git.sesse.net Git - vlc/blob - modules/video_filter/deinterlace.c

   1 /*****************************************************************************
   2  * deinterlace.c : deinterlacer plugin for vlc
   3  *****************************************************************************
   4  * Copyright (C) 2000, 2001, 2002, 2003 the VideoLAN team
   5  * $Id$
   6  *
   7  * Author: Sam Hocevar <sam@zoy.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 /*****************************************************************************
  25  * Preamble
  26  *****************************************************************************/
  27 #include <errno.h>
  28 #include <stdlib.h>                                      /* malloc(), free() */
  29 #include <string.h>
  30
  31 #include <vlc/vlc.h>
  32 #include <vlc/vout.h>
  33 #include <vlc/sout.h>
  34 #include "vlc_filter.h"
  35
  36 #ifdef HAVE_ALTIVEC_H
  37 #   include <altivec.h>
  38 #endif
  39
  40 #ifdef CAN_COMPILE_MMXEXT
  41 #   include "mmx.h"
  42 #endif
  43
  44 #include "filter_common.h"
  45
  46 #define DEINTERLACE_DISCARD 1
  47 #define DEINTERLACE_MEAN    2
  48 #define DEINTERLACE_BLEND   3
  49 #define DEINTERLACE_BOB     4
  50 #define DEINTERLACE_LINEAR  5
  51 #define DEINTERLACE_X       6
  52
  53 /*****************************************************************************
  54  * Local protypes
  55  *****************************************************************************/
  56 static int  Create    ( vlc_object_t * );
  57 static void Destroy   ( vlc_object_t * );
  58
  59 static int  Init      ( vout_thread_t * );
  60 static void End       ( vout_thread_t * );
  61 static void Render    ( vout_thread_t *, picture_t * );
  62
  63 static void RenderDiscard( vout_thread_t *, picture_t *, picture_t *, int );
  64 static void RenderBob    ( vout_thread_t *, picture_t *, picture_t *, int );
  65 static void RenderMean   ( vout_thread_t *, picture_t *, picture_t * );
  66 static void RenderBlend  ( vout_thread_t *, picture_t *, picture_t * );
  67 static void RenderLinear ( vout_thread_t *, picture_t *, picture_t *, int );
  68 static void RenderX      ( vout_thread_t *, picture_t *, picture_t * );
  69
  70 static void MergeGeneric ( void *, const void *, const void *, size_t );
  71 #if defined(CAN_COMPILE_C_ALTIVEC)
  72 static void MergeAltivec ( void *, const void *, const void *, size_t );
  73 #endif
  74 #if defined(CAN_COMPILE_MMXEXT)
  75 static void MergeMMX     ( void *, const void *, const void *, size_t );
  76 #endif
  77 #if defined(CAN_COMPILE_SSE)
  78 static void MergeSSE2    ( void *, const void *, const void *, size_t );
  79 #endif
  80 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
  81 static void EndMMX       ( void );
  82 #endif
  83
  84 static int  SendEvents   ( vlc_object_t *, char const *,
  85                            vlc_value_t, vlc_value_t, void * );
  86
  87 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method );
  88 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout );
  89
  90 static int OpenFilter( vlc_object_t *p_this );
  91 static void CloseFilter( vlc_object_t *p_this );
  92
  93 /*****************************************************************************
  94  * Callback prototypes
  95  *****************************************************************************/
  96 static int FilterCallback ( vlc_object_t *, char const *,
  97                             vlc_value_t, vlc_value_t, void * );
  98
  99 /*****************************************************************************
 100  * Module descriptor
 101  *****************************************************************************/
 102 #define MODE_TEXT N_("Deinterlace mode")
 103 #define MODE_LONGTEXT N_("Default deinterlace method to use for local playback")
 104
 105 #define SOUT_MODE_TEXT N_("Deinterlace mode")
 106 #define SOUT_MODE_LONGTEXT N_("Default deinterlace methode to use for streaming")
 107
 108 #define FILTER_CFG_PREFIX "sout-deinterlace-"
 109
 110 static char *mode_list[] = { "discard", "blend", "mean", "bob", "linear", "x" };
 111 static char *mode_list_text[] = { N_("Discard"), N_("Blend"), N_("Mean"),
 112                                   N_("Bob"), N_("Linear"), "X" };
 113
 114 vlc_module_begin();
 115     set_description( _("Deinterlacing video filter") );
 116     set_shortname( N_("Deinterlace" ));
 117     set_capability( "video filter", 0 );
 118     set_category( CAT_VIDEO );
 119     set_subcategory( SUBCAT_VIDEO_VFILTER );
 120
 121     set_section( N_("Display"),NULL);
 122     add_string( "deinterlace-mode", "discard", NULL, MODE_TEXT,
 123                 MODE_LONGTEXT, VLC_FALSE );
 124         change_string_list( mode_list, mode_list_text, 0 );
 125
 126     add_shortcut( "deinterlace" );
 127     set_callbacks( Create, Destroy );
 128
 129     add_submodule();
 130     set_capability( "video filter2", 0 );
 131     set_section( N_("Streaming"),NULL);
 132     add_string( FILTER_CFG_PREFIX "mode", "blend", NULL, SOUT_MODE_TEXT,
 133                 SOUT_MODE_LONGTEXT, VLC_FALSE );
 134         change_string_list( mode_list, mode_list_text, 0 );
 135     set_callbacks( OpenFilter, CloseFilter );
 136 vlc_module_end();
 137
 138 static const char *ppsz_filter_options[] = {
 139     "mode", NULL
 140 };
 141
 142 /*****************************************************************************
 143  * vout_sys_t: Deinterlace video output method descriptor
 144  *****************************************************************************
 145  * This structure is part of the video output thread descriptor.
 146  * It describes the Deinterlace specific properties of an output thread.
 147  *****************************************************************************/
 148 struct vout_sys_t
 149 {
 150     int        i_mode;        /* Deinterlace mode */
 151     vlc_bool_t b_double_rate; /* Shall we double the framerate? */
 152
 153     mtime_t    last_date;
 154     mtime_t    next_date;
 155
 156     vout_thread_t *p_vout;
 157
 158     vlc_mutex_t filter_lock;
 159
 160     void (*pf_merge) ( void *, const void *, const void *, size_t );
 161     void (*pf_end_merge) ( void );
 162 };
 163
 164 /*****************************************************************************
 165  * Control: control facility for the vout (forwards to child vout)
 166  *****************************************************************************/
 167 static int Control( vout_thread_t *p_vout, int i_query, va_list args )
 168 {
 169     return vout_vaControl( p_vout->p_sys->p_vout, i_query, args );
 170 }
 171
 172 /*****************************************************************************
 173  * Create: allocates Deinterlace video thread output method
 174  *****************************************************************************
 175  * This function allocates and initializes a Deinterlace vout method.
 176  *****************************************************************************/
 177 static int Create( vlc_object_t *p_this )
 178 {
 179     vout_thread_t *p_vout = (vout_thread_t *)p_this;
 180     vlc_value_t val;
 181
 182     /* Allocate structure */
 183     p_vout->p_sys = malloc( sizeof( vout_sys_t ) );
 184     if( p_vout->p_sys == NULL )
 185     {
 186         msg_Err( p_vout, "out of memory" );
 187         return VLC_ENOMEM;
 188     }
 189
 190     p_vout->pf_init = Init;
 191     p_vout->pf_end = End;
 192     p_vout->pf_manage = NULL;
 193     p_vout->pf_render = Render;
 194     p_vout->pf_display = NULL;
 195     p_vout->pf_control = Control;
 196
 197     p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
 198     p_vout->p_sys->b_double_rate = VLC_FALSE;
 199     p_vout->p_sys->last_date = 0;
 200     p_vout->p_sys->p_vout = 0;
 201     vlc_mutex_init( p_vout, &p_vout->p_sys->filter_lock );
 202
 203 #if defined(CAN_COMPILE_C_ALTIVEC)
 204     if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_ALTIVEC )
 205     {
 206         p_vout->p_sys->pf_merge = MergeAltivec;
 207         p_vout->p_sys->pf_end_merge = NULL;
 208     }
 209     else
 210 #endif
 211 #if defined(CAN_COMPILE_SSE)
 212     if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_SSE2 )
 213     {
 214         p_vout->p_sys->pf_merge = MergeSSE2;
 215         p_vout->p_sys->pf_end_merge = EndMMX;
 216     }
 217     else
 218 #endif
 219 #if defined(CAN_COMPILE_MMXEXT)
 220     if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMX )
 221     {
 222         p_vout->p_sys->pf_merge = MergeMMX;
 223         p_vout->p_sys->pf_end_merge = EndMMX;
 224     }
 225     else
 226 #endif
 227     {
 228         p_vout->p_sys->pf_merge = MergeGeneric;
 229         p_vout->p_sys->pf_end_merge = NULL;
 230     }
 231
 232     /* Look what method was requested */
 233     var_Create( p_vout, "deinterlace-mode", VLC_VAR_STRING );
 234     var_Change( p_vout, "deinterlace-mode", VLC_VAR_INHERITVALUE, &val, NULL );
 235
 236     if( val.psz_string == NULL )
 237     {
 238         msg_Err( p_vout, "configuration variable deinterlace-mode empty" );
 239         msg_Err( p_vout, "no deinterlace mode provided, using \"discard\"" );
 240
 241         val.psz_string = strdup( "discard" );
 242     }
 243
 244     msg_Dbg( p_vout, "using %s deinterlace mode", val.psz_string );
 245
 246     SetFilterMethod( p_vout, val.psz_string );
 247
 248     free( val.psz_string );
 249
 250     return VLC_SUCCESS;
 251 }
 252
 253 /*****************************************************************************
 254  * SetFilterMethod: setup the deinterlace method to use.
