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., 59 Temple Place - Suite 330, Boston, MA 02111, 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;
374 fmt.i_aspect /= 2; fmt.i_sar_den *= 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 picture_t *pp_outpic[2];
446 pp_outpic[0] = pp_outpic[1] = NULL;
448 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
450 /* Get a new picture */
451 while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
455 if( p_vout->b_die || p_vout->b_error )
457 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
460 msleep( VOUT_OUTMEM_SLEEP );
463 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
465 /* If we are using double rate, get an additional new picture */
466 if( p_vout->p_sys->b_double_rate )
468 while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
472 if( p_vout->b_die || p_vout->b_error )
474 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
475 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
478 msleep( VOUT_OUTMEM_SLEEP );
481 /* 20ms is a bit arbitrary, but it's only for the first image we get */
482 if( !p_vout->p_sys->last_date )
484 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
485 p_pic->date + 20000 );
489 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
490 (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
492 p_vout->p_sys->last_date = p_pic->date;
495 switch( p_vout->p_sys->i_mode )
497 case DEINTERLACE_DISCARD:
498 RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
499 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
502 case DEINTERLACE_BOB:
503 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
504 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
505 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
506 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
509 case DEINTERLACE_LINEAR:
510 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
511 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
512 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
513 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
516 case DEINTERLACE_MEAN:
517 RenderMean( p_vout, pp_outpic[0], p_pic );
518 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
521 case DEINTERLACE_BLEND:
522 RenderBlend( p_vout, pp_outpic[0], p_pic );
523 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
527 RenderX( p_vout, pp_outpic[0], p_pic );
528 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
531 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
534 /*****************************************************************************
535 * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
536 *****************************************************************************/
537 static void RenderDiscard( vout_thread_t *p_vout,
538 picture_t *p_outpic, picture_t *p_pic, int i_field )
542 /* Copy image and skip lines */
543 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
545 uint8_t *p_in, *p_out_end, *p_out;
548 p_in = p_pic->p[i_plane].p_pixels
549 + i_field * p_pic->p[i_plane].i_pitch;
551 p_out = p_outpic->p[i_plane].p_pixels;
552 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
553 * p_outpic->p[i_plane].i_visible_lines;
555 switch( p_vout->render.i_chroma )
557 case VLC_FOURCC('I','4','2','0'):
558 case VLC_FOURCC('I','Y','U','V'):
559 case VLC_FOURCC('Y','V','1','2'):
561 for( ; p_out < p_out_end ; )
563 p_vout->p_vlc->pf_memcpy( p_out, p_in,
564 p_pic->p[i_plane].i_pitch );
566 p_out += p_pic->p[i_plane].i_pitch;
567 p_in += 2 * p_pic->p[i_plane].i_pitch;
571 case VLC_FOURCC('I','4','2','2'):
573 i_increment = 2 * p_pic->p[i_plane].i_pitch;
575 if( i_plane == Y_PLANE )
577 for( ; p_out < p_out_end ; )
579 p_vout->p_vlc->pf_memcpy( p_out, p_in,
580 p_pic->p[i_plane].i_pitch );
581 p_out += p_pic->p[i_plane].i_pitch;
582 p_vout->p_vlc->pf_memcpy( p_out, p_in,
583 p_pic->p[i_plane].i_pitch );
584 p_out += p_pic->p[i_plane].i_pitch;
590 for( ; p_out < p_out_end ; )
592 p_vout->p_vlc->pf_memcpy( p_out, p_in,
593 p_pic->p[i_plane].i_pitch );
594 p_out += p_pic->p[i_plane].i_pitch;
606 /*****************************************************************************
607 * RenderBob: renders a BOB picture - simple copy
