1 /*****************************************************************************
2 * deinterlace.c : deinterlacer plugin for vlc
3 *****************************************************************************
4 * Copyright (C) 2000, 2001, 2002, 2003 the VideoLAN team
7 * Author: Sam Hocevar <sam@zoy.org>
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.
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.
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 *****************************************************************************/
24 /*****************************************************************************
26 *****************************************************************************/
28 #include <stdlib.h> /* malloc(), free() */
34 #include "vlc_filter.h"
40 #ifdef CAN_COMPILE_MMXEXT
44 #include "filter_common.h"
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
53 /*****************************************************************************
55 *****************************************************************************/
56 static int Create ( vlc_object_t * );
57 static void Destroy ( vlc_object_t * );
59 static int Init ( vout_thread_t * );
60 static void End ( vout_thread_t * );
61 static void Render ( vout_thread_t *, picture_t * );
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 * );
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 );
74 #if defined(CAN_COMPILE_MMXEXT)
75 static void MergeMMX ( void *, const void *, const void *, size_t );
77 #if defined(CAN_COMPILE_SSE)
78 static void MergeSSE2 ( void *, const void *, const void *, size_t );
80 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
81 static void EndMMX ( void );
84 static int SendEvents ( vlc_object_t *, char const *,
85 vlc_value_t, vlc_value_t, void * );
87 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method );
88 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout );
90 static int OpenFilter( vlc_object_t *p_this );
91 static void CloseFilter( vlc_object_t *p_this );
93 /*****************************************************************************
95 *****************************************************************************/
96 static int FilterCallback ( vlc_object_t *, char const *,
97 vlc_value_t, vlc_value_t, void * );
99 /*****************************************************************************
101 *****************************************************************************/
102 #define MODE_TEXT N_("Deinterlace mode")
103 #define MODE_LONGTEXT N_("Default deinterlace method to use for local playback")
105 #define SOUT_MODE_TEXT N_("Deinterlace mode")
106 #define SOUT_MODE_LONGTEXT N_("Default deinterlace methode to use for streaming")
108 #define FILTER_CFG_PREFIX "sout-deinterlace-"
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" };
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 );
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 );
126 add_shortcut( "deinterlace" );
127 set_callbacks( Create, Destroy );
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 );
138 static const char *ppsz_filter_options[] = {
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 *****************************************************************************/
150 int i_mode; /* Deinterlace mode */
151 vlc_bool_t b_double_rate; /* Shall we double the framerate? */
156 vout_thread_t *p_vout;
158 vlc_mutex_t filter_lock;
160 void (*pf_merge) ( void *, const void *, const void *, size_t );
161 void (*pf_end_merge) ( void );
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 )
169 return vout_vaControl( p_vout->p_sys->p_vout, i_query, args );
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 )
179 vout_thread_t *p_vout = (vout_thread_t *)p_this;
182 /* Allocate structure */
183 p_vout->p_sys = malloc( sizeof( vout_sys_t ) );
184 if( p_vout->p_sys == NULL )
186 msg_Err( p_vout, "out of memory" );
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;
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 );
203 #if defined(CAN_COMPILE_C_ALTIVEC)
204 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_ALTIVEC )
206 p_vout->p_sys->pf_merge = MergeAltivec;
207 p_vout->p_sys->pf_end_merge = NULL;
211 #if defined(CAN_COMPILE_SSE)
212 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_SSE2 )
214 p_vout->p_sys->pf_merge = MergeSSE2;
215 p_vout->p_sys->pf_end_merge = EndMMX;
219 #if defined(CAN_COMPILE_MMXEXT)
220 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMX )
222 p_vout->p_sys->pf_merge = MergeMMX;
223 p_vout->p_sys->pf_end_merge = EndMMX;
228 p_vout->p_sys->pf_merge = MergeGeneric;
229 p_vout->p_sys->pf_end_merge = NULL;
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 );
236 if( val.psz_string == NULL )
238 msg_Err( p_vout, "configuration variable deinterlace-mode empty" );
239 msg_Err( p_vout, "no deinterlace mode provided, using \"discard\"" );
241 val.psz_string = strdup( "discard" );
244 msg_Dbg( p_vout, "using %s deinterlace mode", val.psz_string );
246 SetFilterMethod( p_vout, val.psz_string );
248 free( val.psz_string );
253 /*****************************************************************************
254 * SetFilterMethod: setup the deinterlace method to use.
255 *****************************************************************************/
256 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method )
258 if( !strcmp( psz_method, "discard" ) )
260 p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
261 p_vout->p_sys->b_double_rate = VLC_FALSE;
263 else if( !strcmp( psz_method, "mean" ) )
265 p_vout->p_sys->i_mode = DEINTERLACE_MEAN;
266 p_vout->p_sys->b_double_rate = VLC_FALSE;
268 else if( !strcmp( psz_method, "blend" )
269 || !strcmp( psz_method, "average" )
270 || !strcmp( psz_method, "combine-fields" ) )
272 p_vout->p_sys->i_mode = DEINTERLACE_BLEND;
273 p_vout->p_sys->b_double_rate = VLC_FALSE;
275 else if( !strcmp( psz_method, "bob" )
276 || !strcmp( psz_method, "progressive-scan" ) )
278 p_vout->p_sys->i_mode = DEINTERLACE_BOB;
279 p_vout->p_sys->b_double_rate = VLC_TRUE;
281 else if( !strcmp( psz_method, "linear" ) )
283 p_vout->p_sys->i_mode = DEINTERLACE_LINEAR;
284 p_vout->p_sys->b_double_rate = VLC_TRUE;
286 else if( !strcmp( psz_method, "x" ) )
288 p_vout->p_sys->i_mode = DEINTERLACE_X;
289 p_vout->p_sys->b_double_rate = VLC_FALSE;
293 msg_Err( p_vout, "no valid deinterlace mode provided, "
294 "using \"discard\"" );
297 msg_Dbg( p_vout, "using %s deinterlace method", psz_method );
300 /*****************************************************************************
301 * Init: initialize Deinterlace video thread output method
302 *****************************************************************************/
303 static int Init( vout_thread_t *p_vout )
308 I_OUTPUTPICTURES = 0;
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 )
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;
326 return VLC_EGENERIC; /* unknown chroma */
330 /* Try to open the real video output */
331 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
333 if( p_vout->p_sys->p_vout == NULL )
335 /* Everything failed */
336 msg_Err( p_vout, "cannot open vout, aborting" );
341 var_AddCallback( p_vout, "deinterlace-mode", FilterCallback, NULL );
343 ALLOCATE_DIRECTBUFFERS( VOUT_MAX_PICTURES );
345 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
347 ADD_PARENT_CALLBACKS( SendEventsToChild );
352 /*****************************************************************************
353 * SpawnRealVout: spawn the real video output.