 255  *****************************************************************************/
 256 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method )
 257 {
 258     if( !strcmp( psz_method, "discard" ) )
 259     {
 260         p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
 261         p_vout->p_sys->b_double_rate = VLC_FALSE;
 262     }
 263     else if( !strcmp( psz_method, "mean" ) )
 264     {
 265         p_vout->p_sys->i_mode = DEINTERLACE_MEAN;
 266         p_vout->p_sys->b_double_rate = VLC_FALSE;
 267     }
 268     else if( !strcmp( psz_method, "blend" )
 269              || !strcmp( psz_method, "average" )
 270              || !strcmp( psz_method, "combine-fields" ) )
 271     {
 272         p_vout->p_sys->i_mode = DEINTERLACE_BLEND;
 273         p_vout->p_sys->b_double_rate = VLC_FALSE;
 274     }
 275     else if( !strcmp( psz_method, "bob" )
 276              || !strcmp( psz_method, "progressive-scan" ) )
 277     {
 278         p_vout->p_sys->i_mode = DEINTERLACE_BOB;
 279         p_vout->p_sys->b_double_rate = VLC_TRUE;
 280     }
 281     else if( !strcmp( psz_method, "linear" ) )
 282     {
 283         p_vout->p_sys->i_mode = DEINTERLACE_LINEAR;
 284         p_vout->p_sys->b_double_rate = VLC_TRUE;
 285     }
 286     else if( !strcmp( psz_method, "x" ) )
 287     {
 288         p_vout->p_sys->i_mode = DEINTERLACE_X;
 289         p_vout->p_sys->b_double_rate = VLC_FALSE;
 290     }
 291     else
 292     {
 293         msg_Err( p_vout, "no valid deinterlace mode provided, "
 294                  "using \"discard\"" );
 295     }
 296
 297     msg_Dbg( p_vout, "using %s deinterlace method", psz_method );
 298 }
 299
 300 /*****************************************************************************
 301  * Init: initialize Deinterlace video thread output method
 302  *****************************************************************************/
 303 static int Init( vout_thread_t *p_vout )
 304 {
 305     int i_index;
 306     picture_t *p_pic;
 307
 308     I_OUTPUTPICTURES = 0;
 309
 310     /* Initialize the output structure, full of directbuffers since we want
 311      * the decoder to output directly to our structures. */
 312     switch( p_vout->render.i_chroma )
 313     {
 314         case VLC_FOURCC('I','4','2','0'):
 315         case VLC_FOURCC('I','Y','U','V'):
 316         case VLC_FOURCC('Y','V','1','2'):
 317         case VLC_FOURCC('I','4','2','2'):
 318             p_vout->output.i_chroma = p_vout->render.i_chroma;
 319             p_vout->output.i_width  = p_vout->render.i_width;
 320             p_vout->output.i_height = p_vout->render.i_height;
 321             p_vout->output.i_aspect = p_vout->render.i_aspect;
 322             p_vout->fmt_out = p_vout->fmt_in;
 323             break;
 324
 325         default:
 326             return VLC_EGENERIC; /* unknown chroma */
 327             break;
 328     }
 329
 330     /* Try to open the real video output */
 331     p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
 332
 333     if( p_vout->p_sys->p_vout == NULL )
 334     {
 335         /* Everything failed */
 336         msg_Err( p_vout, "cannot open vout, aborting" );
 337
 338         return VLC_EGENERIC;
 339     }
 340
 341     var_AddCallback( p_vout, "deinterlace-mode", FilterCallback, NULL );
 342
 343     ALLOCATE_DIRECTBUFFERS( VOUT_MAX_PICTURES );
 344
 345     ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
 346
 347     ADD_PARENT_CALLBACKS( SendEventsToChild );
 348
 349     return VLC_SUCCESS;
 350 }
 351
 352 /*****************************************************************************
 353  * SpawnRealVout: spawn the real video output.
 354  *****************************************************************************/
 355 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout )
 356 {
 357     vout_thread_t *p_real_vout = NULL;
 358     video_format_t fmt = {0};
 359
 360     msg_Dbg( p_vout, "spawning the real video output" );
 361
 362     fmt = p_vout->fmt_out;
 363
 364     switch( p_vout->render.i_chroma )
 365     {
 366     case VLC_FOURCC('I','4','2','0'):
 367     case VLC_FOURCC('I','Y','U','V'):
 368     case VLC_FOURCC('Y','V','1','2'):
 369         switch( p_vout->p_sys->i_mode )
 370         {
 371         case DEINTERLACE_MEAN:
 372         case DEINTERLACE_DISCARD:
 373             fmt.i_height /= 2; fmt.i_visible_height /= 2; fmt.i_y_offset /= 2;
 374             fmt.i_sar_den *= 2;
 375             p_real_vout = vout_Create( p_vout, &fmt );
 376             break;
 377
 378         case DEINTERLACE_BOB:
 379         case DEINTERLACE_BLEND:
 380         case DEINTERLACE_LINEAR:
 381         case DEINTERLACE_X:
 382             p_real_vout = vout_Create( p_vout, &fmt );
 383             break;
 384         }
 385         break;
 386
 387     case VLC_FOURCC('I','4','2','2'):
 388         fmt.i_chroma = VLC_FOURCC('I','4','2','0');
 389         p_real_vout = vout_Create( p_vout, &fmt );
 390         break;
 391
 392     default:
 393         break;
 394     }
 395
 396     return p_real_vout;
 397 }
 398
 399 /*****************************************************************************
 400  * End: terminate Deinterlace video thread output method
 401  *****************************************************************************/
 402 static void End( vout_thread_t *p_vout )
 403 {
 404     int i_index;
 405
 406     /* Free the fake output buffers we allocated */
 407     for( i_index = I_OUTPUTPICTURES ; i_index ; )
 408     {
 409         i_index--;
 410         free( PP_OUTPUTPICTURE[ i_index ]->p_data_orig );
 411     }
 412
 413     if( p_vout->p_sys->p_vout )
 414     {
 415         DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
 416         vlc_object_detach( p_vout->p_sys->p_vout );
 417         vout_Destroy( p_vout->p_sys->p_vout );
 418     }
 419
 420     DEL_PARENT_CALLBACKS( SendEventsToChild );
 421 }
 422
 423 /*****************************************************************************
 424  * Destroy: destroy Deinterlace video thread output method
 425  *****************************************************************************
 426  * Terminate an output method created by DeinterlaceCreateOutputMethod
 427  *****************************************************************************/
 428 static void Destroy( vlc_object_t *p_this )
 429 {
 430     vout_thread_t *p_vout = (vout_thread_t *)p_this;
 431     vlc_mutex_destroy( &p_vout->p_sys->filter_lock );
 432     free( p_vout->p_sys );
 433 }
 434
 435 /*****************************************************************************
 436  * Render: displays previously rendered output
 437  *****************************************************************************
 438  * This function send the currently rendered image to Deinterlace image,
 439  * waits until it is displayed and switch the two rendering buffers, preparing
 440  * next frame.
