X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fh264.h;h=caea7ba7eb0200820c742e672f6121cb926ab6f9;hb=f666276fa61623f4d6fa97b99d6b336ec0eba8c3;hp=ba813b66ba5077bca7b5c73002330a4b2a71df2e;hpb=32e543f866d9d4b450729e93cd81dacd8c457971;p=ffmpeg

diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index ba813b66ba5..caea7ba7eb0 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -2,25 +2,25 @@
  * H.26L/H.264/AVC/JVT/14496-10/... encoder/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
  */
 
 /**
- * @file libavcodec/h264.h
+ * @file
  * H.264 / AVC / MPEG4 part10 codec.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
@@ -39,16 +39,6 @@
 #define interlaced_dct interlaced_dct_is_a_bad_name
 #define mb_intra mb_intra_is_not_initialized_see_mb_type
 
-#define LUMA_DC_BLOCK_INDEX   25
-#define CHROMA_DC_BLOCK_INDEX 26
-
-#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
-#define COEFF_TOKEN_VLC_BITS           8
-#define TOTAL_ZEROS_VLC_BITS           9
-#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
-#define RUN_VLC_BITS                   3
-#define RUN7_VLC_BITS                  6
-
 #define MAX_SPS_COUNT 32
 #define MAX_PPS_COUNT 256
 
@@ -60,8 +50,6 @@
  * of progressive decoding by about 2%. */
 #define ALLOW_INTERLACE
 
-#define ALLOW_NOCHROMA
-
 #define FMO 0
 
 /**
@@ -75,6 +63,10 @@
 #define MB_FIELD h->mb_field_decoding_flag
 #define FRAME_MBAFF h->mb_aff_frame
 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
+#define LEFT_MBS 2
+#define LTOP 0
+#define LBOT 1
+#define LEFT(i) (i)
 #else
 #define MB_MBAFF 0
 #define MB_FIELD 0
@@ -82,19 +74,20 @@
 #define FIELD_PICTURE 0
 #undef  IS_INTERLACED
 #define IS_INTERLACED(mb_type) 0
+#define LEFT_MBS 1
+#define LTOP 0
+#define LBOT 0
+#define LEFT(i) 0
 #endif
 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
 
-#ifdef ALLOW_NOCHROMA
-#define CHROMA h->sps.chroma_format_idc
-#else
-#define CHROMA 1
-#endif
-
 #ifndef CABAC
 #define CABAC h->pps.cabac
 #endif
 
+#define CHROMA422 (h->sps.chroma_format_idc == 2)
+#define CHROMA444 (h->sps.chroma_format_idc == 3)
+
 #define EXTENDED_SAR          255
 
 #define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
@@ -108,6 +101,7 @@
  */
 #define DELAYED_PIC_REF 4
 
+#define QP_MAX_NUM (51 + 2*6)           // The maximum supported qp
 
 /* NAL unit types */
 enum {
@@ -197,7 +191,7 @@ typedef struct SPS{
     int num_reorder_frames;
     int scaling_matrix_present;
     uint8_t scaling_matrix4[6][16];
-    uint8_t scaling_matrix8[2][64];
+    uint8_t scaling_matrix8[6][64];
     int nal_hrd_parameters_present_flag;
     int vcl_hrd_parameters_present_flag;
     int pic_struct_present_flag;
@@ -209,6 +203,7 @@ typedef struct SPS{
     int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
     int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
     int residual_color_transform_flag; ///< residual_colour_transform_flag
+    int constraint_set_flags;          ///< constraint_set[0-3]_flag
 }SPS;
 
 /**
@@ -231,7 +226,7 @@ typedef struct PPS{
     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
     int transform_8x8_mode;     ///< transform_8x8_mode_flag
     uint8_t scaling_matrix4[6][16];
-    uint8_t scaling_matrix8[2][64];
+    uint8_t scaling_matrix8[6][64];
     uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
     int chroma_qp_diff;
 }PPS;
@@ -264,6 +259,7 @@ 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
@@ -278,12 +274,12 @@ typedef struct H264Context{
     int topleft_mb_xy;
     int top_mb_xy;
     int topright_mb_xy;
-    int left_mb_xy[2];
+    int left_mb_xy[LEFT_MBS];
 
     int topleft_type;
     int top_type;
     int topright_type;
-    int left_type[2];
+    int left_type[LEFT_MBS];
 
     const uint8_t * left_block;
     int topleft_partition;
@@ -295,21 +291,15 @@ 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*3)*2];
 
