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., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, 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 MergeMMXEXT  ( void *, const void *, const void *, size_t );
  76 #endif
  77 #if defined(CAN_COMPILE_3DNOW)
  78 static void Merge3DNow   ( void *, const void *, const void *, size_t );
  79 #endif
  80 #if defined(CAN_COMPILE_SSE)
  81 static void MergeSSE2    ( void *, const void *, const void *, size_t );
  82 #endif
  83 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
  84 static void EndMMX       ( void );
  85 #endif
  86 #if defined(CAN_COMPILE_3DNOW)
  87 static void End3DNow     ( void );
  88 #endif
  89
  90 static int  SendEvents   ( vlc_object_t *, char const *,
  91                            vlc_value_t, vlc_value_t, void * );
  92
  93 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method );
  94 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout );
  95
  96 static int OpenFilter( vlc_object_t *p_this );
  97 static void CloseFilter( vlc_object_t *p_this );
  98
  99 /*****************************************************************************
 100  * Callback prototypes
 101  *****************************************************************************/
 102 static int FilterCallback ( vlc_object_t *, char const *,
 103                             vlc_value_t, vlc_value_t, void * );
 104
 105 /*****************************************************************************
 106  * Module descriptor
 107  *****************************************************************************/
 108 #define MODE_TEXT N_("Deinterlace mode")
 109 #define MODE_LONGTEXT N_("Deinterlace method to use for local playback.")
 110
 111 #define SOUT_MODE_TEXT N_("Streaming deinterlace mode")
 112 #define SOUT_MODE_LONGTEXT N_("Deinterlace method to use for streaming.")
 113
 114 #define FILTER_CFG_PREFIX "sout-deinterlace-"
 115
 116 static const char *mode_list[] = { "discard", "blend", "mean", "bob", "linear", "x" };
 117 static const char *mode_list_text[] = { N_("Discard"), N_("Blend"), N_("Mean"),
 118                                   N_("Bob"), N_("Linear"), "X" };
 119
 120 vlc_module_begin();
 121     set_description( _("Deinterlacing video filter") );
 122     set_shortname( _("Deinterlace" ));
 123     set_capability( "video filter", 0 );
 124     set_category( CAT_VIDEO );
 125     set_subcategory( SUBCAT_VIDEO_VFILTER );
 126
 127     set_section( N_("Display"),NULL);
 128     add_string( "deinterlace-mode", "discard", NULL, MODE_TEXT,
 129                 MODE_LONGTEXT, VLC_FALSE );
 130         change_string_list( mode_list, mode_list_text, 0 );
 131
 132     add_shortcut( "deinterlace" );
 133     set_callbacks( Create, Destroy );
 134
 135     add_submodule();
 136     set_capability( "video filter2", 0 );
 137     set_section( N_("Streaming"),NULL);
 138     add_string( FILTER_CFG_PREFIX "mode", "blend", NULL, SOUT_MODE_TEXT,
 139                 SOUT_MODE_LONGTEXT, VLC_FALSE );
 140         change_string_list( mode_list, mode_list_text, 0 );
 141     set_callbacks( OpenFilter, CloseFilter );
 142 vlc_module_end();
 143
 144 static const char *ppsz_filter_options[] = {
 145     "mode", NULL
 146 };
 147
 148 /*****************************************************************************
 149  * vout_sys_t: Deinterlace video output method descriptor
 150  *****************************************************************************
 151  * This structure is part of the video output thread descriptor.
 152  * It describes the Deinterlace specific properties of an output thread.
 153  *****************************************************************************/
 154 struct vout_sys_t
 155 {
 156     int        i_mode;        /* Deinterlace mode */
 157     vlc_bool_t b_double_rate; /* Shall we double the framerate? */
 158
 159     mtime_t    last_date;
 160     mtime_t    next_date;
 161
 162     vout_thread_t *p_vout;
 163
 164     vlc_mutex_t filter_lock;
 165
 166     void (*pf_merge) ( void *, const void *, const void *, size_t );
 167     void (*pf_end_merge) ( void );
 168 };
 169
 170 /*****************************************************************************
 171  * Control: control facility for the vout (forwards to child vout)
 172  *****************************************************************************/
 173 static int Control( vout_thread_t *p_vout, int i_query, va_list args )
 174 {
 175     return vout_vaControl( p_vout->p_sys->p_vout, i_query, args );
 176 }
 177
 178 /*****************************************************************************
 179  * Create: allocates Deinterlace video thread output method
 180  *****************************************************************************
 181  * This function allocates and initializes a Deinterlace vout method.
 182  *****************************************************************************/
 183 static int Create( vlc_object_t *p_this )
 184 {
 185     vout_thread_t *p_vout = (vout_thread_t *)p_this;
 186     vlc_value_t val;
 187
 188     /* Allocate structure */
 189     p_vout->p_sys = malloc( sizeof( vout_sys_t ) );
 190     if( p_vout->p_sys == NULL )
 191     {
 192         msg_Err( p_vout, "out of memory" );
 193         return VLC_ENOMEM;
 194     }
 195
 196     p_vout->pf_init = Init;
 197     p_vout->pf_end = End;
 198     p_vout->pf_manage = NULL;
 199     p_vout->pf_render = Render;
 200     p_vout->pf_display = NULL;
 201     p_vout->pf_control = Control;
 202
 203     p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
 204     p_vout->p_sys->b_double_rate = VLC_FALSE;
 205     p_vout->p_sys->last_date = 0;
 206     p_vout->p_sys->p_vout = 0;
 207     vlc_mutex_init( p_vout, &p_vout->p_sys->filter_lock );
 208
 209 #if defined(CAN_COMPILE_C_ALTIVEC)
 210     if( vlc_CPU() & CPU_CAPABILITY_ALTIVEC )
 211     {
 212         p_vout->p_sys->pf_merge = MergeAltivec;
 213         p_vout->p_sys->pf_end_merge = NULL;
 214     }
 215     else
 216 #endif
 217 #if defined(CAN_COMPILE_SSE)
 218     if( vlc_CPU() & CPU_CAPABILITY_SSE2 )
 219     {
 220         p_vout->p_sys->pf_merge = MergeSSE2;
 221         p_vout->p_sys->pf_end_merge = EndMMX;
 222     }
 223     else
 224 #endif
 225 #if defined(CAN_COMPILE_MMXEXT)
 226     if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
 227     {
 228         p_vout->p_sys->pf_merge = MergeMMXEXT;
 229         p_vout->p_sys->pf_end_merge = EndMMX;
 230     }
 231     else
 232 #endif
 233 #if defined(CAN_COMPILE_3DNOW)
 234     if( vlc_CPU() & CPU_CAPABILITY_3DNOW )
 235     {
 236         p_vout->p_sys->pf_merge = Merge3DNow;
 237         p_vout->p_sys->pf_end_merge = End3DNow;
 238     }
 239     else
 240 #endif
 241     {
 242         p_vout->p_sys->pf_merge = MergeGeneric;
 243         p_vout->p_sys->pf_end_merge = NULL;
 244     }
 245
 246     /* Look what method was requested */
 247     var_Create( p_vout, "deinterlace-mode", VLC_VAR_STRING );
 248     var_Change( p_vout, "deinterlace-mode", VLC_VAR_INHERITVALUE, &val, NULL );
 249
 250     if( val.psz_string == NULL )
 251     {
 252         msg_Err( p_vout, "configuration variable deinterlace-mode empty" );
 253         msg_Err( p_vout, "no deinterlace mode provided, using \"discard\"" );
 254
 255         val.psz_string = strdup( "discard" );
 256     }
 257
 258     msg_Dbg( p_vout, "using %s deinterlace mode", val.psz_string );
 259
 260     SetFilterMethod( p_vout, val.psz_string );
 261
 262     free( val.psz_string );
 263
 264     return VLC_SUCCESS;
 265 }
 266
 267 /*****************************************************************************
 268  * SetFilterMethod: setup the deinterlace method to use.