 441  *****************************************************************************/
 442 static void Render ( vout_thread_t *p_vout, picture_t *p_pic )
 443 {
 444     picture_t *pp_outpic[2];
 445
 446     pp_outpic[0] = pp_outpic[1] = NULL;
 447
 448     vlc_mutex_lock( &p_vout->p_sys->filter_lock );
 449
 450     /* Get a new picture */
 451     while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
 452                                                 0, 0, 0 ) )
 453               == NULL )
 454     {
 455         if( p_vout->b_die || p_vout->b_error )
 456         {
 457             vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 458             return;
 459         }
 460         msleep( VOUT_OUTMEM_SLEEP );
 461     }
 462
 463     vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
 464
 465     /* If we are using double rate, get an additional new picture */
 466     if( p_vout->p_sys->b_double_rate )
 467     {
 468         while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
 469                                                  0, 0, 0 ) )
 470                   == NULL )
 471         {
 472             if( p_vout->b_die || p_vout->b_error )
 473             {
 474                 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 475                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 476                 return;
 477             }
 478             msleep( VOUT_OUTMEM_SLEEP );
 479         }
 480
 481         /* 20ms is a bit arbitrary, but it's only for the first image we get */
 482         if( !p_vout->p_sys->last_date )
 483         {
 484             vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
 485                               p_pic->date + 20000 );
 486         }
 487         else
 488         {
 489             vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
 490                       (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
 491         }
 492         p_vout->p_sys->last_date = p_pic->date;
 493     }
 494
 495     switch( p_vout->p_sys->i_mode )
 496     {
 497         case DEINTERLACE_DISCARD:
 498             RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
 499             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 500             break;
 501
 502         case DEINTERLACE_BOB:
 503             RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
 504             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 505             RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
 506             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
 507             break;
 508
 509         case DEINTERLACE_LINEAR:
 510             RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
 511             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 512             RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
 513             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
 514             break;
 515
 516         case DEINTERLACE_MEAN:
 517             RenderMean( p_vout, pp_outpic[0], p_pic );
 518             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 519             break;
 520
 521         case DEINTERLACE_BLEND:
 522             RenderBlend( p_vout, pp_outpic[0], p_pic );
 523             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 524             break;
 525
 526         case DEINTERLACE_X:
 527             RenderX( p_vout, pp_outpic[0], p_pic );
 528             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 529             break;
 530     }
 531     vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 532 }
 533
 534 /*****************************************************************************
 535  * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
 536  *****************************************************************************/
 537 static void RenderDiscard( vout_thread_t *p_vout,
 538                            picture_t *p_outpic, picture_t *p_pic, int i_field )
 539 {
 540     int i_plane;
 541
 542     /* Copy image and skip lines */
 543     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 544     {
 545         uint8_t *p_in, *p_out_end, *p_out;
 546         int i_increment;
 547
 548         p_in = p_pic->p[i_plane].p_pixels
 549                    + i_field * p_pic->p[i_plane].i_pitch;
 550
 551         p_out = p_outpic->p[i_plane].p_pixels;
 552         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 553                              * p_outpic->p[i_plane].i_visible_lines;
 554
 555         switch( p_vout->render.i_chroma )
 556         {
 557         case VLC_FOURCC('I','4','2','0'):
 558         case VLC_FOURCC('I','Y','U','V'):
 559         case VLC_FOURCC('Y','V','1','2'):
 560
 561             for( ; p_out < p_out_end ; )
 562             {
 563                 p_vout->p_vlc->pf_memcpy( p_out, p_in,
 564                                           p_pic->p[i_plane].i_pitch );
 565
 566                 p_out += p_pic->p[i_plane].i_pitch;
 567                 p_in += 2 * p_pic->p[i_plane].i_pitch;
 568             }
 569             break;
 570
 571         case VLC_FOURCC('I','4','2','2'):
 572
 573             i_increment = 2 * p_pic->p[i_plane].i_pitch;
 574
 575             if( i_plane == Y_PLANE )
 576             {
 577                 for( ; p_out < p_out_end ; )
 578                 {
 579                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 580                                               p_pic->p[i_plane].i_pitch );
 581                     p_out += p_pic->p[i_plane].i_pitch;
 582                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 583                                               p_pic->p[i_plane].i_pitch );
 584                     p_out += p_pic->p[i_plane].i_pitch;
 585                     p_in += i_increment;
 586                 }
 587             }
 588             else
 589             {
 590                 for( ; p_out < p_out_end ; )
 591                 {
 592                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 593                                               p_pic->p[i_plane].i_pitch );
 594                     p_out += p_pic->p[i_plane].i_pitch;
 595                     p_in += i_increment;
 596                 }
 597             }
 598             break;
 599
 600         default:
 601             break;
 602         }
 603     }
 604 }
 605
 606 /*****************************************************************************
 607  * RenderBob: renders a BOB picture - simple copy
 608  *****************************************************************************/
 609 static void RenderBob( vout_thread_t *p_vout,
 610                        picture_t *p_outpic, picture_t *p_pic, int i_field )
 611 {
 612     int i_plane;
 613
 614     /* Copy image and skip lines */
 615     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 616     {
 617         uint8_t *p_in, *p_out_end, *p_out;
 618
 619         p_in = p_pic->p[i_plane].p_pixels;
 620         p_out = p_outpic->p[i_plane].p_pixels;
 621         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 622                              * p_outpic->p[i_plane].i_visible_lines;
 623
 624         switch( p_vout->render.i_chroma )
 625         {
 626             case VLC_FOURCC('I','4','2','0'):
 627             case VLC_FOURCC('I','Y','U','V'):
 628             case VLC_FOURCC('Y','V','1','2'):
 629                 /* For BOTTOM field we need to add the first line */
 630                 if( i_field == 1 )
 631                 {
 632                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 633                                               p_pic->p[i_plane].i_pitch );
 634                     p_in += p_pic->p[i_plane].i_pitch;
 635                     p_out += p_pic->p[i_plane].i_pitch;
 636                 }
 637
 638                 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 639
 640                 for( ; p_out < p_out_end ; )
 641                 {
 642                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 643                                               p_pic->p[i_plane].i_pitch );
 644
 645                     p_out += p_pic->p[i_plane].i_pitch;
 646
 647                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 648                                               p_pic->p[i_plane].i_pitch );
 649
 650                     p_in += 2 * p_pic->p[i_plane].i_pitch;
 651                     p_out += p_pic->p[i_plane].i_pitch;
 652                 }
 653
 654                 p_vout->p_vlc->pf_memcpy( p_out, p_in,
 655                                           p_pic->p[i_plane].i_pitch );
 656
 657                 /* For TOP field we need to add the last line */
 658                 if( i_field == 0 )
 659                 {
 660                     p_in += p_pic->p[i_plane].i_pitch;
 661                     p_out += p_pic->p[i_plane].i_pitch;
 662                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 663                                               p_pic->p[i_plane].i_pitch );
 664                 }
 665                 break;
 666
 667             case VLC_FOURCC('I','4','2','2'):
 668                 /* For BOTTOM field we need to add the first line */
 669                 if( i_field == 1 )
 670                 {
 671                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 672                                               p_pic->p[i_plane].i_pitch );
 673                     p_in += p_pic->p[i_plane].i_pitch;
 674                     p_out += p_pic->p[i_plane].i_pitch;
 675                 }
 676
 677                 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 678
 679                 if( i_plane == Y_PLANE )
 680                 {
 681                     for( ; p_out < p_out_end ; )
 682                     {
 683                         p_vout->p_vlc->pf_memcpy( p_out, p_in,
 684                                                   p_pic->p[i_plane].i_pitch );
