesize);
-                                }
-                            }
-                        }
-                    }else{
-                        if(transform_bypass){
-                            idct_dc_add =
-                            idct_add    = s->dsp.add_pixels4;
-                        }else{
-                            idct_dc_add = h->h264dsp.h264_idct_dc_add;
-                            idct_add    = h->h264dsp.h264_idct_add;
-                        }
-                        for(i=0; i<16; i++){
-                            uint8_t * const ptr= dest_y + block_offset[i];
-                            const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
-
-                            if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
-                                h->hpc.pred4x4_add[dir](ptr, h->mb + i*16, linesize);
-                            }else{
-                                uint8_t *topright;
-                                int nnz, tr;
-                                if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
-                                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
-                                    assert(mb_y || linesize <= block_offset[i]);
-                                    if(!topright_avail){
-                                        tr= ptr[3 - linesize]*0x01010101;
-                                        topright= (uint8_t*) &tr;
-                                    }else
-                                        topright= ptr + 4 - linesize;
-                                }else
-                                    topright= NULL;
-
-                                h->hpc.pred4x4[ dir ](ptr, topright, linesize);
-                                nnz = h->non_zero_count_cache[ scan8[i] ];
-                                if(nnz){
-                                    if(is_h264){
-                                        if(nnz == 1 && h->mb[i*16])
-                                            idct_dc_add(ptr, h->mb + i*16, linesize);
-                                        else
-                                            idct_add   (ptr, h->mb + i*16, linesize);
-                                    }else
-                                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
-                                }
-                            }
-                        }
-                    }
-                }
-            }else{
-                h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
-                if(is_h264){
-                    if(h->non_zero_count_cache[ scan8[LUMA_DC_BLOCK_INDEX] ]){
-                        if(!transform_bypass)
-                            h->h264dsp.h264_luma_dc_dequant_idct(h->mb, h->mb_luma_dc, h->dequant4_coeff[0][s->qscale][0]);
-                        else{
-                            static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
-                                                                    8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
-                            for(i = 0; i < 16; i++)
-                                h->mb[dc_mapping[i]] = h->mb_luma_dc[i];
-                        }
-                    }
-                }else
-                    ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
-            }
+            hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
+
             if(h->deblocking_filter)
-                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, simple);
+                xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
         }else if(is_h264){
-            hl_motion(h, dest_y, dest_cb, dest_cr,
-                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                      h->h264dsp.weight_h264_pixels_tab, h->h264dsp.biweight_h264_pixels_tab);
-        }
-
-
-        if(!IS_INTRA4x4(mb_type)){
-            if(is_h264){
-                if(IS_INTRA16x16(mb_type)){
-                    if(transform_bypass){
-                        if(h->sps.profile_idc==244 && (h->intra16x16_pred_mode==VERT_PRED8x8 || h->intra16x16_pred_mode==HOR_PRED8x8)){
-                            h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb, linesize);
-                        }else{
-                            for(i=0; i<16; i++){
-                                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
-                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + i*16, linesize);
-                            }
-                        }
-                    }else{
-                         h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
-                    }
-                }else if(h->cbp&15){
-                    if(transform_bypass){
-                        const int di = IS_8x8DCT(mb_type) ? 4 : 1;
-                        idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
-                        for(i=0; i<16; i+=di){
-                            if(h->non_zero_count_cache[ scan8[i] ]){
-                                idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
-                            }
-                        }
-                    }else{
-                        if(IS_8x8DCT(mb_type)){
-                            h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
-                        }else{
-                            h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb, linesize, h->non_zero_count_cache);
-                        }
-                    }
-                }
-            }else{
-                for(i=0; i<16; i++){
-                    if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
-                        uint8_t * const ptr= dest_y + block_offset[i];
-                        ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
-                    }
-                }
+            if (chroma422) {
+                hl_motion_422(h, dest_y, dest_cb, dest_cr,
+                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                              h->h264dsp.weight_h264_pixels_tab,
+                              h->h264dsp.biweight_h264_pixels_tab,
+                              pixel_shift);
+            } else {
+                hl_motion_420(h, dest_y, dest_cb, dest_cr,
+                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                              h->h264dsp.weight_h264_pixels_tab,
+                              h->h264dsp.biweight_h264_pixels_tab,
+                              pixel_shift);
             }
         }
 
+        hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
+
         if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
             uint8_t *dest[2] = {dest_cb, dest_cr};
             if(transform_bypass){
                 if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
-                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + 16*16, uvlinesize);
-                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + 20*16, uvlinesize);
+                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
+                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
                 }else{
                     idct_add = s->dsp.add_pixels4;
-                    for(i=16; i<16+8; i++){
-                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16])
-                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);
+                    for(j=1; j<3; j++){
+                        for(i=j*16; i<j*16+4; i++){
+                            if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
+                        }
+                        if (chroma422) {
+                            for(i=j*16+4; i<j*16+8; i++){
+                                if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                    idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
+                            }
+                        }
                     }
                 }
             }else{
                 if(is_h264){
+                    int qp[2];
+                    if (chroma422) {
+                        qp[0] = h->chroma_qp[0] + 3;
+                        qp[1] = h->chroma_qp[1] + 3;
+                    } else {
+                        qp[0] = h->chroma_qp[0];
+                        qp[1] = h->chroma_qp[1];
+                    }
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
-                        chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
-                        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
                     h->h264dsp.h264_idct_add8(dest, block_offset,
                                               h->mb, uvlinesize,
                                               h->non_zero_count_cache);
-                }else{
-                    chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
-                    chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
-                    for(i=16; i<16+8; i++){
-                        if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
-                            uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
-                            ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[s->qscale + 12] - 12, 2);
+                } else if (CONFIG_SVQ3_DECODER) {
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                    for(j=1; j<3; j++){
+                        for(i=j*16; i<j*16+4; i++){
+                            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
+                                uint8_t * const ptr= dest[j-1] + block_offset[i];
+                                ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
+                            }
                         }
                     }
                 }
@@ -1310,32 +2096,150 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
         }
     }
     if(h->cbp || IS_INTRA(mb_type))
+    {
+        s->dsp.clear_blocks(h->mb);
+        s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
+    }
+}
+
+static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
+    MpegEncContext * const s = &h->s;
+    const int mb_x= s->mb_x;
+    const int mb_y= s->mb_y;
+    const int mb_xy= h->mb_xy;
+    const int mb_type = s->current_picture.f.mb_type[mb_xy];
+    uint8_t  *dest[3];
+    int linesize;
+    int i, j, p;
+    int *block_offset = &h->block_offset[0];
+    const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
+    const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
+
+    for (p = 0; p < plane_count; p++)
+    {
+        dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
+        s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
+    }
+
+    h->list_counts[mb_xy]= h->list_count;
+
+    if (!simple && MB_FIELD) {
+        linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
+        block_offset = &h->block_offset[48];
+        if(mb_y&1) //FIXME move out of this function?
+            for (p = 0; p < 3; p++)
+                dest[p] -= s->linesize*15;
+        if(FRAME_MBAFF) {
+            int list;
+            for(list=0; list<h->list_count; list++){
+                if(!USES_LIST(mb_type, list))
+                    continue;
+                if(IS_16X16(mb_type)){
+                    int8_t *ref = &h->ref_cache[list][scan8[0]];
+                    fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
+                }else{
+                    for(i=0; i<16; i+=4){
+                        int ref = h->ref_cache[list][scan8[i]];
+                        if(ref >= 0)
+                            fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
+                    }
+                }
+            }
+        }
+    } else {
+        linesize   = h->mb_linesize = h->mb_uvlinesize = s->linesize;
+    }
+
+    if (!simple && IS_INTRA_PCM(mb_type)) {
+        if (pixel_shift) {
+            const int bit_depth = h->sps.bit_depth_luma;
+            GetBitContext gb;
+            init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
+
+            for (p = 0; p < plane_count; p++) {
+                for (i = 0; i < 16; i++) {
+                    uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
+                    for (j = 0; j < 16; j++)
+                        tmp[j] = get_bits(&gb, bit_depth);
+                }
+            }
+        } else {
+            for (p = 0; p < plane_count; p++) {
+                for (i = 0; i < 16; i++) {
+                    memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
+                }
+            }
+        }
+    } else {
+        if(IS_INTRA(mb_type)){
+            if(h->deblocking_filter)
+                xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
+
+            for (p = 0; p < plane_count; p++)
+                hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
+
+            if(h->deblocking_filter)
+                xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
+        }else{
+            hl_motion(h, dest[0], dest[1], dest[2],
+                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                      h->h264dsp.weight_h264_pixels_tab,
+                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
+        }
+
+        for (p = 0; p < plane_count; p++)
+            hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
+    }
+    if(h->cbp || IS_INTRA(mb_type))
+    {
         s->dsp.clear_blocks(h->mb);
+        s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
+    }
 }
 