354 *****************************************************************************/
355 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout )
357 vout_thread_t *p_real_vout = NULL;
358 video_format_t fmt = {0};
360 msg_Dbg( p_vout, "spawning the real video output" );
362 fmt = p_vout->fmt_out;
364 switch( p_vout->render.i_chroma )
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 )
371 case DEINTERLACE_MEAN:
372 case DEINTERLACE_DISCARD:
373 fmt.i_height /= 2; fmt.i_visible_height /= 2; fmt.i_y_offset /= 2;
375 p_real_vout = vout_Create( p_vout, &fmt );
378 case DEINTERLACE_BOB:
379 case DEINTERLACE_BLEND:
380 case DEINTERLACE_LINEAR:
382 p_real_vout = vout_Create( p_vout, &fmt );
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 );
399 /*****************************************************************************
400 * End: terminate Deinterlace video thread output method
401 *****************************************************************************/
402 static void End( vout_thread_t *p_vout )
406 /* Free the fake output buffers we allocated */
407 for( i_index = I_OUTPUTPICTURES ; i_index ; )
410 free( PP_OUTPUTPICTURE[ i_index ]->p_data_orig );
413 if( p_vout->p_sys->p_vout )
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 );
420 DEL_PARENT_CALLBACKS( SendEventsToChild );
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 )
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 );
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
441 *****************************************************************************/
442 static void Render ( vout_thread_t *p_vout, picture_t *p_pic )
444 vout_sys_t *p_sys = p_vout->p_sys;
445 picture_t *pp_outpic[2];
447 p_vout->fmt_out.i_x_offset = p_sys->p_vout->fmt_in.i_x_offset =
448 p_vout->fmt_in.i_x_offset;
449 p_vout->fmt_out.i_y_offset = p_sys->p_vout->fmt_in.i_y_offset =
450 p_vout->fmt_in.i_y_offset;
451 p_vout->fmt_out.i_visible_width = p_sys->p_vout->fmt_in.i_visible_width =
452 p_vout->fmt_in.i_visible_width;
453 p_vout->fmt_out.i_visible_height = p_sys->p_vout->fmt_in.i_visible_height =
454 p_vout->fmt_in.i_visible_height;
455 if( p_vout->p_sys->i_mode == DEINTERLACE_MEAN ||
456 p_vout->p_sys->i_mode == DEINTERLACE_DISCARD )
458 p_vout->fmt_out.i_y_offset /= 2; p_sys->p_vout->fmt_in.i_y_offset /= 2;
459 p_vout->fmt_out.i_visible_height /= 2;
460 p_sys->p_vout->fmt_in.i_visible_height /= 2;
463 pp_outpic[0] = pp_outpic[1] = NULL;
465 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
467 /* Get a new picture */
468 while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
472 if( p_vout->b_die || p_vout->b_error )
474 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
477 msleep( VOUT_OUTMEM_SLEEP );
480 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
482 /* If we are using double rate, get an additional new picture */
483 if( p_vout->p_sys->b_double_rate )
485 while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
489 if( p_vout->b_die || p_vout->b_error )
491 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
492 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
495 msleep( VOUT_OUTMEM_SLEEP );
498 /* 20ms is a bit arbitrary, but it's only for the first image we get */
499 if( !p_vout->p_sys->last_date )
501 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
502 p_pic->date + 20000 );
506 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
507 (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
509 p_vout->p_sys->last_date = p_pic->date;
512 switch( p_vout->p_sys->i_mode )
514 case DEINTERLACE_DISCARD:
515 RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
516 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
519 case DEINTERLACE_BOB:
520 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
521 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
522 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
523 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
526 case DEINTERLACE_LINEAR:
527 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
528 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
529 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
530 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
533 case DEINTERLACE_MEAN:
534 RenderMean( p_vout, pp_outpic[0], p_pic );
535 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
538 case DEINTERLACE_BLEND:
539 RenderBlend( p_vout, pp_outpic[0], p_pic );
540 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
544 RenderX( p_vout, pp_outpic[0], p_pic );
545 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
548 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
551 /*****************************************************************************
552 * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
553 *****************************************************************************/
554 static void RenderDiscard( vout_thread_t *p_vout,
555 picture_t *p_outpic, picture_t *p_pic, int i_field )
559 /* Copy image and skip lines */
560 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
562 uint8_t *p_in, *p_out_end, *p_out;
565 p_in = p_pic->p[i_plane].p_pixels