608 *****************************************************************************/
609 static void RenderBob( vout_thread_t *p_vout,
610 picture_t *p_outpic, picture_t *p_pic, int i_field )
614 /* Copy image and skip lines */
615 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
617 uint8_t *p_in, *p_out_end, *p_out;
619 p_in = p_pic->p[i_plane].p_pixels;
620 p_out = p_outpic->p[i_plane].p_pixels;
621 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
622 * p_outpic->p[i_plane].i_visible_lines;
624 switch( p_vout->render.i_chroma )
626 case VLC_FOURCC('I','4','2','0'):
627 case VLC_FOURCC('I','Y','U','V'):
628 case VLC_FOURCC('Y','V','1','2'):
629 /* For BOTTOM field we need to add the first line */
632 p_vout->p_vlc->pf_memcpy( p_out, p_in,
633 p_pic->p[i_plane].i_pitch );
634 p_in += p_pic->p[i_plane].i_pitch;
635 p_out += p_pic->p[i_plane].i_pitch;
638 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
640 for( ; p_out < p_out_end ; )
642 p_vout->p_vlc->pf_memcpy( p_out, p_in,
643 p_pic->p[i_plane].i_pitch );
645 p_out += p_pic->p[i_plane].i_pitch;
647 p_vout->p_vlc->pf_memcpy( p_out, p_in,
648 p_pic->p[i_plane].i_pitch );
650 p_in += 2 * p_pic->p[i_plane].i_pitch;
651 p_out += p_pic->p[i_plane].i_pitch;
654 p_vout->p_vlc->pf_memcpy( p_out, p_in,
655 p_pic->p[i_plane].i_pitch );
657 /* For TOP field we need to add the last line */
660 p_in += p_pic->p[i_plane].i_pitch;
661 p_out += p_pic->p[i_plane].i_pitch;
662 p_vout->p_vlc->pf_memcpy( p_out, p_in,
663 p_pic->p[i_plane].i_pitch );
667 case VLC_FOURCC('I','4','2','2'):
668 /* For BOTTOM field we need to add the first line */
671 p_vout->p_vlc->pf_memcpy( p_out, p_in,
672 p_pic->p[i_plane].i_pitch );
673 p_in += p_pic->p[i_plane].i_pitch;
674 p_out += p_pic->p[i_plane].i_pitch;
677 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
679 if( i_plane == Y_PLANE )
681 for( ; p_out < p_out_end ; )
683 p_vout->p_vlc->pf_memcpy( p_out, p_in,
684 p_pic->p[i_plane].i_pitch );
686 p_out += p_pic->p[i_plane].i_pitch;
688 p_vout->p_vlc->pf_memcpy( p_out, p_in,
689 p_pic->p[i_plane].i_pitch );
691 p_in += 2 * p_pic->p[i_plane].i_pitch;
692 p_out += p_pic->p[i_plane].i_pitch;
697 for( ; p_out < p_out_end ; )
699 p_vout->p_vlc->pf_memcpy( p_out, p_in,
700 p_pic->p[i_plane].i_pitch );
702 p_out += p_pic->p[i_plane].i_pitch;
703 p_in += 2 * p_pic->p[i_plane].i_pitch;
707 p_vout->p_vlc->pf_memcpy( p_out, p_in,
708 p_pic->p[i_plane].i_pitch );
710 /* For TOP field we need to add the last line */
713 p_in += p_pic->p[i_plane].i_pitch;
714 p_out += p_pic->p[i_plane].i_pitch;
715 p_vout->p_vlc->pf_memcpy( p_out, p_in,
716 p_pic->p[i_plane].i_pitch );
723 #define Merge p_vout->p_sys->pf_merge
724 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
726 /*****************************************************************************
727 * RenderLinear: BOB with linear interpolation
728 *****************************************************************************/
729 static void RenderLinear( vout_thread_t *p_vout,
730 picture_t *p_outpic, picture_t *p_pic, int i_field )
734 /* Copy image and skip lines */
735 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
737 uint8_t *p_in, *p_out_end, *p_out;
739 p_in = p_pic->p[i_plane].p_pixels;
740 p_out = p_outpic->p[i_plane].p_pixels;
741 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
742 * p_outpic->p[i_plane].i_visible_lines;
744 /* For BOTTOM field we need to add the first line */
747 p_vout->p_vlc->pf_memcpy( p_out, p_in,
748 p_pic->p[i_plane].i_pitch );
749 p_in += p_pic->p[i_plane].i_pitch;
750 p_out += p_pic->p[i_plane].i_pitch;
753 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
755 for( ; p_out < p_out_end ; )
757 p_vout->p_vlc->pf_memcpy( p_out, p_in,
758 p_pic->p[i_plane].i_pitch );
760 p_out += p_pic->p[i_plane].i_pitch;
762 Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
763 p_pic->p[i_plane].i_pitch );
765 p_in += 2 * p_pic->p[i_plane].i_pitch;
766 p_out += p_pic->p[i_plane].i_pitch;
769 p_vout->p_vlc->pf_memcpy( p_out, p_in,
770 p_pic->p[i_plane].i_pitch );
772 /* For TOP field we need to add the last line */
775 p_in += p_pic->p[i_plane].i_pitch;
776 p_out += p_pic->p[i_plane].i_pitch;
777 p_vout->p_vlc->pf_memcpy( p_out, p_in,
778 p_pic->p[i_plane].i_pitch );
