Add missing #includes to fix 'make checkheaders'.

[ffmpeg] / libavcodec / motion_est.c

diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c
index d81ab32a4457869c9da5ea4bda7a64d6f10d7bcd..16db0f87c9e75511225636b3db4a229dbc8b850e 100644 (file)
--- a/libavcodec/motion_est.c
+++ b/libavcodec/motion_est.c
@@ -3,6 +3,7 @@
  * Copyright (c) 2000,2001 Fabrice Bellard.
  * Copyright (c) 2002-2004 Michael Niedermayer
  *
+ * new motion estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
  *
  * This file is part of FFmpeg.
  *
@@ -19,8 +20,6 @@
  * You should have received a copy of the GNU Lesser General Public
  * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
- * new Motion Estimation (X1/EPZS) by Michael Niedermayer <michaelni@gmx.at>
  */
 
 /**
@@ -103,6 +102,9 @@ static int get_flags(MotionEstContext *c, int direct, int chroma){
            + (chroma ? FLAG_CHROMA : 0);
 }
 
+/*! \brief compares a block (either a full macroblock or a partition thereof)
+    against a proposed motion-compensated prediction of that block
+ */
 static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, const int subx, const int suby,
                       const int size, const int h, int ref_index, int src_index,
                       me_cmp_func cmp_func, me_cmp_func chroma_cmp_func, const int flags){
@@ -119,6 +121,7 @@ static av_always_inline int cmp(MpegEncContext *s, const int x, const int y, con
     int d;
     //FIXME check chroma 4mv, (no crashes ...)
     if(flags&FLAG_DIRECT){
+        assert(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1));
         if(x >= c->xmin && hx <= c->xmax<<(qpel+1) && y >= c->ymin && hy <= c->ymax<<(qpel+1)){
             const int time_pp= s->pp_time;
             const int time_pb= s->pb_time;
@@ -230,8 +233,14 @@ static void zero_hpel(uint8_t *a, const uint8_t *b, int stride, int h){
 
 void ff_init_me(MpegEncContext *s){
     MotionEstContext * const c= &s->me;
+    int cache_size= FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
+    int dia_size= FFMAX(FFABS(s->avctx->dia_size)&255, FFABS(s->avctx->pre_dia_size)&255);
     c->avctx= s->avctx;
 
+    if(cache_size < 2*dia_size && !c->stride){
+        av_log(s->avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
+    }
+
     ff_set_cmp(&s->dsp, s->dsp.me_pre_cmp, c->avctx->me_pre_cmp);
     ff_set_cmp(&s->dsp, s->dsp.me_cmp, c->avctx->me_cmp);
     ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, c->avctx->me_sub_cmp);
@@ -269,7 +278,9 @@ void ff_init_me(MpegEncContext *s){
         c->uvstride=  8*s->mb_width + 16;
     }
 
-    // 8x8 fullpel search would need a 4x4 chroma compare, which we dont have yet, and even if we had the motion estimation code doesnt expect it
+    /* 8x8 fullpel search would need a 4x4 chroma compare, which we do
+     * not have yet, and even if we had, the motion estimation code
+     * does not expect it. */
     if(s->codec_id != CODEC_ID_SNOW){
         if((c->avctx->me_cmp&FF_CMP_CHROMA)/* && !s->dsp.me_cmp[2]*/){
             s->dsp.me_cmp[2]= zero_cmp;
@@ -1791,15 +1802,15 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
     c->pred_x=0;
     c->pred_y=0;
 
-    P_LEFT[0]        = clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
-    P_LEFT[1]        = clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
+    P_LEFT[0]        = av_clip(mv_table[mot_xy - 1][0], xmin<<shift, xmax<<shift);
+    P_LEFT[1]        = av_clip(mv_table[mot_xy - 1][1], ymin<<shift, ymax<<shift);
 
     /* special case for first line */
-    if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as its clipped
-        P_TOP[0]      = clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);
-        P_TOP[1]      = clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);
-        P_TOPRIGHT[0] = clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);
-        P_TOPRIGHT[1] = clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);
+    if (!s->first_slice_line) { //FIXME maybe allow this over thread boundary as it is clipped
+        P_TOP[0]      = av_clip(mv_table[mot_xy - mot_stride             ][0], xmin<<shift, xmax<<shift);
+        P_TOP[1]      = av_clip(mv_table[mot_xy - mot_stride             ][1], ymin<<shift, ymax<<shift);
+        P_TOPRIGHT[0] = av_clip(mv_table[mot_xy - mot_stride + 1         ][0], xmin<<shift, xmax<<shift);
+        P_TOPRIGHT[1] = av_clip(mv_table[mot_xy - mot_stride + 1         ][1], ymin<<shift, ymax<<shift);
 
         P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
         P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
@@ -1816,8 +1827,8 @@ static inline int direct_search(MpegEncContext * s, int mb_x, int mb_y)
 
     get_limits(s, 16*mb_x, 16*mb_y); //restore c->?min/max, maybe not needed
 
-    s->b_direct_mv_table[mot_xy][0]= mx;
-    s->b_direct_mv_table[mot_xy][1]= my;
+    mv_table[mot_xy][0]= mx;
+    mv_table[mot_xy][1]= my;
     c->flags     &= ~FLAG_DIRECT;
     c->sub_flags &= ~FLAG_DIRECT;
 
@@ -1837,6 +1848,18 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
     get_limits(s, 16*mb_x, 16*mb_y);
 
     c->skip=0;
+
+    if(s->codec_id == CODEC_ID_MPEG4 && s->next_picture.mbskip_table[xy]){
+        int score= direct_search(s, mb_x, mb_y); //FIXME just check 0,0
+
+        score= ((unsigned)(score*score + 128*256))>>16;
+        c->mc_mb_var_sum_temp += score;
+        s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
+        s->mb_type[mb_y*s->mb_stride + mb_x]= CANDIDATE_MB_TYPE_DIRECT0;
+
+        return;
+    }
+
     if(c->avctx->me_threshold){
         int vard= check_input_motion(s, mb_x, mb_y, 0);
 
@@ -1958,7 +1981,9 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
             type |= CANDIDATE_MB_TYPE_BIDIR_I;
         }
          //FIXME something smarter
-        if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
+        if(dmin>256*256*16) type&= ~CANDIDATE_MB_TYPE_DIRECT; //do not try direct mode if it is invalid for this MB
+        if(s->codec_id == CODEC_ID_MPEG4 && type&CANDIDATE_MB_TYPE_DIRECT && s->flags&CODEC_FLAG_MV0 && *(uint32_t*)s->b_direct_mv_table[xy])
+            type |= CANDIDATE_MB_TYPE_DIRECT0;
 #if 0
         if(s->out_format == FMT_MPEG1)
             type |= CANDIDATE_MB_TYPE_INTRA;