566 + i_field * p_pic->p[i_plane].i_pitch;
568 p_out = p_outpic->p[i_plane].p_pixels;
569 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
570 * p_outpic->p[i_plane].i_visible_lines;
572 switch( p_vout->render.i_chroma )
574 case VLC_FOURCC('I','4','2','0'):
575 case VLC_FOURCC('I','Y','U','V'):
576 case VLC_FOURCC('Y','V','1','2'):
578 for( ; p_out < p_out_end ; )
580 p_vout->p_vlc->pf_memcpy( p_out, p_in,
581 p_pic->p[i_plane].i_pitch );
583 p_out += p_pic->p[i_plane].i_pitch;
584 p_in += 2 * p_pic->p[i_plane].i_pitch;
588 case VLC_FOURCC('I','4','2','2'):
590 i_increment = 2 * p_pic->p[i_plane].i_pitch;
592 if( i_plane == Y_PLANE )
594 for( ; p_out < p_out_end ; )
596 p_vout->p_vlc->pf_memcpy( p_out, p_in,
597 p_pic->p[i_plane].i_pitch );
598 p_out += p_pic->p[i_plane].i_pitch;
599 p_vout->p_vlc->pf_memcpy( p_out, p_in,
600 p_pic->p[i_plane].i_pitch );
601 p_out += p_pic->p[i_plane].i_pitch;
607 for( ; p_out < p_out_end ; )
609 p_vout->p_vlc->pf_memcpy( p_out, p_in,
610 p_pic->p[i_plane].i_pitch );
611 p_out += p_pic->p[i_plane].i_pitch;
623 /*****************************************************************************
624 * RenderBob: renders a BOB picture - simple copy
625 *****************************************************************************/
626 static void RenderBob( vout_thread_t *p_vout,
627 picture_t *p_outpic, picture_t *p_pic, int i_field )
631 /* Copy image and skip lines */
632 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
634 uint8_t *p_in, *p_out_end, *p_out;
636 p_in = p_pic->p[i_plane].p_pixels;
637 p_out = p_outpic->p[i_plane].p_pixels;
638 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
639 * p_outpic->p[i_plane].i_visible_lines;
641 switch( p_vout->render.i_chroma )
643 case VLC_FOURCC('I','4','2','0'):
644 case VLC_FOURCC('I','Y','U','V'):
645 case VLC_FOURCC('Y','V','1','2'):
646 /* For BOTTOM field we need to add the first line */
649 p_vout->p_vlc->pf_memcpy( p_out, p_in,
650 p_pic->p[i_plane].i_pitch );
651 p_in += p_pic->p[i_plane].i_pitch;
652 p_out += p_pic->p[i_plane].i_pitch;
655 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
657 for( ; p_out < p_out_end ; )
659 p_vout->p_vlc->pf_memcpy( p_out, p_in,
660 p_pic->p[i_plane].i_pitch );
662 p_out += p_pic->p[i_plane].i_pitch;
664 p_vout->p_vlc->pf_memcpy( p_out, p_in,
665 p_pic->p[i_plane].i_pitch );
667 p_in += 2 * p_pic->p[i_plane].i_pitch;
668 p_out += p_pic->p[i_plane].i_pitch;
671 p_vout->p_vlc->pf_memcpy( p_out, p_in,
672 p_pic->p[i_plane].i_pitch );
674 /* For TOP field we need to add the last line */
677 p_in += p_pic->p[i_plane].i_pitch;
678 p_out += p_pic->p[i_plane].i_pitch;
679 p_vout->p_vlc->pf_memcpy( p_out, p_in,
680 p_pic->p[i_plane].i_pitch );
684 case VLC_FOURCC('I','4','2','2'):
685 /* For BOTTOM field we need to add the first line */
688 p_vout->p_vlc->pf_memcpy( p_out, p_in,
689 p_pic->p[i_plane].i_pitch );
690 p_in += p_pic->p[i_plane].i_pitch;
691 p_out += p_pic->p[i_plane].i_pitch;
694 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
696 if( i_plane == Y_PLANE )
698 for( ; p_out < p_out_end ; )
700 p_vout->p_vlc->pf_memcpy( p_out, p_in,
701 p_pic->p[i_plane].i_pitch );
703 p_out += p_pic->p[i_plane].i_pitch;
705 p_vout->p_vlc->pf_memcpy( p_out, p_in,
706 p_pic->p[i_plane].i_pitch );
708 p_in += 2 * p_pic->p[i_plane].i_pitch;
709 p_out += p_pic->p[i_plane].i_pitch;
714 for( ; p_out < p_out_end ; )
716 p_vout->p_vlc->pf_memcpy( p_out, p_in,
717 p_pic->p[i_plane].i_pitch );
719 p_out += p_pic->p[i_plane].i_pitch;
720 p_in += 2 * p_pic->p[i_plane].i_pitch;
724 p_vout->p_vlc->pf_memcpy( p_out, p_in,
725 p_pic->p[i_plane].i_pitch );
727 /* For TOP field we need to add the last line */
730 p_in += p_pic->p[i_plane].i_pitch;
731 p_out += p_pic->p[i_plane].i_pitch;
732 p_vout->p_vlc->pf_memcpy( p_out, p_in,
733 p_pic->p[i_plane].i_pitch );
740 #define Merge p_vout->p_sys->pf_merge
741 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
743 /*****************************************************************************
744 * RenderLinear: BOB with linear interpolation
745 *****************************************************************************/
746 static void RenderLinear( vout_thread_t *p_vout,
747 picture_t *p_outpic, picture_t *p_pic, int i_field )
751 /* Copy image and skip lines */
752 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
754 uint8_t *p_in, *p_out_end, *p_out;
756 p_in = p_pic->p[i_plane].p_pixels;
757 p_out = p_outpic->p[i_plane].p_pixels;
758 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
759 * p_outpic->p[i_plane].i_visible_lines;
761 /* For BOTTOM field we need to add the first line */
764 p_vout->p_vlc->pf_memcpy( p_out, p_in,
765 p_pic->p[i_plane].i_pitch );
766 p_in += p_pic->p[i_plane].i_pitch;
767 p_out += p_pic->p[i_plane].i_pitch;
770 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
772 for( ; p_out < p_out_end ; )
774 p_vout->p_vlc->pf_memcpy( p_out, p_in,
775 p_pic->p[i_plane].i_pitch );
777 p_out += p_pic->p[i_plane].i_pitch;
779 Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
780 p_pic->p[i_plane].i_pitch );
782 p_in += 2 * p_pic->p[i_plane].i_pitch;
