passing AVFrame instead of AVPicture around in ffmpeg, that way stuff like motion vectors can be passed from the decoder to the encoder

moving ref_index from Picture to AVFrame

Originally committed as revision 3028 to svn://svn.ffmpeg.org/ffmpeg/trunk

4 files changed, 62 insertions, 30 deletions

diff --git a/ffmpeg.c b/ffmpeg.c
index 6168547b26..5c8a84d010 100644
--- a/ffmpeg.c
+++ b/ffmpeg.c
@@ -223,7 +223,7 @@ typedef struct AVOutputStream {
     int64_t sync_opts;
     /* video only */
     int video_resample;      /* video_resample and video_crop are mutually exclusive */
-    AVPicture pict_tmp;      /* temporary image for resampling */
+    AVFrame pict_tmp;      /* temporary image for resampling */
     ImgReSampleContext *img_resample_ctx; /* for image resampling */
 
     int video_crop;          /* video_resample and video_crop are mutually exclusive */
@@ -554,15 +554,18 @@ static void do_video_out(AVFormatContext *s,
                          int *frame_size, AVOutputStream *audio_sync)
 {
     int nb_frames, i, ret;
-    AVPicture *final_picture, *formatted_picture;
-    AVPicture picture_format_temp, picture_crop_temp;
+    AVFrame *final_picture, *formatted_picture;
+    AVFrame picture_format_temp, picture_crop_temp;
     static uint8_t *video_buffer= NULL;
     uint8_t *buf = NULL, *buf1 = NULL;
     AVCodecContext *enc, *dec;
     enum PixelFormat target_pixfmt;
-
+    
 #define VIDEO_BUFFER_SIZE (1024*1024)
 
+    avcodec_get_frame_defaults(&picture_format_temp);
+    avcodec_get_frame_defaults(&picture_crop_temp);
+
     enc = &ost->st->codec;
     dec = &ist->st->codec;
 
@@ -641,9 +644,9 @@ static void do_video_out(AVFormatContext *s,
         if (!buf)
             return;
         formatted_picture = &picture_format_temp;
-        avpicture_fill(formatted_picture, buf, target_pixfmt, dec->width, dec->height);
+        avpicture_fill((AVPicture*)formatted_picture, buf, target_pixfmt, dec->width, dec->height);
         
-        if (img_convert(formatted_picture, target_pixfmt, 
+        if (img_convert((AVPicture*)formatted_picture, target_pixfmt, 
                         (AVPicture *)in_picture, dec->pix_fmt, 
                         dec->width, dec->height) < 0) {
 
@@ -653,7 +656,7 @@ static void do_video_out(AVFormatContext *s,
             goto the_end;
         }
     } else {
-        formatted_picture = (AVPicture *)in_picture;
+        formatted_picture = in_picture;
     }
 
     /* XXX: resampling could be done before raw format conversion in
@@ -661,10 +664,10 @@ static void do_video_out(AVFormatContext *s,
     /* XXX: only works for YUV420P */
     if (ost->video_resample) {
         final_picture = &ost->pict_tmp;
-        img_resample(ost->img_resample_ctx, final_picture, formatted_picture);
+        img_resample(ost->img_resample_ctx, (AVPicture*)final_picture, (AVPicture*)formatted_picture);
        
         if (ost->padtop || ost->padbottom || ost->padleft || ost->padright) {
-            fill_pad_region(final_picture, enc->height, enc->width,
+            fill_pad_region((AVPicture*)final_picture, enc->height, enc->width,
                     ost->padtop, ost->padbottom, ost->padleft, ost->padright,
                     padcolor);
         }
@@ -679,10 +682,10 @@ static void do_video_out(AVFormatContext *s,
             if (!buf)
                 return;
             final_picture = &picture_format_temp;
-            avpicture_fill(final_picture, buf, enc->pix_fmt, enc->width, enc->height);
+            avpicture_fill((AVPicture*)final_picture, buf, enc->pix_fmt, enc->width, enc->height);
         