784 static void RenderMean( vout_thread_t *p_vout,
785 picture_t *p_outpic, picture_t *p_pic )
789 /* Copy image and skip lines */
790 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
792 uint8_t *p_in, *p_out_end, *p_out;
794 p_in = p_pic->p[i_plane].p_pixels;
796 p_out = p_outpic->p[i_plane].p_pixels;
797 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
798 * p_outpic->p[i_plane].i_visible_lines;
800 /* All lines: mean value */
801 for( ; p_out < p_out_end ; )
803 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
804 p_pic->p[i_plane].i_pitch );
806 p_out += p_pic->p[i_plane].i_pitch;
807 p_in += 2 * p_pic->p[i_plane].i_pitch;
813 static void RenderBlend( vout_thread_t *p_vout,
814 picture_t *p_outpic, picture_t *p_pic )
818 /* Copy image and skip lines */
819 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
821 uint8_t *p_in, *p_out_end, *p_out;
823 p_in = p_pic->p[i_plane].p_pixels;
825 p_out = p_outpic->p[i_plane].p_pixels;
826 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
827 * p_outpic->p[i_plane].i_visible_lines;
829 switch( p_vout->render.i_chroma )
831 case VLC_FOURCC('I','4','2','0'):
832 case VLC_FOURCC('I','Y','U','V'):
833 case VLC_FOURCC('Y','V','1','2'):
834 /* First line: simple copy */
835 p_vout->p_vlc->pf_memcpy( p_out, p_in,
836 p_pic->p[i_plane].i_pitch );
837 p_out += p_pic->p[i_plane].i_pitch;
839 /* Remaining lines: mean value */
840 for( ; p_out < p_out_end ; )
842 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
843 p_pic->p[i_plane].i_pitch );
845 p_out += p_pic->p[i_plane].i_pitch;
846 p_in += p_pic->p[i_plane].i_pitch;
850 case VLC_FOURCC('I','4','2','2'):
851 /* First line: simple copy */
852 p_vout->p_vlc->pf_memcpy( p_out, p_in,
853 p_pic->p[i_plane].i_pitch );
854 p_out += p_pic->p[i_plane].i_pitch;
856 /* Remaining lines: mean value */
857 if( i_plane == Y_PLANE )
859 for( ; p_out < p_out_end ; )
861 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
862 p_pic->p[i_plane].i_pitch );
864 p_out += p_pic->p[i_plane].i_pitch;
865 p_in += p_pic->p[i_plane].i_pitch;
871 for( ; p_out < p_out_end ; )
873 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
874 p_pic->p[i_plane].i_pitch );
876 p_out += p_pic->p[i_plane].i_pitch;
877 p_in += 2*p_pic->p[i_plane].i_pitch;
888 static void MergeGeneric( void *_p_dest, const void *_p_s1,
889 const void *_p_s2, size_t i_bytes )
891 uint8_t* p_dest = (uint8_t*)_p_dest;
892 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
893 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
894 uint8_t* p_end = p_dest + i_bytes - 8;
896 while( p_dest < p_end )
898 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
899 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
900 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
901 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
902 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
903 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
904 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
905 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
910 while( p_dest < p_end )
912 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
916 #if defined(CAN_COMPILE_MMXEXT)
917 static void MergeMMX( void *_p_dest, const void *_p_s1, const void *_p_s2,
920 uint8_t* p_dest = (uint8_t*)_p_dest;
921 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
922 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
923 uint8_t* p_end = p_dest + i_bytes - 8;
924 while( p_dest < p_end )
926 __asm__ __volatile__( "movq %2,%%mm1;"
928 "movq %%mm1, %0" :"=m" (*p_dest):
938 while( p_dest < p_end )
940 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
945 #if defined(CAN_COMPILE_SSE)
946 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
949 uint8_t* p_dest = (uint8_t*)_p_dest;
950 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
951 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
953 while( (ptrdiff_t)p_s1 % 16 )
955 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
957 p_end = p_dest + i_bytes - 16;
958 while( p_dest < p_end )
960 __asm__ __volatile__( "movdqu %2,%%xmm1;"
962 "movdqu %%xmm1, %0" :"=m" (*p_dest):
972 while( p_dest < p_end )
974 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