783 p_out += p_pic->p[i_plane].i_pitch;
786 p_vout->p_vlc->pf_memcpy( p_out, p_in,
787 p_pic->p[i_plane].i_pitch );
789 /* For TOP field we need to add the last line */
792 p_in += p_pic->p[i_plane].i_pitch;
793 p_out += p_pic->p[i_plane].i_pitch;
794 p_vout->p_vlc->pf_memcpy( p_out, p_in,
795 p_pic->p[i_plane].i_pitch );
801 static void RenderMean( vout_thread_t *p_vout,
802 picture_t *p_outpic, picture_t *p_pic )
806 /* Copy image and skip lines */
807 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
809 uint8_t *p_in, *p_out_end, *p_out;
811 p_in = p_pic->p[i_plane].p_pixels;
813 p_out = p_outpic->p[i_plane].p_pixels;
814 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
815 * p_outpic->p[i_plane].i_visible_lines;
817 /* All lines: mean value */
818 for( ; p_out < p_out_end ; )
820 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
821 p_pic->p[i_plane].i_pitch );
823 p_out += p_pic->p[i_plane].i_pitch;
824 p_in += 2 * p_pic->p[i_plane].i_pitch;
830 static void RenderBlend( vout_thread_t *p_vout,
831 picture_t *p_outpic, picture_t *p_pic )
835 /* Copy image and skip lines */
836 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
838 uint8_t *p_in, *p_out_end, *p_out;
840 p_in = p_pic->p[i_plane].p_pixels;
842 p_out = p_outpic->p[i_plane].p_pixels;
843 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
844 * p_outpic->p[i_plane].i_visible_lines;
846 switch( p_vout->render.i_chroma )
848 case VLC_FOURCC('I','4','2','0'):
849 case VLC_FOURCC('I','Y','U','V'):
850 case VLC_FOURCC('Y','V','1','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;
856 /* Remaining lines: mean value */
857 for( ; p_out < p_out_end ; )
859 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
860 p_pic->p[i_plane].i_pitch );
862 p_out += p_pic->p[i_plane].i_pitch;
863 p_in += p_pic->p[i_plane].i_pitch;
867 case VLC_FOURCC('I','4','2','2'):
868 /* First line: simple copy */
869 p_vout->p_vlc->pf_memcpy( p_out, p_in,
870 p_pic->p[i_plane].i_pitch );
871 p_out += p_pic->p[i_plane].i_pitch;
873 /* Remaining lines: mean value */
874 if( i_plane == Y_PLANE )
876 for( ; p_out < p_out_end ; )
878 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
879 p_pic->p[i_plane].i_pitch );
881 p_out += p_pic->p[i_plane].i_pitch;
882 p_in += p_pic->p[i_plane].i_pitch;
888 for( ; p_out < p_out_end ; )
890 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
891 p_pic->p[i_plane].i_pitch );
893 p_out += p_pic->p[i_plane].i_pitch;
894 p_in += 2*p_pic->p[i_plane].i_pitch;
905 static void MergeGeneric( void *_p_dest, const void *_p_s1,
906 const void *_p_s2, size_t i_bytes )
908 uint8_t* p_dest = (uint8_t*)_p_dest;
909 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
910 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
911 uint8_t* p_end = p_dest + i_bytes - 8;
913 while( p_dest < p_end )
915 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
916 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
917 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
918 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
919 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
920 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
921 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
922 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
927 while( p_dest < p_end )
929 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
933 #if defined(CAN_COMPILE_MMXEXT)
934 static void MergeMMX( void *_p_dest, const void *_p_s1, const void *_p_s2,
937 uint8_t* p_dest = (uint8_t*)_p_dest;
938 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
939 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
940 uint8_t* p_end = p_dest + i_bytes - 8;
941 while( p_dest < p_end )
943 __asm__ __volatile__( "movq %2,%%mm1;"
945 "movq %%mm1, %0" :"=m" (*p_dest):
955 while( p_dest < p_end )
957 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
962 #if defined(CAN_COMPILE_SSE)
963 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
966 uint8_t* p_dest = (uint8_t*)_p_dest;
967 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
968 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
970 while( (ptrdiff_t)p_s1 % 16 )
972 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
974 p_end = p_dest + i_bytes - 16;
975 while( p_dest < p_end )
977 __asm__ __volatile__( "movdqu %2,%%xmm1;"
979 "movdqu %%xmm1, %0" :"=m" (*p_dest):
989 while( p_dest < p_end )
991 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
996 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
997 static void EndMMX( void )
999 __asm__ __volatile__( "emms" :: );
1003 #ifdef CAN_COMPILE_C_ALTIVEC
1004 static void MergeAltivec( void *_p_dest, const void *_p_s1,
1005 const void *_p_s2, size_t i_bytes )
1007 uint8_t *p_dest = (uint8_t *)_p_dest;
1008 uint8_t *p_s1 = (uint8_t *)_p_s1;
1009 uint8_t *p_s2 = (uint8_t *)_p_s2;
1010 uint8_t *p_end = p_dest + i_bytes - 15;
1012 /* Use C until the first 16-bytes aligned destination pixel */
1013 while( (int)p_dest & 0xF )
1015 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1018 if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1020 /* Unaligned source */
1021 vector unsigned char s1v, s2v, destv;