 269  *****************************************************************************/
 270 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method )
 271 {
 272     if( !strcmp( psz_method, "discard" ) )
 273     {
 274         p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
 275         p_vout->p_sys->b_double_rate = VLC_FALSE;
 276     }
 277     else if( !strcmp( psz_method, "mean" ) )
 278     {
 279         p_vout->p_sys->i_mode = DEINTERLACE_MEAN;
 280         p_vout->p_sys->b_double_rate = VLC_FALSE;
 281     }
 282     else if( !strcmp( psz_method, "blend" )
 283              || !strcmp( psz_method, "average" )
 284              || !strcmp( psz_method, "combine-fields" ) )
 285     {
 286         p_vout->p_sys->i_mode = DEINTERLACE_BLEND;
 287         p_vout->p_sys->b_double_rate = VLC_FALSE;
 288     }
 289     else if( !strcmp( psz_method, "bob" )
 290              || !strcmp( psz_method, "progressive-scan" ) )
 291     {
 292         p_vout->p_sys->i_mode = DEINTERLACE_BOB;
 293         p_vout->p_sys->b_double_rate = VLC_TRUE;
 294     }
 295     else if( !strcmp( psz_method, "linear" ) )
 296     {
 297         p_vout->p_sys->i_mode = DEINTERLACE_LINEAR;
 298         p_vout->p_sys->b_double_rate = VLC_TRUE;
 299     }
 300     else if( !strcmp( psz_method, "x" ) )
 301     {
 302         p_vout->p_sys->i_mode = DEINTERLACE_X;
 303         p_vout->p_sys->b_double_rate = VLC_FALSE;
 304     }
 305     else
 306     {
 307         msg_Err( p_vout, "no valid deinterlace mode provided, "
 308                  "using \"discard\"" );
 309     }
 310
 311     msg_Dbg( p_vout, "using %s deinterlace method", psz_method );
 312 }
 313
 314 /*****************************************************************************
 315  * Init: initialize Deinterlace video thread output method
 316  *****************************************************************************/
 317 static int Init( vout_thread_t *p_vout )
 318 {
 319     int i_index;
 320     picture_t *p_pic;
 321
 322     I_OUTPUTPICTURES = 0;
 323
 324     /* Initialize the output structure, full of directbuffers since we want
 325      * the decoder to output directly to our structures. */
 326     switch( p_vout->render.i_chroma )
 327     {
 328         case VLC_FOURCC('I','4','2','0'):
 329         case VLC_FOURCC('I','Y','U','V'):
 330         case VLC_FOURCC('Y','V','1','2'):
 331         case VLC_FOURCC('I','4','2','2'):
 332             p_vout->output.i_chroma = p_vout->render.i_chroma;
 333             p_vout->output.i_width  = p_vout->render.i_width;
 334             p_vout->output.i_height = p_vout->render.i_height;
 335             p_vout->output.i_aspect = p_vout->render.i_aspect;
 336             p_vout->fmt_out = p_vout->fmt_in;
 337             break;
 338
 339         default:
 340             return VLC_EGENERIC; /* unknown chroma */
 341             break;
 342     }
 343
 344     /* Try to open the real video output */
 345     p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
 346
 347     if( p_vout->p_sys->p_vout == NULL )
 348     {
 349         /* Everything failed */
 350         msg_Err( p_vout, "cannot open vout, aborting" );
 351
 352         return VLC_EGENERIC;
 353     }
 354
 355     var_AddCallback( p_vout, "deinterlace-mode", FilterCallback, NULL );
 356
 357     ALLOCATE_DIRECTBUFFERS( VOUT_MAX_PICTURES );
 358
 359     ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
 360
 361     ADD_PARENT_CALLBACKS( SendEventsToChild );
 362
 363     return VLC_SUCCESS;
 364 }
 365
 366 /*****************************************************************************
 367  * SpawnRealVout: spawn the real video output.
 368  *****************************************************************************/
 369 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout )
 370 {
 371     vout_thread_t *p_real_vout = NULL;
 372     video_format_t fmt;
 373     memset( &fmt, 0, sizeof( video_format_t ) );
 374
 375     msg_Dbg( p_vout, "spawning the real video output" );
 376
 377     fmt = p_vout->fmt_out;
 378
 379     switch( p_vout->render.i_chroma )
 380     {
 381     case VLC_FOURCC('I','4','2','0'):
 382     case VLC_FOURCC('I','Y','U','V'):
 383     case VLC_FOURCC('Y','V','1','2'):
 384         switch( p_vout->p_sys->i_mode )
 385         {
 386         case DEINTERLACE_MEAN:
 387         case DEINTERLACE_DISCARD:
 388             fmt.i_height /= 2; fmt.i_visible_height /= 2; fmt.i_y_offset /= 2;
 389             fmt.i_sar_den *= 2;
 390             p_real_vout = vout_Create( p_vout, &fmt );
 391             break;
 392
 393         case DEINTERLACE_BOB:
 394         case DEINTERLACE_BLEND:
 395         case DEINTERLACE_LINEAR:
 396         case DEINTERLACE_X:
 397             p_real_vout = vout_Create( p_vout, &fmt );
 398             break;
 399         }
 400         break;
 401
 402     case VLC_FOURCC('I','4','2','2'):
 403         fmt.i_chroma = VLC_FOURCC('I','4','2','0');
 404         p_real_vout = vout_Create( p_vout, &fmt );
 405         break;
 406
 407     default:
 408         break;
 409     }
 410
 411     return p_real_vout;
 412 }
 413
 414 /*****************************************************************************
 415  * End: terminate Deinterlace video thread output method
 416  *****************************************************************************/
 417 static void End( vout_thread_t *p_vout )
 418 {
 419     int i_index;
 420
 421     /* Free the fake output buffers we allocated */
 422     for( i_index = I_OUTPUTPICTURES ; i_index ; )
 423     {
 424         i_index--;
 425         free( PP_OUTPUTPICTURE[ i_index ]->p_data_orig );
 426     }
 427
 428     if( p_vout->p_sys->p_vout )
 429     {
 430         DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
 431         vlc_object_detach( p_vout->p_sys->p_vout );
 432         vout_Destroy( p_vout->p_sys->p_vout );
 433     }
 434
 435     DEL_PARENT_CALLBACKS( SendEventsToChild );
 436 }
 437
 438 /*****************************************************************************
 439  * Destroy: destroy Deinterlace video thread output method
 440  *****************************************************************************
 441  * Terminate an output method created by DeinterlaceCreateOutputMethod
 442  *****************************************************************************/
 443 static void Destroy( vlc_object_t *p_this )
 444 {
 445     vout_thread_t *p_vout = (vout_thread_t *)p_this;
 446     vlc_mutex_destroy( &p_vout->p_sys->filter_lock );
 447     free( p_vout->p_sys );
 448 }
 449
 450 /*****************************************************************************
 451  * Render: displays previously rendered output
 452  *****************************************************************************
 453  * This function send the currently rendered image to Deinterlace image,
 454  * waits until it is displayed and switch the two rendering buffers, preparing
 455  * next frame.