     /**
      * non zero coeff count cache.
      * is 64 if not available.
      */
-    DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
+    DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15*8];
 
-    /*
-    .UU.YYYY
-    .UU.YYYY
-    .vv.YYYY
-    .VV.YYYY
-    */
-    uint8_t (*non_zero_count)[32];
+    uint8_t (*non_zero_count)[48];
 
     /**
      * Motion vector cache.
@@ -319,11 +309,6 @@ typedef struct H264Context{
 #define LIST_NOT_USED -1 //FIXME rename?
 #define PART_NOT_AVAILABLE -2
 
-    /**
-     * is 1 if the specific list MV&references are set to 0,0,-2.
-     */
-    int mv_cache_clean[2];
-
     /**
      * number of neighbors (top and/or left) that used 8x8 dct
      */
@@ -333,7 +318,7 @@ typedef struct H264Context{
      * block_offset[ 0..23] for frame macroblocks
      * block_offset[24..47] for field macroblocks
      */
-    int block_offset[2*(16+8)];
+    int block_offset[2*(16*3)];
 
     uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
     uint32_t *mb2br_xy;
@@ -352,10 +337,10 @@ typedef struct H264Context{
      */
     PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
 
-    uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
-    uint32_t dequant8_buffer[2][52][64];
+    uint32_t dequant4_buffer[6][QP_MAX_NUM+1][16]; //FIXME should these be moved down?
+    uint32_t dequant8_buffer[6][QP_MAX_NUM+1][64];
     uint32_t (*dequant4_coeff[6])[16];
-    uint32_t (*dequant8_coeff[2])[64];
+    uint32_t (*dequant8_coeff[6])[64];
 
     int slice_num;
     uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
@@ -378,7 +363,7 @@ typedef struct H264Context{
     //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
     int luma_weight[48][2][2];
     int chroma_weight[48][2][2][2];
-    int implicit_weight[48][48];
+    int implicit_weight[48][48][2];
 
     int direct_spatial_mv_pred;
     int col_parity;
@@ -405,14 +390,15 @@ typedef struct H264Context{
     GetBitContext *intra_gb_ptr;
     GetBitContext *inter_gb_ptr;
 
-    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
-    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*48*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)[3][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
      */
     CABACContext cabac;
-    uint8_t      cabac_state[460];
+    uint8_t      cabac_state[1024];
 
     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
     uint16_t     *cbp_table;
@@ -464,6 +450,7 @@ typedef struct H264Context{
      */
     int is_avc; ///< this flag is != 0 if codec is avc1
     int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
+    int got_first; ///< this flag is != 0 if we've parsed a frame
 
     SPS *sps_buffers[MAX_SPS_COUNT];
     PPS *pps_buffers[MAX_PPS_COUNT];
@@ -501,7 +488,10 @@ typedef struct H264Context{
     Picture *long_ref[32];
     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
+    int last_pocs[MAX_DELAYED_PIC_COUNT];
+    Picture *next_output_pic;
     int outputed_poc;
+    int next_outputed_poc;
 
     /**
      * memory management control operations buffer.
@@ -515,7 +505,7 @@ typedef struct H264Context{
     int          cabac_init_idc;
 
     /**
-     * @defgroup multithreading Members for slice based multithreading
+     * @name Members for slice based multithreading
      * @{
      */
     struct H264Context *thread_context[MAX_THREADS];
@@ -588,20 +578,11 @@ typedef struct H264Context{
     int sei_buffering_period_present;  ///< Buffering period SEI flag
     int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
 
-    //SVQ3 specific fields
-    int halfpel_flag;
-    int thirdpel_flag;
-    int unknown_svq3_flag;
-    int next_slice_index;
-    uint32_t svq3_watermark_key;
+    int cur_chroma_format_idc;
 }H264Context;
 
 
-extern const uint8_t ff_h264_chroma_qp[52];
-
-void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
-
-void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
+extern const uint8_t ff_h264_chroma_qp[3][QP_MAX_NUM+1]; ///< One chroma qp table for each supported bit depth (8, 9, 10).
 