 /**
  * Process a macroblock; this case avoids checks for expensive uncommon cases.
  */
-static void hl_decode_mb_simple(H264Context *h){
-    hl_decode_mb_internal(h, 1);
+#define hl_decode_mb_simple(sh, bits) \
+static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
+    hl_decode_mb_internal(h, 1, sh); \
 }
+hl_decode_mb_simple(0, 8)
+hl_decode_mb_simple(1, 16)
 
 /**
  * Process a macroblock; this handles edge cases, such as interlacing.
  */
 static void av_noinline hl_decode_mb_complex(H264Context *h){
-    hl_decode_mb_internal(h, 0);
+    hl_decode_mb_internal(h, 0, h->pixel_shift);
+}
+
+static void av_noinline hl_decode_mb_444_complex(H264Context *h){
+    hl_decode_mb_444_internal(h, 0, h->pixel_shift);
+}
+
+static void av_noinline hl_decode_mb_444_simple(H264Context *h){
+    hl_decode_mb_444_internal(h, 1, 0);
 }
 
 void ff_h264_hl_decode_mb(H264Context *h){
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
-    const int mb_type= s->current_picture.mb_type[mb_xy];
+    const int mb_type = s->current_picture.f.mb_type[mb_xy];
     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
 
-    if (is_complex)
+    if (CHROMA444) {
+        if(is_complex || h->pixel_shift)
+            hl_decode_mb_444_complex(h);
+        else
+            hl_decode_mb_444_simple(h);
+    } else if (is_complex) {
         hl_decode_mb_complex(h);
-    else hl_decode_mb_simple(h);
+    } else if (h->pixel_shift) {
+        hl_decode_mb_simple_16(h);
+    } else
+        hl_decode_mb_simple_8(h);
 }
 
 static int pred_weight_table(H264Context *h){
@@ -1346,7 +2250,7 @@ static int pred_weight_table(H264Context *h){
     h->use_weight= 0;
     h->use_weight_chroma= 0;
     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
-    if(CHROMA)
+    if(h->sps.chroma_format_idc)
         h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
     luma_def = 1<<h->luma_log2_weight_denom;
     chroma_def = 1<<h->chroma_log2_weight_denom;
@@ -1371,7 +2275,7 @@ static int pred_weight_table(H264Context *h){
                 h->luma_weight[i][list][1]= 0;
             }
 
-            if(CHROMA){
+            if(h->sps.chroma_format_idc){
                 chroma_weight_flag= get_bits1(&s->gb);
                 if(chroma_weight_flag){
                     int j;
@@ -1393,7 +2297,7 @@ static int pred_weight_table(H264Context *h){
                 }
             }
         }
-        if(h->slice_type_nos != FF_B_TYPE) break;
+        if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
     }
     h->use_weight= h->use_weight || h->use_weight_chroma;
     return 0;
@@ -1414,7 +2318,11 @@ static void implicit_weight_table(H264Context *h, int field){
     }
 
     if(field < 0){
-        cur_poc = s->current_picture_ptr->poc;
+        if (s->picture_structure == PICT_FRAME) {
+            cur_poc = s->current_picture_ptr->poc;
+        } else {
+            cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
+        }
     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
         h->use_weight= 0;
@@ -1439,15 +2347,17 @@ static void implicit_weight_table(H264Context *h, int field){
     for(ref0=ref_start; ref0 < ref_count0; ref0++){
         int poc0 = h->ref_list[0][ref0].poc;
         for(ref1=ref_start; ref1 < ref_count1; ref1++){
-            int poc1 = h->ref_list[1][ref1].poc;
-            int td = av_clip(poc1 - poc0, -128, 127);
-            int w= 32;
-            if(td){
-                int tb = av_clip(cur_poc - poc0, -128, 127);
-                int tx = (16384 + (FFABS(td) >> 1)) / td;
-                int dist_scale_factor = (tb*tx + 32) >> 8;
-                if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
-                    w = 64 - dist_scale_factor;
+            int w = 32;
+            if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
+                int poc1 = h->ref_list[1][ref1].poc;
+                int td = av_clip(poc1 - poc0, -128, 127);
+                if(td){
+                    int tb = av_clip(cur_poc - poc0, -128, 127);
+                    int tx = (16384 + (FFABS(td) >> 1)) / td;
+                    int dist_scale_factor = (tb*tx + 32) >> 8;
+                    if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
+                        w = 64 - dist_scale_factor;
+                }
             }
             if(field<0){
                 h->implicit_weight[ref0][ref1][0]=
@@ -1476,14 +2386,16 @@ static void flush_dpb(AVCodecContext *avctx){
     int i;
     for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
         if(h->delayed_pic[i])
-            h->delayed_pic[i]->reference= 0;
+            h->delayed_pic[i]->f.reference = 0;
         h->delayed_pic[i]= NULL;
     }
-    h->outputed_poc= INT_MIN;
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+        h->last_pocs[i] = INT_MIN;
+    h->outputed_poc=h->next_outputed_poc= INT_MIN;
     h->prev_interlaced_frame = 1;
     idr(h);
     if(h->s.current_picture_ptr)
-        h->s.current_picture_ptr->reference= 0;
+        h->s.current_picture_ptr->f.reference = 0;
     h->s.first_field= 0;
     ff_h264_reset_sei(h);
     ff_mpeg_flush(avctx);
@@ -1603,24 +2515,29 @@ static void init_scan_tables(H264Context *h){
     }
 }
 
-static void field_end(H264Context *h){
+static int field_end(H264Context *h, int in_setup){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
+    int err = 0;
     s->mb_y= 0;
 
-    s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;
-    s->current_picture_ptr->pict_type= s->pict_type;
+    if (!in_setup && !s->dropable)
+        ff_thread_report_progress((AVFrame*)s->current_picture_ptr, (16*s->mb_height >> FIELD_PICTURE) - 1,
+                                 s->picture_structure==PICT_BOTTOM_FIELD);
 
     if (CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
         ff_vdpau_h264_set_reference_frames(s);
 
-    if(!s->dropable) {
-        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
-        h->prev_poc_msb= h->poc_msb;
-        h->prev_poc_lsb= h->poc_lsb;
+    if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
+        if(!s->dropable) {
+            err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+            h->prev_poc_msb= h->poc_msb;
+            h->prev_poc_lsb= h->poc_lsb;
+        }
+        h->prev_frame_num_offset= h->frame_num_offset;
+        h->prev_frame_num= h->frame_num;
+        h->outputed_poc = h->next_outputed_poc;
     }
-    h->prev_frame_num_offset= h->frame_num_offset;
-    h->prev_frame_num= h->frame_num;
 
     if (avctx->hwaccel) {
         if (avctx->hwaccel->end_frame(avctx) < 0)
@@ -1648,6 +2565,8 @@ static void field_end(H264Context *h){
     MPV_frame_end(s);
 
     h->current_slice=0;
+
+    return err;
 }
 
 /**
@@ -1673,12 +2592,39 @@ static void clone_slice(H264Context *dst, H264Context *src)
     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
 
-    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
-    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
+    memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
+    memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
+}
+
+/**
+ * Compute profile from profile_idc and constraint_set?_flags.
+ *
+ * @param sps SPS
+ *
+ * @return profile as defined by FF_PROFILE_H264_*
+ */
+int ff_h264_get_profile(SPS *sps)
+{
+    int profile = sps->profile_idc;
+
+    switch(sps->profile_idc) {
+    case FF_PROFILE_H264_BASELINE:
+        // constraint_set1_flag set to 1
+        profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
+        break;
+    case FF_PROFILE_H264_HIGH_10:
+    case FF_PROFILE_H264_HIGH_422:
+    case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
+        // constraint_set3_flag set to 1
+        profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
+        break;
+    }
+
+    return profile;
 }
 
 /**
- * decodes a slice header.
+ * Decode a slice header.
  * This will also call MPV_common_init() and frame_start() as needed.
  *
  * @param h h264context
@@ -1698,7 +2644,8 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
     s->dropable= h->nal_ref_idc == 0;
 
-    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc){
+    /* FIXME: 2tap qpel isn't implemented for high bit depth. */
+    if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;
         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;
     }else{
@@ -1710,7 +2657,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
     if(first_mb_in_slice == 0){ //FIXME better field boundary detection
         if(h0->current_slice && FIELD_PICTURE){
-            field_end(h);
+            field_end(h, 1);
         }
 
         h0->current_slice = 0;
@@ -1730,7 +2677,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         h->slice_type_fixed=0;
 
     slice_type= golomb_to_pict_type[ slice_type ];
-    if (slice_type == FF_I_TYPE
+    if (slice_type == AV_PICTURE_TYPE_I
         || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
         default_ref_list_done = 1;
     }
@@ -1756,38 +2703,39 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     }
     h->sps = *h0->sps_buffers[h->pps.sps_id];
 