 456  *****************************************************************************/
 457 static void Render ( vout_thread_t *p_vout, picture_t *p_pic )
 458 {
 459     vout_sys_t *p_sys = p_vout->p_sys;
 460     picture_t *pp_outpic[2];
 461
 462     p_vout->fmt_out.i_x_offset = p_sys->p_vout->fmt_in.i_x_offset =
 463         p_vout->fmt_in.i_x_offset;
 464     p_vout->fmt_out.i_y_offset = p_sys->p_vout->fmt_in.i_y_offset =
 465         p_vout->fmt_in.i_y_offset;
 466     p_vout->fmt_out.i_visible_width = p_sys->p_vout->fmt_in.i_visible_width =
 467         p_vout->fmt_in.i_visible_width;
 468     p_vout->fmt_out.i_visible_height = p_sys->p_vout->fmt_in.i_visible_height =
 469         p_vout->fmt_in.i_visible_height;
 470     if( p_vout->p_sys->i_mode == DEINTERLACE_MEAN ||
 471         p_vout->p_sys->i_mode == DEINTERLACE_DISCARD )
 472     {
 473         p_vout->fmt_out.i_y_offset /= 2; p_sys->p_vout->fmt_in.i_y_offset /= 2;
 474         p_vout->fmt_out.i_visible_height /= 2;
 475         p_sys->p_vout->fmt_in.i_visible_height /= 2;
 476     }
 477
 478     pp_outpic[0] = pp_outpic[1] = NULL;
 479
 480     vlc_mutex_lock( &p_vout->p_sys->filter_lock );
 481
 482     /* Get a new picture */
 483     while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
 484                                                 0, 0, 0 ) )
 485               == NULL )
 486     {
 487         if( p_vout->b_die || p_vout->b_error )
 488         {
 489             vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 490             return;
 491         }
 492         msleep( VOUT_OUTMEM_SLEEP );
 493     }
 494
 495     vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
 496
 497     /* If we are using double rate, get an additional new picture */
 498     if( p_vout->p_sys->b_double_rate )
 499     {
 500         while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
 501                                                  0, 0, 0 ) )
 502                   == NULL )
 503         {
 504             if( p_vout->b_die || p_vout->b_error )
 505             {
 506                 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 507                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 508                 return;
 509             }
 510             msleep( VOUT_OUTMEM_SLEEP );
 511         }
 512
 513         /* 20ms is a bit arbitrary, but it's only for the first image we get */
 514         if( !p_vout->p_sys->last_date )
 515         {
 516             vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
 517                               p_pic->date + 20000 );
 518         }
 519         else
 520         {
 521             vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
 522                       (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
 523         }
 524         p_vout->p_sys->last_date = p_pic->date;
 525     }
 526
 527     switch( p_vout->p_sys->i_mode )
 528     {
 529         case DEINTERLACE_DISCARD:
 530             RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
 531             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 532             break;
 533
 534         case DEINTERLACE_BOB:
 535             RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
 536             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 537             RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
 538             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
 539             break;
 540
 541         case DEINTERLACE_LINEAR:
 542             RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
 543             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 544             RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
 545             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
 546             break;
 547
 548         case DEINTERLACE_MEAN:
 549             RenderMean( p_vout, pp_outpic[0], p_pic );
 550             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 551             break;
 552
 553         case DEINTERLACE_BLEND:
 554             RenderBlend( p_vout, pp_outpic[0], p_pic );
 555             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 556             break;
 557
 558         case DEINTERLACE_X:
 559             RenderX( p_vout, pp_outpic[0], p_pic );
 560             vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
 561             break;
 562     }
 563     vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
 564 }
 565
 566 /*****************************************************************************
 567  * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
 568  *****************************************************************************/
 569 static void RenderDiscard( vout_thread_t *p_vout,
 570                            picture_t *p_outpic, picture_t *p_pic, int i_field )
 571 {
 572     int i_plane;
 573
 574     /* Copy image and skip lines */
 575     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 576     {
 577         uint8_t *p_in, *p_out_end, *p_out;
 578         int i_increment;
 579
 580         p_in = p_pic->p[i_plane].p_pixels
 581                    + i_field * p_pic->p[i_plane].i_pitch;
 582
 583         p_out = p_outpic->p[i_plane].p_pixels;
 584         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 585                              * p_outpic->p[i_plane].i_visible_lines;
 586
 587         switch( p_vout->render.i_chroma )
 588         {
 589         case VLC_FOURCC('I','4','2','0'):
 590         case VLC_FOURCC('I','Y','U','V'):
 591         case VLC_FOURCC('Y','V','1','2'):
 592
 593             for( ; p_out < p_out_end ; )
 594             {
 595                 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 596                                           p_pic->p[i_plane].i_pitch );
 597
 598                 p_out += p_outpic->p[i_plane].i_pitch;
 599                 p_in += 2 * p_pic->p[i_plane].i_pitch;
 600             }
 601             break;
 602
 603         case VLC_FOURCC('I','4','2','2'):
 604
 605             i_increment = 2 * p_pic->p[i_plane].i_pitch;
 606
 607             if( i_plane == Y_PLANE )
 608             {
 609                 for( ; p_out < p_out_end ; )
 610                 {
 611                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 612                                               p_pic->p[i_plane].i_pitch );
 613                     p_out += p_outpic->p[i_plane].i_pitch;
 614                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 615                                               p_pic->p[i_plane].i_pitch );
 616                     p_out += p_outpic->p[i_plane].i_pitch;
 617                     p_in += i_increment;
 618                 }
 619             }
 620             else
 621             {
 622                 for( ; p_out < p_out_end ; )
 623                 {
 624                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 625                                               p_pic->p[i_plane].i_pitch );
 626                     p_out += p_outpic->p[i_plane].i_pitch;
 627                     p_in += i_increment;
 628                 }
 629             }
 630             break;
 631
 632         default:
 633             break;
 634         }
 635     }
 636 }
 637
 638 /*****************************************************************************
 639  * RenderBob: renders a BOB picture - simple copy
 640  *****************************************************************************/
 641 static void RenderBob( vout_thread_t *p_vout,
 642                        picture_t *p_outpic, picture_t *p_pic, int i_field )
 643 {
 644     int i_plane;
 645
 646     /* Copy image and skip lines */
 647     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 648     {
 649         uint8_t *p_in, *p_out_end, *p_out;
 650
 651         p_in = p_pic->p[i_plane].p_pixels;
 652         p_out = p_outpic->p[i_plane].p_pixels;
 653         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 654                              * p_outpic->p[i_plane].i_visible_lines;
 655
 656         switch( p_vout->render.i_chroma )
 657         {
 658             case VLC_FOURCC('I','4','2','0'):
 659             case VLC_FOURCC('I','Y','U','V'):
 660             case VLC_FOURCC('Y','V','1','2'):
 661                 /* For BOTTOM field we need to add the first line */
 662                 if( i_field == 1 )
 663                 {
 664                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 665                                               p_pic->p[i_plane].i_pitch );
 666                     p_in += p_pic->p[i_plane].i_pitch;
 667                     p_out += p_outpic->p[i_plane].i_pitch;
 668                 }
 669
 670                 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 671
 672                 for( ; p_out < p_out_end ; )
 673                 {
 674                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 675                                               p_pic->p[i_plane].i_pitch );
 676
 677                     p_out += p_outpic->p[i_plane].i_pitch;
 678
 679                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 680                                               p_pic->p[i_plane].i_pitch );
 681
 682                     p_in += 2 * p_pic->p[i_plane].i_pitch;
 683                     p_out += p_outpic->p[i_plane].i_pitch;
 684                 }
 685
 686                 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 687                                           p_pic->p[i_plane].i_pitch );
 688
 689                 /* For TOP field we need to add the last line */
 690                 if( i_field == 0 )
 691                 {
 692                     p_in += p_pic->p[i_plane].i_pitch;
 693                     p_out += p_outpic->p[i_plane].i_pitch;
 694                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 695                                               p_pic->p[i_plane].i_pitch );
 696                 }
 697                 break;
 698
 699             case VLC_FOURCC('I','4','2','2'):
 700                 /* For BOTTOM field we need to add the first line */
 701                 if( i_field == 1 )
 702                 {
 703                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 704                                               p_pic->p[i_plane].i_pitch );
 705                     p_in += p_pic->p[i_plane].i_pitch;
 706                     p_out += p_outpic->p[i_plane].i_pitch;
 707                 }
 708
 709                 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 710
 711                 if( i_plane == Y_PLANE )
 712                 {
 713                     for( ; p_out < p_out_end ; )
 714                     {
