* include/video_output.h, include/vlc_video.h, src/video_output/*:

   + When dealing with slow direct buffers, use a temporary picture to overlay subtitles (avoids slow reads during alpha-blending).
* modules/video_output/directx/directx.c: signal slow picture buffers.

include/video_output.h

@@ -123,7 +123,7 @@ struct vout_thread_t
     /**@}*/
 
     /* Picture and subpicture heaps */
-    picture_t           p_picture[2*VOUT_MAX_PICTURES];        /**< pictures */
+    picture_t           p_picture[2*VOUT_MAX_PICTURES+1];      /**< pictures */
     subpicture_t        p_subpicture[VOUT_MAX_PICTURES];    /**< subpictures */
     subpicture_t        *p_default_channel;   /**< subpicture in the default
                                                    channel */

include/vlc_video.h

@@ -74,6 +74,7 @@ struct picture_t
      * @{*/
     int             i_status;                             /**< picture flags */
     int             i_type;                /**< is picture a direct buffer ? */
+    vlc_bool_t      b_slow;                 /**< is picture in slow memory ? */
     int             i_matrix_coefficients;   /**< in YUV type, encoding type */
     /**@}*/
 

modules/video_output/directx/directx.c

@@ -1533,6 +1533,7 @@ static int NewPictureVec( vout_thread_t *p_vout, picture_t *p_pic,
     {
         p_pic[i].i_status = DESTROYED_PICTURE;
         p_pic[i].i_type   = DIRECT_PICTURE;
+        p_pic[i].b_slow   = VLC_TRUE;
         p_pic[i].pf_lock  = DirectXLockSurface;
         p_pic[i].pf_unlock = DirectXUnlockSurface;
         PP_OUTPUTPICTURE[i] = &p_pic[i];

src/video_output/video_output.c

@@ -236,12 +236,13 @@ vout_thread_t * __vout_Create( vlc_object_t *p_parent,
 
     /* Initialize pictures and subpictures - translation tables and functions
      * will be initialized later in InitThread */
-    for( i_index = 0; i_index < 2 * VOUT_MAX_PICTURES; i_index++)
+    for( i_index = 0; i_index < 2 * VOUT_MAX_PICTURES + 1; i_index++)
     {
         p_vout->p_picture[i_index].pf_lock = NULL;
         p_vout->p_picture[i_index].pf_unlock = NULL;
         p_vout->p_picture[i_index].i_status = FREE_PICTURE;
         p_vout->p_picture[i_index].i_type   = EMPTY_PICTURE;
+        p_vout->p_picture[i_index].b_slow   = 0;
     }
 
     for( i_index = 0; i_index < VOUT_MAX_SUBPICTURES; i_index++)
@@ -1114,7 +1115,7 @@ static void EndThread( vout_thread_t *p_vout )
     }
 
     /* Destroy all remaining pictures */
-    for( i_index = 0; i_index < 2 * VOUT_MAX_PICTURES; i_index++ )
+    for( i_index = 0; i_index < 2 * VOUT_MAX_PICTURES + 1; i_index++ )
     {
         if ( p_vout->p_picture[i_index].i_type == MEMORY_PICTURE )
         {

src/video_output/vout_pictures.c

@@ -178,6 +178,7 @@ picture_t *vout_CreatePicture( vout_thread_t *p_vout,
             /* Copy picture information, set some default values */
             p_freepic->i_status   = RESERVED_PICTURE;
             p_freepic->i_type     = MEMORY_PICTURE;
+            p_freepic->b_slow     = 0;
 
             p_freepic->i_refcount = 0;
             p_freepic->b_force = 0;
@@ -346,9 +347,41 @@ picture_t * vout_RenderPicture( vout_thread_t *p_vout, picture_t *p_pic,
                 return NULL;
             }
 
