Add the notion of pixel size in h264 related functions.

In high bit depth the pixels will not be stored in uint8_t like in the
normal case, but in uint16_t. The pixel size is thus 1 in normal bit
depth and 2 in high bit depth.

Preparatory patch for high bit depth h264 decoding support.

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

6 files changed, 335 insertions, 195 deletions

diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index 142a07807a..e02b277769 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -314,12 +314,13 @@ static void chroma_dc_dct_c(DCTELEM *block){
 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){
+                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
+                           int pixel_shift){
     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;
+    uint8_t * src_y = pic->data[0] + ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
     uint8_t * src_cb, * src_cr;
     int extra_width= h->emu_edge_width;
     int extra_height= h->emu_edge_height;
@@ -336,8 +337,8 @@ 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;
     }
 
@@ -353,8 +354,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
         my += 2 * ((s->mb_y & 1) - (pic->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->data[1] + ((mx>>3) << pixel_shift) + (my>>3)*h->mb_uvlinesize;
+    src_cr= pic->data[2] + ((mx>>3) << pixel_shift) + (my>>3)*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);
@@ -374,14 +375,14 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
                            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 list0, int list1, int pixel_shift){
     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;
+    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);
 
@@ -389,7 +390,7 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift);
 
         qpix_op=  qpix_avg;
         chroma_op= chroma_avg;
@@ -399,7 +400,7 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op);
+                           qpix_op, chroma_op, pixel_shift);
     }
 }
 
@@ -409,12 +410,12 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
                            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 list0, int list1, int pixel_shift){
     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;
+    dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->  mb_linesize;
+    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);
 
@@ -422,17 +423,17 @@ 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_cr = s->obmc_scratchpad + (8 << pixel_shift);
         uint8_t *tmp_y  = s->obmc_scratchpad + 8*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,
                     dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put);
+                    x_offset, y_offset, qpix_put, chroma_put, pixel_shift);
         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_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);
 
         if(h->use_weight == 2){
             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
@@ -457,7 +458,7 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         Picture *ref= &h->ref_list[list][refn];
         mc_dir_part(h, ref, n, square, chroma_height, delta, list,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                    qpix_put, chroma_put);
+                    qpix_put, chroma_put, pixel_shift);
 
         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
@@ -476,19 +477,21 @@ static inline void mc_part(H264Context *h, int n, int square, int chroma_height,
                            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 list0, int list1, int pixel_shift){
     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,
                          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[3], weight_avg[0],
+                         weight_avg[3], list0, list1, pixel_shift);
     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);
+                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
+                    chroma_avg, list0, list1, pixel_shift);
 }
 
-static inline void prefetch_motion(H264Context *h, int list){
+static inline void prefetch_motion(H264Context *h, int list, int pixel_shift){
     /* fetch pixels for estimated mv 4 macroblocks ahead
      * optimized for 64byte cache lines */
     MpegEncContext * const s = &h->s;
@@ -497,48 +500,54 @@ static inline void prefetch_motion(H264Context *h, int list){
         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;
+        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;
+        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){
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
     const int mb_type= s->current_picture.mb_type[mb_xy];
 
     assert(IS_INTER(mb_type));
 
-    prefetch_motion(h, 0);
+    prefetch_motion(h, 0, pixel_shift);
 
     if(IS_16X16(mb_type)){
         mc_part(h, 0, 1, 8, 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);
     }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, 4, 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,
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift);
+        mc_part(h, 8, 0, 4, 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));
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift);
     }else if(IS_8X16(mb_type)){
         mc_part(h, 0, 0, 8, 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));
+                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
+                pixel_shift);
         mc_part(h, 4, 0, 8, 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));
+                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
+                pixel_shift);
     }else{
         int i;
 
@@ -554,25 +563,30 @@ static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t
                 mc_part(h, n, 1, 4, 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));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }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, 2, 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,
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
+                mc_part(h, n+2, 0, 2, 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));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }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,
                     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));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
                 mc_part(h, n+1, 0, 4, 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));
+                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                    pixel_shift);
             }else{
                 int j;
                 assert(IS_SUB_4X4(sub_mb_type));
@@ -582,15 +596,32 @@ static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t
                     mc_part(h, n+j, 1, 2, 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));
+                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
+                        pixel_shift);
                 }
             }
         }
     }
 