979 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
980 static void EndMMX( void )
982 __asm__ __volatile__( "emms" :: );
986 #ifdef CAN_COMPILE_C_ALTIVEC
987 static void MergeAltivec( void *_p_dest, const void *_p_s1,
988 const void *_p_s2, size_t i_bytes )
990 uint8_t *p_dest = (uint8_t *)_p_dest;
991 uint8_t *p_s1 = (uint8_t *)_p_s1;
992 uint8_t *p_s2 = (uint8_t *)_p_s2;
993 uint8_t *p_end = p_dest + i_bytes - 15;
995 /* Use C until the first 16-bytes aligned destination pixel */
996 while( (int)p_dest & 0xF )
998 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1001 if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1003 /* Unaligned source */
1004 vector unsigned char s1v, s2v, destv;
1005 vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1006 vector unsigned char perm1v, perm2v;
1008 perm1v = vec_lvsl( 0, p_s1 );
1009 perm2v = vec_lvsl( 0, p_s2 );
1010 s1oldv = vec_ld( 0, p_s1 );
1011 s2oldv = vec_ld( 0, p_s2 );
1013 while( p_dest < p_end )
1015 s1newv = vec_ld( 16, p_s1 );
1016 s2newv = vec_ld( 16, p_s2 );
1017 s1v = vec_perm( s1oldv, s1newv, perm1v );
1018 s2v = vec_perm( s2oldv, s2newv, perm2v );
1021 destv = vec_avg( s1v, s2v );
1022 vec_st( destv, 0, p_dest );
1031 /* Aligned source */
1032 vector unsigned char s1v, s2v, destv;
1034 while( p_dest < p_end )
1036 s1v = vec_ld( 0, p_s1 );
1037 s2v = vec_ld( 0, p_s2 );
1038 destv = vec_avg( s1v, s2v );
1039 vec_st( destv, 0, p_dest );
1049 while( p_dest < p_end )
1051 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1056 /*****************************************************************************
1057 * RenderX: This algo works on a 8x8 block basic, it copies the top field
1058 * and apply a process to recreate the bottom field :
1059 * If a 8x8 block is classified as :
1060 * - progressive: it applies a small blend (1,6,1)
1062 * * in the MMX version: we do a ME between the 2 fields, if there is a
1063 * good match we use MC to recreate the bottom field (with a small
1065 * * otherwise: it recreates the bottom field by an edge oriented
1067 *****************************************************************************/
1069 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1070 * XXX: It need to access to 8x10
1071 * We use more than 8 lines to help with scrolling (text)
1072 * (and because XDeint8x8Frame use line 9)
1073 * XXX: smooth/uniform area with noise detection doesn't works well
1074 * but it's not really a problem because they don't have much details anyway
1076 static inline int ssd( int a ) { return a*a; }
1077 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1083 /* Detect interlacing */
1085 for( y = 0; y < 7; y += 2 )
1088 for( x = 0; x < 8; x++ )
1090 fr += ssd(src[ x] - src[1*i_src+x]) +
1091 ssd(src[i_src+x] - src[2*i_src+x]);
1092 ff += ssd(src[ x] - src[2*i_src+x]) +
1093 ssd(src[i_src+x] - src[3*i_src+x]);
1095 if( ff < 6*fr/8 && fr > 32 )
1101 return fc < 1 ? VLC_FALSE : VLC_TRUE;
1103 #ifdef CAN_COMPILE_MMXEXT
1104 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1111 /* Detect interlacing */
1113 pxor_r2r( mm7, mm7 );
1114 for( y = 0; y < 9; y += 2 )
1117 pxor_r2r( mm5, mm5 );
1118 pxor_r2r( mm6, mm6 );
1119 for( x = 0; x < 8; x+=4 )
1121 movd_m2r( src[ x], mm0 );
1122 movd_m2r( src[1*i_src+x], mm1 );
1123 movd_m2r( src[2*i_src+x], mm2 );
1124 movd_m2r( src[3*i_src+x], mm3 );
1126 punpcklbw_r2r( mm7, mm0 );
1127 punpcklbw_r2r( mm7, mm1 );
1128 punpcklbw_r2r( mm7, mm2 );
1129 punpcklbw_r2r( mm7, mm3 );
1131 movq_r2r( mm0, mm4 );
1133 psubw_r2r( mm1, mm0 );
1134 psubw_r2r( mm2, mm4 );
1136 psubw_r2r( mm1, mm2 );
1137 psubw_r2r( mm1, mm3 );
1139 pmaddwd_r2r( mm0, mm0 );
1140 pmaddwd_r2r( mm4, mm4 );
1141 pmaddwd_r2r( mm2, mm2 );
1142 pmaddwd_r2r( mm3, mm3 );
1143 paddd_r2r( mm0, mm2 );
1144 paddd_r2r( mm4, mm3 );
1145 paddd_r2r( mm2, mm5 );
1146 paddd_r2r( mm3, mm6 );
1149 movq_r2r( mm5, mm0 );
1150 psrlq_i2r( 32, mm0 );
1151 paddd_r2r( mm0, mm5 );
1152 movd_r2m( mm5, fr );
1154 movq_r2r( mm6, mm0 );
1155 psrlq_i2r( 32, mm0 );
1156 paddd_r2r( mm0, mm6 );
1157 movd_r2m( mm6, ff );
1159 if( ff < 6*fr/8 && fr > 32 )
1168 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1169 * This won't destroy details, and help if there is a bit of interlacing.