1022 vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1023 vector unsigned char perm1v, perm2v;
1025 perm1v = vec_lvsl( 0, p_s1 );
1026 perm2v = vec_lvsl( 0, p_s2 );
1027 s1oldv = vec_ld( 0, p_s1 );
1028 s2oldv = vec_ld( 0, p_s2 );
1030 while( p_dest < p_end )
1032 s1newv = vec_ld( 16, p_s1 );
1033 s2newv = vec_ld( 16, p_s2 );
1034 s1v = vec_perm( s1oldv, s1newv, perm1v );
1035 s2v = vec_perm( s2oldv, s2newv, perm2v );
1038 destv = vec_avg( s1v, s2v );
1039 vec_st( destv, 0, p_dest );
1048 /* Aligned source */
1049 vector unsigned char s1v, s2v, destv;
1051 while( p_dest < p_end )
1053 s1v = vec_ld( 0, p_s1 );
1054 s2v = vec_ld( 0, p_s2 );
1055 destv = vec_avg( s1v, s2v );
1056 vec_st( destv, 0, p_dest );
1066 while( p_dest < p_end )
1068 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1073 /*****************************************************************************
1074 * RenderX: This algo works on a 8x8 block basic, it copies the top field
1075 * and apply a process to recreate the bottom field :
1076 * If a 8x8 block is classified as :
1077 * - progressive: it applies a small blend (1,6,1)
1079 * * in the MMX version: we do a ME between the 2 fields, if there is a
1080 * good match we use MC to recreate the bottom field (with a small
1082 * * otherwise: it recreates the bottom field by an edge oriented
1084 *****************************************************************************/
1086 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1087 * XXX: It need to access to 8x10
1088 * We use more than 8 lines to help with scrolling (text)
1089 * (and because XDeint8x8Frame use line 9)
1090 * XXX: smooth/uniform area with noise detection doesn't works well
1091 * but it's not really a problem because they don't have much details anyway
1093 static inline int ssd( int a ) { return a*a; }
1094 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1100 /* Detect interlacing */
1102 for( y = 0; y < 7; y += 2 )
1105 for( x = 0; x < 8; x++ )
1107 fr += ssd(src[ x] - src[1*i_src+x]) +
1108 ssd(src[i_src+x] - src[2*i_src+x]);
1109 ff += ssd(src[ x] - src[2*i_src+x]) +
1110 ssd(src[i_src+x] - src[3*i_src+x]);
1112 if( ff < 6*fr/8 && fr > 32 )
1118 return fc < 1 ? VLC_FALSE : VLC_TRUE;
1120 #ifdef CAN_COMPILE_MMXEXT
1121 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1128 /* Detect interlacing */
1130 pxor_r2r( mm7, mm7 );
1131 for( y = 0; y < 9; y += 2 )
1134 pxor_r2r( mm5, mm5 );
1135 pxor_r2r( mm6, mm6 );
1136 for( x = 0; x < 8; x+=4 )
1138 movd_m2r( src[ x], mm0 );
1139 movd_m2r( src[1*i_src+x], mm1 );
1140 movd_m2r( src[2*i_src+x], mm2 );
1141 movd_m2r( src[3*i_src+x], mm3 );
1143 punpcklbw_r2r( mm7, mm0 );
1144 punpcklbw_r2r( mm7, mm1 );
1145 punpcklbw_r2r( mm7, mm2 );
1146 punpcklbw_r2r( mm7, mm3 );
1148 movq_r2r( mm0, mm4 );
1150 psubw_r2r( mm1, mm0 );
1151 psubw_r2r( mm2, mm4 );
1153 psubw_r2r( mm1, mm2 );
1154 psubw_r2r( mm1, mm3 );
1156 pmaddwd_r2r( mm0, mm0 );
1157 pmaddwd_r2r( mm4, mm4 );
1158 pmaddwd_r2r( mm2, mm2 );
1159 pmaddwd_r2r( mm3, mm3 );
1160 paddd_r2r( mm0, mm2 );
1161 paddd_r2r( mm4, mm3 );
1162 paddd_r2r( mm2, mm5 );
1163 paddd_r2r( mm3, mm6 );
1166 movq_r2r( mm5, mm0 );
1167 psrlq_i2r( 32, mm0 );
1168 paddd_r2r( mm0, mm5 );
1169 movd_r2m( mm5, fr );
1171 movq_r2r( mm6, mm0 );
1172 psrlq_i2r( 32, mm0 );
1173 paddd_r2r( mm0, mm6 );
1174 movd_r2m( mm6, ff );
1176 if( ff < 6*fr/8 && fr > 32 )
1185 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1186 * This won't destroy details, and help if there is a bit of interlacing.
1187 * (It helps with paning to avoid flickers)
1191 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1192 uint8_t *src, int i_src )
1197 for( y = 0; y < 8; y += 2 )
1199 memcpy( dst, src, 8 );
1202 for( x = 0; x < 8; x++ )
1203 dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1209 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1210 uint8_t *src1, int i_src1,
1211 uint8_t *src2, int i_src2 )
1216 for( y = 0; y < 8; y += 2 )
1218 memcpy( dst, src1, 8 );
1221 for( x = 0; x < 8; x++ )
1222 dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1230 #ifdef CAN_COMPILE_MMXEXT
1231 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1232 uint8_t *src1, int i_src1,
1233 uint8_t *src2, int i_src2 )
1235 static const uint64_t m_4 = I64C(0x0004000400040004);
1239 pxor_r2r( mm7, mm7 );
1240 for( y = 0; y < 8; y += 2 )
1242 for( x = 0; x < 8; x +=4 )
1244 movd_m2r( src1[x], mm0 );
1245 movd_r2m( mm0, dst[x] );
1247 movd_m2r( src2[x], mm1 );
1248 movd_m2r( src1[i_src1+x], mm2 );
1250 punpcklbw_r2r( mm7, mm0 );
1251 punpcklbw_r2r( mm7, mm1 );
1252 punpcklbw_r2r( mm7, mm2 );
1253 paddw_r2r( mm1, mm1 );
1254 movq_r2r( mm1, mm3 );
1255 paddw_r2r( mm3, mm3 );
1256 paddw_r2r( mm2, mm0 );
1257 paddw_r2r( mm3, mm1 );
1258 paddw_m2r( m_4, mm1 );
1259 paddw_r2r( mm1, mm0 );
1260 psraw_i2r( 3, mm0 );
1261 packuswb_r2r( mm7, mm0 );
1262 movd_r2m( mm0, dst[i_dst+x] );
1273 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1276 for( y = 0; y < 8; y++ )
1277 memset( &dst[y*i_dst], v, 8 );
1280 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1283 * TODO: a better one for the inner part.