 685
 686                         p_out += p_pic->p[i_plane].i_pitch;
 687
 688                         p_vout->p_vlc->pf_memcpy( p_out, p_in,
 689                                                   p_pic->p[i_plane].i_pitch );
 690
 691                         p_in += 2 * p_pic->p[i_plane].i_pitch;
 692                         p_out += p_pic->p[i_plane].i_pitch;
 693                     }
 694                 }
 695                 else
 696                 {
 697                     for( ; p_out < p_out_end ; )
 698                     {
 699                         p_vout->p_vlc->pf_memcpy( p_out, p_in,
 700                                                   p_pic->p[i_plane].i_pitch );
 701
 702                         p_out += p_pic->p[i_plane].i_pitch;
 703                         p_in += 2 * p_pic->p[i_plane].i_pitch;
 704                     }
 705                 }
 706
 707                 p_vout->p_vlc->pf_memcpy( p_out, p_in,
 708                                           p_pic->p[i_plane].i_pitch );
 709
 710                 /* For TOP field we need to add the last line */
 711                 if( i_field == 0 )
 712                 {
 713                     p_in += p_pic->p[i_plane].i_pitch;
 714                     p_out += p_pic->p[i_plane].i_pitch;
 715                     p_vout->p_vlc->pf_memcpy( p_out, p_in,
 716                                               p_pic->p[i_plane].i_pitch );
 717                 }
 718                 break;
 719         }
 720     }
 721 }
 722
 723 #define Merge p_vout->p_sys->pf_merge
 724 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
 725
 726 /*****************************************************************************
 727  * RenderLinear: BOB with linear interpolation
 728  *****************************************************************************/
 729 static void RenderLinear( vout_thread_t *p_vout,
 730                           picture_t *p_outpic, picture_t *p_pic, int i_field )
 731 {
 732     int i_plane;
 733
 734     /* Copy image and skip lines */
 735     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 736     {
 737         uint8_t *p_in, *p_out_end, *p_out;
 738
 739         p_in = p_pic->p[i_plane].p_pixels;
 740         p_out = p_outpic->p[i_plane].p_pixels;
 741         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 742                              * p_outpic->p[i_plane].i_visible_lines;
 743
 744         /* For BOTTOM field we need to add the first line */
 745         if( i_field == 1 )
 746         {
 747             p_vout->p_vlc->pf_memcpy( p_out, p_in,
 748                                       p_pic->p[i_plane].i_pitch );
 749             p_in += p_pic->p[i_plane].i_pitch;
 750             p_out += p_pic->p[i_plane].i_pitch;
 751         }
 752
 753         p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 754
 755         for( ; p_out < p_out_end ; )
 756         {
 757             p_vout->p_vlc->pf_memcpy( p_out, p_in,
 758                                       p_pic->p[i_plane].i_pitch );
 759
 760             p_out += p_pic->p[i_plane].i_pitch;
 761
 762             Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
 763                    p_pic->p[i_plane].i_pitch );
 764
 765             p_in += 2 * p_pic->p[i_plane].i_pitch;
 766             p_out += p_pic->p[i_plane].i_pitch;
 767         }
 768
 769         p_vout->p_vlc->pf_memcpy( p_out, p_in,
 770                                   p_pic->p[i_plane].i_pitch );
 771
 772         /* For TOP field we need to add the last line */
 773         if( i_field == 0 )
 774         {
 775             p_in += p_pic->p[i_plane].i_pitch;
 776             p_out += p_pic->p[i_plane].i_pitch;
 777             p_vout->p_vlc->pf_memcpy( p_out, p_in,
 778                                       p_pic->p[i_plane].i_pitch );
 779         }
 780     }
 781     EndMerge();
 782 }
 783
 784 static void RenderMean( vout_thread_t *p_vout,
 785                         picture_t *p_outpic, picture_t *p_pic )
 786 {
 787     int i_plane;
 788
 789     /* Copy image and skip lines */
 790     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 791     {
 792         uint8_t *p_in, *p_out_end, *p_out;
 793
 794         p_in = p_pic->p[i_plane].p_pixels;
 795
 796         p_out = p_outpic->p[i_plane].p_pixels;
 797         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 798                              * p_outpic->p[i_plane].i_visible_lines;
 799
 800         /* All lines: mean value */
 801         for( ; p_out < p_out_end ; )
 802         {
 803             Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 804                    p_pic->p[i_plane].i_pitch );
 805
 806             p_out += p_pic->p[i_plane].i_pitch;
 807             p_in += 2 * p_pic->p[i_plane].i_pitch;
 808         }
 809     }
 810     EndMerge();
 811 }
 812
 813 static void RenderBlend( vout_thread_t *p_vout,
 814                          picture_t *p_outpic, picture_t *p_pic )
 815 {
 816     int i_plane;
 817
 818     /* Copy image and skip lines */
 819     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 820     {
 821         uint8_t *p_in, *p_out_end, *p_out;
 822
 823         p_in = p_pic->p[i_plane].p_pixels;
 824
 825         p_out = p_outpic->p[i_plane].p_pixels;
 826         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 827                              * p_outpic->p[i_plane].i_visible_lines;
 828
 829         switch( p_vout->render.i_chroma )
 830         {
 831             case VLC_FOURCC('I','4','2','0'):
 832             case VLC_FOURCC('I','Y','U','V'):
 833             case VLC_FOURCC('Y','V','1','2'):
 834                 /* First line: simple copy */
 835                 p_vout->p_vlc->pf_memcpy( p_out, p_in,
 836                                           p_pic->p[i_plane].i_pitch );
 837                 p_out += p_pic->p[i_plane].i_pitch;
 838
 839                 /* Remaining lines: mean value */
 840                 for( ; p_out < p_out_end ; )
 841                 {
 842                     Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 843                            p_pic->p[i_plane].i_pitch );
 844
 845                     p_out += p_pic->p[i_plane].i_pitch;
 846                     p_in += p_pic->p[i_plane].i_pitch;
 847                 }
 848                 break;
 849
 850             case VLC_FOURCC('I','4','2','2'):
 851                 /* First line: simple copy */
 852                 p_vout->p_vlc->pf_memcpy( p_out, p_in,
 853                                           p_pic->p[i_plane].i_pitch );
 854                 p_out += p_pic->p[i_plane].i_pitch;
 855
 856                 /* Remaining lines: mean value */
 857                 if( i_plane == Y_PLANE )
 858                 {
 859                     for( ; p_out < p_out_end ; )
 860                     {
 861                         Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 862                                p_pic->p[i_plane].i_pitch );
 863
 864                         p_out += p_pic->p[i_plane].i_pitch;
 865                         p_in += p_pic->p[i_plane].i_pitch;
 866                     }
 867                 }
 868
 869                 else
 870                 {
 871                     for( ; p_out < p_out_end ; )
 872                     {
 873                         Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 874                                p_pic->p[i_plane].i_pitch );
 875
 876                         p_out += p_pic->p[i_plane].i_pitch;
 877                         p_in += 2*p_pic->p[i_plane].i_pitch;
 878                     }
 879                 }
 880                 break;
 881         }
 882     }
 883     EndMerge();
 884 }
 885
 886 #undef Merge
 887
 888 static void MergeGeneric( void *_p_dest, const void *_p_s1,
 889                           const void *_p_s2, size_t i_bytes )
 890 {
 891     uint8_t* p_dest = (uint8_t*)_p_dest;
 892     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 893     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 894     uint8_t* p_end = p_dest + i_bytes - 8;
 895
 896     while( p_dest < p_end )
 897     {
 898         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 899         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 900         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 901         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 902         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 903         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 904         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 905         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 906     }
 907
 908     p_end += 8;
 909
 910     while( p_dest < p_end )
 911     {
 912         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 913     }
 914 }
 915
 916 #if defined(CAN_COMPILE_MMXEXT)
 917 static void MergeMMX( void *_p_dest, const void *_p_s1, const void *_p_s2,
 918                       size_t i_bytes )
 919 {
 920     uint8_t* p_dest = (uint8_t*)_p_dest;
 921     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 922     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 923     uint8_t* p_end = p_dest + i_bytes - 8;
 924     while( p_dest < p_end )
 925     {
 926         __asm__  __volatile__( "movq %2,%%mm1;"
 927                                "pavgb %1, %%mm1;"
 928                                "movq %%mm1, %0" :"=m" (*p_dest):
 929                                                  "m" (*p_s1),
 930                                                  "m" (*p_s2) );
 931         p_dest += 8;
 932         p_s1 += 8;
 933         p_s2 += 8;
 934     }
 935
 936     p_end += 8;
 937
 938     while( p_dest < p_end )
 939     {
 940         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 941     }
 942 }
 943 #endif
 944
 945 #if defined(CAN_COMPILE_SSE)
 946 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
 947                        size_t i_bytes )
 948 {
 949     uint8_t* p_dest = (uint8_t*)_p_dest;