1170 * (It helps with paning to avoid flickers)
1174 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1175 uint8_t *src, int i_src )
1180 for( y = 0; y < 8; y += 2 )
1182 memcpy( dst, src, 8 );
1185 for( x = 0; x < 8; x++ )
1186 dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1192 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1193 uint8_t *src1, int i_src1,
1194 uint8_t *src2, int i_src2 )
1199 for( y = 0; y < 8; y += 2 )
1201 memcpy( dst, src1, 8 );
1204 for( x = 0; x < 8; x++ )
1205 dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1213 #ifdef CAN_COMPILE_MMXEXT
1214 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1215 uint8_t *src1, int i_src1,
1216 uint8_t *src2, int i_src2 )
1218 static const uint64_t m_4 = I64C(0x0004000400040004);
1222 pxor_r2r( mm7, mm7 );
1223 for( y = 0; y < 8; y += 2 )
1225 for( x = 0; x < 8; x +=4 )
1227 movd_m2r( src1[x], mm0 );
1228 movd_r2m( mm0, dst[x] );
1230 movd_m2r( src2[x], mm1 );
1231 movd_m2r( src1[i_src1+x], mm2 );
1233 punpcklbw_r2r( mm7, mm0 );
1234 punpcklbw_r2r( mm7, mm1 );
1235 punpcklbw_r2r( mm7, mm2 );
1236 paddw_r2r( mm1, mm1 );
1237 movq_r2r( mm1, mm3 );
1238 paddw_r2r( mm3, mm3 );
1239 paddw_r2r( mm2, mm0 );
1240 paddw_r2r( mm3, mm1 );
1241 paddw_m2r( m_4, mm1 );
1242 paddw_r2r( mm1, mm0 );
1243 psraw_i2r( 3, mm0 );
1244 packuswb_r2r( mm7, mm0 );
1245 movd_r2m( mm0, dst[i_dst+x] );
1256 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1259 for( y = 0; y < 8; y++ )
1260 memset( &dst[y*i_dst], v, 8 );
1263 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1266 * TODO: a better one for the inner part.
1268 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1269 uint8_t *src, int i_src )
1274 for( y = 0; y < 8; y += 2 )
1276 memcpy( dst, src, 8 );
1279 for( x = 0; x < 8; x++ )
1280 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1285 #ifdef CAN_COMPILE_MMXEXT
1286 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1287 uint8_t *src, int i_src )
1292 for( y = 0; y < 8; y += 2 )
1294 movq_m2r( src[0], mm0 );
1295 movq_r2m( mm0, dst[0] );
1298 movq_m2r( src[2*i_src], mm1 );
1299 pavgb_r2r( mm1, mm0 );
1301 movq_r2m( mm0, dst[0] );
1309 /* XDeint8x8Field: Edge oriented interpolation
1310 * (Need -4 and +5 pixels H, +1 line)
1312 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1313 uint8_t *src, int i_src )
1318 for( y = 0; y < 8; y += 2 )
1320 memcpy( dst, src, 8 );
1323 for( x = 0; x < 8; x++ )
1325 uint8_t *src2 = &src[2*i_src];
1326 /* I use 8 pixels just to match the MMX version, but it's overkill
1327 * 5 would be enough (less isn't good) */
1328 const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1329 abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1330 abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1331 abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1333 const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1334 abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1335 abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1336 abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1338 const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1339 abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1340 abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1341 abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1343 if( c0 < c1 && c1 <= c2 )
1344 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1345 else if( c2 < c1 && c1 <= c0 )
1346 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1348 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1355 #ifdef CAN_COMPILE_MMXEXT
1356 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1357 uint8_t *src, int i_src )
1362 for( y = 0; y < 8; y += 2 )
1364 memcpy( dst, src, 8 );
1367 for( x = 0; x < 8; x++ )
1369 uint8_t *src2 = &src[2*i_src];
1372 movq_m2r( src[x-2], mm0 );
1373 movq_m2r( src[x-3], mm1 );
1374 movq_m2r( src[x-4], mm2 );
1376 psadbw_m2r( src2[x-4], mm0 );
1377 psadbw_m2r( src2[x-3], mm1 );
1378 psadbw_m2r( src2[x-2], mm2 );
1380 movd_r2m( mm0, c2 );
1381 movd_r2m( mm1, c1 );
1382 movd_r2m( mm2, c0 );
1384 if( c0 < c1 && c1 <= c2 )
1385 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1386 else if( c2 < c1 && c1 <= c0 )
1387 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1389 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1399 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1400 uint8_t *pix2, int i_pix2 )
1405 for( y = 0; y < 8; y++ )
1406 for( x = 0; x < 8; x++ )
1407 s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1411 #ifdef CAN_COMPILE_MMXEXT
1412 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1413 uint8_t *pix2, int i_pix2 )
1418 pxor_r2r( mm7, mm7 );
