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 *****************************************************************************/
32 #include "vlc_filter.h"
38 #ifdef CAN_COMPILE_MMXEXT
42 #include "filter_common.h"
44 #define DEINTERLACE_DISCARD 1
45 #define DEINTERLACE_MEAN 2
46 #define DEINTERLACE_BLEND 3
47 #define DEINTERLACE_BOB 4
48 #define DEINTERLACE_LINEAR 5
49 #define DEINTERLACE_X 6
51 /*****************************************************************************
53 *****************************************************************************/
54 static int Create ( vlc_object_t * );
55 static void Destroy ( vlc_object_t * );
57 static int Init ( vout_thread_t * );
58 static void End ( vout_thread_t * );
59 static void Render ( vout_thread_t *, picture_t * );
61 static void RenderDiscard( vout_thread_t *, picture_t *, picture_t *, int );
62 static void RenderBob ( vout_thread_t *, picture_t *, picture_t *, int );
63 static void RenderMean ( vout_thread_t *, picture_t *, picture_t * );
64 static void RenderBlend ( vout_thread_t *, picture_t *, picture_t * );
65 static void RenderLinear ( vout_thread_t *, picture_t *, picture_t *, int );
66 static void RenderX ( vout_thread_t *, picture_t *, picture_t * );
68 static void MergeGeneric ( void *, const void *, const void *, size_t );
69 #if defined(CAN_COMPILE_C_ALTIVEC)
70 static void MergeAltivec ( void *, const void *, const void *, size_t );
72 #if defined(CAN_COMPILE_MMXEXT)
73 static void MergeMMXEXT ( void *, const void *, const void *, size_t );
75 #if defined(CAN_COMPILE_3DNOW)
76 static void Merge3DNow ( void *, const void *, const void *, size_t );
78 #if defined(CAN_COMPILE_SSE)
79 static void MergeSSE2 ( void *, const void *, const void *, size_t );
81 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
82 static void EndMMX ( void );
84 #if defined(CAN_COMPILE_3DNOW)
85 static void End3DNow ( void );
88 static int SendEvents ( vlc_object_t *, char const *,
89 vlc_value_t, vlc_value_t, void * );
91 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method );
92 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout );
94 static int OpenFilter( vlc_object_t *p_this );
95 static void CloseFilter( vlc_object_t *p_this );
97 /*****************************************************************************
99 *****************************************************************************/
100 static int FilterCallback ( vlc_object_t *, char const *,
101 vlc_value_t, vlc_value_t, void * );
103 /*****************************************************************************
105 *****************************************************************************/
106 #define MODE_TEXT N_("Deinterlace mode")
107 #define MODE_LONGTEXT N_("Deinterlace method to use for local playback.")
109 #define SOUT_MODE_TEXT N_("Streaming deinterlace mode")
110 #define SOUT_MODE_LONGTEXT N_("Deinterlace method to use for streaming.")
112 #define FILTER_CFG_PREFIX "sout-deinterlace-"
114 static const char *mode_list[] = { "discard", "blend", "mean", "bob", "linear", "x" };
115 static const char *mode_list_text[] = { N_("Discard"), N_("Blend"), N_("Mean"),
116 N_("Bob"), N_("Linear"), "X" };
119 set_description( _("Deinterlacing video filter") );
120 set_shortname( _("Deinterlace" ));
121 set_capability( "video filter", 0 );
122 set_category( CAT_VIDEO );
123 set_subcategory( SUBCAT_VIDEO_VFILTER );
125 set_section( N_("Display"),NULL);
126 add_string( "deinterlace-mode", "discard", NULL, MODE_TEXT,
127 MODE_LONGTEXT, VLC_FALSE );
128 change_string_list( mode_list, mode_list_text, 0 );
130 add_shortcut( "deinterlace" );
131 set_callbacks( Create, Destroy );
134 set_capability( "video filter2", 0 );
135 set_section( N_("Streaming"),NULL);
136 add_string( FILTER_CFG_PREFIX "mode", "blend", NULL, SOUT_MODE_TEXT,
137 SOUT_MODE_LONGTEXT, VLC_FALSE );
138 change_string_list( mode_list, mode_list_text, 0 );
139 set_callbacks( OpenFilter, CloseFilter );
142 static const char *ppsz_filter_options[] = {
146 /*****************************************************************************
147 * vout_sys_t: Deinterlace video output method descriptor
148 *****************************************************************************
149 * This structure is part of the video output thread descriptor.
150 * It describes the Deinterlace specific properties of an output thread.