 715                         p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 716                                                   p_pic->p[i_plane].i_pitch );
 717
 718                         p_out += p_outpic->p[i_plane].i_pitch;
 719
 720                         p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 721                                                   p_pic->p[i_plane].i_pitch );
 722
 723                         p_in += 2 * p_pic->p[i_plane].i_pitch;
 724                         p_out += p_outpic->p[i_plane].i_pitch;
 725                     }
 726                 }
 727                 else
 728                 {
 729                     for( ; p_out < p_out_end ; )
 730                     {
 731                         p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 732                                                   p_pic->p[i_plane].i_pitch );
 733
 734                         p_out += p_outpic->p[i_plane].i_pitch;
 735                         p_in += 2 * p_pic->p[i_plane].i_pitch;
 736                     }
 737                 }
 738
 739                 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 740                                           p_pic->p[i_plane].i_pitch );
 741
 742                 /* For TOP field we need to add the last line */
 743                 if( i_field == 0 )
 744                 {
 745                     p_in += p_pic->p[i_plane].i_pitch;
 746                     p_out += p_outpic->p[i_plane].i_pitch;
 747                     p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 748                                               p_pic->p[i_plane].i_pitch );
 749                 }
 750                 break;
 751         }
 752     }
 753 }
 754
 755 #define Merge p_vout->p_sys->pf_merge
 756 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
 757
 758 /*****************************************************************************
 759  * RenderLinear: BOB with linear interpolation
 760  *****************************************************************************/
 761 static void RenderLinear( vout_thread_t *p_vout,
 762                           picture_t *p_outpic, picture_t *p_pic, int i_field )
 763 {
 764     int i_plane;
 765
 766     /* Copy image and skip lines */
 767     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 768     {
 769         uint8_t *p_in, *p_out_end, *p_out;
 770
 771         p_in = p_pic->p[i_plane].p_pixels;
 772         p_out = p_outpic->p[i_plane].p_pixels;
 773         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 774                              * p_outpic->p[i_plane].i_visible_lines;
 775
 776         /* For BOTTOM field we need to add the first line */
 777         if( i_field == 1 )
 778         {
 779             p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 780                                       p_pic->p[i_plane].i_pitch );
 781             p_in += p_pic->p[i_plane].i_pitch;
 782             p_out += p_outpic->p[i_plane].i_pitch;
 783         }
 784
 785         p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
 786
 787         for( ; p_out < p_out_end ; )
 788         {
 789             p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 790                                       p_pic->p[i_plane].i_pitch );
 791
 792             p_out += p_outpic->p[i_plane].i_pitch;
 793
 794             Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
 795                    p_pic->p[i_plane].i_pitch );
 796
 797             p_in += 2 * p_pic->p[i_plane].i_pitch;
 798             p_out += p_outpic->p[i_plane].i_pitch;
 799         }
 800
 801         p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 802                                   p_pic->p[i_plane].i_pitch );
 803
 804         /* For TOP field we need to add the last line */
 805         if( i_field == 0 )
 806         {
 807             p_in += p_pic->p[i_plane].i_pitch;
 808             p_out += p_outpic->p[i_plane].i_pitch;
 809             p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 810                                       p_pic->p[i_plane].i_pitch );
 811         }
 812     }
 813     EndMerge();
 814 }
 815
 816 static void RenderMean( vout_thread_t *p_vout,
 817                         picture_t *p_outpic, picture_t *p_pic )
 818 {
 819     int i_plane;
 820
 821     /* Copy image and skip lines */
 822     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 823     {
 824         uint8_t *p_in, *p_out_end, *p_out;
 825
 826         p_in = p_pic->p[i_plane].p_pixels;
 827
 828         p_out = p_outpic->p[i_plane].p_pixels;
 829         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 830                              * p_outpic->p[i_plane].i_visible_lines;
 831
 832         /* All lines: mean value */
 833         for( ; p_out < p_out_end ; )
 834         {
 835             Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 836                    p_pic->p[i_plane].i_pitch );
 837
 838             p_out += p_outpic->p[i_plane].i_pitch;
 839             p_in += 2 * p_pic->p[i_plane].i_pitch;
 840         }
 841     }
 842     EndMerge();
 843 }
 844
 845 static void RenderBlend( vout_thread_t *p_vout,
 846                          picture_t *p_outpic, picture_t *p_pic )
 847 {
 848     int i_plane;
 849
 850     /* Copy image and skip lines */
 851     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
 852     {
 853         uint8_t *p_in, *p_out_end, *p_out;
 854
 855         p_in = p_pic->p[i_plane].p_pixels;
 856
 857         p_out = p_outpic->p[i_plane].p_pixels;
 858         p_out_end = p_out + p_outpic->p[i_plane].i_pitch
 859                              * p_outpic->p[i_plane].i_visible_lines;
 860
 861         switch( p_vout->render.i_chroma )
 862         {
 863             case VLC_FOURCC('I','4','2','0'):
 864             case VLC_FOURCC('I','Y','U','V'):
 865             case VLC_FOURCC('Y','V','1','2'):
 866                 /* First line: simple copy */
 867                 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 868                                           p_pic->p[i_plane].i_pitch );
 869                 p_out += p_outpic->p[i_plane].i_pitch;
 870
 871                 /* Remaining lines: mean value */
 872                 for( ; p_out < p_out_end ; )
 873                 {
 874                     Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 875                            p_pic->p[i_plane].i_pitch );
 876
 877                     p_out += p_outpic->p[i_plane].i_pitch;
 878                     p_in += p_pic->p[i_plane].i_pitch;
 879                 }
 880                 break;
 881
 882             case VLC_FOURCC('I','4','2','2'):
 883                 /* First line: simple copy */
 884                 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
 885                                           p_pic->p[i_plane].i_pitch );
 886                 p_out += p_outpic->p[i_plane].i_pitch;
 887
 888                 /* Remaining lines: mean value */
 889                 if( i_plane == Y_PLANE )
 890                 {
 891                     for( ; p_out < p_out_end ; )
 892                     {
 893                         Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 894                                p_pic->p[i_plane].i_pitch );
 895
 896                         p_out += p_outpic->p[i_plane].i_pitch;
 897                         p_in += p_pic->p[i_plane].i_pitch;
 898                     }
 899                 }
 900
 901                 else
 902                 {
 903                     for( ; p_out < p_out_end ; )
 904                     {
 905                         Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
 906                                p_pic->p[i_plane].i_pitch );
 907
 908                         p_out += p_outpic->p[i_plane].i_pitch;
 909                         p_in += 2*p_pic->p[i_plane].i_pitch;
 910                     }
 911                 }
 912                 break;
 913         }
 914     }
 915     EndMerge();
 916 }
 917
 918 #undef Merge
 919
 920 static void MergeGeneric( void *_p_dest, const void *_p_s1,
 921                           const void *_p_s2, size_t i_bytes )
 922 {
 923     uint8_t* p_dest = (uint8_t*)_p_dest;
 924     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 925     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 926     uint8_t* p_end = p_dest + i_bytes - 8;
 927
 928     while( p_dest < p_end )
 929     {
 930         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 931         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 932         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 933         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 934         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 935         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 936         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 937         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 938     }
 939
 940     p_end += 8;
 941
 942     while( p_dest < p_end )
 943     {
 944         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 945     }
 946 }
 947
 948 #if defined(CAN_COMPILE_MMXEXT)
 949 static void MergeMMXEXT( void *_p_dest, const void *_p_s1, const void *_p_s2,
 950                          size_t i_bytes )
 951 {
 952     uint8_t* p_dest = (uint8_t*)_p_dest;
 953     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 954     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 955     uint8_t* p_end = p_dest + i_bytes - 8;
 956     while( p_dest < p_end )
 957     {
 958         __asm__  __volatile__( "movq %2,%%mm1;"
 959                                "pavgb %1, %%mm1;"
 960                                "movq %%mm1, %0" :"=m" (*p_dest):
 961                                                  "m" (*p_s1),
 962                                                  "m" (*p_s2) );
 963         p_dest += 8;
 964         p_s1 += 8;
 965         p_s2 += 8;
 966     }
 967
 968     p_end += 8;
 969
 970     while( p_dest < p_end )
 971     {
 972         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
 973     }
 974 }
 975 #endif
 976
 977 #if defined(CAN_COMPILE_3DNOW)
 978 static void Merge3DNow( void *_p_dest, const void *_p_s1, const void *_p_s2,
 979                         size_t i_bytes )
 980 {
 981     uint8_t* p_dest = (uint8_t*)_p_dest;
 982     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
 983     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
 984     uint8_t* p_end = p_dest + i_bytes - 8;
 985     while( p_dest < p_end )
 986     {
 987         __asm__  __volatile__( "movq %2,%%mm1;"
 988                                "pavgusb %1, %%mm1;"