1419 pxor_r2r( mm6, mm6 );
1421 for( y = 0; y < 8; y++ )
1423 movq_m2r( pix1[0], mm0 );
1424 movq_m2r( pix2[0], mm1 );
1426 movq_r2r( mm0, mm2 );
1427 movq_r2r( mm1, mm3 );
1429 punpcklbw_r2r( mm7, mm0 );
1430 punpckhbw_r2r( mm7, mm2 );
1431 punpcklbw_r2r( mm7, mm1 );
1432 punpckhbw_r2r( mm7, mm3 );
1434 psubw_r2r( mm1, mm0 );
1435 psubw_r2r( mm3, mm2 );
1437 pmaddwd_r2r( mm0, mm0 );
1438 pmaddwd_r2r( mm2, mm2 );
1440 paddd_r2r( mm2, mm0 );
1441 paddd_r2r( mm0, mm6 );
1447 movq_r2r( mm6, mm7 );
1448 psrlq_i2r( 32, mm7 );
1449 paddd_r2r( mm6, mm7 );
1458 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1459 #ifdef CAN_COMPILE_MMXEXT
1462 * TODO: mmx version (easier in sse2)
1464 static inline void XDeintMC( uint8_t *dst, int i_dst,
1465 uint8_t *src, int i_src,
1467 int i_width, int i_height )
1469 const int d4x = mvx&0x03;
1470 const int d4y = mvy&0x03;
1472 const int cA = (4-d4x)*(4-d4y);
1473 const int cB = d4x *(4-d4y);
1474 const int cC = (4-d4x)*d4y;
1475 const int cD = d4x *d4y;
1481 src += (mvy >> 2) * i_src + (mvx >> 2);
1484 for( y = 0; y < i_height; y++ )
1486 for( x = 0; x < i_width; x++ )
1488 dst[x] = ( cA*src[x] + cB*src[x+1] +
1489 cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1497 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1498 uint8_t *pix2, int i_pix2 )
1502 movq_m2r( pix1[0*i_pix1], mm0 );
1503 movq_m2r( pix1[1*i_pix1], mm1 );
1505 psadbw_m2r( pix2[0*i_pix2], mm0 );
1506 psadbw_m2r( pix2[1*i_pix2], mm1 );
1508 movq_m2r( pix1[2*i_pix1], mm2 );
1509 movq_m2r( pix1[3*i_pix1], mm3 );
1510 psadbw_m2r( pix2[2*i_pix2], mm2 );
1511 psadbw_m2r( pix2[3*i_pix2], mm3 );
1513 paddd_r2r( mm1, mm0 );
1514 paddd_r2r( mm3, mm2 );
1515 paddd_r2r( mm2, mm0 );
1521 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1522 uint8_t *ref, int i_stride,
1524 int xmax, int ymax )
1526 uint8_t buffer[8*4];
1528 if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1531 XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1532 return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1534 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1535 uint8_t *ref, int i_stride,
1537 int xmax, int ymax )
1539 if( abs(mx) >= xmax || abs(my) >= ymax )
1542 return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1545 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1546 uint8_t *src, int i_src,
1548 int xmax, int ymax )
1550 static const int dx[8] = { 0, 0, -1, 1, -1, -1, 1, 1 };
1551 static const int dy[8] = {-1, 1, 0, 0, -1, 1, -1, 1 };
1552 uint8_t *next = &src[i_src];
1553 const int i_src2 = 2*i_src;
1558 uint8_t *rec = &dst[i_dst];
1560 /* We construct with intra method the missing field */
1561 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1563 /* Now we will try to find a match with ME with the other field */
1565 /* ME: A small/partial EPZS
1566 * We search only for small MV (with high motion intra will be perfect */
1567 if( xmax > 4 ) xmax = 4;
1568 if( ymax > 4 ) ymax = 4;
1570 /* Init with NULL Mv */
1572 mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1574 /* Try predicted Mv */
1575 if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1581 /* Search interger pel (small mv) */
1582 for( i_step = 0; i_step < 4; i_step++ )
1588 for( i = 0; i < 4; i++ )
1590 s = XDeint8x4TestInt( rec, i_src2,
1591 next, i_src2, mvx+dx[i], mvy+dy[i],
1611 if( mvs > 4 && mvs < 256 )
1614 /* XXX: for now only HPEL (too slow) */
1615 for( i_step = 0; i_step < 4; i_step++ )
1621 for( i = 0; i < 8; i++ )
1623 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1624 mvx+dx[i], mvy+dy[i],
1642 uint8_t buffer[8*4];
1643 XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1644 XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1646 //XDeint8x8Set( dst, i_dst, 0 );
1653 /* Kernel interpolation (1,-5,20,20,-5,1)
1654 * Loose a bit more details+add aliasing than edge interpol but avoid
1657 static inline uint8_t clip1( int a )
1666 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1667 uint8_t *src, int i_src )
1672 for( y = 0; y < 8; y += 2 )
1674 const int i_src2 = i_src*2;
1676 memcpy( dst, src, 8 );
1679 for( x = 0; x < 8; x++ )
1683 pix = 1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1684 -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1685 +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1687 dst[x] = clip1( ( pix + 16 ) >> 5 );
1697 /* NxN arbitray size (and then only use pixel in the NxN block)
1699 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1700 int i_height, int i_width )
1707 /* Detect interlacing */
1708 /* FIXME way too simple, need to be more like XDeint8x8Detect */
1711 for( y = 0; y < i_height - 2; y += 2 )