151 *****************************************************************************/
154 int i_mode; /* Deinterlace mode */
155 vlc_bool_t b_double_rate; /* Shall we double the framerate? */
160 vout_thread_t *p_vout;
162 vlc_mutex_t filter_lock;
164 void (*pf_merge) ( void *, const void *, const void *, size_t );
165 void (*pf_end_merge) ( void );
168 /*****************************************************************************
169 * Control: control facility for the vout (forwards to child vout)
170 *****************************************************************************/
171 static int Control( vout_thread_t *p_vout, int i_query, va_list args )
173 return vout_vaControl( p_vout->p_sys->p_vout, i_query, args );
176 /*****************************************************************************
177 * Create: allocates Deinterlace video thread output method
178 *****************************************************************************
179 * This function allocates and initializes a Deinterlace vout method.
180 *****************************************************************************/
181 static int Create( vlc_object_t *p_this )
183 vout_thread_t *p_vout = (vout_thread_t *)p_this;
186 /* Allocate structure */
187 p_vout->p_sys = malloc( sizeof( vout_sys_t ) );
188 if( p_vout->p_sys == NULL )
190 msg_Err( p_vout, "out of memory" );
194 p_vout->pf_init = Init;
195 p_vout->pf_end = End;
196 p_vout->pf_manage = NULL;
197 p_vout->pf_render = Render;
198 p_vout->pf_display = NULL;
199 p_vout->pf_control = Control;
201 p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
202 p_vout->p_sys->b_double_rate = VLC_FALSE;
203 p_vout->p_sys->last_date = 0;
204 p_vout->p_sys->p_vout = 0;
205 vlc_mutex_init( p_vout, &p_vout->p_sys->filter_lock );
207 #if defined(CAN_COMPILE_C_ALTIVEC)
208 if( vlc_CPU() & CPU_CAPABILITY_ALTIVEC )
210 p_vout->p_sys->pf_merge = MergeAltivec;
211 p_vout->p_sys->pf_end_merge = NULL;
215 #if defined(CAN_COMPILE_SSE)
216 if( vlc_CPU() & CPU_CAPABILITY_SSE2 )
218 p_vout->p_sys->pf_merge = MergeSSE2;
219 p_vout->p_sys->pf_end_merge = EndMMX;
223 #if defined(CAN_COMPILE_MMXEXT)
224 if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
226 p_vout->p_sys->pf_merge = MergeMMXEXT;
227 p_vout->p_sys->pf_end_merge = EndMMX;
231 #if defined(CAN_COMPILE_3DNOW)
232 if( vlc_CPU() & CPU_CAPABILITY_3DNOW )
234 p_vout->p_sys->pf_merge = Merge3DNow;
235 p_vout->p_sys->pf_end_merge = End3DNow;
240 p_vout->p_sys->pf_merge = MergeGeneric;
241 p_vout->p_sys->pf_end_merge = NULL;
244 /* Look what method was requested */
245 var_Create( p_vout, "deinterlace-mode", VLC_VAR_STRING );
246 var_Change( p_vout, "deinterlace-mode", VLC_VAR_INHERITVALUE, &val, NULL );
248 if( val.psz_string == NULL )
250 msg_Err( p_vout, "configuration variable deinterlace-mode empty" );
251 msg_Err( p_vout, "no deinterlace mode provided, using \"discard\"" );
253 val.psz_string = strdup( "discard" );
256 msg_Dbg( p_vout, "using %s deinterlace mode", val.psz_string );
258 SetFilterMethod( p_vout, val.psz_string );
260 free( val.psz_string );
265 /*****************************************************************************
266 * SetFilterMethod: setup the deinterlace method to use.