 950     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 951     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 952     uint8_t* p_end;
 953     while( (ptrdiff_t)p_s1 % 16 )
 954     {
 955         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 956     }
 957     p_end = p_dest + i_bytes - 16;
 958     while( p_dest < p_end )
 959     {
 960         __asm__  __volatile__( "movdqu %2,%%xmm1;"
 961                                "pavgb %1, %%xmm1;"
 962                                "movdqu %%xmm1, %0" :"=m" (*p_dest):
 963                                                  "m" (*p_s1),
 964                                                  "m" (*p_s2) );
 965         p_dest += 16;
 966         p_s1 += 16;
 967         p_s2 += 16;
 968     }
 969
 970     p_end += 16;
 971
 972     while( p_dest < p_end )
 973     {
 974         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 975     }
 976 }
 977 #endif
 978
 979 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
 980 static void EndMMX( void )
 981 {
 982     __asm__ __volatile__( "emms" :: );
 983 }
 984 #endif
 985
 986 #ifdef CAN_COMPILE_C_ALTIVEC
 987 static void MergeAltivec( void *_p_dest, const void *_p_s1,
 988                           const void *_p_s2, size_t i_bytes )
 989 {
 990     uint8_t *p_dest = (uint8_t *)_p_dest;
 991     uint8_t *p_s1   = (uint8_t *)_p_s1;
 992     uint8_t *p_s2   = (uint8_t *)_p_s2;
 993     uint8_t *p_end  = p_dest + i_bytes - 15;
 994
 995     /* Use C until the first 16-bytes aligned destination pixel */
 996     while( (int)p_dest & 0xF )
 997     {
 998         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 999     }
1000
1001     if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1002     {
1003         /* Unaligned source */
1004         vector unsigned char s1v, s2v, destv;
1005         vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1006         vector unsigned char perm1v, perm2v;
1007
1008         perm1v = vec_lvsl( 0, p_s1 );
1009         perm2v = vec_lvsl( 0, p_s2 );
1010         s1oldv = vec_ld( 0, p_s1 );
1011         s2oldv = vec_ld( 0, p_s2 );
1012
1013         while( p_dest < p_end )
1014         {
1015             s1newv = vec_ld( 16, p_s1 );
1016             s2newv = vec_ld( 16, p_s2 );
1017             s1v    = vec_perm( s1oldv, s1newv, perm1v );
1018             s2v    = vec_perm( s2oldv, s2newv, perm2v );
1019             s1oldv = s1newv;
1020             s2oldv = s2newv;
1021             destv  = vec_avg( s1v, s2v );
1022             vec_st( destv, 0, p_dest );
1023
1024             p_s1   += 16;
1025             p_s2   += 16;
1026             p_dest += 16;
1027         }
1028     }
1029     else
1030     {
1031         /* Aligned source */
1032         vector unsigned char s1v, s2v, destv;
1033
1034         while( p_dest < p_end )
1035         {
1036             s1v   = vec_ld( 0, p_s1 );
1037             s2v   = vec_ld( 0, p_s2 );
1038             destv = vec_avg( s1v, s2v );
1039             vec_st( destv, 0, p_dest );
1040
1041             p_s1   += 16;
1042             p_s2   += 16;
1043             p_dest += 16;
1044         }
1045     }
1046
1047     p_end += 15;
1048
1049     while( p_dest < p_end )
1050     {
1051         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1052     }
1053 }
1054 #endif
1055
1056 /*****************************************************************************
1057  * RenderX: This algo works on a 8x8 block basic, it copies the top field
1058  * and apply a process to recreate the bottom field :
1059  *  If a 8x8 block is classified as :
1060  *   - progressive: it applies a small blend (1,6,1)
1061  *   - interlaced:
1062  *    * in the MMX version: we do a ME between the 2 fields, if there is a
1063  *    good match we use MC to recreate the bottom field (with a small
1064  *    blend (1,6,1) )
1065  *    * otherwise: it recreates the bottom field by an edge oriented
1066  *    interpolation.
1067   *****************************************************************************/
1068
1069 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1070  * XXX: It need to access to 8x10
1071  * We use more than 8 lines to help with scrolling (text)
1072  * (and because XDeint8x8Frame use line 9)
1073  * XXX: smooth/uniform area with noise detection doesn't works well
1074  * but it's not really a problem because they don't have much details anyway
1075  */
1076 static inline int ssd( int a ) { return a*a; }
1077 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1078 {
1079     int y, x;
1080     int ff, fr;
1081     int fc;
1082
1083     /* Detect interlacing */
1084     fc = 0;
1085     for( y = 0; y < 7; y += 2 )
1086     {
1087         ff = fr = 0;
1088         for( x = 0; x < 8; x++ )
1089         {
1090             fr += ssd(src[      x] - src[1*i_src+x]) +
1091                   ssd(src[i_src+x] - src[2*i_src+x]);
1092             ff += ssd(src[      x] - src[2*i_src+x]) +
1093                   ssd(src[i_src+x] - src[3*i_src+x]);
1094         }
1095         if( ff < 6*fr/8 && fr > 32 )
1096             fc++;
1097
1098         src += 2*i_src;
1099     }
1100
1101     return fc < 1 ? VLC_FALSE : VLC_TRUE;
1102 }
1103 #ifdef CAN_COMPILE_MMXEXT
1104 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1105 {
1106
1107     int y, x;
1108     int32_t ff, fr;
1109     int fc;
1110
1111     /* Detect interlacing */
1112     fc = 0;
1113     pxor_r2r( mm7, mm7 );
1114     for( y = 0; y < 9; y += 2 )
1115     {
1116         ff = fr = 0;
1117         pxor_r2r( mm5, mm5 );
1118         pxor_r2r( mm6, mm6 );
1119         for( x = 0; x < 8; x+=4 )
1120         {
1121             movd_m2r( src[        x], mm0 );
1122             movd_m2r( src[1*i_src+x], mm1 );
1123             movd_m2r( src[2*i_src+x], mm2 );
1124             movd_m2r( src[3*i_src+x], mm3 );
1125
1126             punpcklbw_r2r( mm7, mm0 );
1127             punpcklbw_r2r( mm7, mm1 );
1128             punpcklbw_r2r( mm7, mm2 );
1129             punpcklbw_r2r( mm7, mm3 );
1130
1131             movq_r2r( mm0, mm4 );
1132
1133             psubw_r2r( mm1, mm0 );
1134             psubw_r2r( mm2, mm4 );
1135
1136             psubw_r2r( mm1, mm2 );
1137             psubw_r2r( mm1, mm3 );
1138
1139             pmaddwd_r2r( mm0, mm0 );
1140             pmaddwd_r2r( mm4, mm4 );
1141             pmaddwd_r2r( mm2, mm2 );
1142             pmaddwd_r2r( mm3, mm3 );
1143             paddd_r2r( mm0, mm2 );
1144             paddd_r2r( mm4, mm3 );
1145             paddd_r2r( mm2, mm5 );
1146             paddd_r2r( mm3, mm6 );
1147         }
1148
1149         movq_r2r( mm5, mm0 );
1150         psrlq_i2r( 32, mm0 );
1151         paddd_r2r( mm0, mm5 );
1152         movd_r2m( mm5, fr );
1153
1154         movq_r2r( mm6, mm0 );
1155         psrlq_i2r( 32, mm0 );
1156         paddd_r2r( mm0, mm6 );
1157         movd_r2m( mm6, ff );
1158
1159         if( ff < 6*fr/8 && fr > 32 )
1160             fc++;
1161
1162         src += 2*i_src;
1163     }
1164     return fc;
1165 }
1166 #endif
1167
1168 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1169  * This won't destroy details, and help if there is a bit of interlacing.
1170  * (It helps with paning to avoid flickers)
1171  * (Use 8x9 pixels)
1172  */
1173 #if 0
1174 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1175                                     uint8_t *src, int i_src )
1176 {
1177     int y, x;
1178
1179     /* Progressive */
1180     for( y = 0; y < 8; y += 2 )
1181     {
1182         memcpy( dst, src, 8 );
1183         dst += i_dst;
1184
1185         for( x = 0; x < 8; x++ )
1186             dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1187         dst += 1*i_dst;
1188         src += 2*i_src;
1189     }
1190 }
1191 #endif
1192 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1193                                     uint8_t *src1, int i_src1,
1194                                     uint8_t *src2, int i_src2 )
1195 {
1196     int y, x;
1197
1198     /* Progressive */
1199     for( y = 0; y < 8; y += 2 )
1200     {
1201         memcpy( dst, src1, 8 );
1202         dst  += i_dst;
1203
1204         for( x = 0; x < 8; x++ )
1205             dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1206         dst += i_dst;
1207
1208         src1 += i_src1;
1209         src2 += i_src2;
1210     }
1211 }
1212
1213 #ifdef CAN_COMPILE_MMXEXT
1214 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1215                                          uint8_t *src1, int i_src1,
1216                                          uint8_t *src2, int i_src2 )
1217 {
1218     static const uint64_t m_4 = I64C(0x0004000400040004);
1219     int y, x;
1220
1221     /* Progressive */
1222     pxor_r2r( mm7, mm7 );
1223     for( y = 0; y < 8; y += 2 )
1224     {
1225         for( x = 0; x < 8; x +=4 )
1226         {
1227             movd_m2r( src1[x], mm0 );
1228             movd_r2m( mm0, dst[x] );
1229
1230             movd_m2r( src2[x], mm1 );
1231             movd_m2r( src1[i_src1+x], mm2 );
1232
1233             punpcklbw_r2r( mm7, mm0 );
1234             punpcklbw_r2r( mm7, mm1 );
1235             punpcklbw_r2r( mm7, mm2 );
1236             paddw_r2r( mm1, mm1 );
1237             movq_r2r( mm1, mm3 );
1238             paddw_r2r( mm3, mm3 );
1239             paddw_r2r( mm2, mm0 );
1240             paddw_r2r( mm3, mm1 );
1241             paddw_m2r( m_4, mm1 );
1242             paddw_r2r( mm1, mm0 );
1243             psraw_i2r( 3, mm0 );
1244             packuswb_r2r( mm7, mm0 );
1245             movd_r2m( mm0, dst[i_dst+x] );
1246         }
1247         dst += 2*i_dst;
1248         src1 += i_src1;
1249         src2 += i_src2;
1250     }
1251 }
1252
1253 #endif
1254
1255 /* For debug */
1256 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1257 {
1258     int y;
1259     for( y = 0; y < 8; y++ )
1260         memset( &dst[y*i_dst], v, 8 );
1261 }
1262
1263 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1264  * neighbour
1265  * (Use 8x9 pixels)
1266  * TODO: a better one for the inner part.