 /**
  * Decode SEI
@@ -613,13 +594,18 @@ int ff_h264_decode_sei(H264Context *h);
  */
 int ff_h264_decode_seq_parameter_set(H264Context *h);
 
+/**
+ * compute profile from sps
+ */
+int ff_h264_get_profile(SPS *sps);
+
 /**
  * Decode PPS
  */
 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
 
 /**
- * Decodes a network abstraction layer unit.
+ * Decode a network abstraction layer unit.
  * @param consumed is the number of bytes used as input
  * @param length is the length of the array
  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
@@ -628,29 +614,23 @@ int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
 
 /**
- * identifies the exact end of the bitstream
- * @return the length of the trailing, or 0 if damaged
- */
-int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
-
-/**
- * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
+ * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
  */
 av_cold void ff_h264_free_context(H264Context *h);
 
 /**
- * reconstructs bitstream slice_type.
+ * Reconstruct bitstream slice_type.
  */
 int ff_h264_get_slice_type(const H264Context *h);
 
 /**
- * allocates tables.
+ * Allocate tables.
  * needs width/height
  */
 int ff_h264_alloc_tables(H264Context *h);
 
 /**
- * fills the default_ref_list.
+ * Fill the default_ref_list.
  */
 int ff_h264_fill_default_ref_list(H264Context *h);
 
@@ -659,38 +639,40 @@ void ff_h264_fill_mbaff_ref_list(H264Context *h);
 void ff_h264_remove_all_refs(H264Context *h);
 
 /**
- * Executes the reference picture marking (memory management control operations).
+ * Execute the reference picture marking (memory management control operations).
  */
 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
 
 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
 
+void ff_generate_sliding_window_mmcos(H264Context *h);
+
 
 /**
- * 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 & change the dc mode so it only uses the available blocks.
  */
 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
 
 /**
- * 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 & change the dc mode so it only uses the available blocks.
  */
 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
 
-void ff_h264_write_back_intra_pred_mode(H264Context *h);
 void ff_h264_hl_decode_mb(H264Context *h);
 int ff_h264_frame_start(H264Context *h);
+int ff_h264_decode_extradata(H264Context *h);
 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
 av_cold void ff_h264_decode_init_vlc(void);
 
 /**
- * decodes a macroblock
+ * Decode a macroblock
  * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
  */
 int ff_h264_decode_mb_cavlc(H264Context *h);
 
 /**
- * decodes a CABAC coded macroblock
+ * Decode a CABAC coded macroblock
  * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
  */
 int ff_h264_decode_mb_cabac(H264Context *h);
@@ -721,16 +703,45 @@ o-o o-o
  / / /
 o-o o-o
 */
+
+/* Scan8 organization:
+ *    0 1 2 3 4 5 6 7
+ * 0  DY    y y y y y
+ * 1        y Y Y Y Y
+ * 2        y Y Y Y Y
+ * 3        y Y Y Y Y
+ * 4        y Y Y Y Y
+ * 5  DU    u u u u u
+ * 6        u U U U U
+ * 7        u U U U U
+ * 8        u U U U U
+ * 9        u U U U U
+ * 10 DV    v v v v v
+ * 11       v V V V V
+ * 12       v V V V V
+ * 13       v V V V V
+ * 14       v V V V V
+ * DY/DU/DV are for luma/chroma DC.
+ */
+
+#define LUMA_DC_BLOCK_INDEX   48
+#define CHROMA_DC_BLOCK_INDEX 49
+
 //This table must be here because scan8[constant] must be known at compiletime
-static const uint8_t scan8[16 + 2*4]={
- 4+1*8, 5+1*8, 4+2*8, 5+2*8,
- 6+1*8, 7+1*8, 6+2*8, 7+2*8,
- 4+3*8, 5+3*8, 4+4*8, 5+4*8,
- 6+3*8, 7+3*8, 6+4*8, 7+4*8,
- 1+1*8, 2+1*8,
- 1+2*8, 2+2*8,
- 1+4*8, 2+4*8,
- 1+5*8, 2+5*8,
+static const uint8_t scan8[16*3 + 3]={
+ 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
+ 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
+ 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
+ 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
+ 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
+ 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
+ 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
+ 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
+ 4+11*8, 5+11*8, 4+12*8, 5+12*8,
+ 6+11*8, 7+11*8, 6+12*8, 7+12*8,
+ 4+13*8, 5+13*8, 4+14*8, 5+14*8,
+ 6+13*8, 7+13*8, 6+14*8, 7+14*8,
+ 0+ 0*8, 0+ 5*8, 0+10*8
 };
 
 static av_always_inline uint32_t pack16to32(int a, int b){
@@ -752,637 +763,14 @@ static av_always_inline uint16_t pack8to16(int a, int b){
 /**
  * gets the chroma qp.
  */
-static inline int get_chroma_qp(H264Context *h, int t, int qscale){
+static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
     return h->pps.chroma_qp_table[t][qscale];
 }
 