267 *****************************************************************************/
268 static void SetFilterMethod( vout_thread_t *p_vout, char *psz_method )
270 if( !strcmp( psz_method, "discard" ) )
272 p_vout->p_sys->i_mode = DEINTERLACE_DISCARD;
273 p_vout->p_sys->b_double_rate = VLC_FALSE;
275 else if( !strcmp( psz_method, "mean" ) )
277 p_vout->p_sys->i_mode = DEINTERLACE_MEAN;
278 p_vout->p_sys->b_double_rate = VLC_FALSE;
280 else if( !strcmp( psz_method, "blend" )
281 || !strcmp( psz_method, "average" )
282 || !strcmp( psz_method, "combine-fields" ) )
284 p_vout->p_sys->i_mode = DEINTERLACE_BLEND;
285 p_vout->p_sys->b_double_rate = VLC_FALSE;
287 else if( !strcmp( psz_method, "bob" )
288 || !strcmp( psz_method, "progressive-scan" ) )
290 p_vout->p_sys->i_mode = DEINTERLACE_BOB;
291 p_vout->p_sys->b_double_rate = VLC_TRUE;
293 else if( !strcmp( psz_method, "linear" ) )
295 p_vout->p_sys->i_mode = DEINTERLACE_LINEAR;
296 p_vout->p_sys->b_double_rate = VLC_TRUE;
298 else if( !strcmp( psz_method, "x" ) )
300 p_vout->p_sys->i_mode = DEINTERLACE_X;
301 p_vout->p_sys->b_double_rate = VLC_FALSE;
305 msg_Err( p_vout, "no valid deinterlace mode provided, "
306 "using \"discard\"" );
309 msg_Dbg( p_vout, "using %s deinterlace method", psz_method );
312 /*****************************************************************************
313 * Init: initialize Deinterlace video thread output method
314 *****************************************************************************/
315 static int Init( vout_thread_t *p_vout )
320 I_OUTPUTPICTURES = 0;
322 /* Initialize the output structure, full of directbuffers since we want
323 * the decoder to output directly to our structures. */
324 switch( p_vout->render.i_chroma )
326 case VLC_FOURCC('I','4','2','0'):
327 case VLC_FOURCC('I','Y','U','V'):
328 case VLC_FOURCC('Y','V','1','2'):
329 case VLC_FOURCC('I','4','2','2'):
330 p_vout->output.i_chroma = p_vout->render.i_chroma;
331 p_vout->output.i_width = p_vout->render.i_width;
332 p_vout->output.i_height = p_vout->render.i_height;
333 p_vout->output.i_aspect = p_vout->render.i_aspect;
334 p_vout->fmt_out = p_vout->fmt_in;
338 return VLC_EGENERIC; /* unknown chroma */
342 /* Try to open the real video output */
343 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
345 if( p_vout->p_sys->p_vout == NULL )
347 /* Everything failed */
348 msg_Err( p_vout, "cannot open vout, aborting" );
353 var_AddCallback( p_vout, "deinterlace-mode", FilterCallback, NULL );
355 ALLOCATE_DIRECTBUFFERS( VOUT_MAX_PICTURES );
357 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
359 ADD_PARENT_CALLBACKS( SendEventsToChild );
364 /*****************************************************************************
365 * SpawnRealVout: spawn the real video output.
366 *****************************************************************************/
367 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout )
369 vout_thread_t *p_real_vout = NULL;
371 memset( &fmt, 0, sizeof( video_format_t ) );
373 msg_Dbg( p_vout, "spawning the real video output" );
375 fmt = p_vout->fmt_out;
377 switch( p_vout->render.i_chroma )
379 case VLC_FOURCC('I','4','2','0'):
380 case VLC_FOURCC('I','Y','U','V'):
381 case VLC_FOURCC('Y','V','1','2'):
382 switch( p_vout->p_sys->i_mode )
384 case DEINTERLACE_MEAN:
385 case DEINTERLACE_DISCARD:
386 fmt.i_height /= 2; fmt.i_visible_height /= 2; fmt.i_y_offset /= 2;
388 p_real_vout = vout_Create( p_vout, &fmt );
391 case DEINTERLACE_BOB:
392 case DEINTERLACE_BLEND:
393 case DEINTERLACE_LINEAR:
395 p_real_vout = vout_Create( p_vout, &fmt );
400 case VLC_FOURCC('I','4','2','2'):
401 fmt.i_chroma = VLC_FOURCC('I','4','2','0');
402 p_real_vout = vout_Create( p_vout, &fmt );
412 /*****************************************************************************
413 * End: terminate Deinterlace video thread output method
414 *****************************************************************************/
415 static void End( vout_thread_t *p_vout )
419 /* Free the fake output buffers we allocated */
420 for( i_index = I_OUTPUTPICTURES ; i_index ; )
423 free( PP_OUTPUTPICTURE[ i_index ]->p_data_orig );
426 if( p_vout->p_sys->p_vout )
428 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
429 vlc_object_detach( p_vout->p_sys->p_vout );
430 vout_Destroy( p_vout->p_sys->p_vout );
433 DEL_PARENT_CALLBACKS( SendEventsToChild );
436 /*****************************************************************************