1267  */
1268 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1269                                      uint8_t *src, int i_src )
1270 {
1271     int y, x;
1272
1273     /* Interlaced */
1274     for( y = 0; y < 8; y += 2 )
1275     {
1276         memcpy( dst, src, 8 );
1277         dst += i_dst;
1278
1279         for( x = 0; x < 8; x++ )
1280             dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1281         dst += 1*i_dst;
1282         src += 2*i_src;
1283     }
1284 }
1285 #ifdef CAN_COMPILE_MMXEXT
1286 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1287                                           uint8_t *src, int i_src )
1288 {
1289     int y;
1290
1291     /* Interlaced */
1292     for( y = 0; y < 8; y += 2 )
1293     {
1294         movq_m2r( src[0], mm0 );
1295         movq_r2m( mm0, dst[0] );
1296         dst += i_dst;
1297
1298         movq_m2r( src[2*i_src], mm1 );
1299         pavgb_r2r( mm1, mm0 );
1300
1301         movq_r2m( mm0, dst[0] );
1302
1303         dst += 1*i_dst;
1304         src += 2*i_src;
1305     }
1306 }
1307 #endif
1308
1309 /* XDeint8x8Field: Edge oriented interpolation
1310  * (Need -4 and +5 pixels H, +1 line)
1311  */
1312 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1313                                     uint8_t *src, int i_src )
1314 {
1315     int y, x;
1316
1317     /* Interlaced */
1318     for( y = 0; y < 8; y += 2 )
1319     {
1320         memcpy( dst, src, 8 );
1321         dst += i_dst;
1322
1323         for( x = 0; x < 8; x++ )
1324         {
1325             uint8_t *src2 = &src[2*i_src];
1326             /* I use 8 pixels just to match the MMX version, but it's overkill
1327              * 5 would be enough (less isn't good) */
1328             const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1329                            abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1330                            abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1331                            abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1332
1333             const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1334                            abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1335                            abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1336                            abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1337
1338             const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1339                            abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1340                            abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1341                            abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1342
1343             if( c0 < c1 && c1 <= c2 )
1344                 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1345             else if( c2 < c1 && c1 <= c0 )
1346                 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1347             else
1348                 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1349         }
1350
1351         dst += 1*i_dst;
1352         src += 2*i_src;
1353     }
1354 }
1355 #ifdef CAN_COMPILE_MMXEXT
1356 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1357                                          uint8_t *src, int i_src )
1358 {
1359     int y, x;
1360
1361     /* Interlaced */
1362     for( y = 0; y < 8; y += 2 )
1363     {
1364         memcpy( dst, src, 8 );
1365         dst += i_dst;
1366
1367         for( x = 0; x < 8; x++ )
1368         {
1369             uint8_t *src2 = &src[2*i_src];
1370             int32_t c0, c1, c2;
1371
1372             movq_m2r( src[x-2], mm0 );
1373             movq_m2r( src[x-3], mm1 );
1374             movq_m2r( src[x-4], mm2 );
1375
1376             psadbw_m2r( src2[x-4], mm0 );
1377             psadbw_m2r( src2[x-3], mm1 );
1378             psadbw_m2r( src2[x-2], mm2 );
1379
1380             movd_r2m( mm0, c2 );
1381             movd_r2m( mm1, c1 );
1382             movd_r2m( mm2, c0 );
1383
1384             if( c0 < c1 && c1 <= c2 )
1385                 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1386             else if( c2 < c1 && c1 <= c0 )
1387                 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1388             else
1389                 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1390         }
1391
1392         dst += 1*i_dst;
1393         src += 2*i_src;
1394     }
1395 }
1396 #endif
1397
1398 #if 0
1399 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1400                                  uint8_t *pix2, int i_pix2 )
1401 {
1402     int y, x;
1403     int s = 0;
1404
1405     for( y = 0; y < 8; y++ )
1406         for( x = 0; x < 8; x++ )
1407             s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1408     return s;
1409 }
1410
1411 #ifdef CAN_COMPILE_MMXEXT
1412 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1413                                       uint8_t *pix2, int i_pix2 )
1414 {
1415     int y;
1416     int32_t s;
1417
1418     pxor_r2r( mm7, mm7 );
1419     pxor_r2r( mm6, mm6 );
1420
1421     for( y = 0; y < 8; y++ )
1422     {
1423         movq_m2r( pix1[0], mm0 );
1424         movq_m2r( pix2[0], mm1 );
1425
1426         movq_r2r( mm0, mm2 );
1427         movq_r2r( mm1, mm3 );
1428
1429         punpcklbw_r2r( mm7, mm0 );
1430         punpckhbw_r2r( mm7, mm2 );
1431         punpcklbw_r2r( mm7, mm1 );
1432         punpckhbw_r2r( mm7, mm3 );
1433
1434         psubw_r2r( mm1, mm0 );
1435         psubw_r2r( mm3, mm2 );
1436
1437         pmaddwd_r2r( mm0, mm0 );
1438         pmaddwd_r2r( mm2, mm2 );
1439
1440         paddd_r2r( mm2, mm0 );
1441         paddd_r2r( mm0, mm6 );
1442
1443         pix1 += i_pix1;
1444         pix2 += i_pix2;
1445     }
1446
1447     movq_r2r( mm6, mm7 );
1448     psrlq_i2r( 32, mm7 );
1449     paddd_r2r( mm6, mm7 );
1450     movd_r2m( mm7, s );
1451
1452     return s;
1453 }
1454 #endif
1455 #endif
1456
1457 #if 0
1458 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1459 #ifdef CAN_COMPILE_MMXEXT
1460 /* XDeintMC:
1461  *  Bilinear MC QPel
1462  *  TODO: mmx version (easier in sse2)
1463  */
1464 static inline void XDeintMC( uint8_t *dst, int i_dst,
1465                              uint8_t *src, int i_src,
1466                              int mvx, int mvy,
1467                              int i_width, int i_height )
1468 {
1469     const int d4x = mvx&0x03;
1470     const int d4y = mvy&0x03;
1471
1472     const int cA = (4-d4x)*(4-d4y);
1473     const int cB = d4x    *(4-d4y);
1474     const int cC = (4-d4x)*d4y;
1475     const int cD = d4x    *d4y;
1476
1477     int y, x;
1478     uint8_t *srcp;
1479
1480
1481     src  += (mvy >> 2) * i_src + (mvx >> 2);
1482     srcp = &src[i_src];
1483
1484     for( y = 0; y < i_height; y++ )
1485     {
1486         for( x = 0; x < i_width; x++ )
1487         {
1488             dst[x] = ( cA*src[x]  + cB*src[x+1] +
1489                        cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1490         }
1491         dst  += i_dst;
1492
1493         src   = srcp;
1494         srcp += i_src;
1495     }
1496 }
1497 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1498                                uint8_t *pix2, int i_pix2 )
1499 {
1500     int32_t s;
1501
1502     movq_m2r( pix1[0*i_pix1], mm0 );
1503     movq_m2r( pix1[1*i_pix1], mm1 );
1504
1505     psadbw_m2r( pix2[0*i_pix2], mm0 );
1506     psadbw_m2r( pix2[1*i_pix2], mm1 );
1507
1508     movq_m2r( pix1[2*i_pix1], mm2 );
1509     movq_m2r( pix1[3*i_pix1], mm3 );
1510     psadbw_m2r( pix2[2*i_pix2], mm2 );
1511     psadbw_m2r( pix2[3*i_pix2], mm3 );
1512
1513     paddd_r2r( mm1, mm0 );
1514     paddd_r2r( mm3, mm2 );
1515     paddd_r2r( mm2, mm0 );
1516     movd_r2m( mm0, s );
1517
1518     return s;
1519 }
1520
1521 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1522                                      uint8_t *ref, int i_stride,
1523                                      int mx, int my,
1524                                      int xmax, int ymax )
1525 {
1526     uint8_t buffer[8*4];
1527
1528     if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1529         return 255*255*255;
1530
1531     XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1532     return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1533 }
1534 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1535                                     uint8_t *ref, int i_stride,
1536                                     int mx, int my,
1537                                     int xmax, int ymax )
1538 {
1539     if( abs(mx) >= xmax || abs(my) >= ymax )
1540         return 255*255*255;
1541
1542     return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1543 }
1544
1545 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1546                                          uint8_t *src, int i_src,
1547                                          int *mpx, int *mpy,
1548                                          int xmax, int ymax )
1549 {
1550     static const int dx[8] = { 0,  0, -1, 1, -1, -1,  1, 1 };
1551     static const int dy[8] = {-1,  1,  0, 0, -1,  1, -1, 1 };
1552     uint8_t *next = &src[i_src];
1553     const int i_src2 = 2*i_src;
1554     int mvx, mvy;
1555     int mvs, s;
1556     int i_step;
1557
1558     uint8_t *rec = &dst[i_dst];
1559
1560     /* We construct with intra method the missing field */
1561     XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1562
1563     /* Now we will try to find a match with ME with the other field */
1564
1565     /* ME: A small/partial EPZS
1566      * We search only for small MV (with high motion intra will be perfect */
1567     if( xmax > 4 ) xmax = 4;
1568     if( ymax > 4 ) ymax = 4;
1569
1570     /* Init with NULL Mv */
1571     mvx = mvy = 0;
1572     mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1573
1574     /* Try predicted Mv */
1575     if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1576     {
1577         mvs = s;
1578         mvx = *mpx;
1579         mvy = *mpy;
1580     }
1581     /* Search interger pel (small mv) */
1582     for( i_step = 0; i_step < 4; i_step++ )
1583     {
1584         int c = 4;
1585         int s;
1586         int i;
1587
1588         for( i = 0; i < 4; i++ )
1589         {
1590             s = XDeint8x4TestInt( rec, i_src2,
1591                                   next, i_src2, mvx+dx[i], mvy+dy[i],
1592                                   xmax, ymax );
1593             if( s < mvs )
1594             {
1595                 mvs = s;
1596                 c = i;
1597             }
1598         }
1599         if( c == 4 )
1600             break;
1601
1602         mvx += dx[c];
1603         mvy += dy[c];
1604     }
