]> git.sesse.net Git - vlc/blobdiff - modules/codec/theora.c
* fix author field of OSX info dialog
[vlc] / modules / codec / theora.c
index ae8ab3bc8ebb4bc33b2f33900543dad01abfb49a..0353b0467ea3022a3cacc0a83fea140f4984f0f5 100644 (file)
@@ -53,6 +53,11 @@ struct decoder_sys_t
     theora_comment   tc;                            /* theora comment header */
     theora_state     td;                   /* theora bitstream user comments */
 
+    /*
+     * Decoding properties
+     */
+    vlc_bool_t b_decoded_first_keyframe;
+
     /*
      * Common properties
      */
@@ -88,6 +93,8 @@ static block_t *Encode( encoder_t *p_enc, picture_t *p_pict );
   "of specifying a particular bitrate. This will produce a VBR stream." )
 
 vlc_module_begin();
+    set_category( CAT_INPUT );
+    set_subcategory( SUBCAT_INPUT_VCODEC );
     set_description( _("Theora video decoder") );
     set_capability( "decoder", 100 );
     set_callbacks( OpenDecoder, CloseDecoder );
@@ -137,6 +144,7 @@ static int OpenDecoder( vlc_object_t *p_this )
     p_dec->p_sys->b_packetizer = VLC_FALSE;
 
     p_sys->i_pts = 0;
+    p_sys->b_decoded_first_keyframe = VLC_FALSE;
 
     /* Set output properties */
     p_dec->fmt_out.i_cat = VIDEO_ES;
@@ -276,6 +284,11 @@ static int ProcessHeaders( decoder_t *p_dec )
     /* Set output properties */
     p_dec->fmt_out.video.i_width = p_sys->ti.width;
     p_dec->fmt_out.video.i_height = p_sys->ti.height;
+    if( p_sys->ti.frame_width && p_sys->ti.frame_height )
+    {
+        p_dec->fmt_out.video.i_width = p_sys->ti.frame_width;
+        p_dec->fmt_out.video.i_height = p_sys->ti.frame_height;
+    }
 
     if( p_sys->ti.aspect_denominator && p_sys->ti.aspect_numerator )
     {
@@ -289,6 +302,12 @@ static int ProcessHeaders( decoder_t *p_dec )
             p_sys->ti.frame_width / p_sys->ti.frame_height;
     }
 
+    if( p_sys->ti.fps_numerator > 0 && p_sys->ti.fps_denominator > 0 )
+    {
+        p_dec->fmt_out.video.i_frame_rate = p_sys->ti.fps_numerator;
+        p_dec->fmt_out.video.i_frame_rate_base = p_sys->ti.fps_denominator;
+    }
+
     msg_Dbg( p_dec, "%dx%d %.02f fps video, frame content "
              "is %dx%d with offset (%d,%d)",
              p_sys->ti.width, p_sys->ti.height,
@@ -367,6 +386,14 @@ static void *ProcessPacket( decoder_t *p_dec, ogg_packet *p_oggpacket,
     block_t *p_block = *pp_block;
     void *p_buf;
 
+    if( ( p_block->i_flags & BLOCK_FLAG_DISCONTINUITY ) != 0 )
+    {
+        /* Don't send the the first packet after a discontinuity to
+         * theora_decode, otherwise we get purple/green display artifacts
+         * appearing in the video output */
+        return NULL;
+    }
+
     /* Date management */
     if( p_block->i_pts > 0 && p_block->i_pts != p_sys->i_pts )
     {
@@ -415,8 +442,20 @@ static picture_t *DecodePacket( decoder_t *p_dec, ogg_packet *p_oggpacket )
 
     theora_decode_packetin( &p_sys->td, p_oggpacket );
 