437 * Destroy: destroy Deinterlace video thread output method
438 *****************************************************************************
439 * Terminate an output method created by DeinterlaceCreateOutputMethod
440 *****************************************************************************/
441 static void Destroy( vlc_object_t *p_this )
443 vout_thread_t *p_vout = (vout_thread_t *)p_this;
444 vlc_mutex_destroy( &p_vout->p_sys->filter_lock );
445 free( p_vout->p_sys );
448 /*****************************************************************************
449 * Render: displays previously rendered output
450 *****************************************************************************
451 * This function send the currently rendered image to Deinterlace image,
452 * waits until it is displayed and switch the two rendering buffers, preparing
454 *****************************************************************************/
455 static void Render ( vout_thread_t *p_vout, picture_t *p_pic )
457 vout_sys_t *p_sys = p_vout->p_sys;
458 picture_t *pp_outpic[2];
460 p_vout->fmt_out.i_x_offset = p_sys->p_vout->fmt_in.i_x_offset =
461 p_vout->fmt_in.i_x_offset;
462 p_vout->fmt_out.i_y_offset = p_sys->p_vout->fmt_in.i_y_offset =
463 p_vout->fmt_in.i_y_offset;
464 p_vout->fmt_out.i_visible_width = p_sys->p_vout->fmt_in.i_visible_width =
465 p_vout->fmt_in.i_visible_width;
466 p_vout->fmt_out.i_visible_height = p_sys->p_vout->fmt_in.i_visible_height =
467 p_vout->fmt_in.i_visible_height;
468 if( p_vout->p_sys->i_mode == DEINTERLACE_MEAN ||
469 p_vout->p_sys->i_mode == DEINTERLACE_DISCARD )
471 p_vout->fmt_out.i_y_offset /= 2; p_sys->p_vout->fmt_in.i_y_offset /= 2;
472 p_vout->fmt_out.i_visible_height /= 2;
473 p_sys->p_vout->fmt_in.i_visible_height /= 2;
476 pp_outpic[0] = pp_outpic[1] = NULL;
478 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
480 /* Get a new picture */
481 while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
485 if( p_vout->b_die || p_vout->b_error )
487 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
490 msleep( VOUT_OUTMEM_SLEEP );
493 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
495 /* If we are using double rate, get an additional new picture */
496 if( p_vout->p_sys->b_double_rate )
498 while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
502 if( p_vout->b_die || p_vout->b_error )
504 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
505 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
508 msleep( VOUT_OUTMEM_SLEEP );
511 /* 20ms is a bit arbitrary, but it's only for the first image we get */
512 if( !p_vout->p_sys->last_date )
514 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
515 p_pic->date + 20000 );
519 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
520 (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
522 p_vout->p_sys->last_date = p_pic->date;
525 switch( p_vout->p_sys->i_mode )
527 case DEINTERLACE_DISCARD:
528 RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
529 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
532 case DEINTERLACE_BOB:
533 RenderBob( p_vout, pp_outpic[0], p_pic, p_pic->b_top_field_first ? 0 : 1 );
534 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
535 RenderBob( p_vout, pp_outpic[1], p_pic, p_pic->b_top_field_first ? 1 : 0 );
536 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
539 case DEINTERLACE_LINEAR:
540 RenderLinear( p_vout, pp_outpic[0], p_pic, p_pic->b_top_field_first ? 0 : 1 );
541 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
542 RenderLinear( p_vout, pp_outpic[1], p_pic, p_pic->b_top_field_first ? 1 : 0 );
543 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
546 case DEINTERLACE_MEAN:
547 RenderMean( p_vout, pp_outpic[0], p_pic );
548 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
551 case DEINTERLACE_BLEND:
552 RenderBlend( p_vout, pp_outpic[0], p_pic );
553 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
557 RenderX( p_vout, pp_outpic[0], p_pic );
558 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
561 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
564 /*****************************************************************************
565 * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
566 *****************************************************************************/
567 static void RenderDiscard( vout_thread_t *p_vout,