1285 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1286 uint8_t *src, int i_src )
1291 for( y = 0; y < 8; y += 2 )
1293 memcpy( dst, src, 8 );
1296 for( x = 0; x < 8; x++ )
1297 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1302 #ifdef CAN_COMPILE_MMXEXT
1303 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1304 uint8_t *src, int i_src )
1309 for( y = 0; y < 8; y += 2 )
1311 movq_m2r( src[0], mm0 );
1312 movq_r2m( mm0, dst[0] );
1315 movq_m2r( src[2*i_src], mm1 );
1316 pavgb_r2r( mm1, mm0 );
1318 movq_r2m( mm0, dst[0] );
1326 /* XDeint8x8Field: Edge oriented interpolation
1327 * (Need -4 and +5 pixels H, +1 line)
1329 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1330 uint8_t *src, int i_src )
1335 for( y = 0; y < 8; y += 2 )
1337 memcpy( dst, src, 8 );
1340 for( x = 0; x < 8; x++ )
1342 uint8_t *src2 = &src[2*i_src];
1343 /* I use 8 pixels just to match the MMX version, but it's overkill
1344 * 5 would be enough (less isn't good) */
1345 const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1346 abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1347 abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1348 abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1350 const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1351 abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1352 abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1353 abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1355 const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1356 abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1357 abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1358 abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1360 if( c0 < c1 && c1 <= c2 )
1361 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1362 else if( c2 < c1 && c1 <= c0 )
1363 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1365 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1372 #ifdef CAN_COMPILE_MMXEXT
1373 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1374 uint8_t *src, int i_src )
1379 for( y = 0; y < 8; y += 2 )
1381 memcpy( dst, src, 8 );
1384 for( x = 0; x < 8; x++ )
1386 uint8_t *src2 = &src[2*i_src];
1389 movq_m2r( src[x-2], mm0 );
1390 movq_m2r( src[x-3], mm1 );
1391 movq_m2r( src[x-4], mm2 );
1393 psadbw_m2r( src2[x-4], mm0 );
1394 psadbw_m2r( src2[x-3], mm1 );
1395 psadbw_m2r( src2[x-2], mm2 );
1397 movd_r2m( mm0, c2 );
1398 movd_r2m( mm1, c1 );
1399 movd_r2m( mm2, c0 );
1401 if( c0 < c1 && c1 <= c2 )
1402 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1403 else if( c2 < c1 && c1 <= c0 )
1404 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1406 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1416 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1417 uint8_t *pix2, int i_pix2 )
1422 for( y = 0; y < 8; y++ )
1423 for( x = 0; x < 8; x++ )
1424 s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1428 #ifdef CAN_COMPILE_MMXEXT
1429 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1430 uint8_t *pix2, int i_pix2 )
1435 pxor_r2r( mm7, mm7 );
1436 pxor_r2r( mm6, mm6 );
1438 for( y = 0; y < 8; y++ )
1440 movq_m2r( pix1[0], mm0 );
1441 movq_m2r( pix2[0], mm1 );
1443 movq_r2r( mm0, mm2 );
1444 movq_r2r( mm1, mm3 );
1446 punpcklbw_r2r( mm7, mm0 );
1447 punpckhbw_r2r( mm7, mm2 );
1448 punpcklbw_r2r( mm7, mm1 );
1449 punpckhbw_r2r( mm7, mm3 );
1451 psubw_r2r( mm1, mm0 );
1452 psubw_r2r( mm3, mm2 );
1454 pmaddwd_r2r( mm0, mm0 );
1455 pmaddwd_r2r( mm2, mm2 );
1457 paddd_r2r( mm2, mm0 );
1458 paddd_r2r( mm0, mm6 );
1464 movq_r2r( mm6, mm7 );
1465 psrlq_i2r( 32, mm7 );
1466 paddd_r2r( mm6, mm7 );
1475 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1476 #ifdef CAN_COMPILE_MMXEXT
1479 * TODO: mmx version (easier in sse2)
1481 static inline void XDeintMC( uint8_t *dst, int i_dst,
1482 uint8_t *src, int i_src,
1484 int i_width, int i_height )
1486 const int d4x = mvx&0x03;
1487 const int d4y = mvy&0x03;
1489 const int cA = (4-d4x)*(4-d4y);
1490 const int cB = d4x *(4-d4y);
1491 const int cC = (4-d4x)*d4y;
1492 const int cD = d4x *d4y;
1498 src += (mvy >> 2) * i_src + (mvx >> 2);
1501 for( y = 0; y < i_height; y++ )
1503 for( x = 0; x < i_width; x++ )
1505 dst[x] = ( cA*src[x] + cB*src[x+1] +
1506 cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1514 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1515 uint8_t *pix2, int i_pix2 )
1519 movq_m2r( pix1[0*i_pix1], mm0 );
1520 movq_m2r( pix1[1*i_pix1], mm1 );
1522 psadbw_m2r( pix2[0*i_pix2], mm0 );
1523 psadbw_m2r( pix2[1*i_pix2], mm1 );
1525 movq_m2r( pix1[2*i_pix1], mm2 );
1526 movq_m2r( pix1[3*i_pix1], mm3 );
1527 psadbw_m2r( pix2[2*i_pix2], mm2 );
1528 psadbw_m2r( pix2[3*i_pix2], mm3 );
1530 paddd_r2r( mm1, mm0 );
1531 paddd_r2r( mm3, mm2 );
1532 paddd_r2r( mm2, mm0 );
1538 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1539 uint8_t *ref, int i_stride,
1541 int xmax, int ymax )
1543 uint8_t buffer[8*4];
1545 if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1548 XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1549 return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1551 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1552 uint8_t *ref, int i_stride,
1554 int xmax, int ymax )
1556 if( abs(mx) >= xmax || abs(my) >= ymax )
1559 return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1562 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1563 uint8_t *src, int i_src,
1565 int xmax, int ymax )
1567 static const int dx[8] = { 0, 0, -1, 1, -1, -1, 1, 1 };
1568 static const int dy[8] = {-1, 1, 0, 0, -1, 1, -1, 1 };
1569 uint8_t *next = &src[i_src];
1570 const int i_src2 = 2*i_src;
1575 uint8_t *rec = &dst[i_dst];
1577 /* We construct with intra method the missing field */
1578 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1580 /* Now we will try to find a match with ME with the other field */
1582 /* ME: A small/partial EPZS
1583 * We search only for small MV (with high motion intra will be perfect */
1584 if( xmax > 4 ) xmax = 4;
1585 if( ymax > 4 ) ymax = 4;
1587 /* Init with NULL Mv */
1589 mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1591 /* Try predicted Mv */