-static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
-
-static void fill_decode_neighbors(H264Context *h, int mb_type){
-    MpegEncContext * const s = &h->s;
-    const int mb_xy= h->mb_xy;
-    int topleft_xy, top_xy, topright_xy, left_xy[2];
-    static const uint8_t left_block_options[4][16]={
-        {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8},
-        {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8},
-        {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8},
-        {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}
-    };
-
-    h->topleft_partition= -1;
-
-    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
-
-    /* Wow, what a mess, why didn't they simplify the interlacing & intra
-     * stuff, I can't imagine that these complex rules are worth it. */
-
-    topleft_xy = top_xy - 1;
-    topright_xy= top_xy + 1;
-    left_xy[1] = left_xy[0] = mb_xy-1;
-    h->left_block = left_block_options[0];
-    if(FRAME_MBAFF){
-        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.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[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
-                if (curr_mb_field_flag) {
-                    left_xy[1] += s->mb_stride;
-                    h->left_block = left_block_options[3];
-                } else {
-                    topleft_xy += s->mb_stride;
-                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
-                    h->topleft_partition = 0;
-                    h->left_block = left_block_options[1];
-                }
-            }
-        }else{
-            if(curr_mb_field_flag){
-                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
-                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
-                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
-            }
-            if (left_mb_field_flag != curr_mb_field_flag) {
-                if (curr_mb_field_flag) {
-                    left_xy[1] += s->mb_stride;
-                    h->left_block = left_block_options[3];
-                } else {
-                    h->left_block = left_block_options[2];
-                }
-            }
-        }
-    }
-
-    h->topleft_mb_xy = topleft_xy;
-    h->top_mb_xy     = top_xy;
-    h->topright_mb_xy= topright_xy;
-    h->left_mb_xy[0] = left_xy[0];
-    h->left_mb_xy[1] = left_xy[1];
-    //FIXME do we need all in the context?
-
-    h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
-    h->top_type     = s->current_picture.mb_type[top_xy]     ;
-    h->topright_type= s->current_picture.mb_type[topright_xy];
-    h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
-    h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
-
-    if(FMO){
-    if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
-    if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
-    if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
-    }else{
-        if(h->slice_table[topleft_xy ] != h->slice_num){
-            h->topleft_type = 0;
-            if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
-            if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
-        }
-    }
-    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
-}
-
-static void fill_decode_caches(H264Context *h, int mb_type){
-    MpegEncContext * const s = &h->s;
-    int topleft_xy, top_xy, topright_xy, left_xy[2];
-    int topleft_type, top_type, topright_type, left_type[2];
-    const uint8_t * left_block= h->left_block;
-    int i;
-
-    topleft_xy   = h->topleft_mb_xy ;
-    top_xy       = h->top_mb_xy     ;
-    topright_xy  = h->topright_mb_xy;
-    left_xy[0]   = h->left_mb_xy[0] ;
-    left_xy[1]   = h->left_mb_xy[1] ;
-    topleft_type = h->topleft_type  ;
-    top_type     = h->top_type      ;
-    topright_type= h->topright_type ;
-    left_type[0] = h->left_type[0]  ;
-    left_type[1] = h->left_type[1]  ;
-
-    if(!IS_SKIP(mb_type)){
-        if(IS_INTRA(mb_type)){
-            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
-            h->topleft_samples_available=
-            h->top_samples_available=
-            h->left_samples_available= 0xFFFF;
-            h->topright_samples_available= 0xEEEA;
-
-            if(!(top_type & type_mask)){
-                h->topleft_samples_available= 0xB3FF;
-                h->top_samples_available= 0x33FF;
-                h->topright_samples_available= 0x26EA;
-            }
-            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
-                if(IS_INTERLACED(mb_type)){
-                    if(!(left_type[0] & type_mask)){
-                        h->topleft_samples_available&= 0xDFFF;
-                        h->left_samples_available&= 0x5FFF;
-                    }
-                    if(!(left_type[1] & type_mask)){
-                        h->topleft_samples_available&= 0xFF5F;
-                        h->left_samples_available&= 0xFF5F;
-                    }
-                }else{
-                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
-
-                    assert(left_xy[0] == left_xy[1]);
-                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
-                        h->topleft_samples_available&= 0xDF5F;
-                        h->left_samples_available&= 0x5F5F;
-                    }
-                }
-            }else{
-                if(!(left_type[0] & type_mask)){
-                    h->topleft_samples_available&= 0xDF5F;
-                    h->left_samples_available&= 0x5F5F;
-                }
-            }
-
-            if(!(topleft_type & type_mask))
-                h->topleft_samples_available&= 0x7FFF;
-
-            if(!(topright_type & type_mask))
-                h->topright_samples_available&= 0xFBFF;
-
-            if(IS_INTRA4x4(mb_type)){
-                if(IS_INTRA4x4(top_type)){
-                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
-                }else{
-                    h->intra4x4_pred_mode_cache[4+8*0]=
-                    h->intra4x4_pred_mode_cache[5+8*0]=
-                    h->intra4x4_pred_mode_cache[6+8*0]=
-                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
-                }
-                for(i=0; i<2; i++){
-                    if(IS_INTRA4x4(left_type[i])){
-                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
-                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
-                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
-                    }else{
-                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
-                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
-                    }
-                }
-            }
-        }
-
-
-/*
-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]);
-            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
-            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
-
-            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
-            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
-    }else {
-            h->non_zero_count_cache[1+8*0]=
-            h->non_zero_count_cache[2+8*0]=
-
-            h->non_zero_count_cache[1+8*3]=
-            h->non_zero_count_cache[2+8*3]=
-            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
-    }
-
-    for (i=0; i<2; i++) {
-        if(left_type[i]){
-            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
-            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
-                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
-                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
-        }else{
-                h->non_zero_count_cache[3+8*1 + 2*8*i]=
-                h->non_zero_count_cache[3+8*2 + 2*8*i]=
-                h->non_zero_count_cache[0+8*1 +   8*i]=
-                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
-        }
-    }
-
-    if( CABAC ) {
-        // top_cbp
-        if(top_type) {
-            h->top_cbp = h->cbp_table[top_xy];
-        } else {
-            h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
-        }
-        // left_cbp
-        if (left_type[0]) {
-            h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
-                        |  ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
-                        | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
-        } else {
-            h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
-        }
-    }
-    }
-
-#if 1
-    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
-        int list;
-        for(list=0; list<h->list_count; list++){
-            if(!USES_LIST(mb_type, list)){
-                /*if(!h->mv_cache_clean[list]){
-                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
-                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
-                    h->mv_cache_clean[list]= 1;
-                }*/
-                continue;
-            }
-            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
-
-            h->mv_cache_clean[list]= 0;
-
-            if(USES_LIST(top_type, list)){
-                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
-                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]= s->current_picture.ref_index[list][4*top_xy + 2];
-                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