568 picture_t *p_outpic, picture_t *p_pic, int i_field )
572 /* Copy image and skip lines */
573 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
575 uint8_t *p_in, *p_out_end, *p_out;
578 p_in = p_pic->p[i_plane].p_pixels
579 + i_field * p_pic->p[i_plane].i_pitch;
581 p_out = p_outpic->p[i_plane].p_pixels;
582 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
583 * p_outpic->p[i_plane].i_visible_lines;
585 switch( p_vout->render.i_chroma )
587 case VLC_FOURCC('I','4','2','0'):
588 case VLC_FOURCC('I','Y','U','V'):
589 case VLC_FOURCC('Y','V','1','2'):
591 for( ; p_out < p_out_end ; )
593 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
594 p_pic->p[i_plane].i_pitch );
596 p_out += p_outpic->p[i_plane].i_pitch;
597 p_in += 2 * p_pic->p[i_plane].i_pitch;
601 case VLC_FOURCC('I','4','2','2'):
603 i_increment = 2 * p_pic->p[i_plane].i_pitch;
605 if( i_plane == Y_PLANE )
607 for( ; p_out < p_out_end ; )
609 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
610 p_pic->p[i_plane].i_pitch );
611 p_out += p_outpic->p[i_plane].i_pitch;
612 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
613 p_pic->p[i_plane].i_pitch );
614 p_out += p_outpic->p[i_plane].i_pitch;
620 for( ; p_out < p_out_end ; )
622 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
623 p_pic->p[i_plane].i_pitch );
624 p_out += p_outpic->p[i_plane].i_pitch;
636 /*****************************************************************************
637 * RenderBob: renders a BOB picture - simple copy
638 *****************************************************************************/
639 static void RenderBob( vout_thread_t *p_vout,
640 picture_t *p_outpic, picture_t *p_pic, int i_field )
644 /* Copy image and skip lines */
645 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
647 uint8_t *p_in, *p_out_end, *p_out;
649 p_in = p_pic->p[i_plane].p_pixels;
650 p_out = p_outpic->p[i_plane].p_pixels;
651 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
652 * p_outpic->p[i_plane].i_visible_lines;
654 switch( p_vout->render.i_chroma )
656 case VLC_FOURCC('I','4','2','0'):
657 case VLC_FOURCC('I','Y','U','V'):
658 case VLC_FOURCC('Y','V','1','2'):
659 /* For BOTTOM field we need to add the first line */
662 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
663 p_pic->p[i_plane].i_pitch );
664 p_in += p_pic->p[i_plane].i_pitch;
665 p_out += p_outpic->p[i_plane].i_pitch;
668 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
670 for( ; p_out < p_out_end ; )
672 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
673 p_pic->p[i_plane].i_pitch );
675 p_out += p_outpic->p[i_plane].i_pitch;
677 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
678 p_pic->p[i_plane].i_pitch );
680 p_in += 2 * p_pic->p[i_plane].i_pitch;
681 p_out += p_outpic->p[i_plane].i_pitch;
684 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
685 p_pic->p[i_plane].i_pitch );
687 /* For TOP field we need to add the last line */
690 p_in += p_pic->p[i_plane].i_pitch;
691 p_out += p_outpic->p[i_plane].i_pitch;
692 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
693 p_pic->p[i_plane].i_pitch );
697 case VLC_FOURCC('I','4','2','2'):
698 /* For BOTTOM field we need to add the first line */
701 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
702 p_pic->p[i_plane].i_pitch );
703 p_in += p_pic->p[i_plane].i_pitch;
704 p_out += p_outpic->p[i_plane].i_pitch;
707 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
709 if( i_plane == Y_PLANE )
711 for( ; p_out < p_out_end ; )
713 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
714 p_pic->p[i_plane].i_pitch );
716 p_out += p_outpic->p[i_plane].i_pitch;
718 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
719 p_pic->p[i_plane].i_pitch );
721 p_in += 2 * p_pic->p[i_plane].i_pitch;
722 p_out += p_outpic->p[i_plane].i_pitch;
727 for( ; p_out < p_out_end ; )
729 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
730 p_pic->p[i_plane].i_pitch );
732 p_out += p_outpic->p[i_plane].i_pitch;
733 p_in += 2 * p_pic->p[i_plane].i_pitch;
737 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
738 p_pic->p[i_plane].i_pitch );
740 /* For TOP field we need to add the last line */
743 p_in += p_pic->p[i_plane].i_pitch;
744 p_out += p_outpic->p[i_plane].i_pitch;
745 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
746 p_pic->p[i_plane].i_pitch );
753 #define Merge p_vout->p_sys->pf_merge
754 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