-            if (img_convert(final_picture, enc->pix_fmt, 
-                            &ost->pict_tmp, PIX_FMT_YUV420P, 
+            if (img_convert((AVPicture*)final_picture, enc->pix_fmt, 
+                            (AVPicture*)&ost->pict_tmp, PIX_FMT_YUV420P, 
                             enc->width, enc->height) < 0) {
 
                 if (verbose >= 0)
@@ -729,7 +732,7 @@ static void do_video_out(AVFormatContext *s,
             }
         }
 
-        fill_pad_region(final_picture, enc->height, enc->width,
+        fill_pad_region((AVPicture*)final_picture, enc->height, enc->width,
                 ost->padtop, ost->padbottom, ost->padleft, ost->padright,
                 padcolor);
         
@@ -743,10 +746,10 @@ static void do_video_out(AVFormatContext *s,
             if (!buf)
                 return;
             final_picture = &picture_format_temp;
-            avpicture_fill(final_picture, buf, enc->pix_fmt, enc->width, enc->height);
+            avpicture_fill((AVPicture*)final_picture, buf, enc->pix_fmt, enc->width, enc->height);
 
-            if (img_convert(final_picture, enc->pix_fmt, 
-                        &ost->pict_tmp, PIX_FMT_YUV420P, 
+            if (img_convert((AVPicture*)final_picture, enc->pix_fmt, 
+                        (AVPicture*)&ost->pict_tmp, PIX_FMT_YUV420P, 
                         enc->width, enc->height) < 0) {
 
                 if (verbose >= 0)
@@ -772,9 +775,8 @@ static void do_video_out(AVFormatContext *s,
 	    enc->coded_frame = old_frame;
         } else {
             AVFrame big_picture;
-            
-            avcodec_get_frame_defaults(&big_picture);
-            *(AVPicture*)&big_picture= *final_picture;
+
+            big_picture= *final_picture;
             /* better than nothing: use input picture interlaced
                settings */
             big_picture.interlaced_frame = in_picture->interlaced_frame;
@@ -791,6 +793,7 @@ static void do_video_out(AVFormatContext *s,
                 big_picture.quality = ist->st->quality;
             }else
                 big_picture.quality = ost->st->quality;
+            big_picture.pict_type = 0;
             ret = avcodec_encode_video(enc, 
                                        video_buffer, VIDEO_BUFFER_SIZE,
                                        &big_picture);
@@ -1419,15 +1422,16 @@ static int av_encode(AVFormatContext **output_files,
                     ost->padleft = frame_padleft;
                     ost->padbottom = frame_padbottom;
                     ost->padright = frame_padright;
-                    if( avpicture_alloc( &ost->pict_tmp, PIX_FMT_YUV420P,
+                    if( avpicture_alloc( (AVPicture*)&ost->pict_tmp, PIX_FMT_YUV420P,
                                 codec->width, codec->height ) )
                         goto fail;
                 } else {
                     ost->video_resample = 1;
                     ost->video_crop = 0; // cropping is handled as part of resample
-                    if( avpicture_alloc( &ost->pict_tmp, PIX_FMT_YUV420P,
+                    if( avpicture_alloc( (AVPicture*)&ost->pict_tmp, PIX_FMT_YUV420P,
                                          codec->width, codec->height ) )
                         goto fail;
+                    avcodec_get_frame_defaults(&ost->pict_tmp);
 
                     ost->img_resample_ctx = img_resample_full_init( 
                                       ost->st->codec.width, ost->st->codec.height,
diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index 396d399ef9..a8acae326a 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -17,7 +17,7 @@ extern "C" {
 
 #define FFMPEG_VERSION_INT     0x000408
 #define FFMPEG_VERSION         "0.4.8"
-#define LIBAVCODEC_BUILD       4708
+#define LIBAVCODEC_BUILD       4709
 