-                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
-            }else{
-                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
-                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
-            }
-
-            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
-            for(i=0; i<2; i++){
-                int cache_idx = scan8[0] - 1 + i*2*8;
-                if(USES_LIST(left_type[i], list)){
-                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
-                    const int b8_xy= 4*left_xy[i] + 1;
-                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
-                    AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]);
-                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
-                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
-                }else{
-                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
-                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
-                    h->ref_cache[list][cache_idx  ]=
-                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-                }
-            }
-            }else{
-                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;
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
-                    h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
-                }else{
-                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
-                    h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-                }
-            }
-
-            if(USES_LIST(topright_type, list)){
-                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
-                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
-                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
-            }else{
-                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
-                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-            }
-            if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
-                if(USES_LIST(topleft_type, list)){
-                    const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
-                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
-                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
-                    h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
-                }else{
-                    AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
-                    h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
-                }
-            }
-
-            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
-                continue;
-
-            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
-            h->ref_cache[list][scan8[4 ]] =
-            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
-            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
-            AV_ZERO32(h->mv_cache [list][scan8[12]]);
-
-            if( CABAC ) {
-                /* XXX beurk, Load mvd */
-                if(USES_LIST(top_type, list)){
-                    const int b_xy= h->mb2br_xy[top_xy];
-                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
-                }else{
-                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
-                }
-                if(USES_LIST(left_type[0], list)){
-                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
-                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
-                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
-                }else{
-                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
-                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
-                }
-                if(USES_LIST(left_type[1], list)){
-                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
-                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
-                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
-                }else{
-                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
-                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
-                }
-                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
-                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
-                if(h->slice_type_nos == FF_B_TYPE){
-                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
-
-                    if(IS_DIRECT(top_type)){
-                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
-                    }else if(IS_8X8(top_type)){
-                        int b8_xy = 4*top_xy;
-                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
-                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
-                    }else{
-                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
-                    }
-
-                    if(IS_DIRECT(left_type[0]))
-                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
-                    else if(IS_8X8(left_type[0]))
-                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
-                    else
-                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
-
-                    if(IS_DIRECT(left_type[1]))
-                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
-                    else if(IS_8X8(left_type[1]))
-                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
-                    else
-                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
-                }
-            }
-            }
-            if(FRAME_MBAFF){
-#define MAP_MVS\
-                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
-                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
-                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
-                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
-                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
-                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
-                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
-                if(MB_FIELD){
-#define MAP_F2F(idx, mb_type)\
-                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
-                        h->ref_cache[list][idx] <<= 1;\
-                        h->mv_cache[list][idx][1] /= 2;\
-                        h->mvd_cache[list][idx][1] >>=1;\
-                    }
-                    MAP_MVS
-#undef MAP_F2F
-                }else{
-#define MAP_F2F(idx, mb_type)\
-                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
-                        h->ref_cache[list][idx] >>= 1;\
-                        h->mv_cache[list][idx][1] <<= 1;\
-                        h->mvd_cache[list][idx][1] <<= 1;\
-                    }
-                    MAP_MVS
-#undef MAP_F2F
-                }
-            }
-        }
-    }
-#endif
-
-        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
-}
-
-/**
- *
- * @return non zero if the loop filter can be skiped
- */
-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];
-
-    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;
-    if(FRAME_MBAFF){
-        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.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;
-            }
-        }else{
-            if(curr_mb_field_flag){
-                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
-            }
-            if (left_mb_field_flag != curr_mb_field_flag) {
-                left_xy[1] += 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];
-    {
-        //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];
-        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)){
-            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))
-                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]];
-    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;
-    }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;
-    }
-    h->top_type    = top_type    ;
-    h->left_type[0]= left_type[0];
-    h->left_type[1]= left_type[1];
-
-    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]);
-
-    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]);
-    }
-
-    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];
-    }
-
-    // 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;
-        }
-        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[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(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;
-
-            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;
-
-            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;
-
-            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;
-                }
-            }
-        }
-    }
-
-    return 0;
-}
-
 /**
  * gets the predicted intra4x4 prediction mode.
  */
-static inline int pred_intra_mode(H264Context *h, int n){
+static av_always_inline int pred_intra_mode(H264Context *h, int n){
     const int index8= scan8[n];
     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
@@ -1394,62 +782,87 @@ static inline int pred_intra_mode(H264Context *h, int n){
     else      return min;
 }
 