756 /*****************************************************************************
757 * RenderLinear: BOB with linear interpolation
758 *****************************************************************************/
759 static void RenderLinear( vout_thread_t *p_vout,
760 picture_t *p_outpic, picture_t *p_pic, int i_field )
764 /* Copy image and skip lines */
765 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
767 uint8_t *p_in, *p_out_end, *p_out;
769 p_in = p_pic->p[i_plane].p_pixels;
770 p_out = p_outpic->p[i_plane].p_pixels;
771 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
772 * p_outpic->p[i_plane].i_visible_lines;
774 /* For BOTTOM field we need to add the first line */
777 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
778 p_pic->p[i_plane].i_pitch );
779 p_in += p_pic->p[i_plane].i_pitch;
780 p_out += p_outpic->p[i_plane].i_pitch;
783 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
785 for( ; p_out < p_out_end ; )
787 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
788 p_pic->p[i_plane].i_pitch );
790 p_out += p_outpic->p[i_plane].i_pitch;
792 Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
793 p_pic->p[i_plane].i_pitch );
795 p_in += 2 * p_pic->p[i_plane].i_pitch;
796 p_out += p_outpic->p[i_plane].i_pitch;
799 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
800 p_pic->p[i_plane].i_pitch );
802 /* For TOP field we need to add the last line */
805 p_in += p_pic->p[i_plane].i_pitch;
806 p_out += p_outpic->p[i_plane].i_pitch;
807 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
808 p_pic->p[i_plane].i_pitch );
814 static void RenderMean( vout_thread_t *p_vout,
815 picture_t *p_outpic, picture_t *p_pic )
819 /* Copy image and skip lines */
820 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
822 uint8_t *p_in, *p_out_end, *p_out;
824 p_in = p_pic->p[i_plane].p_pixels;
826 p_out = p_outpic->p[i_plane].p_pixels;
827 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
828 * p_outpic->p[i_plane].i_visible_lines;
830 /* All lines: mean value */
831 for( ; p_out < p_out_end ; )
833 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
834 p_pic->p[i_plane].i_pitch );
836 p_out += p_outpic->p[i_plane].i_pitch;
837 p_in += 2 * p_pic->p[i_plane].i_pitch;
843 static void RenderBlend( vout_thread_t *p_vout,
844 picture_t *p_outpic, picture_t *p_pic )
848 /* Copy image and skip lines */
849 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
851 uint8_t *p_in, *p_out_end, *p_out;
853 p_in = p_pic->p[i_plane].p_pixels;
855 p_out = p_outpic->p[i_plane].p_pixels;
856 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
857 * p_outpic->p[i_plane].i_visible_lines;
859 switch( p_vout->render.i_chroma )
861 case VLC_FOURCC('I','4','2','0'):
862 case VLC_FOURCC('I','Y','U','V'):
863 case VLC_FOURCC('Y','V','1','2'):
864 /* First line: simple copy */
865 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
866 p_pic->p[i_plane].i_pitch );
867 p_out += p_outpic->p[i_plane].i_pitch;
869 /* Remaining lines: mean value */
870 for( ; p_out < p_out_end ; )
872 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
873 p_pic->p[i_plane].i_pitch );
875 p_out += p_outpic->p[i_plane].i_pitch;
876 p_in += p_pic->p[i_plane].i_pitch;
880 case VLC_FOURCC('I','4','2','2'):
881 /* First line: simple copy */
882 p_vout->p_libvlc->pf_memcpy( p_out, p_in,
883 p_pic->p[i_plane].i_pitch );
884 p_out += p_outpic->p[i_plane].i_pitch;
886 /* Remaining lines: mean value */
887 if( i_plane == Y_PLANE )
889 for( ; p_out < p_out_end ; )
891 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
892 p_pic->p[i_plane].i_pitch );
894 p_out += p_outpic->p[i_plane].i_pitch;
895 p_in += p_pic->p[i_plane].i_pitch;
901 for( ; p_out < p_out_end ; )
903 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
904 p_pic->p[i_plane].i_pitch );
906 p_out += p_outpic->p[i_plane].i_pitch;
907 p_in += 2*p_pic->p[i_plane].i_pitch;
918 static void MergeGeneric( void *_p_dest, const void *_p_s1,
919 const void *_p_s2, size_t i_bytes )
921 uint8_t* p_dest = (uint8_t*)_p_dest;
922 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
923 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
924 uint8_t* p_end = p_dest + i_bytes - 8;
926 while( p_dest < p_end )
928 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
929 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
930 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