 989                                "movq %%mm1, %0" :"=m" (*p_dest):
 990                                                  "m" (*p_s1),
 991                                                  "m" (*p_s2) );
 992         p_dest += 8;
 993         p_s1 += 8;
 994         p_s2 += 8;
 995     }
 996
 997     p_end += 8;
 998
 999     while( p_dest < p_end )
1000     {
1001         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1002     }
1003 }
1004 #endif
1005
1006 #if defined(CAN_COMPILE_SSE)
1007 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
1008                        size_t i_bytes )
1009 {
1010     uint8_t* p_dest = (uint8_t*)_p_dest;
1011     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
1012     const uint8_t *p_s2 = (const uint8_t *)_p_s2;
1013     uint8_t* p_end;
1014     while( (ptrdiff_t)p_s1 % 16 )
1015     {
1016         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1017     }
1018     p_end = p_dest + i_bytes - 16;
1019     while( p_dest < p_end )
1020     {
1021         __asm__  __volatile__( "movdqu %2,%%xmm1;"
1022                                "pavgb %1, %%xmm1;"
1023                                "movdqu %%xmm1, %0" :"=m" (*p_dest):
1024                                                  "m" (*p_s1),
1025                                                  "m" (*p_s2) );
1026         p_dest += 16;
1027         p_s1 += 16;
1028         p_s2 += 16;
1029     }
1030
1031     p_end += 16;
1032
1033     while( p_dest < p_end )
1034     {
1035         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1036     }
1037 }
1038 #endif
1039
1040 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
1041 static void EndMMX( void )
1042 {
1043     __asm__ __volatile__( "emms" :: );
1044 }
1045 #endif
1046
1047 #if defined(CAN_COMPILE_3DNOW)
1048 static void End3DNow( void )
1049 {
1050     __asm__ __volatile__( "femms" :: );
1051 }
1052 #endif
1053
1054 #ifdef CAN_COMPILE_C_ALTIVEC
1055 static void MergeAltivec( void *_p_dest, const void *_p_s1,
1056                           const void *_p_s2, size_t i_bytes )
1057 {
1058     uint8_t *p_dest = (uint8_t *)_p_dest;
1059     uint8_t *p_s1   = (uint8_t *)_p_s1;
1060     uint8_t *p_s2   = (uint8_t *)_p_s2;
1061     uint8_t *p_end  = p_dest + i_bytes - 15;
1062
1063     /* Use C until the first 16-bytes aligned destination pixel */
1064     while( (int)p_dest & 0xF )
1065     {
1066         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1067     }
1068
1069     if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1070     {
1071         /* Unaligned source */
1072         vector unsigned char s1v, s2v, destv;
1073         vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1074         vector unsigned char perm1v, perm2v;
1075
1076         perm1v = vec_lvsl( 0, p_s1 );
1077         perm2v = vec_lvsl( 0, p_s2 );
1078         s1oldv = vec_ld( 0, p_s1 );
1079         s2oldv = vec_ld( 0, p_s2 );
1080
1081         while( p_dest < p_end )
1082         {
1083             s1newv = vec_ld( 16, p_s1 );
1084             s2newv = vec_ld( 16, p_s2 );
1085             s1v    = vec_perm( s1oldv, s1newv, perm1v );
1086             s2v    = vec_perm( s2oldv, s2newv, perm2v );
1087             s1oldv = s1newv;
1088             s2oldv = s2newv;
1089             destv  = vec_avg( s1v, s2v );
1090             vec_st( destv, 0, p_dest );
1091
1092             p_s1   += 16;
1093             p_s2   += 16;
1094             p_dest += 16;
1095         }
1096     }
1097     else
1098     {
1099         /* Aligned source */
1100         vector unsigned char s1v, s2v, destv;
1101
1102         while( p_dest < p_end )
1103         {
1104             s1v   = vec_ld( 0, p_s1 );
1105             s2v   = vec_ld( 0, p_s2 );
1106             destv = vec_avg( s1v, s2v );
1107             vec_st( destv, 0, p_dest );
1108
1109             p_s1   += 16;
1110             p_s2   += 16;
1111             p_dest += 16;
1112         }
1113     }
1114
1115     p_end += 15;
1116
1117     while( p_dest < p_end )
1118     {
1119         *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1120     }
1121 }
1122 #endif
1123
1124 /*****************************************************************************
1125  * RenderX: This algo works on a 8x8 block basic, it copies the top field
1126  * and apply a process to recreate the bottom field :
1127  *  If a 8x8 block is classified as :
1128  *   - progressive: it applies a small blend (1,6,1)
1129  *   - interlaced:
1130  *    * in the MMX version: we do a ME between the 2 fields, if there is a
1131  *    good match we use MC to recreate the bottom field (with a small
1132  *    blend (1,6,1) )
1133  *    * otherwise: it recreates the bottom field by an edge oriented
1134  *    interpolation.
1135   *****************************************************************************/
1136
1137 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1138  * XXX: It need to access to 8x10
1139  * We use more than 8 lines to help with scrolling (text)
1140  * (and because XDeint8x8Frame use line 9)
1141  * XXX: smooth/uniform area with noise detection doesn't works well
1142  * but it's not really a problem because they don't have much details anyway
1143  */
1144 static inline int ssd( int a ) { return a*a; }
1145 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1146 {
1147     int y, x;
1148     int ff, fr;
1149     int fc;
1150
1151     /* Detect interlacing */
1152     fc = 0;
1153     for( y = 0; y < 7; y += 2 )
1154     {
1155         ff = fr = 0;
1156         for( x = 0; x < 8; x++ )
1157         {
1158             fr += ssd(src[      x] - src[1*i_src+x]) +
1159                   ssd(src[i_src+x] - src[2*i_src+x]);
1160             ff += ssd(src[      x] - src[2*i_src+x]) +
1161                   ssd(src[i_src+x] - src[3*i_src+x]);
1162         }
1163         if( ff < 6*fr/8 && fr > 32 )
1164             fc++;
1165
1166         src += 2*i_src;
1167     }
1168
1169     return fc < 1 ? VLC_FALSE : VLC_TRUE;
1170 }
1171 #ifdef CAN_COMPILE_MMXEXT
1172 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1173 {
1174
1175     int y, x;
1176     int32_t ff, fr;
1177     int fc;
1178
1179     /* Detect interlacing */
1180     fc = 0;
1181     pxor_r2r( mm7, mm7 );
1182     for( y = 0; y < 9; y += 2 )
1183     {
1184         ff = fr = 0;
1185         pxor_r2r( mm5, mm5 );
1186         pxor_r2r( mm6, mm6 );
1187         for( x = 0; x < 8; x+=4 )
1188         {
1189             movd_m2r( src[        x], mm0 );
1190             movd_m2r( src[1*i_src+x], mm1 );
1191             movd_m2r( src[2*i_src+x], mm2 );
1192             movd_m2r( src[3*i_src+x], mm3 );
1193
1194             punpcklbw_r2r( mm7, mm0 );
1195             punpcklbw_r2r( mm7, mm1 );
1196             punpcklbw_r2r( mm7, mm2 );
1197             punpcklbw_r2r( mm7, mm3 );
1198
1199             movq_r2r( mm0, mm4 );
1200
1201             psubw_r2r( mm1, mm0 );
1202             psubw_r2r( mm2, mm4 );
1203
1204             psubw_r2r( mm1, mm2 );
1205             psubw_r2r( mm1, mm3 );
1206
1207             pmaddwd_r2r( mm0, mm0 );
1208             pmaddwd_r2r( mm4, mm4 );
1209             pmaddwd_r2r( mm2, mm2 );
1210             pmaddwd_r2r( mm3, mm3 );
1211             paddd_r2r( mm0, mm2 );
1212             paddd_r2r( mm4, mm3 );
1213             paddd_r2r( mm2, mm5 );
1214             paddd_r2r( mm3, mm6 );
1215         }
1216
1217         movq_r2r( mm5, mm0 );
1218         psrlq_i2r( 32, mm0 );
1219         paddd_r2r( mm0, mm5 );
1220         movd_r2m( mm5, fr );
1221
1222         movq_r2r( mm6, mm0 );
1223         psrlq_i2r( 32, mm0 );
1224         paddd_r2r( mm0, mm6 );
1225         movd_r2m( mm6, ff );
1226
1227         if( ff < 6*fr/8 && fr > 32 )
1228             fc++;
1229
1230         src += 2*i_src;
1231     }
1232     return fc;
1233 }
1234 #endif
1235
1236 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1237  * This won't destroy details, and help if there is a bit of interlacing.
1238  * (It helps with paning to avoid flickers)
1239  * (Use 8x9 pixels)
1240  */
1241 #if 0
1242 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1243                                     uint8_t *src, int i_src )
1244 {
1245     int y, x;
1246
1247     /* Progressive */
1248     for( y = 0; y < 8; y += 2 )
1249     {
1250         memcpy( dst, src, 8 );
1251         dst += i_dst;
1252
1253         for( x = 0; x < 8; x++ )
1254             dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1255         dst += 1*i_dst;
1256         src += 2*i_src;
1257     }
1258 }
1259 #endif
1260 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1261                                     uint8_t *src1, int i_src1,
1262                                     uint8_t *src2, int i_src2 )
1263 {
1264     int y, x;
1265
1266     /* Progressive */
1267     for( y = 0; y < 8; y += 2 )
1268     {
1269         memcpy( dst, src1, 8 );
1270         dst  += i_dst;
1271
1272         for( x = 0; x < 8; x++ )
1273             dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1274         dst += i_dst;
1275
1276         src1 += i_src1;
1277         src2 += i_src2;
1278     }
1279 }
1280
1281 #ifdef CAN_COMPILE_MMXEXT
1282 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1283                                          uint8_t *src1, int i_src1,
1284                                          uint8_t *src2, int i_src2 )
1285 {
1286     static const uint64_t m_4 = I64C(0x0004000400040004);
1287     int y, x;
1288
1289     /* Progressive */
1290     pxor_r2r( mm7, mm7 );
1291     for( y = 0; y < 8; y += 2 )
1292     {
1293         for( x = 0; x < 8; x +=4 )
1294         {
1295             movd_m2r( src1[x], mm0 );
1296             movd_r2m( mm0, dst[x] );
1297
1298             movd_m2r( src2[x], mm1 );
1299             movd_m2r( src1[i_src1+x], mm2 );
1300
1301             punpcklbw_r2r( mm7, mm0 );
1302             punpcklbw_r2r( mm7, mm1 );
1303             punpcklbw_r2r( mm7, mm2 );
1304             paddw_r2r( mm1, mm1 );
1305             movq_r2r( mm1, mm3 );
1306             paddw_r2r( mm3, mm3 );
1307             paddw_r2r( mm2, mm0 );
1308             paddw_r2r( mm3, mm1 );
1309             paddw_m2r( m_4, mm1 );
1310             paddw_r2r( mm1, mm0 );
1311             psraw_i2r( 3, mm0 );
1312             packuswb_r2r( mm7, mm0 );
1313             movd_r2m( mm0, dst[i_dst+x] );
1314         }
1315         dst += 2*i_dst;
1316         src1 += i_src1;
1317         src2 += i_src2;
1318     }
1319 }
1320
1321 #endif
1322
1323 /* For debug */
1324 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1325 {
1326     int y;
1327     for( y = 0; y < 8; y++ )
1328         memset( &dst[y*i_dst], v, 8 );
1329 }
1330
1331 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1332  * neighbour
1333  * (Use 8x9 pixels)
1334  * TODO: a better one for the inner part.