-        CopyPicture( p_vout, p_pic, PP_OUTPUTPICTURE[0] );
+        if( p_subpic == NULL ||
+            ( p_vout->render.b_allow_modify_pics && !p_pic->i_refcount ) )
+        {
+            vout_RenderSubPictures( p_vout, p_pic, p_subpic );
+            CopyPicture( p_vout, p_pic, PP_OUTPUTPICTURE[0] );
+	}
+        else if( PP_OUTPUTPICTURE[0]->b_slow )
+        {
+            /* The picture buffer is in slow memory. We'll use
+             * the "2 * VOUT_MAX_PICTURES + 1" picture as a temporary
+             * one for subpictures rendering. */
+            picture_t *p_tmp_pic = &p_vout->p_picture[2 * VOUT_MAX_PICTURES];
+            if( p_tmp_pic->i_status == FREE_PICTURE )
+            {
+                vout_AllocatePicture( VLC_OBJECT(p_vout),
+                                      p_tmp_pic, p_vout->render.i_chroma,
+                                      p_vout->render.i_width,
+                                      p_vout->render.i_height,
+                                      p_vout->render.i_aspect );
+                p_tmp_pic->i_type = MEMORY_PICTURE;
+                p_tmp_pic->i_status = RESERVED_PICTURE;
+            }
+
+            CopyPicture( p_vout, p_pic, p_tmp_pic );
+
+            vout_RenderSubPictures( p_vout, p_tmp_pic, p_subpic );
 
-        vout_RenderSubPictures( p_vout, PP_OUTPUTPICTURE[0], p_subpic );
+            CopyPicture( p_vout, p_tmp_pic, PP_OUTPUTPICTURE[0] );
+        }
+        else
+        {
+            CopyPicture( p_vout, p_pic, PP_OUTPUTPICTURE[0] );
+
+            vout_RenderSubPictures( p_vout, PP_OUTPUTPICTURE[0], p_subpic );
+        }
 
         if( PP_OUTPUTPICTURE[0]->pf_unlock )
             PP_OUTPUTPICTURE[0]->pf_unlock( p_vout, PP_OUTPUTPICTURE[0] );
@@ -361,15 +394,47 @@ picture_t * vout_RenderPicture( vout_thread_t *p_vout, picture_t *p_pic,
      * well. This usually means software YUV, or hardware YUV with a
      * different chroma. */
 
-    if( p_vout->p_picture[0].pf_lock )
-        if( p_vout->p_picture[0].pf_lock( p_vout, &p_vout->p_picture[0] ) )
-            return NULL;
+    if( p_subpic != NULL && p_vout->p_picture[0].b_slow )
+    {
+        /* The picture buffer is in slow memory. We'll use
+         * the "2 * VOUT_MAX_PICTURES + 1" picture as a temporary
+         * one for subpictures rendering. */
+        picture_t *p_tmp_pic = &p_vout->p_picture[2 * VOUT_MAX_PICTURES];
+        if( p_tmp_pic->i_status == FREE_PICTURE )
+        {
+            vout_AllocatePicture( VLC_OBJECT(p_vout),
+                                  p_tmp_pic, p_vout->output.i_chroma,
+                                  p_vout->output.i_width,
+                                  p_vout->output.i_height,
+                                  p_vout->output.i_aspect );
+            p_tmp_pic->i_type = MEMORY_PICTURE;
+            p_tmp_pic->i_status = RESERVED_PICTURE;
+        }
+
+        /* Convert image to the first direct buffer */
+        p_vout->chroma.pf_convert( p_vout, p_pic, p_tmp_pic );
+
+        /* Render subpictures on the first direct buffer */
+        vout_RenderSubPictures( p_vout, p_tmp_pic, p_subpic );
+
+        if( p_vout->p_picture[0].pf_lock )
+            if( p_vout->p_picture[0].pf_lock( p_vout, &p_vout->p_picture[0] ) )
+                return NULL;
+
+        CopyPicture( p_vout, p_tmp_pic, &p_vout->p_picture[0] );
+    }
+    else
+    {
+        if( p_vout->p_picture[0].pf_lock )
+            if( p_vout->p_picture[0].pf_lock( p_vout, &p_vout->p_picture[0] ) )
+                return NULL;
 
-    /* Convert image to the first direct buffer */
-    p_vout->chroma.pf_convert( p_vout, p_pic, &p_vout->p_picture[0] );
+        /* Convert image to the first direct buffer */
+        p_vout->chroma.pf_convert( p_vout, p_pic, &p_vout->p_picture[0] );
 
-    /* Render subpictures on the first direct buffer */
-    vout_RenderSubPictures( p_vout, &p_vout->p_picture[0], p_subpic );
+        /* Render subpictures on the first direct buffer */
+        vout_RenderSubPictures( p_vout, &p_vout->p_picture[0], p_subpic );
+    }
 
     if( p_vout->p_picture[0].pf_unlock )
         p_vout->p_picture[0].pf_unlock( p_vout, &p_vout->p_picture[0] );