931 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
932 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
933 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
934 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
935 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
940 while( p_dest < p_end )
942 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
946 #if defined(CAN_COMPILE_MMXEXT)
947 static void MergeMMXEXT( void *_p_dest, const void *_p_s1, const void *_p_s2,
950 uint8_t* p_dest = (uint8_t*)_p_dest;
951 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
952 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
953 uint8_t* p_end = p_dest + i_bytes - 8;
954 while( p_dest < p_end )
956 __asm__ __volatile__( "movq %2,%%mm1;"
958 "movq %%mm1, %0" :"=m" (*p_dest):
968 while( p_dest < p_end )
970 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
975 #if defined(CAN_COMPILE_3DNOW)
976 static void Merge3DNow( void *_p_dest, const void *_p_s1, const void *_p_s2,
979 uint8_t* p_dest = (uint8_t*)_p_dest;
980 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
981 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
982 uint8_t* p_end = p_dest + i_bytes - 8;
983 while( p_dest < p_end )
985 __asm__ __volatile__( "movq %2,%%mm1;"
987 "movq %%mm1, %0" :"=m" (*p_dest):
997 while( p_dest < p_end )
999 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1004 #if defined(CAN_COMPILE_SSE)
1005 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
1008 uint8_t* p_dest = (uint8_t*)_p_dest;
1009 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
1010 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
1012 while( (uintptr_t)p_s1 % 16 )
1014 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1016 p_end = p_dest + i_bytes - 16;
1017 while( p_dest < p_end )
1019 __asm__ __volatile__( "movdqu %2,%%xmm1;"
1021 "movdqu %%xmm1, %0" :"=m" (*p_dest):
1031 while( p_dest < p_end )
1033 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1038 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
1039 static void EndMMX( void )
1041 __asm__ __volatile__( "emms" :: );
1045 #if defined(CAN_COMPILE_3DNOW)
1046 static void End3DNow( void )
1048 __asm__ __volatile__( "femms" :: );
1052 #ifdef CAN_COMPILE_C_ALTIVEC
1053 static void MergeAltivec( void *_p_dest, const void *_p_s1,
1054 const void *_p_s2, size_t i_bytes )
1056 uint8_t *p_dest = (uint8_t *)_p_dest;
1057 uint8_t *p_s1 = (uint8_t *)_p_s1;
1058 uint8_t *p_s2 = (uint8_t *)_p_s2;
1059 uint8_t *p_end = p_dest + i_bytes - 15;
1061 /* Use C until the first 16-bytes aligned destination pixel */
1062 while( (int)p_dest & 0xF )
1064 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1067 if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1069 /* Unaligned source */
1070 vector unsigned char s1v, s2v, destv;
1071 vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1072 vector unsigned char perm1v, perm2v;
1074 perm1v = vec_lvsl( 0, p_s1 );
1075 perm2v = vec_lvsl( 0, p_s2 );
1076 s1oldv = vec_ld( 0, p_s1 );
1077 s2oldv = vec_ld( 0, p_s2 );
1079 while( p_dest < p_end )
1081 s1newv = vec_ld( 16, p_s1 );
1082 s2newv = vec_ld( 16, p_s2 );
1083 s1v = vec_perm( s1oldv, s1newv, perm1v );
1084 s2v = vec_perm( s2oldv, s2newv, perm2v );
1087 destv = vec_avg( s1v, s2v );
1088 vec_st( destv, 0, p_dest );
1097 /* Aligned source */
1098 vector unsigned char s1v, s2v, destv;
1100 while( p_dest < p_end )
1102 s1v = vec_ld( 0, p_s1 );
1103 s2v = vec_ld( 0, p_s2 );
1104 destv = vec_avg( s1v, s2v );
1105 vec_st( destv, 0, p_dest );
1115 while( p_dest < p_end )
1117 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1122 /*****************************************************************************
1123 * RenderX: This algo works on a 8x8 block basic, it copies the top field
1124 * and apply a process to recreate the bottom field :
1125 * If a 8x8 block is classified as :
1126 * - progressive: it applies a small blend (1,6,1)
1128 * * in the MMX version: we do a ME between the 2 fields, if there is a
1129 * good match we use MC to recreate the bottom field (with a small
1131 * * otherwise: it recreates the bottom field by an edge oriented
1133 *****************************************************************************/
1135 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1136 * XXX: It need to access to 8x10
1137 * We use more than 8 lines to help with scrolling (text)