-    s->avctx->profile = h->sps.profile_idc;
+    s->avctx->profile = ff_h264_get_profile(&h->sps);
     s->avctx->level   = h->sps.level_idc;
     s->avctx->refs    = h->sps.ref_frame_count;
 
-    if(h == h0 && h->dequant_coeff_pps != pps_id){
-        h->dequant_coeff_pps = pps_id;
-        init_dequant_tables(h);
-    }
-
     s->mb_width= h->sps.mb_width;
     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
 
     h->b_stride=  s->mb_width*4;
 
-    s->width = 16*s->mb_width - 2*FFMIN(h->sps.crop_right, 7);
+    s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
+
+    s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
     if(h->sps.frame_mbs_only_flag)
-        s->height= 16*s->mb_height - 2*FFMIN(h->sps.crop_bottom, 7);
+        s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
     else
-        s->height= 16*s->mb_height - 4*FFMIN(h->sps.crop_bottom, 7);
+        s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
 
     if (s->context_initialized
         && (   s->width != s->avctx->width || s->height != s->avctx->height
             || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
-        if(h != h0)
+        if(h != h0) {
+            av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
             return -1;   // width / height changed during parallelized decoding
+        }
         free_tables(h, 0);
         flush_dpb(s->avctx);
         MPV_common_end(s);
     }
     if (!s->context_initialized) {
-        if(h != h0)
-            return -1;  // we cant (re-)initialize context during parallel decoding
+        if (h != h0) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
+            return -1;
+        }
 
         avcodec_set_dimensions(s->avctx, s->width, s->height);
         s->avctx->sample_aspect_ratio= h->sps.sar;
@@ -1809,37 +2757,98 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
                       h->sps.num_units_in_tick, den, 1<<30);
         }
-        s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
-                                                 s->avctx->codec->pix_fmts ?
-                                                 s->avctx->codec->pix_fmts :
-                                                 s->avctx->color_range == AVCOL_RANGE_JPEG ?
-                                                 hwaccel_pixfmt_list_h264_jpeg_420 :
-                                                 ff_hwaccel_pixfmt_list_420);
+
+        switch (h->sps.bit_depth_luma) {
+            case 9 :
+                if (CHROMA444) {
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP9;
+                    } else
+                        s->avctx->pix_fmt = PIX_FMT_YUV444P9;
+                } else if (CHROMA422)
+                    s->avctx->pix_fmt = PIX_FMT_YUV422P9;
+                else
+                    s->avctx->pix_fmt = PIX_FMT_YUV420P9;
+                break;
+            case 10 :
+                if (CHROMA444) {
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP10;
+                    } else
+                        s->avctx->pix_fmt = PIX_FMT_YUV444P10;
+                } else if (CHROMA422)
+                    s->avctx->pix_fmt = PIX_FMT_YUV422P10;
+                else
+                    s->avctx->pix_fmt = PIX_FMT_YUV420P10;
+                break;
+            case 8:
+                if (CHROMA444){
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP;
+                    } else
+                        s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
+                } else if (CHROMA422) {
+                    s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
+                }else{
+                    s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
+                                                             s->avctx->codec->pix_fmts ?
+                                                             s->avctx->codec->pix_fmts :
+                                                             s->avctx->color_range == AVCOL_RANGE_JPEG ?
+                                                             hwaccel_pixfmt_list_h264_jpeg_420 :
+                                                             ff_hwaccel_pixfmt_list_420);
+                }
+                break;
+            default:
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
+                return AVERROR_INVALIDDATA;
+        }
+
         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
 
-        if (MPV_common_init(s) < 0)
+        if (MPV_common_init(s) < 0) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
             return -1;
+        }
         s->first_field = 0;
         h->prev_interlaced_frame = 1;
 
         init_scan_tables(h);
-        ff_h264_alloc_tables(h);
-
-        for(i = 1; i < s->avctx->thread_count; i++) {
-            H264Context *c;
-            c = h->thread_context[i] = av_malloc(sizeof(H264Context));
-            memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
-            memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
-            c->h264dsp = h->h264dsp;
-            c->sps = h->sps;
-            c->pps = h->pps;
-            init_scan_tables(c);
-            clone_tables(c, h, i);
-        }
-
-        for(i = 0; i < s->avctx->thread_count; i++)
-            if(context_init(h->thread_context[i]) < 0)
+        if (ff_h264_alloc_tables(h) < 0) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
+            return AVERROR(ENOMEM);
+        }
+
+        if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
+            if (context_init(h) < 0) {
+                av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
                 return -1;
+            }
+        } else {
+            for(i = 1; i < s->slice_context_count; i++) {
+                H264Context *c;
+                c = h->thread_context[i] = av_malloc(sizeof(H264Context));
+                memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
+                memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
+                c->h264dsp = h->h264dsp;
+                c->sps = h->sps;
+                c->pps = h->pps;
+                c->pixel_shift = h->pixel_shift;
+                init_scan_tables(c);
+                clone_tables(c, h, i);
+            }
+
+            for(i = 0; i < s->slice_context_count; i++)
+                if (context_init(h->thread_context[i]) < 0) {
+                    av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
+                    return -1;
+                }
+        }
+    }
+
+    if(h == h0 && h->dequant_coeff_pps != pps_id){
+        h->dequant_coeff_pps = pps_id;
+        init_dequant_tables(h);
     }
 
     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
@@ -1860,6 +2869,21 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->mb_field_decoding_flag= s->picture_structure != PICT_FRAME;
 
     if(h0->current_slice == 0){
+        // Shorten frame num gaps so we don't have to allocate reference frames just to throw them away
+        if(h->frame_num != h->prev_frame_num) {
+            int unwrap_prev_frame_num = h->prev_frame_num, max_frame_num = 1<<h->sps.log2_max_frame_num;
+
+            if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
+
+            if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
+                unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
+                if (unwrap_prev_frame_num < 0)
+                    unwrap_prev_frame_num += max_frame_num;
+
+                h->prev_frame_num = unwrap_prev_frame_num;
+            }
+        }
+
         while(h->frame_num !=  h->prev_frame_num &&
               h->frame_num != (h->prev_frame_num+1)%(1<<h->sps.log2_max_frame_num)){
             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
@@ -1869,18 +2893,22 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             h->prev_frame_num++;
             h->prev_frame_num %= 1<<h->sps.log2_max_frame_num;
             s->current_picture_ptr->frame_num= h->prev_frame_num;
+            ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
+            ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
             ff_generate_sliding_window_mmcos(h);
-            ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+            if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
+                (s->avctx->err_recognition & AV_EF_EXPLODE))
+                return AVERROR_INVALIDDATA;
             /* Error concealment: if a ref is missing, copy the previous ref in its place.
              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
              * about there being no actual duplicates.
              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
-             * concealing a lost frame, this probably isn't noticable by comparison, but it should
+             * concealing a lost frame, this probably isn't noticeable by comparison, but it should
              * be fixed. */
             if (h->short_ref_count) {
                 if (prev) {
-                    av_image_copy(h->short_ref[0]->data, h->short_ref[0]->linesize,
-                                  (const uint8_t**)prev->data, prev->linesize,
+                    av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
+                                  (const uint8_t**)prev->f.data, prev->f.linesize,
                                   s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
                     h->short_ref[0]->poc = prev->poc+2;
                 }
@@ -1891,8 +2919,8 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         /* See if we have a decoded first field looking for a pair... */
         if (s0->first_field) {
             assert(s0->current_picture_ptr);
-            assert(s0->current_picture_ptr->data[0]);
-            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
+            assert(s0->current_picture_ptr->f.data[0]);
+            assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
 
             /* figure out if we have a complementary field pair */
             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
@@ -1905,7 +2933,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
             } else {
                 if (h->nal_ref_idc &&
-                        s0->current_picture_ptr->reference &&
+                        s0->current_picture_ptr->f.reference &&
                         s0->current_picture_ptr->frame_num != h->frame_num) {
                     /*
                      * This and previous field were reference, but had
@@ -1928,9 +2956,13 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             s0->first_field = FIELD_PICTURE;
         }
 
-        if((!FIELD_PICTURE || s0->first_field) && ff_h264_frame_start(h) < 0) {
-            s0->first_field = 0;
-            return -1;
+        if(!FIELD_PICTURE || s0->first_field) {
+            if (ff_h264_frame_start(h) < 0) {
+                s0->first_field = 0;
+                return -1;
+            }
+        } else {
+            ff_release_unused_pictures(s, 0);
         }
     }
     if(h != h0)
@@ -1987,24 +3019,27 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->ref_count[0]= h->pps.ref_count[0];
     h->ref_count[1]= h->pps.ref_count[1];
 