1605     *mpx = mvx;
1606     *mpy = mvy;
1607
1608     mvx <<= 2;
1609     mvy <<= 2;
1610
1611     if( mvs > 4 && mvs < 256 )
1612     {
1613         /* Search Qpel */
1614         /* XXX: for now only HPEL (too slow) */
1615         for( i_step = 0; i_step < 4; i_step++ )
1616         {
1617             int c = 8;
1618             int s;
1619             int i;
1620
1621             for( i = 0; i < 8; i++ )
1622             {
1623                 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1624                                        mvx+dx[i], mvy+dy[i],
1625                                        xmax, ymax );
1626                 if( s < mvs )
1627                 {
1628                     mvs = s;
1629                     c = i;
1630                 }
1631             }
1632             if( c == 8 )
1633                 break;
1634
1635             mvx += dx[c];
1636             mvy += dy[c];
1637         }
1638     }
1639
1640     if( mvs < 128 )
1641     {
1642         uint8_t buffer[8*4];
1643         XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1644         XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1645
1646         //XDeint8x8Set( dst, i_dst, 0 );
1647     }
1648 }
1649 #endif
1650 #endif
1651
1652 #if 0
1653 /* Kernel interpolation (1,-5,20,20,-5,1)
1654  * Loose a bit more details+add aliasing than edge interpol but avoid
1655  * more artifacts
1656  */
1657 static inline uint8_t clip1( int a )
1658 {
1659     if( a <= 0 )
1660         return 0;
1661     else if( a >= 255 )
1662         return 255;
1663     else
1664         return a;
1665 }
1666 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1667                                    uint8_t *src, int i_src )
1668 {
1669     int y, x;
1670
1671     /* Interlaced */
1672     for( y = 0; y < 8; y += 2 )
1673     {
1674         const int i_src2 = i_src*2;
1675
1676         memcpy( dst, src, 8 );
1677         dst += i_dst;
1678
1679         for( x = 0; x < 8; x++ )
1680         {
1681             int pix;
1682
1683             pix =   1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1684                    -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1685                   +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1686
1687             dst[x] = clip1( ( pix + 16 ) >> 5 );
1688         }
1689
1690         dst += 1*i_dst;
1691         src += 2*i_src;
1692     }
1693 }
1694
1695 #endif
1696
1697 /* NxN arbitray size (and then only use pixel in the NxN block)
1698  */
1699 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1700                                    int i_height, int i_width )
1701 {
1702     int y, x;
1703     int ff, fr;
1704     int fc;
1705
1706
1707     /* Detect interlacing */
1708     /* FIXME way too simple, need to be more like XDeint8x8Detect */
1709     ff = fr = 0;
1710     fc = 0;
1711     for( y = 0; y < i_height - 2; y += 2 )
1712     {
1713         const uint8_t *s = &src[y*i_src];
1714         for( x = 0; x < i_width; x++ )
1715         {
1716             fr += ssd(s[      x] - s[1*i_src+x]);
1717             ff += ssd(s[      x] - s[2*i_src+x]);
1718         }
1719         if( ff < fr && fr > i_width / 2 )
1720             fc++;
1721     }
1722
1723     return fc < 2 ? VLC_FALSE : VLC_TRUE;
1724 }
1725
1726 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1727                                    uint8_t *src, int i_src,
1728                                    int i_width, int i_height )
1729 {
1730     int y, x;
1731
1732     /* Progressive */
1733     for( y = 0; y < i_height; y += 2 )
1734     {
1735         memcpy( dst, src, i_width );
1736         dst += i_dst;
1737
1738         if( y < i_height - 2 )
1739         {
1740             for( x = 0; x < i_width; x++ )
1741                 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1742         }
1743         else
1744         {
1745             /* Blend last line */
1746             for( x = 0; x < i_width; x++ )
1747                 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1748         }
1749         dst += 1*i_dst;
1750         src += 2*i_src;
1751     }
1752 }
1753
1754 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1755                                    uint8_t *src, int i_src,
1756                                    int i_width, int i_height )
1757 {
1758     int y, x;
1759
1760     /* Interlaced */
1761     for( y = 0; y < i_height; y += 2 )
1762     {
1763         memcpy( dst, src, i_width );
1764         dst += i_dst;
1765
1766         if( y < i_height - 2 )
1767         {
1768             for( x = 0; x < i_width; x++ )
1769                 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1770         }
1771         else
1772         {
1773             /* Blend last line */
1774             for( x = 0; x < i_width; x++ )
1775                 dst[x] = (src[x] + src[i_src+x]) >> 1;
1776         }
1777         dst += 1*i_dst;
1778         src += 2*i_src;
1779     }
1780 }
1781
1782 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1783                               int i_width, int i_height )
1784 {
1785     if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1786         XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1787     else
1788         XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1789 }
1790
1791
1792 static inline int median( int a, int b, int c )
1793 {
1794     int min = a, max =a;
1795     if( b < min )
1796         min = b;
1797     else
1798         max = b;
1799
1800     if( c < min )
1801         min = c;
1802     else if( c > max )
1803         max = c;
1804
1805     return a + b + c - min - max;
1806 }
1807
1808
1809 /* XDeintBand8x8:
1810  */
1811 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1812                                    uint8_t *src, int i_src,
1813                                    const int i_mbx, int i_modx )
1814 {
1815     int x;
1816
1817     for( x = 0; x < i_mbx; x++ )
1818     {
1819         int s;
1820         if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1821         {
1822             if( x == 0 || x == i_mbx - 1 )
1823                 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1824             else
1825                 XDeint8x8FieldC( dst, i_dst, src, i_src );
1826         }
1827         else
1828         {
1829             XDeint8x8MergeC( dst, i_dst,
1830                              &src[0*i_src], 2*i_src,
1831                              &src[1*i_src], 2*i_src );
1832         }
1833
1834         dst += 8;
1835         src += 8;
1836     }
1837
1838     if( i_modx )
1839         XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1840 }
1841 #ifdef CAN_COMPILE_MMXEXT
1842 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1843                                         uint8_t *src, int i_src,
1844                                         const int i_mbx, int i_modx )
1845 {
1846     int x;
1847
1848     /* Reset current line */
1849     for( x = 0; x < i_mbx; x++ )
1850     {
1851         int s;
1852         if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1853         {
1854             if( x == 0 || x == i_mbx - 1 )
1855                 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1856             else
1857                 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1858         }
1859         else
1860         {
1861             XDeint8x8MergeMMXEXT( dst, i_dst,
1862                                   &src[0*i_src], 2*i_src,
1863                                   &src[1*i_src], 2*i_src );
1864         }
1865
1866         dst += 8;
1867         src += 8;
1868     }
1869
1870     if( i_modx )
1871         XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1872 }
1873 #endif
1874
1875 static void RenderX( vout_thread_t *p_vout,
1876                      picture_t *p_outpic, picture_t *p_pic )
1877 {
1878     int i_plane;
1879
1880     /* Copy image and skip lines */
1881     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1882     {
1883         const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1884         const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1885
1886         const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1887         const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1888
1889         const int i_dst = p_outpic->p[i_plane].i_pitch;
1890         const int i_src = p_pic->p[i_plane].i_pitch;
1891
1892         int y, x;
1893
1894         for( y = 0; y < i_mby; y++ )
1895         {
1896             uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1897             uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1898
1899 #ifdef CAN_COMPILE_MMXEXT
1900             if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1901                 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1902             else
1903 #endif
1904                 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1905         }
1906
1907         /* Last line (C only)*/
1908         if( i_mody )
1909         {
1910             uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1911             uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1912
1913             for( x = 0; x < i_mbx; x++ )
1914             {
1915                 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1916
1917                 dst += 8;
1918                 src += 8;
1919             }
1920
1921             if( i_modx )
1922                 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1923         }
1924     }
1925
1926 #ifdef CAN_COMPILE_MMXEXT
1927     if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1928         emms();
1929 #endif
1930 }
1931
1932 /*****************************************************************************
1933  * SendEvents: forward mouse and keyboard events to the parent p_vout
1934  *****************************************************************************/
1935 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
1936                        vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
1937 {
1938     vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
1939     vlc_value_t sentval = newval;
1940
1941     if( !strcmp( psz_var, "mouse-y" ) )
1942     {
1943         switch( p_vout->p_sys->i_mode )
1944         {
1945             case DEINTERLACE_MEAN:
1946             case DEINTERLACE_DISCARD:
1947                 sentval.i_int *= 2;
1948                 break;
1949         }
1950     }
1951
1952     var_Set( p_vout, psz_var, sentval );
1953
1954     return VLC_SUCCESS;
1955 }
1956
1957 /*****************************************************************************
1958  * FilterCallback: called when changing the deinterlace method on the fly.