-    prefetch_motion(h, 1);
+    prefetch_motion(h, 1, pixel_shift);
 }
 
+#define hl_motion_fn(sh, bits) \
+static av_always_inline void hl_motion_ ## bits(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) \
+{ \
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put, \
+              qpix_avg, chroma_avg, weight_op, weight_avg, sh); \
+}
+hl_motion_fn(0, 8);
+hl_motion_fn(1, 16);
 
 static void free_tables(H264Context *h, int free_rbsp){
     int i;
@@ -758,8 +789,8 @@ 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+8+8) * sizeof(uint8_t)*2, fail)
+    FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * (16+8+8) * 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;
@@ -861,6 +892,8 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 
     ff_h264_decode_init_vlc();
 
+    h->pixel_shift = 0;
+
     h->thread_context[0] = h;
     h->outputed_poc = INT_MIN;
     h->prev_poc_msb= 1<<16;
@@ -888,6 +921,7 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 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;
@@ -904,14 +938,14 @@ int ff_h264_frame_start(H264Context *h){
     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[24+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
     }
     for(i=0; i<4; 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[20+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 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[24+20+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.
@@ -945,6 +979,7 @@ static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src
     MpegEncContext * const s = &h->s;
     uint8_t *top_border;
     int top_idx = 1;
+    const int pixel_shift = h->pixel_shift;
 
     src_y  -=   linesize;
     src_cb -= uvlinesize;
@@ -955,9 +990,16 @@ 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)){
+                    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){
@@ -970,14 +1012,24 @@ 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)){
+        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 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, int pixel_shift){
     MpegEncContext * const s = &h->s;
     int deblock_left;
     int deblock_top;
@@ -1002,41 +1054,62 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_c
         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);
+            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);
+                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, src_cb+1, 1);
-            XCHG(top_border+24, src_cr+1, 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);
         }
     }
 }
 
-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_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;
@@ -1052,12 +1125,12 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
 
-    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.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+    dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
+    dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
 
-    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;
 
@@ -1094,6 +1167,28 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
     }
 
     if (!simple && IS_INTRA_PCM(mb_type)) {
+        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);
+            }
+            for (i = 0; i < 8; 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 < 8; 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);
         }
@@ -1101,10 +1196,11 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
             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, simple, pixel_shift);
 
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
                 h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
@@ -1125,16 +1221,16 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             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);
+                                h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16 << pixel_shift), 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);
+                                    if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
+                                        idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
                                     else
-                                        idct_add   (ptr, h->mb + i*16, linesize);
+                                        idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
                                 }
                             }
                         }
@@ -1151,18 +1247,24 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             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);
+                                h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16 << 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(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]*0x01010101;
                                         topright= (uint8_t*) &tr;
+                                        }
                                     }else
-                                        topright= ptr + 4 - linesize;
+                                        topright= ptr + (4 << pixel_shift) - linesize;
                                 }else
                                     topright= NULL;
 
@@ -1170,10 +1272,10 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                                 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);
+                                        if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16))
+                                            idct_dc_add(ptr, h->mb + (i*16 << pixel_shift), linesize);
                                         else
-                                            idct_add   (ptr, h->mb + i*16, linesize);
+                                            idct_add   (ptr, h->mb + (i*16 << pixel_shift), linesize);
                                     }else
                                         ff_svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
                                 }
@@ -1191,19 +1293,27 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             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];
+                                dctcoef_set(h->mb, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc, pixel_shift, i));
                         }
                     }
                 }else
                     ff_svq3_luma_dc_dequant_idct_c(h->mb, h->mb_luma_dc, s->qscale);
             }
             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, 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 (pixel_shift) {
+                hl_motion_16(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);
+            } else
+                hl_motion_8(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);
         }
 
 
@@ -1215,8 +1325,8 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                             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);
+                                if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                    s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
                             }
                         }
                     }else{
@@ -1228,7 +1338,7 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                         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);
+                                idct_add(dest_y + block_offset[i], h->mb + (i*16 << pixel_shift), linesize);
                             }
                         }
                     }else{
@@ -1253,21 +1363,21 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
             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 << pixel_shift), uvlinesize);
+                    h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 20, h->mb + (20*16 << 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);
+                        if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                            idct_add   (dest[(i&4)>>2] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
                     }
                 }
             }else{
                 if(is_h264){
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(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 << pixel_shift)       , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(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+4*16) << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
                     h->h264dsp.h264_idct_add8(dest, block_offset,
                                               h->mb, uvlinesize,
                                               h->non_zero_count_cache);
@@ -1291,15 +1401,18 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
 /**
  * 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);
 }
 
 void ff_h264_hl_decode_mb(H264Context *h){
@@ -1308,9 +1421,12 @@ void ff_h264_hl_decode_mb(H264Context *h){
     const int mb_type= s->current_picture.mb_type[mb_xy];
     int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
 