1713 const uint8_t *s = &src[y*i_src];
1714 for( x = 0; x < i_width; x++ )
1716 fr += ssd(s[ x] - s[1*i_src+x]);
1717 ff += ssd(s[ x] - s[2*i_src+x]);
1719 if( ff < fr && fr > i_width / 2 )
1723 return fc < 2 ? VLC_FALSE : VLC_TRUE;
1726 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1727 uint8_t *src, int i_src,
1728 int i_width, int i_height )
1733 for( y = 0; y < i_height; y += 2 )
1735 memcpy( dst, src, i_width );
1738 if( y < i_height - 2 )
1740 for( x = 0; x < i_width; x++ )
1741 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1745 /* Blend last line */
1746 for( x = 0; x < i_width; x++ )
1747 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1754 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1755 uint8_t *src, int i_src,
1756 int i_width, int i_height )
1761 for( y = 0; y < i_height; y += 2 )
1763 memcpy( dst, src, i_width );
1766 if( y < i_height - 2 )
1768 for( x = 0; x < i_width; x++ )
1769 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1773 /* Blend last line */
1774 for( x = 0; x < i_width; x++ )
1775 dst[x] = (src[x] + src[i_src+x]) >> 1;
1782 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1783 int i_width, int i_height )
1785 if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1786 XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1788 XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1792 static inline int median( int a, int b, int c )
1794 int min = a, max =a;
1805 return a + b + c - min - max;
1811 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1812 uint8_t *src, int i_src,
1813 const int i_mbx, int i_modx )
1817 for( x = 0; x < i_mbx; x++ )
1820 if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1822 if( x == 0 || x == i_mbx - 1 )
1823 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1825 XDeint8x8FieldC( dst, i_dst, src, i_src );
1829 XDeint8x8MergeC( dst, i_dst,
1830 &src[0*i_src], 2*i_src,
1831 &src[1*i_src], 2*i_src );
1839 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1841 #ifdef CAN_COMPILE_MMXEXT
1842 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1843 uint8_t *src, int i_src,
1844 const int i_mbx, int i_modx )
1848 /* Reset current line */
1849 for( x = 0; x < i_mbx; x++ )
1852 if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1854 if( x == 0 || x == i_mbx - 1 )
1855 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1857 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1861 XDeint8x8MergeMMXEXT( dst, i_dst,
1862 &src[0*i_src], 2*i_src,
1863 &src[1*i_src], 2*i_src );
1871 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1875 static void RenderX( vout_thread_t *p_vout,
1876 picture_t *p_outpic, picture_t *p_pic )
1880 /* Copy image and skip lines */
1881 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1883 const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1884 const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1886 const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1887 const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1889 const int i_dst = p_outpic->p[i_plane].i_pitch;
1890 const int i_src = p_pic->p[i_plane].i_pitch;
1894 for( y = 0; y < i_mby; y++ )
1896 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1897 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1899 #ifdef CAN_COMPILE_MMXEXT
1900 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1901 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1904 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1907 /* Last line (C only)*/
1910 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1911 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1913 for( x = 0; x < i_mbx; x++ )
1915 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1922 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1926 #ifdef CAN_COMPILE_MMXEXT
1927 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1932 /*****************************************************************************
1933 * SendEvents: forward mouse and keyboard events to the parent p_vout
1934 *****************************************************************************/
1935 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
1936 vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
1938 vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
1939 vlc_value_t sentval = newval;
1941 if( !strcmp( psz_var, "mouse-y" ) )
1943 switch( p_vout->p_sys->i_mode )
1945 case DEINTERLACE_MEAN:
1946 case DEINTERLACE_DISCARD:
1952 var_Set( p_vout, psz_var, sentval );
1957 /*****************************************************************************
1958 * FilterCallback: called when changing the deinterlace method on the fly.