1592 if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1598 /* Search interger pel (small mv) */
1599 for( i_step = 0; i_step < 4; i_step++ )
1605 for( i = 0; i < 4; i++ )
1607 s = XDeint8x4TestInt( rec, i_src2,
1608 next, i_src2, mvx+dx[i], mvy+dy[i],
1628 if( mvs > 4 && mvs < 256 )
1631 /* XXX: for now only HPEL (too slow) */
1632 for( i_step = 0; i_step < 4; i_step++ )
1638 for( i = 0; i < 8; i++ )
1640 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1641 mvx+dx[i], mvy+dy[i],
1659 uint8_t buffer[8*4];
1660 XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1661 XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1663 //XDeint8x8Set( dst, i_dst, 0 );
1670 /* Kernel interpolation (1,-5,20,20,-5,1)
1671 * Loose a bit more details+add aliasing than edge interpol but avoid
1674 static inline uint8_t clip1( int a )
1683 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1684 uint8_t *src, int i_src )
1689 for( y = 0; y < 8; y += 2 )
1691 const int i_src2 = i_src*2;
1693 memcpy( dst, src, 8 );
1696 for( x = 0; x < 8; x++ )
1700 pix = 1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1701 -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1702 +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1704 dst[x] = clip1( ( pix + 16 ) >> 5 );
1714 /* NxN arbitray size (and then only use pixel in the NxN block)
1716 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1717 int i_height, int i_width )
1724 /* Detect interlacing */
1725 /* FIXME way too simple, need to be more like XDeint8x8Detect */
1728 for( y = 0; y < i_height - 2; y += 2 )
1730 const uint8_t *s = &src[y*i_src];
1731 for( x = 0; x < i_width; x++ )
1733 fr += ssd(s[ x] - s[1*i_src+x]);
1734 ff += ssd(s[ x] - s[2*i_src+x]);
1736 if( ff < fr && fr > i_width / 2 )
1740 return fc < 2 ? VLC_FALSE : VLC_TRUE;
1743 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1744 uint8_t *src, int i_src,
1745 int i_width, int i_height )
1750 for( y = 0; y < i_height; y += 2 )
1752 memcpy( dst, src, i_width );
1755 if( y < i_height - 2 )
1757 for( x = 0; x < i_width; x++ )
1758 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1762 /* Blend last line */
1763 for( x = 0; x < i_width; x++ )
1764 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1771 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1772 uint8_t *src, int i_src,
1773 int i_width, int i_height )
1778 for( y = 0; y < i_height; y += 2 )
1780 memcpy( dst, src, i_width );
1783 if( y < i_height - 2 )
1785 for( x = 0; x < i_width; x++ )
1786 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1790 /* Blend last line */
1791 for( x = 0; x < i_width; x++ )
1792 dst[x] = (src[x] + src[i_src+x]) >> 1;
1799 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1800 int i_width, int i_height )
1802 if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1803 XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1805 XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1809 static inline int median( int a, int b, int c )
1811 int min = a, max =a;
1822 return a + b + c - min - max;
1828 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1829 uint8_t *src, int i_src,
1830 const int i_mbx, int i_modx )
1834 for( x = 0; x < i_mbx; x++ )
1837 if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1839 if( x == 0 || x == i_mbx - 1 )
1840 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1842 XDeint8x8FieldC( dst, i_dst, src, i_src );
1846 XDeint8x8MergeC( dst, i_dst,
1847 &src[0*i_src], 2*i_src,
1848 &src[1*i_src], 2*i_src );
1856 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1858 #ifdef CAN_COMPILE_MMXEXT
1859 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1860 uint8_t *src, int i_src,
1861 const int i_mbx, int i_modx )
1865 /* Reset current line */
1866 for( x = 0; x < i_mbx; x++ )
1869 if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1871 if( x == 0 || x == i_mbx - 1 )
1872 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1874 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1878 XDeint8x8MergeMMXEXT( dst, i_dst,
1879 &src[0*i_src], 2*i_src,
1880 &src[1*i_src], 2*i_src );
1888 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1892 static void RenderX( vout_thread_t *p_vout,
1893 picture_t *p_outpic, picture_t *p_pic )
1897 /* Copy image and skip lines */
1898 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1900 const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1901 const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1903 const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1904 const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1906 const int i_dst = p_outpic->p[i_plane].i_pitch;
1907 const int i_src = p_pic->p[i_plane].i_pitch;
1911 for( y = 0; y < i_mby; y++ )
1913 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1914 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1916 #ifdef CAN_COMPILE_MMXEXT
1917 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1918 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1921 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1924 /* Last line (C only)*/
1927 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1928 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1930 for( x = 0; x < i_mbx; x++ )
1932 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1939 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1943 #ifdef CAN_COMPILE_MMXEXT
1944 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1949 /*****************************************************************************
1950 * SendEvents: forward mouse and keyboard events to the parent p_vout
1951 *****************************************************************************/
1952 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
1953 vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
1955 vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
1956 vlc_value_t sentval = newval;
1958 if( !strcmp( psz_var, "mouse-y" ) )
1960 switch( p_vout->p_sys->i_mode )
1962 case DEINTERLACE_MEAN:
1963 case DEINTERLACE_DISCARD:
1969 var_Set( p_vout, psz_var, sentval );
1974 /*****************************************************************************
1975 * FilterCallback: called when changing the deinterlace method on the fly.