-    if (is_complex)
+    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){
@@ -2432,6 +2548,7 @@ static void loop_filter(H264Context *h){
     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;
 
     if(h->deblocking_filter) {
         for(mb_x= 0; mb_x<s->mb_width; mb_x++){
@@ -2447,9 +2564,9 @@ 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.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
+                dest_cb = s->current_picture.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
+                dest_cr = s->current_picture.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * 8;
                     //FIXME simplify above
 
                 if (MB_FIELD) {
diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index 0a0ec36649..470d9bee63 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -265,6 +265,7 @@ typedef struct MMCO{
 typedef struct H264Context{
     MpegEncContext s;
     H264DSPContext h264dsp;
+    int pixel_shift;    ///< 0 for 8-bit H264, 1 for high-bit-depth H264
     int chroma_qp[2]; //QPc
 
     int qp_thresh;      ///< QP threshold to skip loopfilter
@@ -296,7 +297,7 @@ typedef struct H264Context{
     unsigned int top_samples_available;
     unsigned int topright_samples_available;
     unsigned int left_samples_available;
-    uint8_t (*top_borders[2])[16+2*8];
+    uint8_t (*top_borders[2])[(16+2*8)*2];
 
     /**
      * non zero coeff count cache.
@@ -406,9 +407,9 @@ typedef struct H264Context{
     GetBitContext *intra_gb_ptr;
     GetBitContext *inter_gb_ptr;
 
-    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
-    DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16];
-    DCTELEM mb_padding[256];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
+    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24*2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
+    DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16*2];
+    DCTELEM mb_padding[256*2];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
 
     /**
      * Cabac
diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index 25215fabe8..61ad4493f9 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -1100,47 +1100,54 @@ static av_always_inline void decode_cabac_residual_internal( H264Context *h, DCT
         }
     }
 
-    do {
-        uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base;
-
-        int j= scantable[index[--coeff_count]];
-
-        if( get_cabac( CC, ctx ) == 0 ) {
-            node_ctx = coeff_abs_level_transition[0][node_ctx];
-            if( is_dc ) {
-                block[j] = get_cabac_bypass_sign( CC, -1);
-            }else{
-                block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;
-            }
-        } else {
-            int coeff_abs = 2;
-            ctx = coeff_abs_levelgt1_ctx[node_ctx] + abs_level_m1_ctx_base;
-            node_ctx = coeff_abs_level_transition[1][node_ctx];
-
-            while( coeff_abs < 15 && get_cabac( CC, ctx ) ) {
-                coeff_abs++;
-            }
-
-            if( coeff_abs >= 15 ) {
-                int j = 0;
-                while( get_cabac_bypass( CC ) ) {
-                    j++;
-                }
-
-                coeff_abs=1;
-                while( j-- ) {
-                    coeff_abs += coeff_abs + get_cabac_bypass( CC );
-                }
-                coeff_abs+= 14;
-            }
+#define STORE_BLOCK(type) \
+    do { \
+        uint8_t *ctx = coeff_abs_level1_ctx[node_ctx] + abs_level_m1_ctx_base; \
+ \
+        int j= scantable[index[--coeff_count]]; \
+ \
+        if( get_cabac( CC, ctx ) == 0 ) { \
+            node_ctx = coeff_abs_level_transition[0][node_ctx]; \
+            if( is_dc ) { \
+                ((type*)block)[j] = get_cabac_bypass_sign( CC, -1); \
+            }else{ \
+                ((type*)block)[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6; \
+            } \
+        } else { \
+            int coeff_abs = 2; \
+            ctx = coeff_abs_levelgt1_ctx[node_ctx] + abs_level_m1_ctx_base; \
+            node_ctx = coeff_abs_level_transition[1][node_ctx]; \
+\
+            while( coeff_abs < 15 && get_cabac( CC, ctx ) ) { \
+                coeff_abs++; \
+            } \
+\
+            if( coeff_abs >= 15 ) { \
+                int j = 0; \
+                while( get_cabac_bypass( CC ) ) { \
+                    j++; \
+                } \
+\
+                coeff_abs=1; \
+                while( j-- ) { \
+                    coeff_abs += coeff_abs + get_cabac_bypass( CC ); \
+                } \
+                coeff_abs+= 14; \
+            } \
+\
+            if( is_dc ) { \
+                ((type*)block)[j] = get_cabac_bypass_sign( CC, -coeff_abs ); \
+            }else{ \
+                ((type*)block)[j] = ((int)(get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32)) >> 6; \
+            } \
+        } \
+    } while ( coeff_count );
 