1959  *****************************************************************************/
1960 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
1961                            vlc_value_t oldval, vlc_value_t newval,
1962                            void *p_data )
1963 {
1964     vout_thread_t * p_vout = (vout_thread_t *)p_this;
1965     int i_old_mode = p_vout->p_sys->i_mode;
1966
1967     msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
1968
1969     vlc_mutex_lock( &p_vout->p_sys->filter_lock );
1970
1971     SetFilterMethod( p_vout, newval.psz_string );
1972
1973     switch( p_vout->render.i_chroma )
1974     {
1975     case VLC_FOURCC('I','4','2','2'):
1976         vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1977         return VLC_SUCCESS;
1978         break;
1979
1980     case VLC_FOURCC('I','4','2','0'):
1981     case VLC_FOURCC('I','Y','U','V'):
1982     case VLC_FOURCC('Y','V','1','2'):
1983         switch( p_vout->p_sys->i_mode )
1984         {
1985         case DEINTERLACE_MEAN:
1986         case DEINTERLACE_DISCARD:
1987             if( ( i_old_mode == DEINTERLACE_MEAN )
1988                 || ( i_old_mode == DEINTERLACE_DISCARD ) )
1989             {
1990                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1991                 return VLC_SUCCESS;
1992             }
1993             break;
1994
1995         case DEINTERLACE_BOB:
1996         case DEINTERLACE_BLEND:
1997         case DEINTERLACE_LINEAR:
1998             if( ( i_old_mode == DEINTERLACE_BOB )
1999                 || ( i_old_mode == DEINTERLACE_BLEND )
2000                 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2001             {
2002                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2003                 return VLC_SUCCESS;
2004             }
2005             break;
2006         }
2007         break;
2008
2009     default:
2010         break;
2011     }
2012
2013     /* We need to kill the old vout */
2014
2015     DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2016
2017     vlc_object_detach( p_vout->p_sys->p_vout );
2018     vout_Destroy( p_vout->p_sys->p_vout );
2019
2020     /* Try to open a new video output */
2021     p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2022
2023     if( p_vout->p_sys->p_vout == NULL )
2024     {
2025         /* Everything failed */
2026         msg_Err( p_vout, "cannot open vout, aborting" );
2027
2028         vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2029         return VLC_EGENERIC;
2030     }
2031
2032     ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2033
2034     vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2035     return VLC_SUCCESS;
2036 }
2037
2038 /*****************************************************************************
2039  * SendEventsToChild: forward events to the child/children vout
2040  *****************************************************************************/
2041 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2042                        vlc_value_t oldval, vlc_value_t newval, void *p_data )
2043 {
2044     vout_thread_t *p_vout = (vout_thread_t *)p_this;
2045     var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2046     return VLC_SUCCESS;
2047 }
2048
2049
2050 /*****************************************************************************
2051  * video filter2 functions
2052  *****************************************************************************/
2053 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2054 {
2055     vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2056     picture_t *p_pic_dst;
2057
2058     /* Request output picture */
2059     p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2060     if( p_pic_dst == NULL )
2061     {
2062         msg_Warn( p_filter, "can't get output picture" );
2063         return NULL;
2064     }
2065
2066     switch( p_vout->p_sys->i_mode )
2067     {
2068         case DEINTERLACE_DISCARD:
2069 #if 0
2070             RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2071 #endif
2072             msg_Err( p_vout, "discarding lines is not supported yet" );
2073             p_pic_dst->pf_release( p_pic_dst );
2074             return p_pic;
2075             break;
2076
2077         case DEINTERLACE_BOB:
2078 #if 0
2079             RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2080             RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2081             break;
2082 #endif
2083
2084         case DEINTERLACE_LINEAR:
2085 #if 0
2086             RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2087             RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2088 #endif
2089             msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2090             p_pic_dst->pf_release( p_pic_dst );
2091             return p_pic;
2092             break;
2093
2094         case DEINTERLACE_MEAN:
2095             RenderMean( p_vout, p_pic_dst, p_pic );
2096             break;
2097
2098         case DEINTERLACE_BLEND:
2099             RenderBlend( p_vout, p_pic_dst, p_pic );
2100             break;
2101
2102         case DEINTERLACE_X:
2103             RenderX( p_vout, p_pic_dst, p_pic );
2104             break;
2105     }
2106
2107     p_pic_dst->date = p_pic->date;
2108     p_pic_dst->b_force = p_pic->b_force;
2109     p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2110     p_pic_dst->b_progressive = VLC_TRUE;
2111     p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2112
2113     p_pic->pf_release( p_pic );
2114     return p_pic_dst;
2115 }
2116
2117 /*****************************************************************************
2118  * OpenFilter:
2119  *****************************************************************************/
2120 static int OpenFilter( vlc_object_t *p_this )
2121 {
2122     filter_t *p_filter = (filter_t*)p_this;
2123     vout_thread_t *p_vout;
2124     vlc_value_t val;
2125
2126     if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2127           p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2128           p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2129         p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2130     {
2131         return VLC_EGENERIC;
2132     }
2133
2134     /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2135      * by spu filters */
2136     p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2137     vlc_object_attach( p_vout, p_filter );
2138     p_filter->p_sys = (filter_sys_t *)p_vout;
2139     p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2140
2141     sout_CfgParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2142                    p_filter->p_cfg );
2143     var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2144     var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2145     var_Set( p_filter, "deinterlace-mode", val );
2146
2147     if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2148     {
2149         vlc_object_detach( p_vout );
2150         vlc_object_release( p_vout );
2151         return VLC_EGENERIC;
2152     }
2153
2154     p_filter->pf_video_filter = Deinterlace;
2155
2156     msg_Dbg( p_filter, "deinterlacing" );
2157
2158     return VLC_SUCCESS;
2159 }
2160
2161 /*****************************************************************************
2162  * CloseFilter: clean up the filter
2163  *****************************************************************************/
2164 static void CloseFilter( vlc_object_t *p_this )
2165 {
2166     filter_t *p_filter = (filter_t*)p_this;
2167     vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2168
2169     Destroy( VLC_OBJECT(p_vout) );
2170     vlc_object_detach( p_vout );
2171     vlc_object_release( p_vout );
2172 }
2173