1959 *****************************************************************************/
1960 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
1961 vlc_value_t oldval, vlc_value_t newval,
1964 vout_thread_t * p_vout = (vout_thread_t *)p_this;
1965 int i_old_mode = p_vout->p_sys->i_mode;
1967 msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
1969 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
1971 SetFilterMethod( p_vout, newval.psz_string );
1973 switch( p_vout->render.i_chroma )
1975 case VLC_FOURCC('I','4','2','2'):
1976 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1980 case VLC_FOURCC('I','4','2','0'):
1981 case VLC_FOURCC('I','Y','U','V'):
1982 case VLC_FOURCC('Y','V','1','2'):
1983 switch( p_vout->p_sys->i_mode )
1985 case DEINTERLACE_MEAN:
1986 case DEINTERLACE_DISCARD:
1987 if( ( i_old_mode == DEINTERLACE_MEAN )
1988 || ( i_old_mode == DEINTERLACE_DISCARD ) )
1990 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1995 case DEINTERLACE_BOB:
1996 case DEINTERLACE_BLEND:
1997 case DEINTERLACE_LINEAR:
1998 if( ( i_old_mode == DEINTERLACE_BOB )
1999 || ( i_old_mode == DEINTERLACE_BLEND )
2000 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2002 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2013 /* We need to kill the old vout */
2015 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2017 vlc_object_detach( p_vout->p_sys->p_vout );
2018 vout_Destroy( p_vout->p_sys->p_vout );
2020 /* Try to open a new video output */
2021 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2023 if( p_vout->p_sys->p_vout == NULL )
2025 /* Everything failed */
2026 msg_Err( p_vout, "cannot open vout, aborting" );
2028 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2029 return VLC_EGENERIC;
2032 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2034 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2038 /*****************************************************************************
2039 * SendEventsToChild: forward events to the child/children vout
2040 *****************************************************************************/
2041 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2042 vlc_value_t oldval, vlc_value_t newval, void *p_data )
2044 vout_thread_t *p_vout = (vout_thread_t *)p_this;
2045 var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2050 /*****************************************************************************
2051 * video filter2 functions
2052 *****************************************************************************/
2053 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2055 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2056 picture_t *p_pic_dst;
2058 /* Request output picture */
2059 p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2060 if( p_pic_dst == NULL )
2062 msg_Warn( p_filter, "can't get output picture" );
2066 switch( p_vout->p_sys->i_mode )
2068 case DEINTERLACE_DISCARD:
2070 RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2072 msg_Err( p_vout, "discarding lines is not supported yet" );
2073 p_pic_dst->pf_release( p_pic_dst );
2077 case DEINTERLACE_BOB:
2079 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2080 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2084 case DEINTERLACE_LINEAR:
2086 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2087 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2089 msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2090 p_pic_dst->pf_release( p_pic_dst );
2094 case DEINTERLACE_MEAN:
2095 RenderMean( p_vout, p_pic_dst, p_pic );
2098 case DEINTERLACE_BLEND:
2099 RenderBlend( p_vout, p_pic_dst, p_pic );
2103 RenderX( p_vout, p_pic_dst, p_pic );
2107 p_pic_dst->date = p_pic->date;
2108 p_pic_dst->b_force = p_pic->b_force;
2109 p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2110 p_pic_dst->b_progressive = VLC_TRUE;
2111 p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2113 p_pic->pf_release( p_pic );
2117 /*****************************************************************************
2119 *****************************************************************************/
2120 static int OpenFilter( vlc_object_t *p_this )
2122 filter_t *p_filter = (filter_t*)p_this;
2123 vout_thread_t *p_vout;
2126 if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2127 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2128 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2129 p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2131 return VLC_EGENERIC;
2134 /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2136 p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2137 vlc_object_attach( p_vout, p_filter );
2138 p_filter->p_sys = (filter_sys_t *)p_vout;
2139 p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2141 sout_CfgParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2143 var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2144 var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2145 var_Set( p_filter, "deinterlace-mode", val );
2147 if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2149 vlc_object_detach( p_vout );
2150 vlc_object_release( p_vout );
2151 return VLC_EGENERIC;
2154 p_filter->pf_video_filter = Deinterlace;
2156 msg_Dbg( p_filter, "deinterlacing" );
2161 /*****************************************************************************
2162 * CloseFilter: clean up the filter
2163 *****************************************************************************/
2164 static void CloseFilter( vlc_object_t *p_this )
2166 filter_t *p_filter = (filter_t*)p_this;
2167 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2169 Destroy( VLC_OBJECT(p_vout) );
2170 vlc_object_detach( p_vout );
2171 vlc_object_release( p_vout );