-            if( is_dc ) {
-                block[j] = get_cabac_bypass_sign( CC, -coeff_abs );
-            }else{
-                block[j] = (get_cabac_bypass_sign( CC, -coeff_abs ) * qmul[j] + 32) >> 6;
-            }
-        }
-    } while( coeff_count );
+    if (h->pixel_shift) {
+        STORE_BLOCK(int32_t)
+    } else {
+        STORE_BLOCK(int16_t)
+    }
 #ifdef CABAC_ON_STACK
             h->cabac.range     = cc.range     ;
             h->cabac.low       = cc.low       ;
@@ -1196,6 +1203,7 @@ int ff_h264_decode_mb_cabac(H264Context *h) {
     int mb_xy;
     int mb_type, partition_count, cbp = 0;
     int dct8x8_allowed= h->pps.transform_8x8_mode;
+    const int pixel_shift = h->pixel_shift;
 
     mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
 
@@ -1304,6 +1312,7 @@ decode_intra_mb:
     h->slice_table[ mb_xy ]= h->slice_num;
 
     if(IS_INTRA_PCM(mb_type)) {
+        const int mb_size = (384*h->sps.bit_depth_luma) >> 3;
         const uint8_t *ptr;
 
         // We assume these blocks are very rare so we do not optimize it.
@@ -1316,9 +1325,9 @@ decode_intra_mb:
         }
 
         // The pixels are stored in the same order as levels in h->mb array.
-        memcpy(h->mb, ptr, 256); ptr+=256;
+        memcpy(h->mb, ptr, 2*mb_size/3); ptr+=2*mb_size/3;
         if(CHROMA){
-            memcpy(h->mb+128, ptr, 128); ptr+=128;
+            memcpy(h->mb+mb_size/3, ptr, mb_size/3); ptr+=mb_size/3;
         }
 
         ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
@@ -1652,13 +1661,15 @@ decode_intra_mb:
             //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
             AV_ZERO128(h->mb_luma_dc+0);
             AV_ZERO128(h->mb_luma_dc+8);
+            AV_ZERO128(h->mb_luma_dc+16);
+            AV_ZERO128(h->mb_luma_dc+24);
             decode_cabac_residual_dc( h, h->mb_luma_dc, 0, LUMA_DC_BLOCK_INDEX, scan, 16);
 
             if( cbp&15 ) {
                 qmul = h->dequant4_coeff[0][s->qscale];
                 for( i = 0; i < 16; i++ ) {
                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
-                    decode_cabac_residual_nondc(h, h->mb + 16*i, 1, i, scan + 1, qmul, 15);
+                    decode_cabac_residual_nondc(h, h->mb + (16*i << pixel_shift), 1, i, scan + 1, qmul, 15);
                 }
             } else {
                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
@@ -1668,7 +1679,7 @@ decode_intra_mb:
             for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
                 if( cbp & (1<<i8x8) ) {
                     if( IS_8x8DCT(mb_type) ) {
-                        decode_cabac_residual_nondc(h, h->mb + 64*i8x8, 5, 4*i8x8,
+                        decode_cabac_residual_nondc(h, h->mb + (64*i8x8 << pixel_shift), 5, 4*i8x8,
                             scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64);
                     } else {
                         qmul = h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale];
@@ -1676,7 +1687,7 @@ decode_intra_mb:
                             const int index = 4*i8x8 + i4x4;
                             //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
 //START_TIMER
-                            decode_cabac_residual_nondc(h, h->mb + 16*index, 2, index, scan, qmul, 16);
+                            decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), 2, index, scan, qmul, 16);
 //STOP_TIMER("decode_residual")
                         }
                     }
@@ -1691,7 +1702,7 @@ decode_intra_mb:
             int c;
             for( c = 0; c < 2; c++ ) {
                 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
-                decode_cabac_residual_dc(h, h->mb + 256 + 16*4*c, 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
+                decode_cabac_residual_dc(h, h->mb + ((256 + 16*4*c) << pixel_shift), 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
             }
         }
 
