added skip macroblock optimization (big perf win on black regions for example)

Originally committed as revision 13 to svn://svn.ffmpeg.org/ffmpeg/trunk

5 files changed, 38 insertions, 0 deletions

diff --git a/libavcodec/h263.c b/libavcodec/h263.c
index 35ecb15c2e..03e4aa381b 100644
--- a/libavcodec/h263.c
+++ b/libavcodec/h263.c
@@ -738,6 +738,7 @@ int h263_decode_mb(MpegEncContext *s,
             s->mv_type = MV_TYPE_16X16;
             s->mv[0][0][0] = 0;
             s->mv[0][0][1] = 0;
+            s->mb_skiped = 1;
             return 0;
         }
         cbpc = get_vlc(&s->gb, &inter_MCBPC_vlc);
diff --git a/libavcodec/mpeg12.c b/libavcodec/mpeg12.c
index 5b5411b1d4..e9e1e27e1a 100644
--- a/libavcodec/mpeg12.c
+++ b/libavcodec/mpeg12.c
@@ -542,6 +542,7 @@ static int mpeg_decode_mb(MpegEncContext *s,
             s->mv[1][0][0] = s->last_mv[1][0][0];
             s->mv[1][0][1] = s->last_mv[1][0][1];
         }
+        s->mb_skiped = 1;
         return 0;
     }
 
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index 467062ddf6..dc053de426 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -171,6 +171,13 @@ int MPV_common_init(MpegEncContext *s)
         s->non_intra_matrix[i] = default_non_intra_matrix[i];
         s->chroma_non_intra_matrix[i] = default_non_intra_matrix[i];
     }
+    /* init macroblock skip table */
+    if (!s->encoding) {
+        s->mbskip_table = av_mallocz(s->mb_width * s->mb_height);
+        if (!s->mbskip_table)
+            goto fail;
+    }
+
     s->context_initialized = 1;
     return 0;
  fail:
@@ -182,6 +189,8 @@ int MPV_common_init(MpegEncContext *s)
         free(s->ac_val[0]);
     if (s->coded_block)
         free(s->coded_block);
+    if (s->mbskip_table)
+        free(s->mbskip_table);
     for(i=0;i<3;i++) {
         if (s->last_picture_base[i])
             free(s->last_picture_base[i]);
@@ -205,6 +214,8 @@ void MPV_common_end(MpegEncContext *s)
         free(s->ac_val[0]);
         free(s->coded_block);
     }
+    if (s->mbskip_table)
+        free(s->mbskip_table);
     for(i=0;i<3;i++) {
         free(s->last_picture_base[i]);
         free(s->next_picture_base[i]);
@@ -276,6 +287,8 @@ int MPV_encode_init(AVCodecContext *avctx)
     if (s->out_format == FMT_H263)
         h263_encode_init_vlc(s);
 
+    s->encoding = 1;
+
     /* init */
     if (MPV_common_init(s) < 0)
         return -1;
@@ -712,6 +725,21 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
     
     if (!s->intra_only) {
         UINT8 *dest_y, *dest_cb, *dest_cr;
+        UINT8 *mbskip_ptr;
+
+        /* avoid copy if macroblock skipped in last frame too */
+        if (!s->encoding) {
+            mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x];
+            if (s->mb_skiped) {
+                s->mb_skiped = 0;
+                /* if previous was skipped too, then nothing to do ! */
+                if (*mbskip_ptr != 0) 
+                    goto the_end;
+                *mbskip_ptr = 1; /* indicate that this time we skiped it */
+            } else {
+                *mbskip_ptr = 0; /* not skipped */
+            }
+        }
 
         dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize) + mb_x * 16;
         dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
@@ -762,6 +790,8 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
             put_dct(s, block[5], 5, dest_cr, dct_linesize >> 1);
         }
     }
+ the_end:
+    emms_c();
 }
 
 static void encode_picture(MpegEncContext *s, int picture_number)
@@ -885,6 +915,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                 ptr = s->last_picture[2] + offset;
                 sub_pixels_2(block[5], ptr, s->linesize >> 1, dxy);
             }
+            emms_c();
 
             /* DCT & quantize */
             if (s->h263_msmpeg4) {
diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h
index b04d7b7693..a097b01b00 100644
--- a/libavcodec/mpegvideo.h
+++ b/libavcodec/mpegvideo.h
@@ -44,6 +44,7 @@ typedef struct MpegEncContext {
     int h263_msmpeg4; /* generate MSMPEG4 compatible stream */
     int h263_intel; /* use I263 intel h263 header */
     int fixed_qscale; /* fixed qscale if non zero */
+    int encoding;     /* true if we are encoding (vs decoding) */
     /* the following fields are managed internally by the encoder */
 
     /* bit output */
@@ -70,6 +71,9 @@ typedef struct MpegEncContext {
     UINT8 *coded_block; /* used for coded block pattern prediction */
     INT16 (*ac_val[3])[16]; /* used for for mpeg4 AC prediction */
     int ac_pred;
+    int mb_skiped;              /* MUST BE SET only during DECODING */
+    UINT8 *mbskip_table;        /* used to avoid copy if macroblock
+                                   skipped (for black regions for example) */
 
     int qscale;
     int pict_type;
diff --git a/libavcodec/msmpeg4.c b/libavcodec/msmpeg4.c
index 044d964ee5..ac14da68f3 100644
--- a/libavcodec/msmpeg4.c
+++ b/libavcodec/msmpeg4.c
@@ -694,6 +694,7 @@ int msmpeg4_decode_mb(MpegEncContext *s,
                 s->mv_type = MV_TYPE_16X16;
                 s->mv[0][0][0] = 0;
                 s->mv[0][0][1] = 0;
+                s->mb_skiped = 1;
                 return 0;
             }
         }