1335  */
1336 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1337                                      uint8_t *src, int i_src )
1338 {
1339     int y, x;
1340
1341     /* Interlaced */
1342     for( y = 0; y < 8; y += 2 )
1343     {
1344         memcpy( dst, src, 8 );
1345         dst += i_dst;
1346
1347         for( x = 0; x < 8; x++ )
1348             dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1349         dst += 1*i_dst;
1350         src += 2*i_src;
1351     }
1352 }
1353 #ifdef CAN_COMPILE_MMXEXT
1354 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1355                                           uint8_t *src, int i_src )
1356 {
1357     int y;
1358
1359     /* Interlaced */
1360     for( y = 0; y < 8; y += 2 )
1361     {
1362         movq_m2r( src[0], mm0 );
1363         movq_r2m( mm0, dst[0] );
1364         dst += i_dst;
1365
1366         movq_m2r( src[2*i_src], mm1 );
1367         pavgb_r2r( mm1, mm0 );
1368
1369         movq_r2m( mm0, dst[0] );
1370
1371         dst += 1*i_dst;
1372         src += 2*i_src;
1373     }
1374 }
1375 #endif
1376
1377 /* XDeint8x8Field: Edge oriented interpolation
1378  * (Need -4 and +5 pixels H, +1 line)
1379  */
1380 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1381                                     uint8_t *src, int i_src )
1382 {
1383     int y, x;
1384
1385     /* Interlaced */
1386     for( y = 0; y < 8; y += 2 )
1387     {
1388         memcpy( dst, src, 8 );
1389         dst += i_dst;
1390
1391         for( x = 0; x < 8; x++ )
1392         {
1393             uint8_t *src2 = &src[2*i_src];
1394             /* I use 8 pixels just to match the MMX version, but it's overkill
1395              * 5 would be enough (less isn't good) */
1396             const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1397                            abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1398                            abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1399                            abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1400
1401             const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1402                            abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1403                            abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1404                            abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1405
1406             const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1407                            abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1408                            abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1409                            abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1410
1411             if( c0 < c1 && c1 <= c2 )
1412                 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1413             else if( c2 < c1 && c1 <= c0 )
1414                 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1415             else
1416                 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1417         }
1418
1419         dst += 1*i_dst;
1420         src += 2*i_src;
1421     }
1422 }
1423 #ifdef CAN_COMPILE_MMXEXT
1424 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1425                                          uint8_t *src, int i_src )
1426 {
1427     int y, x;
1428
1429     /* Interlaced */
1430     for( y = 0; y < 8; y += 2 )
1431     {
1432         memcpy( dst, src, 8 );
1433         dst += i_dst;
1434
1435         for( x = 0; x < 8; x++ )
1436         {
1437             uint8_t *src2 = &src[2*i_src];
1438             int32_t c0, c1, c2;
1439
1440             movq_m2r( src[x-2], mm0 );
1441             movq_m2r( src[x-3], mm1 );
1442             movq_m2r( src[x-4], mm2 );
1443
1444             psadbw_m2r( src2[x-4], mm0 );
1445             psadbw_m2r( src2[x-3], mm1 );
1446             psadbw_m2r( src2[x-2], mm2 );
1447
1448             movd_r2m( mm0, c2 );
1449             movd_r2m( mm1, c1 );
1450             movd_r2m( mm2, c0 );
1451
1452             if( c0 < c1 && c1 <= c2 )
1453                 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1454             else if( c2 < c1 && c1 <= c0 )
1455                 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1456             else
1457                 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1458         }
1459
1460         dst += 1*i_dst;
1461         src += 2*i_src;
1462     }
1463 }
1464 #endif
1465
1466 #if 0
1467 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1468                                  uint8_t *pix2, int i_pix2 )
1469 {
1470     int y, x;
1471     int s = 0;
1472
1473     for( y = 0; y < 8; y++ )
1474         for( x = 0; x < 8; x++ )
1475             s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1476     return s;
1477 }
1478
1479 #ifdef CAN_COMPILE_MMXEXT
1480 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1481                                       uint8_t *pix2, int i_pix2 )
1482 {
1483     int y;
1484     int32_t s;
1485
1486     pxor_r2r( mm7, mm7 );
1487     pxor_r2r( mm6, mm6 );
1488
1489     for( y = 0; y < 8; y++ )
1490     {
1491         movq_m2r( pix1[0], mm0 );
1492         movq_m2r( pix2[0], mm1 );
1493
1494         movq_r2r( mm0, mm2 );
1495         movq_r2r( mm1, mm3 );
1496
1497         punpcklbw_r2r( mm7, mm0 );
1498         punpckhbw_r2r( mm7, mm2 );
1499         punpcklbw_r2r( mm7, mm1 );
1500         punpckhbw_r2r( mm7, mm3 );
1501
1502         psubw_r2r( mm1, mm0 );
1503         psubw_r2r( mm3, mm2 );
1504
1505         pmaddwd_r2r( mm0, mm0 );
1506         pmaddwd_r2r( mm2, mm2 );
1507
1508         paddd_r2r( mm2, mm0 );
1509         paddd_r2r( mm0, mm6 );
1510
1511         pix1 += i_pix1;
1512         pix2 += i_pix2;
1513     }
1514
1515     movq_r2r( mm6, mm7 );
1516     psrlq_i2r( 32, mm7 );
1517     paddd_r2r( mm6, mm7 );
1518     movd_r2m( mm7, s );
1519
1520     return s;
1521 }
1522 #endif
1523 #endif
1524
1525 #if 0
1526 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1527 #ifdef CAN_COMPILE_MMXEXT
1528 /* XDeintMC:
1529  *  Bilinear MC QPel
1530  *  TODO: mmx version (easier in sse2)
1531  */
1532 static inline void XDeintMC( uint8_t *dst, int i_dst,
1533                              uint8_t *src, int i_src,
1534                              int mvx, int mvy,
1535                              int i_width, int i_height )
1536 {
1537     const int d4x = mvx&0x03;
1538     const int d4y = mvy&0x03;
1539
1540     const int cA = (4-d4x)*(4-d4y);
1541     const int cB = d4x    *(4-d4y);
1542     const int cC = (4-d4x)*d4y;
1543     const int cD = d4x    *d4y;
1544
1545     int y, x;
1546     uint8_t *srcp;
1547
1548
1549     src  += (mvy >> 2) * i_src + (mvx >> 2);
1550     srcp = &src[i_src];
1551
1552     for( y = 0; y < i_height; y++ )
1553     {
1554         for( x = 0; x < i_width; x++ )
1555         {
1556             dst[x] = ( cA*src[x]  + cB*src[x+1] +
1557                        cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1558         }
1559         dst  += i_dst;
1560
1561         src   = srcp;
1562         srcp += i_src;
1563     }
1564 }
1565 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1566                                uint8_t *pix2, int i_pix2 )
1567 {
1568     int32_t s;
1569
1570     movq_m2r( pix1[0*i_pix1], mm0 );
1571     movq_m2r( pix1[1*i_pix1], mm1 );
1572
1573     psadbw_m2r( pix2[0*i_pix2], mm0 );
1574     psadbw_m2r( pix2[1*i_pix2], mm1 );
1575
1576     movq_m2r( pix1[2*i_pix1], mm2 );
1577     movq_m2r( pix1[3*i_pix1], mm3 );
1578     psadbw_m2r( pix2[2*i_pix2], mm2 );
1579     psadbw_m2r( pix2[3*i_pix2], mm3 );
1580
1581     paddd_r2r( mm1, mm0 );
1582     paddd_r2r( mm3, mm2 );
1583     paddd_r2r( mm2, mm0 );
1584     movd_r2m( mm0, s );
1585
1586     return s;
1587 }
1588
1589 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1590                                      uint8_t *ref, int i_stride,
1591                                      int mx, int my,
1592                                      int xmax, int ymax )
1593 {
1594     uint8_t buffer[8*4];
1595
1596     if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1597         return 255*255*255;
1598
1599     XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1600     return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1601 }
1602 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1603                                     uint8_t *ref, int i_stride,
1604                                     int mx, int my,
1605                                     int xmax, int ymax )
1606 {
1607     if( abs(mx) >= xmax || abs(my) >= ymax )
1608         return 255*255*255;
1609
1610     return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1611 }
1612
1613 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1614                                          uint8_t *src, int i_src,
1615                                          int *mpx, int *mpy,
1616                                          int xmax, int ymax )
1617 {
1618     static const int dx[8] = { 0,  0, -1, 1, -1, -1,  1, 1 };
1619     static const int dy[8] = {-1,  1,  0, 0, -1,  1, -1, 1 };
1620     uint8_t *next = &src[i_src];
1621     const int i_src2 = 2*i_src;
1622     int mvx, mvy;
1623     int mvs, s;
1624     int i_step;
1625
1626     uint8_t *rec = &dst[i_dst];
1627
1628     /* We construct with intra method the missing field */
1629     XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1630
1631     /* Now we will try to find a match with ME with the other field */
1632
1633     /* ME: A small/partial EPZS
1634      * We search only for small MV (with high motion intra will be perfect */
1635     if( xmax > 4 ) xmax = 4;
1636     if( ymax > 4 ) ymax = 4;
1637
1638     /* Init with NULL Mv */
1639     mvx = mvy = 0;
1640     mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1641
1642     /* Try predicted Mv */
1643     if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1644     {
1645         mvs = s;
1646         mvx = *mpx;
1647         mvy = *mpy;
1648     }
1649     /* Search interger pel (small mv) */
1650     for( i_step = 0; i_step < 4; i_step++ )
1651     {
1652         int c = 4;
1653         int s;
1654         int i;
1655
1656         for( i = 0; i < 4; i++ )
1657         {
1658             s = XDeint8x4TestInt( rec, i_src2,
1659                                   next, i_src2, mvx+dx[i], mvy+dy[i],
1660                                   xmax, ymax );
1661             if( s < mvs )
1662             {
1663                 mvs = s;
1664                 c = i;
1665             }
1666         }
1667         if( c == 4 )
1668             break;
1669
1670         mvx += dx[c];
1671         mvy += dy[c];
1672     }