-    if(h->slice_type_nos != FF_I_TYPE){
-        if(h->slice_type_nos == FF_B_TYPE){
+    if(h->slice_type_nos != AV_PICTURE_TYPE_I){
+        int max_refs = s->picture_structure == PICT_FRAME ? 16 : 32;
+
+        if(h->slice_type_nos == AV_PICTURE_TYPE_B){
             h->direct_spatial_mv_pred= get_bits1(&s->gb);
         }
         num_ref_idx_active_override_flag= get_bits1(&s->gb);
 
         if(num_ref_idx_active_override_flag){
             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
-            if(h->slice_type_nos==FF_B_TYPE)
+            if(h->slice_type_nos==AV_PICTURE_TYPE_B)
                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
+        }
 
-            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
-                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
-                h->ref_count[0]= h->ref_count[1]= 1;
-                return -1;
-            }
+        if (h->ref_count[0] > max_refs || h->ref_count[1] > max_refs) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
+            h->ref_count[0] = h->ref_count[1] = 1;
+            return AVERROR_INVALIDDATA;
         }
-        if(h->slice_type_nos == FF_B_TYPE)
+
+        if(h->slice_type_nos == AV_PICTURE_TYPE_B)
             h->list_count= 2;
         else
             h->list_count= 1;
@@ -2015,22 +3050,24 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         ff_h264_fill_default_ref_list(h);
     }
 
-    if(h->slice_type_nos!=FF_I_TYPE && ff_h264_decode_ref_pic_list_reordering(h) < 0)
+    if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
+        h->ref_count[1]= h->ref_count[0]= 0;
         return -1;
+    }
 
-    if(h->slice_type_nos!=FF_I_TYPE){
+    if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
         s->last_picture_ptr= &h->ref_list[0][0];
         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
     }
-    if(h->slice_type_nos==FF_B_TYPE){
+    if(h->slice_type_nos==AV_PICTURE_TYPE_B){
         s->next_picture_ptr= &h->ref_list[1][0];
         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
     }
 
-    if(   (h->pps.weighted_pred          && h->slice_type_nos == FF_P_TYPE )
-       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== FF_B_TYPE ) )
+    if(   (h->pps.weighted_pred          && h->slice_type_nos == AV_PICTURE_TYPE_P )
+       ||  (h->pps.weighted_bipred_idc==1 && h->slice_type_nos== AV_PICTURE_TYPE_B ) )
         pred_weight_table(h);
-    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
+    else if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
         implicit_weight_table(h, -1);
     }else {
         h->use_weight = 0;
@@ -2040,23 +3077,24 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         }
     }
 
-    if(h->nal_ref_idc)
-        ff_h264_decode_ref_pic_marking(h0, &s->gb);
+    if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
+       (s->avctx->err_recognition & AV_EF_EXPLODE))
+        return AVERROR_INVALIDDATA;
 
     if(FRAME_MBAFF){
         ff_h264_fill_mbaff_ref_list(h);
 
-        if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== FF_B_TYPE){
+        if(h->pps.weighted_bipred_idc==2 && h->slice_type_nos== AV_PICTURE_TYPE_B){
             implicit_weight_table(h, 0);
             implicit_weight_table(h, 1);
         }
     }
 
-    if(h->slice_type_nos==FF_B_TYPE && !h->direct_spatial_mv_pred)
+    if(h->slice_type_nos==AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
         ff_h264_direct_dist_scale_factor(h);
     ff_h264_direct_ref_list_init(h);
 
-    if( h->slice_type_nos != FF_I_TYPE && h->pps.cabac ){
+    if( h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac ){
         tmp = get_ue_golomb_31(&s->gb);
         if(tmp > 2){
             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
@@ -2067,7 +3105,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
     h->last_qscale_diff = 0;
     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
-    if(tmp>51){
+    if(tmp>51+6*(h->sps.bit_depth_luma-8)){
         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
         return -1;
     }
@@ -2075,10 +3113,10 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
     //FIXME qscale / qp ... stuff
-    if(h->slice_type == FF_SP_TYPE){
+    if(h->slice_type == AV_PICTURE_TYPE_SP){
         get_bits1(&s->gb); /* sp_for_switch_flag */
     }
-    if(h->slice_type==FF_SP_TYPE || h->slice_type == FF_SI_TYPE){
+    if(h->slice_type==AV_PICTURE_TYPE_SP || h->slice_type == AV_PICTURE_TYPE_SI){
         get_se_golomb(&s->gb); /* slice_qs_delta */
     }
 
@@ -2107,8 +3145,8 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     }
 
     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL
-       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != FF_I_TYPE)
-       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == FF_B_TYPE)
+       ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type_nos != AV_PICTURE_TYPE_I)
+       ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type_nos == AV_PICTURE_TYPE_B)
        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
         h->deblocking_filter= 0;
 
@@ -2123,11 +3161,15 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                 av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
                 h0->single_decode_warning = 1;
             }
-            if(h != h0)
-                return 1; // deblocking switched inside frame
+            if (h != h0) {
+                av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
+                return 1;
+            }
         }
     }
-    h->qp_thresh= 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]);
+    h->qp_thresh = 15 + 52 - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset)
+                 - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1])
+                 + 6 * (h->sps.bit_depth_luma - 8);
 
 #if 0 //FMO
     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
@@ -2145,16 +3187,16 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
         for(i=0; i<16; i++){
             id_list[i]= 60;
-            if(h->ref_list[j][i].data[0]){
+            if (h->ref_list[j][i].f.data[0]) {
                 int k;
-                uint8_t *base= h->ref_list[j][i].base[0];
+                uint8_t *base = h->ref_list[j][i].f.base[0];
                 for(k=0; k<h->short_ref_count; k++)
-                    if(h->short_ref[k]->base[0] == base){
+                    if (h->short_ref[k]->f.base[0] == base) {
                         id_list[i]= k;
                         break;
                     }
                 for(k=0; k<h->long_ref_count; k++)
-                    if(h->long_ref[k] && h->long_ref[k]->base[0] == base){
+                    if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
                         id_list[i]= h->short_ref_count + k;
                         break;
                     }
@@ -2165,15 +3207,16 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         ref2frm[1]= -1;
         for(i=0; i<16; i++)
             ref2frm[i+2]= 4*id_list[i]
-                          +(h->ref_list[j][i].reference&3);
+                          + (h->ref_list[j][i].f.reference & 3);
         ref2frm[18+0]=
         ref2frm[18+1]= -1;
         for(i=16; i<48; i++)
             ref2frm[i+4]= 4*id_list[(i-16)>>1]
-                          +(h->ref_list[j][i].reference&3);
+                          + (h->ref_list[j][i].f.reference & 3);
     }
 
-    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
+    //FIXME: fix draw_edges+PAFF+frame threads
+    h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE || (!h->sps.frame_mbs_only_flag && s->avctx->active_thread_type)) ? 0 : 16;
     h->emu_edge_height= (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
 
     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
@@ -2181,7 +3224,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                h->slice_num,
                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
                first_mb_in_slice,
-               av_get_pict_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
+               av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
                pps_id, h->frame_num,
                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
                h->ref_count[0], h->ref_count[1],
@@ -2189,7 +3232,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                h->deblocking_filter, h->slice_alpha_c0_offset/2-26, h->slice_beta_offset/2-26,
                h->use_weight,
                h->use_weight==1 && h->use_weight_chroma ? "c" : "",
-               h->slice_type == FF_B_TYPE ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
+               h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
                );
     }
 
@@ -2199,245 +3242,238 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 int ff_h264_get_slice_type(const H264Context *h)
 {
     switch (h->slice_type) {
-    case FF_P_TYPE:  return 0;
-    case FF_B_TYPE:  return 1;
-    case FF_I_TYPE:  return 2;
-    case FF_SP_TYPE: return 3;
-    case FF_SI_TYPE: return 4;
+    case AV_PICTURE_TYPE_P:  return 0;
+    case AV_PICTURE_TYPE_B:  return 1;
+    case AV_PICTURE_TYPE_I:  return 2;
+    case AV_PICTURE_TYPE_SP: return 3;
+    case AV_PICTURE_TYPE_SI: return 4;
     default:         return -1;
     }
 }
 