-    /* Decode */
-    theora_decode_YUVout( &p_sys->td, &yuv );
+    /* Check for keyframe */
+    if( !(p_oggpacket->packet[0] & 0x80) /* data packet */ &&
+        !(p_oggpacket->packet[0] & 0x40) /* intra frame */ )
+        p_sys->b_decoded_first_keyframe = VLC_TRUE;
+
+    /* If we haven't seen a single keyframe yet, don't let Theora decode
+     * anything, otherwise we'll get display artifacts.  (This is impossible
+     * in the general case, but can happen if e.g. we play a network stream
+     * using a timed URL, such that the server doesn't start the video with a
+     * keyframe). */
+    if( p_sys->b_decoded_first_keyframe )
+        theora_decode_YUVout( &p_sys->td, &yuv );
+    else
+        return NULL;
 
     /* Get a new picture */
     p_pic = p_dec->pf_vout_buffer_new( p_dec );
@@ -504,7 +543,7 @@ static void theora_CopyPicture( decoder_t *p_dec, picture_t *p_pic,
 
         p_src += (i_src_yoffset * i_src_stride + i_src_yoffset);
 
-        for( i_line = 0; i_line < p_pic->p[i_plane].i_lines; i_line++ )
+        for( i_line = 0; i_line < p_pic->p[i_plane].i_visible_lines; i_line++ )
         {
             p_dec->p_vlc->pf_memcpy( p_dst, p_src, i_width );
             p_src += i_src_stride;
@@ -530,10 +569,7 @@ struct encoder_sys_t
     theora_comment   tc;                            /* theora comment header */
     theora_state     td;                   /* theora bitstream user comments */
 
-    /*
-     * Common properties
-     */
-    mtime_t i_pts;
+    int i_width, i_height;
 };
 
 /*****************************************************************************
@@ -554,15 +590,6 @@ static int OpenEncoder( vlc_object_t *p_this )
         return VLC_EGENERIC;
     }
 
-    if( p_enc->fmt_in.video.i_width % 16 ||
-        p_enc->fmt_in.video.i_height % 16 )
-    {
-        msg_Err( p_enc, "Theora video encoding requires dimensions which are "
-                 "multiples of 16. Which is not the case here (%dx%d).",
-                 p_enc->fmt_in.video.i_width, p_enc->fmt_in.video.i_height );
-        return VLC_EGENERIC;
-    }
-
     /* Allocate the memory needed to store the decoder's structure */
     if( ( p_sys = (encoder_sys_t *)malloc(sizeof(encoder_sys_t)) ) == NULL )
     {
@@ -586,10 +613,26 @@ static int OpenEncoder( vlc_object_t *p_this )
 
     p_sys->ti.width = p_enc->fmt_in.video.i_width;
     p_sys->ti.height = p_enc->fmt_in.video.i_height;
+
+    if( p_sys->ti.width % 16 || p_sys->ti.height % 16 )
+    {
+        /* Pictures from the transcoder should always have a pitch
+         * which is a multiple of 16 */
+        p_sys->ti.width = (p_sys->ti.width + 15) >> 4 << 4;
+        p_sys->ti.height = (p_sys->ti.height + 15) >> 4 << 4;
+
+        msg_Dbg( p_enc, "padding video from %dx%d to %dx%d",
+                 p_enc->fmt_in.video.i_width, p_enc->fmt_in.video.i_height,
+                 p_sys->ti.width, p_sys->ti.height );
+    }
+
     p_sys->ti.frame_width = p_enc->fmt_in.video.i_width;
     p_sys->ti.frame_height = p_enc->fmt_in.video.i_height;
     p_sys->ti.offset_x = 0 /*frame_x_offset*/;
-    p_sys->ti.offset_y = 0/*frame_y_offset*/;
+    p_sys->ti.offset_y = 0 /*frame_y_offset*/;
+
+    p_sys->i_width = p_sys->ti.width;
+    p_sys->i_height = p_sys->ti.height;
 
     if( !p_enc->fmt_in.video.i_frame_rate ||
         !p_enc->fmt_in.video.i_frame_rate_base )
@@ -605,9 +648,14 @@ static int OpenEncoder( vlc_object_t *p_this )
 