@@ -1702,7 +1713,7 @@ decode_intra_mb:
                 for( i = 0; i < 4; i++ ) {
                     const int index = 16 + 4 * c + i;
                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
-                    decode_cabac_residual_nondc(h, h->mb + 16*index, 4, index, scan + 1, qmul, 15);
+                    decode_cabac_residual_nondc(h, h->mb + (16*index << pixel_shift), 4, index, scan + 1, qmul, 15);
                 }
             }
         } else {
diff --git a/libavcodec/h264_cavlc.c b/libavcodec/h264_cavlc.c
index 15d84ac32c..61b057176b 100644
--- a/libavcodec/h264_cavlc.c
+++ b/libavcodec/h264_cavlc.c
@@ -488,37 +488,44 @@ static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, in
             zeros_left= get_vlc2(gb, (total_zeros_vlc-1)[ total_coeff ].table, TOTAL_ZEROS_VLC_BITS, 1);
     }
 
-    scantable += zeros_left + total_coeff - 1;
-    if(n >= LUMA_DC_BLOCK_INDEX){
-        block[*scantable] = level[0];
-        for(i=1;i<total_coeff && zeros_left > 0;i++) {
-            if(zeros_left < 7)
-                run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1);
-            else
-                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
-            zeros_left -= run_before;
-            scantable -= 1 + run_before;
-            block[*scantable]= level[i];
-        }
-        for(;i<total_coeff;i++) {
-            scantable--;
-            block[*scantable]= level[i];
-        }
-    }else{
-        block[*scantable] = (level[0] * qmul[*scantable] + 32)>>6;
-        for(i=1;i<total_coeff && zeros_left > 0;i++) {
-            if(zeros_left < 7)
-                run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1);
-            else
-                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
-            zeros_left -= run_before;
-            scantable -= 1 + run_before;
-            block[*scantable]= (level[i] * qmul[*scantable] + 32)>>6;
-        }
-        for(;i<total_coeff;i++) {
-            scantable--;
-            block[*scantable]= (level[i] * qmul[*scantable] + 32)>>6;
-        }
+#define STORE_BLOCK(type) \
+    scantable += zeros_left + total_coeff - 1; \
+    if(n >= LUMA_DC_BLOCK_INDEX){ \
+        ((type*)block)[*scantable] = level[0]; \
+        for(i=1;i<total_coeff && zeros_left > 0;i++) { \
+            if(zeros_left < 7) \
+                run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
+            else \
+                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
+            zeros_left -= run_before; \
+            scantable -= 1 + run_before; \
+            ((type*)block)[*scantable]= level[i]; \
+        } \
+        for(;i<total_coeff;i++) { \
+            scantable--; \
+            ((type*)block)[*scantable]= level[i]; \
+        } \
+    }else{ \
+        ((type*)block)[*scantable] = ((int)(level[0] * qmul[*scantable] + 32))>>6; \
+        for(i=1;i<total_coeff && zeros_left > 0;i++) { \
+            if(zeros_left < 7) \
+                run_before= get_vlc2(gb, (run_vlc-1)[zeros_left].table, RUN_VLC_BITS, 1); \
+            else \
+                run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); \
+            zeros_left -= run_before; \
+            scantable -= 1 + run_before; \
+            ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
+        } \
+        for(;i<total_coeff;i++) { \
+            scantable--; \
+            ((type*)block)[*scantable]= ((int)(level[i] * qmul[*scantable] + 32))>>6; \
+        } \
+    }
+
+    if (h->pixel_shift) {
+        STORE_BLOCK(int32_t)
+    } else {
+        STORE_BLOCK(int16_t)
     }
 
     if(zeros_left<0){
@@ -535,6 +542,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h){
     int partition_count;
     unsigned int mb_type, cbp;
     int dct8x8_allowed= h->pps.transform_8x8_mode;
+    const int pixel_shift = h->pixel_shift;
 
     mb_xy = h->mb_xy = s->mb_x + s->mb_y*s->mb_stride;
 
@@ -605,7 +613,7 @@ decode_intra_mb:
         align_get_bits(&s->gb);
 
         // The pixels are stored in the same order as levels in h->mb array.
-        for(x=0; x < (CHROMA ? 384 : 256); x++){
+        for(x=0; x < (CHROMA ? 384 : 256)*h->sps.bit_depth_luma/8; x++){
             ((uint8_t*)h->mb)[x]= get_bits(&s->gb, 8);
         }
 