+static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
+                                                      int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
+{
+    int b_stride = h->b_stride;
+    int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
+    int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
+    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
+        if(USES_LIST(top_type, list)){
+            const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
+            const int b8_xy= 4*top_xy + 2;
+            int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+            AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
+            ref_cache[0 - 1*8]=
+            ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
+            ref_cache[2 - 1*8]=
+            ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
+        }else{
+            AV_ZERO128(mv_dst - 1*8);
+            AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+        }
+
+        if(!IS_INTERLACED(mb_type^left_type[LTOP])){
+            if(USES_LIST(left_type[LTOP], list)){
+                const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
+                const int b8_xy= 4*left_xy[LTOP] + 1;
+                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+                AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
+                AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
+                AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
+                AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
+                ref_cache[-1 +  0]=
+                ref_cache[-1 +  8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
+                ref_cache[-1 + 16]=
+                ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
+            }else{
+                AV_ZERO32(mv_dst - 1 + 0);
+                AV_ZERO32(mv_dst - 1 + 8);
+                AV_ZERO32(mv_dst - 1 +16);
+                AV_ZERO32(mv_dst - 1 +24);
+                ref_cache[-1 +  0]=
+                ref_cache[-1 +  8]=
+                ref_cache[-1 + 16]=
+                ref_cache[-1 + 24]= LIST_NOT_USED;
+            }
+        }
+    }
+
+    if(!USES_LIST(mb_type, list)){
+        fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
+        AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+        AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+        AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+        AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
+        return;
+    }
+
+    {
+        int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
+        int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
+        uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
+        uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
+        AV_WN32A(&ref_cache[0*8], ref01);
+        AV_WN32A(&ref_cache[1*8], ref01);
+        AV_WN32A(&ref_cache[2*8], ref23);
+        AV_WN32A(&ref_cache[3*8], ref23);
+    }
+
+    {
+        int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
+        AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
+        AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
+        AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
+        AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
+    }
+}
+
 /**
  *
- * @return non zero if the loop filter can be skiped
+ * @return non zero if the loop filter can be skipped
  */
 static int fill_filter_caches(H264Context *h, int mb_type){
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
-    int top_xy, left_xy[2];
-    int top_type, left_type[2];
+    int top_xy, left_xy[LEFT_MBS];
+    int top_type, left_type[LEFT_MBS];
+    uint8_t *nnz;
+    uint8_t *nnz_cache;
 
     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
 
-    //FIXME deblocking could skip the intra and nnz parts.
-
     /* Wow, what a mess, why didn't they simplify the interlacing & intra
      * stuff, I can't imagine that these complex rules are worth it. */
 
-    left_xy[1] = left_xy[0] = mb_xy-1;
+    left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
     if(FRAME_MBAFF){
-        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
+        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
         if(s->mb_y&1){
             if (left_mb_field_flag != curr_mb_field_flag) {
-                left_xy[0] -= s->mb_stride;
+                left_xy[LTOP] -= s->mb_stride;
             }
         }else{
             if(curr_mb_field_flag){
-                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
+                top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
             }
             if (left_mb_field_flag != curr_mb_field_flag) {
-                left_xy[1] += s->mb_stride;
+                left_xy[LBOT] += s->mb_stride;
             }
         }
     }
 
     h->top_mb_xy = top_xy;
-    h->left_mb_xy[0] = left_xy[0];
-    h->left_mb_xy[1] = left_xy[1];
+    h->left_mb_xy[LTOP] = left_xy[LTOP];
+    h->left_mb_xy[LBOT] = left_xy[LBOT];
     {
         //for sufficiently low qp, filtering wouldn't do anything
         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
-        int qp = s->current_picture.qscale_table[mb_xy];
+        int qp = s->current_picture.f.qscale_table[mb_xy];
         if(qp <= qp_thresh
-           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
-           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
+           && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
+           && (top_xy        < 0 || ((qp + s->current_picture.f.qscale_table[top_xy       ] + 1) >> 1) <= qp_thresh)) {
             if(!FRAME_MBAFF)
                 return 1;
-            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
-               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
+            if ((left_xy[LTOP] < 0            || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]        ] + 1) >> 1) <= qp_thresh) &&
+                (top_xy        < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
                 return 1;
         }
     }
 
-    top_type     = s->current_picture.mb_type[top_xy]    ;
-    left_type[0] = s->current_picture.mb_type[left_xy[0]];
-    left_type[1] = s->current_picture.mb_type[left_xy[1]];
+    top_type        = s->current_picture.f.mb_type[top_xy];
+    left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
+    left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
     if(h->deblocking_filter == 2){
-        if(h->slice_table[top_xy     ] != h->slice_num) top_type= 0;
-        if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
+        if(h->slice_table[top_xy       ] != h->slice_num) top_type= 0;
+        if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
     }else{
-        if(h->slice_table[top_xy     ] == 0xFFFF) top_type= 0;
-        if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
+        if(h->slice_table[top_xy       ] == 0xFFFF) top_type= 0;
+        if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
     }
-    h->top_type    = top_type    ;
-    h->left_type[0]= left_type[0];
-    h->left_type[1]= left_type[1];
+    h->top_type       = top_type;
+    h->left_type[LTOP]= left_type[LTOP];
+    h->left_type[LBOT]= left_type[LBOT];
 
     if(IS_INTRA(mb_type))
         return 0;
 
-    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
-    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
-    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
-    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
-    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
+    fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
+    if(h->list_count == 2)
+        fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
 
+    nnz = h->non_zero_count[mb_xy];
+    nnz_cache = h->non_zero_count_cache;
+    AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
+    AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
+    AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
+    AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
     h->cbp= h->cbp_table[mb_xy];
 
-    {
-        int list;
-        for(list=0; list<h->list_count; list++){
-            int8_t *ref;
-            int y, b_stride;
-            int16_t (*mv_dst)[2];
-            int16_t (*mv_src)[2];
-
-            if(!USES_LIST(mb_type, list)){
-                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
-                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
-                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
-                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
-                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
-                continue;
-            }
-
-            ref = &s->current_picture.ref_index[list][4*mb_xy];
-            {
-                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
-                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
-                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
-                ref += 2;
-                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
-                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
-            }
-
-            b_stride = h->b_stride;
-            mv_dst   = &h->mv_cache[list][scan8[0]];
-            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
-            for(y=0; y<4; y++){
-                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
-            }
-
-        }
-    }
-
-
-/*
-0 . T T. T T T T
-1 L . .L . . . .
-2 L . .L . . . .
-3 . T TL . . . .
-4 L . .L . . . .
-5 L . .. . . . .
-*/
-//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
     if(top_type){
-        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
+        nnz = h->non_zero_count[top_xy];
+        AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
     }
 
-    if(left_type[0]){
-        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
-        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
-        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
-        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
+    if(left_type[LTOP]){
+        nnz = h->non_zero_count[left_xy[LTOP]];
+        nnz_cache[3+8*1]= nnz[3+0*4];
+        nnz_cache[3+8*2]= nnz[3+1*4];
+        nnz_cache[3+8*3]= nnz[3+2*4];
+        nnz_cache[3+8*4]= nnz[3+3*4];
     }
 
     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
     if(!CABAC && h->pps.transform_8x8_mode){
         if(IS_8x8DCT(top_type)){
-            h->non_zero_count_cache[4+8*0]=
-            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
-            h->non_zero_count_cache[6+8*0]=
-            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
+            nnz_cache[4+8*0]=
+            nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
+            nnz_cache[6+8*0]=
+            nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
         }
-        if(IS_8x8DCT(left_type[0])){
-            h->non_zero_count_cache[3+8*1]=
-            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
+        if(IS_8x8DCT(left_type[LTOP])){
+            nnz_cache[3+8*1]=
+            nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
         }
-        if(IS_8x8DCT(left_type[1])){
-            h->non_zero_count_cache[3+8*3]=
-            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
+        if(IS_8x8DCT(left_type[LBOT])){
+            nnz_cache[3+8*3]=
+            nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
         }
 
         if(IS_8x8DCT(mb_type)){
-            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
-            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
+            nnz_cache[scan8[0   ]]= nnz_cache[scan8[1   ]]=
+            nnz_cache[scan8[2   ]]= nnz_cache[scan8[3   ]]= (h->cbp & 0x1000) >> 12;
 
-            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
-            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
+            nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
+            nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
 
-            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
-            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
+            nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
+            nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
 