1976 *****************************************************************************/
1977 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
1978 vlc_value_t oldval, vlc_value_t newval,
1981 vout_thread_t * p_vout = (vout_thread_t *)p_this;
1982 int i_old_mode = p_vout->p_sys->i_mode;
1984 msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
1986 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
1988 SetFilterMethod( p_vout, newval.psz_string );
1990 switch( p_vout->render.i_chroma )
1992 case VLC_FOURCC('I','4','2','2'):
1993 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1997 case VLC_FOURCC('I','4','2','0'):
1998 case VLC_FOURCC('I','Y','U','V'):
1999 case VLC_FOURCC('Y','V','1','2'):
2000 switch( p_vout->p_sys->i_mode )
2002 case DEINTERLACE_MEAN:
2003 case DEINTERLACE_DISCARD:
2004 if( ( i_old_mode == DEINTERLACE_MEAN )
2005 || ( i_old_mode == DEINTERLACE_DISCARD ) )
2007 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2012 case DEINTERLACE_BOB:
2013 case DEINTERLACE_BLEND:
2014 case DEINTERLACE_LINEAR:
2015 if( ( i_old_mode == DEINTERLACE_BOB )
2016 || ( i_old_mode == DEINTERLACE_BLEND )
2017 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2019 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2030 /* We need to kill the old vout */
2032 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2034 vlc_object_detach( p_vout->p_sys->p_vout );
2035 vout_Destroy( p_vout->p_sys->p_vout );
2037 /* Try to open a new video output */
2038 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2040 if( p_vout->p_sys->p_vout == NULL )
2042 /* Everything failed */
2043 msg_Err( p_vout, "cannot open vout, aborting" );
2045 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2046 return VLC_EGENERIC;
2049 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2051 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2055 /*****************************************************************************
2056 * SendEventsToChild: forward events to the child/children vout
2057 *****************************************************************************/
2058 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2059 vlc_value_t oldval, vlc_value_t newval, void *p_data )
2061 vout_thread_t *p_vout = (vout_thread_t *)p_this;
2062 var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2067 /*****************************************************************************
2068 * video filter2 functions
2069 *****************************************************************************/
2070 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2072 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2073 picture_t *p_pic_dst;
2075 /* Request output picture */
2076 p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2077 if( p_pic_dst == NULL )
2079 msg_Warn( p_filter, "can't get output picture" );
2083 switch( p_vout->p_sys->i_mode )
2085 case DEINTERLACE_DISCARD:
2087 RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2089 msg_Err( p_vout, "discarding lines is not supported yet" );
2090 p_pic_dst->pf_release( p_pic_dst );
2094 case DEINTERLACE_BOB:
2096 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2097 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2101 case DEINTERLACE_LINEAR:
2103 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2104 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2106 msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2107 p_pic_dst->pf_release( p_pic_dst );
2111 case DEINTERLACE_MEAN:
2112 RenderMean( p_vout, p_pic_dst, p_pic );
2115 case DEINTERLACE_BLEND:
2116 RenderBlend( p_vout, p_pic_dst, p_pic );
2120 RenderX( p_vout, p_pic_dst, p_pic );
2124 p_pic_dst->date = p_pic->date;
2125 p_pic_dst->b_force = p_pic->b_force;
2126 p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2127 p_pic_dst->b_progressive = VLC_TRUE;
2128 p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2130 p_pic->pf_release( p_pic );
2134 /*****************************************************************************
2136 *****************************************************************************/
2137 static int OpenFilter( vlc_object_t *p_this )
2139 filter_t *p_filter = (filter_t*)p_this;
2140 vout_thread_t *p_vout;
2143 if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2144 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2145 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2146 p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2148 return VLC_EGENERIC;
2151 /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2153 p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2154 vlc_object_attach( p_vout, p_filter );
2155 p_filter->p_sys = (filter_sys_t *)p_vout;
2156 p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2158 sout_CfgParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2160 var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2161 var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2162 var_Set( p_filter, "deinterlace-mode", val );
2164 if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2166 vlc_object_detach( p_vout );
2167 vlc_object_release( p_vout );
2168 return VLC_EGENERIC;
2171 p_filter->pf_video_filter = Deinterlace;
2173 msg_Dbg( p_filter, "deinterlacing" );
2178 /*****************************************************************************
2179 * CloseFilter: clean up the filter
2180 *****************************************************************************/
2181 static void CloseFilter( vlc_object_t *p_this )
2183 filter_t *p_filter = (filter_t*)p_this;
2184 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2186 Destroy( VLC_OBJECT(p_vout) );
2187 vlc_object_detach( p_vout );
2188 vlc_object_release( p_vout );