     if( p_enc->fmt_in.video.i_aspect )
     {
-        p_sys->ti.aspect_numerator =
-            p_enc->fmt_in.video.i_aspect * p_sys->ti.height / p_sys->ti.width;
-        p_sys->ti.aspect_denominator = VOUT_ASPECT_FACTOR;
+        int64_t i_num, i_den;
+        int i_dst_num, i_dst_den;
+
+        i_num = p_enc->fmt_in.video.i_aspect * (int64_t)p_sys->ti.height;
+        i_den = VOUT_ASPECT_FACTOR * p_sys->ti.width;
+        vlc_reduce( &i_dst_num, &i_dst_den, i_num, i_den, 0 );
+        p_sys->ti.aspect_numerator = i_dst_num;
+        p_sys->ti.aspect_denominator = i_dst_den;
     }
     else
     {
@@ -667,16 +715,70 @@ static block_t *Encode( encoder_t *p_enc, picture_t *p_pict )
     ogg_packet oggpacket;
     block_t *p_block;
     yuv_buffer yuv;
+    int i;
+
+    /* Sanity check */
+    if( p_pict->p[0].i_pitch < (int)p_sys->i_width ||
+        p_pict->p[0].i_lines < (int)p_sys->i_height )
+    {
+        msg_Warn( p_enc, "frame is smaller than encoding size"
+                  "(%ix%i->%ix%i) -> dropping frame",
+                  p_pict->p[0].i_pitch, p_pict->p[0].i_lines,
+                  p_sys->i_width, p_sys->i_height );
+        return NULL;
+    }
+
+    /* Fill padding */
+    if( p_pict->p[0].i_visible_pitch < (int)p_sys->i_width )
+    {
+        for( i = 0; i < p_sys->i_height; i++ )
+        {
+            memset( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch +
+                    p_pict->p[0].i_visible_pitch,
+                    *( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch +
+                       p_pict->p[0].i_visible_pitch - 1 ),
+                    p_sys->i_width - p_pict->p[0].i_visible_pitch );
+        }
+        for( i = 0; i < p_sys->i_height / 2; i++ )
+        {
+            memset( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch +
+                    p_pict->p[1].i_visible_pitch,
+                    *( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch +
+                       p_pict->p[1].i_visible_pitch - 1 ),
+                    p_sys->i_width / 2 - p_pict->p[1].i_visible_pitch );
+            memset( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch +
+                    p_pict->p[2].i_visible_pitch,
+                    *( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch +
+                       p_pict->p[2].i_visible_pitch - 1 ),
+                    p_sys->i_width / 2 - p_pict->p[2].i_visible_pitch );
+        }
+    }
+
+    if( p_pict->p[0].i_visible_lines < (int)p_sys->i_height )
+    {
+        for( i = p_pict->p[0].i_visible_lines; i < p_sys->i_height; i++ )
+        {
+            memset( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch, 0,
+                    p_sys->i_width );
+        }
+        for( i = p_pict->p[1].i_visible_lines; i < p_sys->i_height / 2; i++ )
+        {
+            memset( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch, 0x80,
+                    p_sys->i_width / 2 );
+            memset( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch, 0x80,
+                    p_sys->i_width / 2 );
+        }
+    }
 
     /* Theora is a one-frame-in, one-frame-out system. Submit a frame
      * for compression and pull out the packet. */
 
-    yuv.y_width  = p_pict->p[0].i_visible_pitch;
-    yuv.y_height = p_pict->p[0].i_lines;
+    yuv.y_width  = p_sys->i_width;
+    yuv.y_height = p_sys->i_height;
     yuv.y_stride = p_pict->p[0].i_pitch;
 
-    yuv.uv_width  = p_pict->p[1].i_visible_pitch;
-    yuv.uv_height = p_pict->p[1].i_lines;
+    yuv.uv_width  = p_sys->i_width / 2;
+    yuv.uv_height = p_sys->i_height / 2;
     yuv.uv_stride = p_pict->p[1].i_pitch;
 
     yuv.y = p_pict->p[0].p_pixels;