-            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
-            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
-        }
-    }
-
-    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
-        int list;
-        for(list=0; list<h->list_count; list++){
-            if(USES_LIST(top_type, list)){
-                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
-                const int b8_xy= 4*top_xy + 2;
-                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
-                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
-                h->ref_cache[list][scan8[0] + 0 - 1*8]=
-                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
-                h->ref_cache[list][scan8[0] + 2 - 1*8]=
-                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
-            }else{
-                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
-                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
-            }
-
-            if(!IS_INTERLACED(mb_type^left_type[0])){
-                if(USES_LIST(left_type[0], list)){
-                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
-                    const int b8_xy= 4*left_xy[0] + 1;
-                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
-                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
-                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
-                    h->ref_cache[list][scan8[0] - 1 +16 ]=
-                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
-                }else{
-                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
-                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
-                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
-                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
-                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
-                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
-                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
-                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
-                }
-            }
+            nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
+            nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
         }
     }
 
     return 0;
 }
 
-static void loop_filter(H264Context *h){
+static void loop_filter(H264Context *h, int start_x, int end_x){
     MpegEncContext * const s = &h->s;
     uint8_t  *dest_y, *dest_cb, *dest_cr;
     int linesize, uvlinesize, mb_x, mb_y;
     const int end_mb_y= s->mb_y + FRAME_MBAFF;
     const int old_slice_type= h->slice_type;
+    const int pixel_shift = h->pixel_shift;
+    const int block_h = 16 >> s->chroma_y_shift;
 
     if(h->deblocking_filter) {
-        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
+        for(mb_x= start_x; mb_x<end_x; mb_x++){
             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
                 int mb_xy, mb_type;
                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
                 h->slice_num= h->slice_table[mb_xy];
-                mb_type= s->current_picture.mb_type[mb_xy];
+                mb_type = s->current_picture.f.mb_type[mb_xy];
                 h->list_count= h->list_counts[mb_xy];
 
                 if(FRAME_MBAFF)
@@ -2445,9 +3481,9 @@ static void loop_filter(H264Context *h){
 
                 s->mb_x= mb_x;
                 s->mb_y= mb_y;
-                dest_y  = s->current_picture.data[0] + (mb_x + mb_y * s->linesize  ) * 16;
-                dest_cb = s->current_picture.data[1] + (mb_x + mb_y * s->uvlinesize) * 8;
-                dest_cr = s->current_picture.data[2] + (mb_x + mb_y * s->uvlinesize) * 8;
+                dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+                dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
+                dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
                     //FIXME simplify above
 
                 if (MB_FIELD) {
@@ -2455,8 +3491,8 @@ static void loop_filter(H264Context *h){
                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
                     if(mb_y&1){ //FIXME move out of this function?
                         dest_y -= s->linesize*15;
-                        dest_cb-= s->uvlinesize*7;
-                        dest_cr-= s->uvlinesize*7;
+                        dest_cb-= s->uvlinesize * (block_h - 1);
+                        dest_cr-= s->uvlinesize * (block_h - 1);
                     }
                 } else {
                     linesize   = h->mb_linesize   = s->linesize;
@@ -2465,8 +3501,8 @@ static void loop_filter(H264Context *h){
                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
                 if(fill_filter_caches(h, mb_type))
                     continue;
-                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
-                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
+                h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
+                h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
 
                 if (FRAME_MBAFF) {
                     ff_h264_filter_mb     (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
@@ -2477,7 +3513,7 @@ static void loop_filter(H264Context *h){
         }
     }
     h->slice_type= old_slice_type;
-    s->mb_x= 0;
+    s->mb_x= end_x;
     s->mb_y= end_mb_y - FRAME_MBAFF;
     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
@@ -2487,17 +3523,52 @@ static void predict_field_decoding_flag(H264Context *h){
     MpegEncContext * const s = &h->s;
     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
-                ? s->current_picture.mb_type[mb_xy-1]
+                ? s->current_picture.f.mb_type[mb_xy - 1]
                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
-                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
+                ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
                 : 0;
     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
 }
 
+/**
+ * Draw edges and report progress for the last MB row.
+ */
+static void decode_finish_row(H264Context *h){
+    MpegEncContext * const s = &h->s;
+    int top = 16*(s->mb_y >> FIELD_PICTURE);
+    int height = 16 << FRAME_MBAFF;
+    int deblock_border = (16 + 4) << FRAME_MBAFF;
+    int pic_height = 16*s->mb_height >> FIELD_PICTURE;
+
+    if (h->deblocking_filter) {
+        if((top + height) >= pic_height)
+            height += deblock_border;
+
+        top -= deblock_border;
+    }
+
+    if (top >= pic_height || (top + height) < h->emu_edge_height)
+        return;
+
+    height = FFMIN(height, pic_height - top);
+    if (top < h->emu_edge_height) {
+        height = top+height;
+        top = 0;
+    }
+
+    ff_draw_horiz_band(s, top, height);
+
+    if (s->dropable) return;
+
+    ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
+                             s->picture_structure==PICT_BOTTOM_FIELD);
+}
+
 static int decode_slice(struct AVCodecContext *avctx, void *arg){
     H264Context *h = *(void**)arg;
     MpegEncContext * const s = &h->s;
-    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+    const int part_mask= s->partitioned_frame ? (ER_AC_END|ER_AC_ERROR) : 0x7F;
+    int lf_x_start = s->mb_x;
 
     s->mb_skip_run= -1;
 
@@ -2535,19 +3606,20 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             eos = get_cabac_terminate( &h->cabac );
 
             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
+                if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
                 return 0;
             }
             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
                 return -1;
             }
 
             if( ++s->mb_x >= s->mb_width ) {
-                s->mb_x = 0;
-                loop_filter(h);
-                ff_draw_horiz_band(s, 16*s->mb_y, 16);
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
+                decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
                     ++s->mb_y;
@@ -2558,7 +3630,8 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if( eos || s->mb_y >= s->mb_height ) {
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
+                if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
                 return 0;
             }
         }
@@ -2579,15 +3652,14 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if(ret<0){
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
                 return -1;
             }
 
             if(++s->mb_x >= s->mb_width){
-                s->mb_x=0;
-                loop_filter(h);
-                ff_draw_horiz_band(s, 16*s->mb_y, 16);
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
+                decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
                     ++s->mb_y;
@@ -2597,79 +3669,33 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
                 if(s->mb_y >= s->mb_height){
                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
 
-                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                    if (get_bits_left(&s->gb) == 0) {
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
 
                         return 0;
                     }else{
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_END&part_mask);
 
                         return -1;
                     }
                 }
             }
 
-            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
+            if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0){
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
-                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                    return 0;
-                }else{
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
-                    return -1;
-                }
-            }
-        }
-    }
-
-#if 0
-    for(;s->mb_y < s->mb_height; s->mb_y++){
-        for(;s->mb_x < s->mb_width; s->mb_x++){
-            int ret= decode_mb(h);
-
-            ff_h264_hl_decode_mb(h);
-
-            if(ret<0){
-                av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
-                return -1;
-            }
-
-            if(++s->mb_x >= s->mb_width){
-                s->mb_x=0;
-                if(++s->mb_y >= s->mb_height){
-                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                        return 0;
-                    }else{
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
-
-                        return -1;
-                    }
-                }
-            }
-
-            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
-                if(get_bits_count(s->gb) == s->gb.size_in_bits){
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (get_bits_left(&s->gb) == 0) {
+                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
+                    if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
 
                     return 0;
                 }else{
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
+                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
 
                     return -1;
                 }
             }
         }
-        s->mb_x=0;
-        ff_draw_horiz_band(s, 16*s->mb_y, 16);
     }
-#endif
-    return -1; //not reached
 }
 
 /**
@@ -2678,26 +3704,24 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
  * @param h h264 master context
  * @param context_count number of contexts to execute
  */
-static void execute_decode_slices(H264Context *h, int context_count){
+static int execute_decode_slices(H264Context *h, int context_count){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
     H264Context *hx;
     int i;
 
-    if (s->avctx->hwaccel)
-        return;
-    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
-        return;
+    if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
+        return 0;
     if(context_count == 1) {
-        decode_slice(avctx, &h);
+        return decode_slice(avctx, &h);
     } else {
         for(i = 1; i < context_count; i++) {
             hx = h->thread_context[i];
-            hx->s.error_recognition = avctx->error_recognition;
+            hx->s.err_recognition = avctx->err_recognition;
             hx->s.error_count = 0;
         }
 
-        avctx->execute(avctx, (void *)decode_slice,
+        avctx->execute(avctx, decode_slice,
                        h->thread_context, NULL, context_count, sizeof(void*));
 
         /* pull back stuff from slices to master context */
@@ -2709,24 +3733,23 @@ static void execute_decode_slices(H264Context *h, int context_count){
         for(i = 1; i < context_count; i++)
             h->s.error_count += h->thread_context[i]->s.error_count;
     }
+
+    return 0;
 }
 
 
 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
-    int buf_index=0;
     H264Context *hx; ///< thread context
-    int context_count = 0;
-    int next_avc= h->is_avc ? 0 : buf_size;
+    int buf_index;
+    int context_count;
+    int next_avc;
+    int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
+    int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
+    int nal_index;
 
-    h->max_contexts = avctx->thread_count;
-#if 0
-    int i;
-    for(i=0; i<50; i++){
-        av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);
-    }
-#endif
+    h->max_contexts = s->slice_context_count;
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
         h->current_slice = 0;
         if (!s->first_field)
@@ -2734,6 +3757,11 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
         ff_h264_reset_sei(h);
     }
 