-static inline void write_back_non_zero_count(H264Context *h){
+static av_always_inline void write_back_intra_pred_mode(H264Context *h){
+    int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
+    int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
+
+    AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
+    i4x4[4]= i4x4_cache[7+8*3];
+    i4x4[5]= i4x4_cache[7+8*2];
+    i4x4[6]= i4x4_cache[7+8*1];
+}
+
+static av_always_inline void write_back_non_zero_count(H264Context *h){
     const int mb_xy= h->mb_xy;
+    uint8_t *nnz = h->non_zero_count[mb_xy];
+    uint8_t *nnz_cache = h->non_zero_count_cache;
+
+    AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
+    AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
+    AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
+    AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
+    AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
+    AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
+    AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
+    AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
+
+    if(!h->s.chroma_y_shift){
+        AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
+        AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
+        AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
+        AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
+    }
+}
 
-    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
-    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
-    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
-    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
-    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
+static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
+                                                    int b_xy, int b8_xy, int mb_type, int list )
+{
+    int16_t (*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
+    int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
+    AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
+    AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
+    AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
+    AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
+    if( CABAC ) {
+        uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
+        uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
+        if(IS_SKIP(mb_type))
+            AV_ZERO128(mvd_dst);
+        else{
+            AV_COPY64(mvd_dst, mvd_src + 8*3);
+            AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
+            AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
+            AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
+        }
+    }
+
+    {
+        int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
+        int8_t *ref_cache = h->ref_cache[list];
+        ref_index[0+0*2]= ref_cache[scan8[0]];
+        ref_index[1+0*2]= ref_cache[scan8[4]];
+        ref_index[0+1*2]= ref_cache[scan8[8]];
+        ref_index[1+1*2]= ref_cache[scan8[12]];
+    }
 }
 