1673     *mpx = mvx;
1674     *mpy = mvy;
1675
1676     mvx <<= 2;
1677     mvy <<= 2;
1678
1679     if( mvs > 4 && mvs < 256 )
1680     {
1681         /* Search Qpel */
1682         /* XXX: for now only HPEL (too slow) */
1683         for( i_step = 0; i_step < 4; i_step++ )
1684         {
1685             int c = 8;
1686             int s;
1687             int i;
1688
1689             for( i = 0; i < 8; i++ )
1690             {
1691                 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1692                                        mvx+dx[i], mvy+dy[i],
1693                                        xmax, ymax );
1694                 if( s < mvs )
1695                 {
1696                     mvs = s;
1697                     c = i;
1698                 }
1699             }
1700             if( c == 8 )
1701                 break;
1702
1703             mvx += dx[c];
1704             mvy += dy[c];
1705         }
1706     }
1707
1708     if( mvs < 128 )
1709     {
1710         uint8_t buffer[8*4];
1711         XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1712         XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1713
1714         //XDeint8x8Set( dst, i_dst, 0 );
1715     }
1716 }
1717 #endif
1718 #endif
1719
1720 #if 0
1721 /* Kernel interpolation (1,-5,20,20,-5,1)
1722  * Lose a bit more details+add aliasing than edge interpol but avoid
1723  * more artifacts
1724  */
1725 static inline uint8_t clip1( int a )
1726 {
1727     if( a <= 0 )
1728         return 0;
1729     else if( a >= 255 )
1730         return 255;
1731     else
1732         return a;
1733 }
1734 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1735                                    uint8_t *src, int i_src )
1736 {
1737     int y, x;
1738
1739     /* Interlaced */
1740     for( y = 0; y < 8; y += 2 )
1741     {
1742         const int i_src2 = i_src*2;
1743
1744         memcpy( dst, src, 8 );
1745         dst += i_dst;
1746
1747         for( x = 0; x < 8; x++ )
1748         {
1749             int pix;
1750
1751             pix =   1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1752                    -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1753                   +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1754
1755             dst[x] = clip1( ( pix + 16 ) >> 5 );
1756         }
1757
1758         dst += 1*i_dst;
1759         src += 2*i_src;
1760     }
1761 }
1762
1763 #endif
1764
1765 /* NxN arbitray size (and then only use pixel in the NxN block)
1766  */
1767 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1768                                    int i_height, int i_width )
1769 {
1770     int y, x;
1771     int ff, fr;
1772     int fc;
1773
1774
1775     /* Detect interlacing */
1776     /* FIXME way too simple, need to be more like XDeint8x8Detect */
1777     ff = fr = 0;
1778     fc = 0;
1779     for( y = 0; y < i_height - 2; y += 2 )
1780     {
1781         const uint8_t *s = &src[y*i_src];
1782         for( x = 0; x < i_width; x++ )
1783         {
1784             fr += ssd(s[      x] - s[1*i_src+x]);
1785             ff += ssd(s[      x] - s[2*i_src+x]);
1786         }
1787         if( ff < fr && fr > i_width / 2 )
1788             fc++;
1789     }
1790
1791     return fc < 2 ? VLC_FALSE : VLC_TRUE;
1792 }
1793
1794 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1795                                    uint8_t *src, int i_src,
1796                                    int i_width, int i_height )
1797 {
1798     int y, x;
1799
1800     /* Progressive */
1801     for( y = 0; y < i_height; y += 2 )
1802     {
1803         memcpy( dst, src, i_width );
1804         dst += i_dst;
1805
1806         if( y < i_height - 2 )
1807         {
1808             for( x = 0; x < i_width; x++ )
1809                 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1810         }
1811         else
1812         {
1813             /* Blend last line */
1814             for( x = 0; x < i_width; x++ )
1815                 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1816         }
1817         dst += 1*i_dst;
1818         src += 2*i_src;
1819     }
1820 }
1821
1822 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1823                                    uint8_t *src, int i_src,
1824                                    int i_width, int i_height )
1825 {
1826     int y, x;
1827
1828     /* Interlaced */
1829     for( y = 0; y < i_height; y += 2 )
1830     {
1831         memcpy( dst, src, i_width );
1832         dst += i_dst;
1833
1834         if( y < i_height - 2 )
1835         {
1836             for( x = 0; x < i_width; x++ )
1837                 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1838         }
1839         else
1840         {
1841             /* Blend last line */
1842             for( x = 0; x < i_width; x++ )
1843                 dst[x] = (src[x] + src[i_src+x]) >> 1;
1844         }
1845         dst += 1*i_dst;
1846         src += 2*i_src;
1847     }
1848 }
1849
1850 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1851                               int i_width, int i_height )
1852 {
1853     if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1854         XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1855     else
1856         XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1857 }
1858
1859
1860 static inline int median( int a, int b, int c )
1861 {
1862     int min = a, max =a;
1863     if( b < min )
1864         min = b;
1865     else
1866         max = b;
1867
1868     if( c < min )
1869         min = c;
1870     else if( c > max )
1871         max = c;
1872
1873     return a + b + c - min - max;
1874 }
1875
1876
1877 /* XDeintBand8x8:
1878  */
1879 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1880                                    uint8_t *src, int i_src,
1881                                    const int i_mbx, int i_modx )
1882 {
1883     int x;
1884
1885     for( x = 0; x < i_mbx; x++ )
1886     {
1887         int s;
1888         if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1889         {
1890             if( x == 0 || x == i_mbx - 1 )
1891                 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1892             else
1893                 XDeint8x8FieldC( dst, i_dst, src, i_src );
1894         }
1895         else
1896         {
1897             XDeint8x8MergeC( dst, i_dst,
1898                              &src[0*i_src], 2*i_src,
1899                              &src[1*i_src], 2*i_src );
1900         }
1901
1902         dst += 8;
1903         src += 8;
1904     }
1905
1906     if( i_modx )
1907         XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1908 }
1909 #ifdef CAN_COMPILE_MMXEXT
1910 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1911                                         uint8_t *src, int i_src,
1912                                         const int i_mbx, int i_modx )
1913 {
1914     int x;
1915
1916     /* Reset current line */
1917     for( x = 0; x < i_mbx; x++ )
1918     {
1919         int s;
1920         if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1921         {
1922             if( x == 0 || x == i_mbx - 1 )
1923                 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1924             else
1925                 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1926         }
1927         else
1928         {
1929             XDeint8x8MergeMMXEXT( dst, i_dst,
1930                                   &src[0*i_src], 2*i_src,
1931                                   &src[1*i_src], 2*i_src );
1932         }
1933
1934         dst += 8;
1935         src += 8;
1936     }
1937
1938     if( i_modx )
1939         XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1940 }
1941 #endif
1942
1943 static void RenderX( vout_thread_t *p_vout,
1944                      picture_t *p_outpic, picture_t *p_pic )
1945 {
1946     int i_plane;
1947
1948     /* Copy image and skip lines */
1949     for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1950     {
1951         const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1952         const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1953
1954         const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1955         const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1956
1957         const int i_dst = p_outpic->p[i_plane].i_pitch;
1958         const int i_src = p_pic->p[i_plane].i_pitch;
1959
1960         int y, x;
1961
1962         for( y = 0; y < i_mby; y++ )
1963         {
1964             uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1965             uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1966
1967 #ifdef CAN_COMPILE_MMXEXT
1968             if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
1969                 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1970             else
1971 #endif
1972                 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1973         }
1974
1975         /* Last line (C only)*/
1976         if( i_mody )
1977         {
1978             uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1979             uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1980
1981             for( x = 0; x < i_mbx; x++ )
1982             {
1983                 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1984
1985                 dst += 8;
1986                 src += 8;
1987             }
1988
1989             if( i_modx )
1990                 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1991         }
1992     }
1993
1994 #ifdef CAN_COMPILE_MMXEXT
1995     if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
1996         emms();
1997 #endif
1998 }
1999
2000 /*****************************************************************************
2001  * SendEvents: forward mouse and keyboard events to the parent p_vout
2002  *****************************************************************************/
2003 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
2004                        vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
2005 {
2006     vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
2007     vlc_value_t sentval = newval;
2008
2009     if( !strcmp( psz_var, "mouse-y" ) )
2010     {
2011         switch( p_vout->p_sys->i_mode )
2012         {
2013             case DEINTERLACE_MEAN:
2014             case DEINTERLACE_DISCARD:
2015                 sentval.i_int *= 2;
2016                 break;
2017         }
2018     }
2019
2020     var_Set( p_vout, psz_var, sentval );
2021
2022     return VLC_SUCCESS;
2023 }
2024
2025 /*****************************************************************************
2026  * FilterCallback: called when changing the deinterlace method on the fly.