+    for(;pass <= 1;pass++){
+        buf_index = 0;
+        context_count = 0;
+        next_avc = h->is_avc ? 0 : buf_size;
+        nal_index = 0;
     for(;;){
         int consumed;
         int dst_length;
@@ -2743,7 +3771,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
         int err;
 
         if(buf_index >= next_avc) {
-            if(buf_index >= buf_size) break;
+            if (buf_index >= buf_size - h->nal_length_size) break;
             nalsize = 0;
             for(i = 0; i < h->nal_length_size; i++)
                 nalsize = (nalsize << 8) | buf[buf_index++];
@@ -2792,9 +3820,28 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
         }
 
         buf_index += consumed;
+        nal_index++;
+
+        if(pass == 0) {
+            // packets can sometimes contain multiple PPS/SPS
+            // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
+            // if so, when frame threading we can't start the next thread until we've read all of them
+            switch (hx->nal_unit_type) {
+                case NAL_SPS:
+                case NAL_PPS:
+                    nals_needed = nal_index;
+                    break;
+                case NAL_IDR_SLICE:
+                case NAL_SLICE:
+                    init_get_bits(&hx->s.gb, ptr, bit_length);
+                    if (!get_ue_golomb(&hx->s.gb))
+                        nals_needed = nal_index;
+            }
+            continue;
+        }
 
-        if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME do not discard SEI id
-           ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))
+        //FIXME do not discard SEI id
+        if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0)
             continue;
 
       again:
@@ -2805,7 +3852,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
                 return -1;
             }
-            idr(h); //FIXME ensure we don't loose some frames if there is reordering
+            idr(h); // FIXME ensure we don't lose some frames if there is reordering
         case NAL_SLICE:
             init_get_bits(&hx->s.gb, ptr, bit_length);
             hx->intra_gb_ptr=
@@ -2815,20 +3862,25 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             if((err = decode_slice_header(hx, h)))
                break;
 
+            s->current_picture_ptr->f.key_frame |=
+                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
+                    (h->sei_recovery_frame_cnt >= 0);
+
             if (h->current_slice == 1) {
+                if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
+                    decode_postinit(h, nal_index >= nals_needed);
+                }
+
                 if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
                     return -1;
                 if(CONFIG_H264_VDPAU_DECODER && s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
                     ff_vdpau_h264_picture_start(s);
             }
 
-            s->current_picture_ptr->key_frame |=
-                    (hx->nal_unit_type == NAL_IDR_SLICE) ||
-                    (h->sei_recovery_frame_cnt >= 0);
-            if(hx->redundant_pic_count==0 && hx->s.hurry_up < 5
+            if(hx->redundant_pic_count==0
                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
-               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
-               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
+               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
+               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
                && avctx->skip_frame < AVDISCARD_ALL){
                 if(avctx->hwaccel) {
                     if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
@@ -2863,10 +3915,9 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
 
             if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
                && s->context_initialized
-               && s->hurry_up < 5
                && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
-               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=FF_B_TYPE)
-               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==FF_I_TYPE)
+               && (avctx->skip_frame < AVDISCARD_BIDIR  || hx->slice_type_nos!=AV_PICTURE_TYPE_B)
+               && (avctx->skip_frame < AVDISCARD_NONKEY || hx->slice_type_nos==AV_PICTURE_TYPE_I)
                && avctx->skip_frame < AVDISCARD_ALL)
                 context_count++;
             break;
@@ -2876,13 +3927,38 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             break;
         case NAL_SPS:
             init_get_bits(&s->gb, ptr, bit_length);
-            ff_h264_decode_seq_parameter_set(h);
+            if (ff_h264_decode_seq_parameter_set(h) < 0 &&
+                h->is_avc && (nalsize != consumed) && nalsize) {
+                av_log(h->s.avctx, AV_LOG_DEBUG, "SPS decoding failure, "
+                       "try parsing the coomplete NAL\n");
+                init_get_bits(&s->gb, buf + buf_index + 1 - consumed,
+                              8 * (nalsize - 1));
+                ff_h264_decode_seq_parameter_set(h);
+            }
 
-            if(s->flags& CODEC_FLAG_LOW_DELAY)
+            if (s->flags& CODEC_FLAG_LOW_DELAY ||
+                (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
                 s->low_delay=1;
 
             if(avctx->has_b_frames < 2)
                 avctx->has_b_frames= !s->low_delay;
+
+            if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
+                h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
+                if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
+                    avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
+                    h->cur_chroma_format_idc = h->sps.chroma_format_idc;
+                    h->pixel_shift = h->sps.bit_depth_luma > 8;
+
+                    ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
+                    ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
+                    s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
+                    dsputil_init(&s->dsp, s->avctx);
+                } else {
+                    av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
+                    return -1;
+                }
+            }
             break;
         case NAL_PPS:
             init_get_bits(&s->gb, ptr, bit_length);
@@ -2919,13 +3995,14 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             goto again;
         }
     }
+    }
     if(context_count)
         execute_decode_slices(h, context_count);
     return buf_index;
 }
 
 /**
- * returns the number of bytes consumed for building the current frame
+ * Return the number of bytes consumed for building the current frame.
  */
 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
@@ -2943,7 +4020,7 @@ static int decode_frame(AVCodecContext *avctx,
     H264Context *h = avctx->priv_data;
     MpegEncContext *s = &h->s;
     AVFrame *pict = data;
-    int buf_index;
+    int buf_index = 0;
 
     s->flags= avctx->flags;
     s->flags2= avctx->flags2;
@@ -2954,10 +4031,12 @@ static int decode_frame(AVCodecContext *avctx,
         Picture *out;
         int i, out_idx;
 
+        s->current_picture_ptr = NULL;
+
 //FIXME factorize this with the output code below
         out = h->delayed_pic[0];
         out_idx = 0;
-        for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
+        for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
             if(h->delayed_pic[i]->poc < out->poc){
                 out = h->delayed_pic[i];
                 out_idx = i;
@@ -2971,7 +4050,7 @@ static int decode_frame(AVCodecContext *avctx,
             *pict= *(AVFrame*)out;
         }
 
-        return 0;
+        return buf_index;
     }
 
     buf_index=decode_nal_units(h, buf, buf_size);
@@ -2984,149 +4063,25 @@ static int decode_frame(AVCodecContext *avctx,
     }
 
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
-        if (avctx->skip_frame >= AVDISCARD_NONREF || s->hurry_up) return 0;
+        if (avctx->skip_frame >= AVDISCARD_NONREF)
+            return 0;
         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
         return -1;
     }
 
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
-        Picture *out = s->current_picture_ptr;
-        Picture *cur = s->current_picture_ptr;
-        int i, pics, out_of_order, out_idx;
 
-        field_end(h);
+        if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
 
-        if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
+        field_end(h, 0);
+
+        if (!h->next_output_pic) {
             /* Wait for second field. */
             *data_size = 0;
 