-static inline void write_back_motion(H264Context *h, int mb_type){
+static av_always_inline void write_back_motion(H264Context *h, int mb_type){
     MpegEncContext * const s = &h->s;
+    const int b_stride = h->b_stride;
     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
     const int b8_xy= 4*h->mb_xy;
-    int list;
-
-    if(!USES_LIST(mb_type, 0))
-        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
 
-    for(list=0; list<h->list_count; list++){
-        int y, b_stride;
-        int16_t (*mv_dst)[2];
-        int16_t (*mv_src)[2];
-
-        if(!USES_LIST(mb_type, list))
-            continue;
-
-        b_stride = h->b_stride;
-        mv_dst   = &s->current_picture.motion_val[list][b_xy];
-        mv_src   = &h->mv_cache[list][scan8[0]];
-        for(y=0; y<4; y++){
-            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
-        }
-        if( CABAC ) {
-            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
-            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
-            if(IS_SKIP(mb_type))
-                AV_ZERO128(mvd_dst);
-            else{
-            AV_COPY64(mvd_dst, mvd_src + 8*3);
-                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
-                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
-                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
-            }
-        }
-
-        {
-            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
-            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
-            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
-            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
-            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
-        }
+    if(USES_LIST(mb_type, 0)){
+        write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
+    }else{
+        fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
+                       2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
+    }
+    if(USES_LIST(mb_type, 1)){
+        write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
     }
 
-    if(h->slice_type_nos == FF_B_TYPE && CABAC){
+    if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
         if(IS_8X8(mb_type)){
             uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
             direct_table[1] = h->sub_mb_type[1]>>1;
@@ -1459,57 +872,11 @@ static inline void write_back_motion(H264Context *h, int mb_type){
     }
 }
 
-static inline int get_dct8x8_allowed(H264Context *h){
+static av_always_inline int get_dct8x8_allowed(H264Context *h){
     if(h->sps.direct_8x8_inference_flag)
         return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
     else
         return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
 }
 
-/**
- * decodes a P_SKIP or B_SKIP macroblock
- */
-static void decode_mb_skip(H264Context *h){
-    MpegEncContext * const s = &h->s;
-    const int mb_xy= h->mb_xy;
-    int mb_type=0;
-
-    memset(h->non_zero_count[mb_xy], 0, 32);
-    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
-
-    if(MB_FIELD)
-        mb_type|= MB_TYPE_INTERLACED;
-
-    if( h->slice_type_nos == FF_B_TYPE )
-    {
-        // just for fill_caches. pred_direct_motion will set the real mb_type
-        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
-        if(h->direct_spatial_mv_pred){
-            fill_decode_neighbors(h, mb_type);
-        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
-        }
-        ff_h264_pred_direct_motion(h, &mb_type);
-        mb_type|= MB_TYPE_SKIP;
-    }
-    else
-    {
-        int mx, my;
-        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
-
-        fill_decode_neighbors(h, mb_type);
-        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
-        pred_pskip_motion(h, &mx, &my);
-        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
-        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
-    }
-
-    write_back_motion(h, mb_type);
-    s->current_picture.mb_type[mb_xy]= mb_type;
-    s->current_picture.qscale_table[mb_xy]= s->qscale;
-    h->slice_table[ mb_xy ]= h->slice_num;
-    h->prev_mb_skipped= 1;
-}
-
-#include "h264_mvpred.h" //For pred_pskip_motion()
-
 #endif /* AVCODEC_H264_H */