 #define LIBAVCODEC_VERSION_INT FFMPEG_VERSION_INT
 #define LIBAVCODEC_VERSION     FFMPEG_VERSION
@@ -442,7 +442,7 @@ typedef struct AVPanScan{
 \
     /**\
      * Motion vector table\
-     * - encoding: unused\
+     * - encoding: set by user\
      * - decoding: set by lavc\
      */\
     int16_t (*motion_val[2])[2];\
@@ -450,7 +450,7 @@ typedef struct AVPanScan{
     /**\
      * Macroblock type table\
      * mb_type_base + mb_width + 2\
-     * - encoding: unused\
+     * - encoding: set by user\
      * - decoding: set by lavc\
      */\
     uint32_t *mb_type;\
@@ -538,13 +538,20 @@ typedef struct AVPanScan{
      * - decoding: set by lavc\
      */\
     short *dct_coeff;\
+\
+    /**\
+     * Motion referece frame index\
+     * - encoding: set by user\
+     * - decoding: set by lavc\
+     */\
+    int8_t *ref_index[2];
 
 #define FF_QSCALE_TYPE_MPEG1	0
 #define FF_QSCALE_TYPE_MPEG2	1
 
 #define FF_BUFFER_TYPE_INTERNAL 1
 #define FF_BUFFER_TYPE_USER     2 ///< Direct rendering buffers (image is (de)allocated by user)
-#define FF_BUFFER_TYPE_SHARED   4 ///< buffer from somewher else, dont dealloc image (data/base)
+#define FF_BUFFER_TYPE_SHARED   4 ///< buffer from somewher else, dont dealloc image (data/base), all other tables are not shared
 #define FF_BUFFER_TYPE_COPY     8 ///< just a (modified) copy of some other buffer, dont dealloc anything
 
 
@@ -1561,6 +1568,14 @@ typedef struct AVCodecContext {
      * - decoding: set by execute()
      */
     void *thread_opaque;
+
+    /**
+     * Motion estimation threshold.
+     * 
+     * - encoding: set by user
+     * - decoding: set by user
+     */
+    void *me_threshold;
 } AVCodecContext;
 
 
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index c15c55fac4..e88db4cc2a 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -287,15 +287,28 @@ static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *sr
     dst->interlaced_frame       = src->interlaced_frame;
     dst->top_field_first        = src->top_field_first;
 
-    if(src->motion_val[0] && src->motion_val[0] != dst->motion_val[0]){
+    if(s->avctx->me_threshold){
+        if(!src->motion_val[0])
+            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
+        if(!src->mb_type)
+            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
+        if(!src->ref_index[0])
+            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
         if(src->motion_subsample_log2 != dst->motion_subsample_log2)
             av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesnt match!\n");
-        else{
+
+        memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
+        
+        for(i=0; i<2; i++){
             int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
             int height= ((16*s->mb_height)>>src->motion_subsample_log2);
 
-            for(i=0; i<2; i++)
-                memcpy(dst->motion_val[i], src->motion_val[i], stride*height*sizeof(int16_t));
+            if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
+                memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
+            }
+            if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
+                memcpy(dst->ref_index[i], src->ref_index[i], s->mb_stride*s->mb_height*sizeof(int8_t)); //FIXME init this too
+            }
         }
     }
 }
@@ -363,6 +376,7 @@ static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
             for(i=0; i<2; i++){
                 CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b8_array_size+2) * sizeof(int16_t))
                 pic->motion_val[i]= pic->motion_val_base[i]+2;
+                CHECKED_ALLOCZ(pic->ref_index[i], mb_array_size * sizeof(int8_t))
             }
             pic->motion_subsample_log2= 3;
         }
diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h
index 51153fd38b..a759feae74 100644
--- a/libavcodec/mpegvideo.h
+++ b/libavcodec/mpegvideo.h
@@ -138,7 +138,6 @@ typedef struct Picture{
      */
     uint8_t *interpolated[3];
     int16_t (*motion_val_base[2])[2];
-    int8_t *ref_index[2];
     uint32_t *mb_type_base;
 #define MB_TYPE_INTRA MB_TYPE_INTRA4x4 //default mb_type if theres just one type
 #define IS_INTRA4x4(a)   ((a)&MB_TYPE_INTRA4x4)