         } else {
-            cur->interlaced_frame = 0;
-            cur->repeat_pict = 0;
-
-            /* Signal interlacing information externally. */
-            /* Prioritize picture timing SEI information over used decoding process if it exists. */
-
-            if(h->sps.pic_struct_present_flag){
-                switch (h->sei_pic_struct)
-                {
-                case SEI_PIC_STRUCT_FRAME:
-                    break;
-                case SEI_PIC_STRUCT_TOP_FIELD:
-                case SEI_PIC_STRUCT_BOTTOM_FIELD:
-                    cur->interlaced_frame = 1;
-                    break;
-                case SEI_PIC_STRUCT_TOP_BOTTOM:
-                case SEI_PIC_STRUCT_BOTTOM_TOP:
-                    if (FIELD_OR_MBAFF_PICTURE)
-                        cur->interlaced_frame = 1;
-                    else
-                        // try to flag soft telecine progressive
-                        cur->interlaced_frame = h->prev_interlaced_frame;
-                    break;
-                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
-                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
-                    // Signal the possibility of telecined film externally (pic_struct 5,6)
-                    // From these hints, let the applications decide if they apply deinterlacing.
-                    cur->repeat_pict = 1;
-                    break;
-                case SEI_PIC_STRUCT_FRAME_DOUBLING:
-                    // Force progressive here, as doubling interlaced frame is a bad idea.
-                    cur->repeat_pict = 2;
-                    break;
-                case SEI_PIC_STRUCT_FRAME_TRIPLING:
-                    cur->repeat_pict = 4;
-                    break;
-                }
-
-                if ((h->sei_ct_type & 3) && h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
-                    cur->interlaced_frame = (h->sei_ct_type & (1<<1)) != 0;
-            }else{
-                /* Derive interlacing flag from used decoding process. */
-                cur->interlaced_frame = FIELD_OR_MBAFF_PICTURE;
-            }
-            h->prev_interlaced_frame = cur->interlaced_frame;
-
-            if (cur->field_poc[0] != cur->field_poc[1]){
-                /* Derive top_field_first from field pocs. */
-                cur->top_field_first = cur->field_poc[0] < cur->field_poc[1];
-            }else{
-                if(cur->interlaced_frame || h->sps.pic_struct_present_flag){
-                    /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
-                    if(h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM
-                      || h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
-                        cur->top_field_first = 1;
-                    else
-                        cur->top_field_first = 0;
-                }else{
-                    /* Most likely progressive */
-                    cur->top_field_first = 0;
-                }
-            }
-
-        //FIXME do something with unavailable reference frames
-
-            /* Sort B-frames into display order */
-
-            if(h->sps.bitstream_restriction_flag
-               && s->avctx->has_b_frames < h->sps.num_reorder_frames){
-                s->avctx->has_b_frames = h->sps.num_reorder_frames;
-                s->low_delay = 0;
-            }
-
-            if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
-               && !h->sps.bitstream_restriction_flag){
-                s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
-                s->low_delay= 0;
-            }
-
-            pics = 0;
-            while(h->delayed_pic[pics]) pics++;
-
-            assert(pics <= MAX_DELAYED_PIC_COUNT);
-
-            h->delayed_pic[pics++] = cur;
-            if(cur->reference == 0)
-                cur->reference = DELAYED_PIC_REF;
-
-            out = h->delayed_pic[0];
-            out_idx = 0;
-            for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame && !h->delayed_pic[i]->mmco_reset; i++)
-                if(h->delayed_pic[i]->poc < out->poc){
-                    out = h->delayed_pic[i];
-                    out_idx = i;
-                }
-            if(s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset))
-                h->outputed_poc= INT_MIN;
-            out_of_order = out->poc < h->outputed_poc;
-
-            if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
-                { }
-            else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
-               || (s->low_delay &&
-                ((h->outputed_poc != INT_MIN && out->poc > h->outputed_poc + 2)
-                 || cur->pict_type == FF_B_TYPE)))
-            {
-                s->low_delay = 0;
-                s->avctx->has_b_frames++;
-            }
-
-            if(out_of_order || pics > s->avctx->has_b_frames){
-                out->reference &= ~DELAYED_PIC_REF;
-                for(i=out_idx; h->delayed_pic[i]; i++)
-                    h->delayed_pic[i] = h->delayed_pic[i+1];
-            }
-            if(!out_of_order && pics > s->avctx->has_b_frames){
-                *data_size = sizeof(AVFrame);
-
-                if(out_idx==0 && h->delayed_pic[0] && (h->delayed_pic[0]->key_frame || h->delayed_pic[0]->mmco_reset)) {
-                    h->outputed_poc = INT_MIN;
-                } else
-                    h->outputed_poc = out->poc;
-                *pict= *(AVFrame*)out;
-            }else{
-                av_log(avctx, AV_LOG_DEBUG, "no picture\n");
-            }
+            *data_size = sizeof(AVFrame);
+            *pict = *(AVFrame*)h->next_output_pic;
         }
     }
 
@@ -3166,10 +4121,10 @@ int main(void){
     uint8_t temp[SIZE];
     PutBitContext pb;
     GetBitContext gb;
-//    int int_temp[10000];
     DSPContext dsp;
     AVCodecContext avctx;
 
+    avctx.av_class = avcodec_get_class();
     dsputil_init(&dsp, &avctx);
 
     init_put_bits(&pb, temp, SIZE);
@@ -3183,9 +4138,7 @@ int main(void){
 
     init_get_bits(&gb, temp, 8*SIZE);
     for(i=0; i<COUNT; i++){
-        int j, s;
-
-        s= show_bits(&gb, 24);
+        int j, s = show_bits(&gb, 24);
 
         START_TIMER
         j= get_ue_golomb(&gb);
@@ -3208,9 +4161,7 @@ int main(void){
 
     init_get_bits(&gb, temp, 8*SIZE);
     for(i=0; i<COUNT; i++){
-        int j, s;
-
-        s= show_bits(&gb, 24);
+        int j, s = show_bits(&gb, 24);
 
         START_TIMER
         j= get_se_golomb(&gb);
@@ -3221,109 +4172,6 @@ int main(void){
         STOP_TIMER("get_se_golomb");
     }
 
-#if 0
-    printf("testing 4x4 (I)DCT\n");
-
-    DCTELEM block[16];
-    uint8_t src[16], ref[16];
-    uint64_t error= 0, max_error=0;
-
-    for(i=0; i<COUNT; i++){
-        int j;
-//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
-        for(j=0; j<16; j++){
-            ref[j]= random()%255;
-            src[j]= random()%255;
-        }
-
-        h264_diff_dct_c(block, src, ref, 4);
-
-        //normalize
-        for(j=0; j<16; j++){
-//            printf("%d ", block[j]);
-            block[j]= block[j]*4;
-            if(j&1) block[j]= (block[j]*4 + 2)/5;
-            if(j&4) block[j]= (block[j]*4 + 2)/5;
-        }
-//        printf("\n");
-
-        h->h264dsp.h264_idct_add(ref, block, 4);
-/*        for(j=0; j<16; j++){
-            printf("%d ", ref[j]);
-        }
-        printf("\n");*/
-
-        for(j=0; j<16; j++){
-            int diff= FFABS(src[j] - ref[j]);
-
-            error+= diff*diff;
-            max_error= FFMAX(max_error, diff);
-        }
-    }
-    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
-    printf("testing quantizer\n");
-    for(qp=0; qp<52; qp++){
-        for(i=0; i<16; i++)
-            src1_block[i]= src2_block[i]= random()%255;
-
-    }
-    printf("Testing NAL layer\n");
-
-    uint8_t bitstream[COUNT];
-    uint8_t nal[COUNT*2];
-    H264Context h;
-    memset(&h, 0, sizeof(H264Context));
-
-    for(i=0; i<COUNT; i++){
-        int zeros= i;
-        int nal_length;
-        int consumed;
-        int out_length;
-        uint8_t *out;
-        int j;
-
-        for(j=0; j<COUNT; j++){
-            bitstream[j]= (random() % 255) + 1;
-        }
-
-        for(j=0; j<zeros; j++){
-            int pos= random() % COUNT;
-            while(bitstream[pos] == 0){
-                pos++;
-                pos %= COUNT;
-            }
-            bitstream[pos]=0;
-        }
-
-        START_TIMER
-
-        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
-        if(nal_length<0){
-            printf("encoding failed\n");
-            return -1;
-        }
-
-        out= ff_h264_decode_nal(&h, nal, &out_length, &consumed, nal_length);
-
-        STOP_TIMER("NAL")
-
-        if(out_length != COUNT){
-            printf("incorrect length %d %d\n", out_length, COUNT);
-            return -1;
-        }
-
-        if(consumed != nal_length){
-            printf("incorrect consumed length %d %d\n", nal_length, consumed);
-            return -1;
-        }
-
-        if(memcmp(bitstream, out, COUNT)){
-            printf("mismatch\n");
-            return -1;
-        }
-    }
-#endif
-
     printf("Testing RBSP\n");
 
 
@@ -3359,34 +4207,53 @@ av_cold int ff_h264_decode_end(AVCodecContext *avctx)
     return 0;
 }
 
+static const AVProfile profiles[] = {
+    { FF_PROFILE_H264_BASELINE,             "Baseline"              },
+    { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
+    { FF_PROFILE_H264_MAIN,                 "Main"                  },
+    { FF_PROFILE_H264_EXTENDED,             "Extended"              },
+    { FF_PROFILE_H264_HIGH,                 "High"                  },
+    { FF_PROFILE_H264_HIGH_10,              "High 10"               },
+    { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
+    { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
+    { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
+    { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
+    { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
+    { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
+    { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
+    { FF_PROFILE_UNKNOWN },
+};
 
 AVCodec ff_h264_decoder = {
-    "h264",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_H264,
-    sizeof(H264Context),
-    ff_h264_decode_init,
-    NULL,
-    ff_h264_decode_end,
-    decode_frame,
-    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY,
+    .name           = "h264",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = CODEC_ID_H264,
+    .priv_data_size = sizeof(H264Context),
+    .init           = ff_h264_decode_init,
+    .close          = ff_h264_decode_end,
+    .decode         = decode_frame,
+    .capabilities   = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
+                      CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
+    .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
+    .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
+    .profiles = NULL_IF_CONFIG_SMALL(profiles),
 };
 
 #if CONFIG_H264_VDPAU_DECODER
 AVCodec ff_h264_vdpau_decoder = {
-    "h264_vdpau",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_H264,
-    sizeof(H264Context),
-    ff_h264_decode_init,
-    NULL,
-    ff_h264_decode_end,
-    decode_frame,
-    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
+    .name           = "h264_vdpau",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = CODEC_ID_H264,
+    .priv_data_size = sizeof(H264Context),
+    .init           = ff_h264_decode_init,
+    .close          = ff_h264_decode_end,
+    .decode         = decode_frame,
+    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
+    .profiles = NULL_IF_CONFIG_SMALL(profiles),
 };
 #endif