2027  *****************************************************************************/
2028 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
2029                            vlc_value_t oldval, vlc_value_t newval,
2030                            void *p_data )
2031 {
2032     vout_thread_t * p_vout = (vout_thread_t *)p_this;
2033     int i_old_mode = p_vout->p_sys->i_mode;
2034
2035     msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
2036
2037     vlc_mutex_lock( &p_vout->p_sys->filter_lock );
2038
2039     SetFilterMethod( p_vout, newval.psz_string );
2040
2041     switch( p_vout->render.i_chroma )
2042     {
2043     case VLC_FOURCC('I','4','2','2'):
2044         vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2045         return VLC_SUCCESS;
2046         break;
2047
2048     case VLC_FOURCC('I','4','2','0'):
2049     case VLC_FOURCC('I','Y','U','V'):
2050     case VLC_FOURCC('Y','V','1','2'):
2051         switch( p_vout->p_sys->i_mode )
2052         {
2053         case DEINTERLACE_MEAN:
2054         case DEINTERLACE_DISCARD:
2055             if( ( i_old_mode == DEINTERLACE_MEAN )
2056                 || ( i_old_mode == DEINTERLACE_DISCARD ) )
2057             {
2058                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2059                 return VLC_SUCCESS;
2060             }
2061             break;
2062
2063         case DEINTERLACE_BOB:
2064         case DEINTERLACE_BLEND:
2065         case DEINTERLACE_LINEAR:
2066             if( ( i_old_mode == DEINTERLACE_BOB )
2067                 || ( i_old_mode == DEINTERLACE_BLEND )
2068                 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2069             {
2070                 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2071                 return VLC_SUCCESS;
2072             }
2073             break;
2074         }
2075         break;
2076
2077     default:
2078         break;
2079     }
2080
2081     /* We need to kill the old vout */
2082
2083     DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2084
2085     vlc_object_detach( p_vout->p_sys->p_vout );
2086     vout_Destroy( p_vout->p_sys->p_vout );
2087
2088     /* Try to open a new video output */
2089     p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2090
2091     if( p_vout->p_sys->p_vout == NULL )
2092     {
2093         /* Everything failed */
2094         msg_Err( p_vout, "cannot open vout, aborting" );
2095
2096         vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2097         return VLC_EGENERIC;
2098     }
2099
2100     ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2101
2102     vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2103     return VLC_SUCCESS;
2104 }
2105
2106 /*****************************************************************************
2107  * SendEventsToChild: forward events to the child/children vout
2108  *****************************************************************************/
2109 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2110                        vlc_value_t oldval, vlc_value_t newval, void *p_data )
2111 {
2112     vout_thread_t *p_vout = (vout_thread_t *)p_this;
2113     var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2114     return VLC_SUCCESS;
2115 }
2116
2117
2118 /*****************************************************************************
2119  * video filter2 functions
2120  *****************************************************************************/
2121 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2122 {
2123     vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2124     picture_t *p_pic_dst;
2125
2126     /* Request output picture */
2127     p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2128     if( p_pic_dst == NULL )
2129     {
2130         msg_Warn( p_filter, "can't get output picture" );
2131         return NULL;
2132     }
2133
2134     switch( p_vout->p_sys->i_mode )
2135     {
2136         case DEINTERLACE_DISCARD:
2137 #if 0
2138             RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2139 #endif
2140             msg_Err( p_vout, "discarding lines is not supported yet" );
2141             p_pic_dst->pf_release( p_pic_dst );
2142             return p_pic;
2143             break;
2144
2145         case DEINTERLACE_BOB:
2146 #if 0
2147             RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2148             RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2149             break;
2150 #endif
2151
2152         case DEINTERLACE_LINEAR:
2153 #if 0
2154             RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2155             RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2156 #endif
2157             msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2158             p_pic_dst->pf_release( p_pic_dst );
2159             return p_pic;
2160             break;
2161
2162         case DEINTERLACE_MEAN:
2163             RenderMean( p_vout, p_pic_dst, p_pic );
2164             break;
2165
2166         case DEINTERLACE_BLEND:
2167             RenderBlend( p_vout, p_pic_dst, p_pic );
2168             break;
2169
2170         case DEINTERLACE_X:
2171             RenderX( p_vout, p_pic_dst, p_pic );
2172             break;
2173     }
2174
2175     p_pic_dst->date = p_pic->date;
2176     p_pic_dst->b_force = p_pic->b_force;
2177     p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2178     p_pic_dst->b_progressive = VLC_TRUE;
2179     p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2180
2181     p_pic->pf_release( p_pic );
2182     return p_pic_dst;
2183 }
2184
2185 /*****************************************************************************
2186  * OpenFilter:
2187  *****************************************************************************/
2188 static int OpenFilter( vlc_object_t *p_this )
2189 {
2190     filter_t *p_filter = (filter_t*)p_this;
2191     vout_thread_t *p_vout;
2192     vlc_value_t val;
2193
2194     if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2195           p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2196           p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2197         p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2198     {
2199         return VLC_EGENERIC;
2200     }
2201
2202     /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2203      * by spu filters */
2204     p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2205     vlc_object_attach( p_vout, p_filter );
2206     p_filter->p_sys = (filter_sys_t *)p_vout;
2207     p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2208
2209     config_ChainParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2210                    p_filter->p_cfg );
2211     var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2212     var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2213     var_Set( p_filter, "deinterlace-mode", val );
2214
2215     if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2216     {
2217         vlc_object_detach( p_vout );
2218         vlc_object_release( p_vout );
2219         return VLC_EGENERIC;
2220     }
2221
2222     p_filter->pf_video_filter = Deinterlace;
2223
2224     msg_Dbg( p_filter, "deinterlacing" );
2225
2226     return VLC_SUCCESS;
2227 }
2228
2229 /*****************************************************************************
2230  * CloseFilter: clean up the filter
2231  *****************************************************************************/
2232 static void CloseFilter( vlc_object_t *p_this )
2233 {
2234     filter_t *p_filter = (filter_t*)p_this;
2235     vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2236
2237     Destroy( VLC_OBJECT(p_vout) );
2238     vlc_object_detach( p_vout );
2239     vlc_object_release( p_vout );
2240 }
2241