@@ -941,6 +949,8 @@ decode_intra_mb:
         if(IS_INTRA16x16(mb_type)){
             AV_ZERO128(h->mb_luma_dc+0);
             AV_ZERO128(h->mb_luma_dc+8);
+            AV_ZERO128(h->mb_luma_dc+16);
+            AV_ZERO128(h->mb_luma_dc+24);
             if( decode_residual(h, h->intra_gb_ptr, h->mb_luma_dc, LUMA_DC_BLOCK_INDEX, scan, h->dequant4_coeff[0][s->qscale], 16) < 0){
                 return -1; //FIXME continue if partitioned and other return -1 too
             }
@@ -951,7 +961,7 @@ decode_intra_mb:
                 for(i8x8=0; i8x8<4; i8x8++){
                     for(i4x4=0; i4x4<4; i4x4++){
                         const int index= i4x4 + 4*i8x8;
-                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
+                        if( decode_residual(h, h->intra_gb_ptr, h->mb + (16*index << pixel_shift), index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){
                             return -1;
                         }
                     }
@@ -963,7 +973,7 @@ decode_intra_mb:
             for(i8x8=0; i8x8<4; i8x8++){
                 if(cbp & (1<<i8x8)){
                     if(IS_8x8DCT(mb_type)){
-                        DCTELEM *buf = &h->mb[64*i8x8];
+                        DCTELEM *buf = &h->mb[64*i8x8 << pixel_shift];
                         uint8_t *nnz;
                         for(i4x4=0; i4x4<4; i4x4++){
                             if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,
@@ -976,7 +986,7 @@ decode_intra_mb:
                         for(i4x4=0; i4x4<4; i4x4++){
                             const int index= i4x4 + 4*i8x8;
 
-                            if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
+                            if( decode_residual(h, gb, h->mb + (16*index << pixel_shift), index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){
                                 return -1;
                             }
                         }
@@ -990,7 +1000,7 @@ decode_intra_mb:
 
         if(cbp&0x30){
             for(chroma_idx=0; chroma_idx<2; chroma_idx++)
-                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
+                if( decode_residual(h, gb, h->mb + ((256 + 16*4*chroma_idx) << pixel_shift), CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
                     return -1;
                 }
         }
@@ -1000,7 +1010,7 @@ decode_intra_mb:
                 const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp[chroma_idx]];
                 for(i4x4=0; i4x4<4; i4x4++){
                     const int index= 16 + 4*chroma_idx + i4x4;
-                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, qmul, 15) < 0){
+                    if( decode_residual(h, gb, h->mb + (16*index << pixel_shift), index, scan + 1, qmul, 15) < 0){
                         return -1;
                     }
                 }
diff --git a/libavcodec/h264_loopfilter.c b/libavcodec/h264_loopfilter.c
index a08ff69931..72b1905936 100644
--- a/libavcodec/h264_loopfilter.c
+++ b/libavcodec/h264_loopfilter.c
@@ -544,10 +544,10 @@ static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, u
         tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
         //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
         if( dir == 0 ) {
-            filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
+            filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, h );
             if( (edge&1) == 0 ) {
-                filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h);
-                filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h);
+                filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
+                filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
             }
         } else {
             filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
diff --git a/libavcodec/utils.c b/libavcodec/utils.c
index e356ca08cb..9a19cde9cc 100644
--- a/libavcodec/utils.c
+++ b/libavcodec/utils.c
@@ -273,6 +273,7 @@ int avcodec_default_get_buffer(AVCodecContext *s, AVFrame *pic){
         int unaligned;
         AVPicture picture;
         int stride_align[4];
+        const int pixel_size = av_pix_fmt_descriptors[s->pix_fmt].comp[0].step_minus1+1;
 
         avcodec_get_chroma_sub_sample(s->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 
@@ -322,7 +323,7 @@ int avcodec_default_get_buffer(AVCodecContext *s, AVFrame *pic){
             if((s->flags&CODEC_FLAG_EMU_EDGE) || !size[2])
                 buf->data[i] = buf->base[i];
             else
-                buf->data[i] = buf->base[i] + FFALIGN((buf->linesize[i]*EDGE_WIDTH>>v_shift) + (EDGE_WIDTH>>h_shift), stride_align[i]);
+                buf->data[i] = buf->base[i] + FFALIGN((buf->linesize[i]*EDGE_WIDTH>>v_shift) + (pixel_size*EDGE_WIDTH>>h_shift), stride_align[i]);
         }
         if(size[1] && !size[2])
             ff_set_systematic_pal2((uint32_t*)buf->data[1], s->pix_fmt);