1138 * (and because XDeint8x8Frame use line 9)
1139 * XXX: smooth/uniform area with noise detection doesn't works well
1140 * but it's not really a problem because they don't have much details anyway
1142 static inline int ssd( int a ) { return a*a; }
1143 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1149 /* Detect interlacing */
1151 for( y = 0; y < 7; y += 2 )
1154 for( x = 0; x < 8; x++ )
1156 fr += ssd(src[ x] - src[1*i_src+x]) +
1157 ssd(src[i_src+x] - src[2*i_src+x]);
1158 ff += ssd(src[ x] - src[2*i_src+x]) +
1159 ssd(src[i_src+x] - src[3*i_src+x]);
1161 if( ff < 6*fr/8 && fr > 32 )
1167 return fc < 1 ? VLC_FALSE : VLC_TRUE;
1169 #ifdef CAN_COMPILE_MMXEXT
1170 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1177 /* Detect interlacing */
1179 pxor_r2r( mm7, mm7 );
1180 for( y = 0; y < 9; y += 2 )
1183 pxor_r2r( mm5, mm5 );
1184 pxor_r2r( mm6, mm6 );
1185 for( x = 0; x < 8; x+=4 )
1187 movd_m2r( src[ x], mm0 );
1188 movd_m2r( src[1*i_src+x], mm1 );
1189 movd_m2r( src[2*i_src+x], mm2 );
1190 movd_m2r( src[3*i_src+x], mm3 );
1192 punpcklbw_r2r( mm7, mm0 );
1193 punpcklbw_r2r( mm7, mm1 );
1194 punpcklbw_r2r( mm7, mm2 );
1195 punpcklbw_r2r( mm7, mm3 );
1197 movq_r2r( mm0, mm4 );
1199 psubw_r2r( mm1, mm0 );
1200 psubw_r2r( mm2, mm4 );
1202 psubw_r2r( mm1, mm2 );
1203 psubw_r2r( mm1, mm3 );
1205 pmaddwd_r2r( mm0, mm0 );
1206 pmaddwd_r2r( mm4, mm4 );
1207 pmaddwd_r2r( mm2, mm2 );
1208 pmaddwd_r2r( mm3, mm3 );
1209 paddd_r2r( mm0, mm2 );
1210 paddd_r2r( mm4, mm3 );
1211 paddd_r2r( mm2, mm5 );
1212 paddd_r2r( mm3, mm6 );
1215 movq_r2r( mm5, mm0 );
1216 psrlq_i2r( 32, mm0 );
1217 paddd_r2r( mm0, mm5 );
1218 movd_r2m( mm5, fr );
1220 movq_r2r( mm6, mm0 );
1221 psrlq_i2r( 32, mm0 );
1222 paddd_r2r( mm0, mm6 );
1223 movd_r2m( mm6, ff );
1225 if( ff < 6*fr/8 && fr > 32 )
1234 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1235 * This won't destroy details, and help if there is a bit of interlacing.
1236 * (It helps with paning to avoid flickers)
1240 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1241 uint8_t *src, int i_src )
1246 for( y = 0; y < 8; y += 2 )
1248 memcpy( dst, src, 8 );
1251 for( x = 0; x < 8; x++ )
1252 dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1258 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1259 uint8_t *src1, int i_src1,
1260 uint8_t *src2, int i_src2 )
1265 for( y = 0; y < 8; y += 2 )
1267 memcpy( dst, src1, 8 );
1270 for( x = 0; x < 8; x++ )
1271 dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1279 #ifdef CAN_COMPILE_MMXEXT
1280 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1281 uint8_t *src1, int i_src1,
1282 uint8_t *src2, int i_src2 )
1284 static const uint64_t m_4 = I64C(0x0004000400040004);
1288 pxor_r2r( mm7, mm7 );
1289 for( y = 0; y < 8; y += 2 )
1291 for( x = 0; x < 8; x +=4 )
1293 movd_m2r( src1[x], mm0 );
1294 movd_r2m( mm0, dst[x] );
1296 movd_m2r( src2[x], mm1 );
1297 movd_m2r( src1[i_src1+x], mm2 );
1299 punpcklbw_r2r( mm7, mm0 );
1300 punpcklbw_r2r( mm7, mm1 );
1301 punpcklbw_r2r( mm7, mm2 );
1302 paddw_r2r( mm1, mm1 );
1303 movq_r2r( mm1, mm3 );
1304 paddw_r2r( mm3, mm3 );
1305 paddw_r2r( mm2, mm0 );
1306 paddw_r2r( mm3, mm1 );
1307 paddw_m2r( m_4, mm1 );
1308 paddw_r2r( mm1, mm0 );
1309 psraw_i2r( 3, mm0 );
1310 packuswb_r2r( mm7, mm0 );
1311 movd_r2m( mm0, dst[i_dst+x] );
1322 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1325 for( y = 0; y < 8; y++ )
1326 memset( &dst[y*i_dst], v, 8 );
1329 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1332 * TODO: a better one for the inner part.
1334 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1335 uint8_t *src, int i_src )
1340 for( y = 0; y < 8; y += 2 )
1342 memcpy( dst, src, 8 );
1345 for( x = 0; x < 8; x++ )
1346 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1351 #ifdef CAN_COMPILE_MMXEXT
1352 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1353 uint8_t *src, int i_src )
1358 for( y = 0; y < 8; y += 2 )
1360 movq_m2r( src[0], mm0 );
1361 movq_r2m( mm0, dst[0] );
1364 movq_m2r( src[2*i_src], mm1 );
1365 pavgb_r2r( mm1, mm0 );
1367 movq_r2m( mm0, dst[0] );
1375 /* XDeint8x8Field: Edge oriented interpolation
1376 * (Need -4 and +5 pixels H, +1 line)
1378 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1379 uint8_t *src, int i_src )
1384 for( y = 0; y < 8; y += 2 )
1386 memcpy( dst, src, 8 );
1389 for( x = 0; x < 8; x++ )
1391 uint8_t *src2 = &src[2*i_src];
1392 /* I use 8 pixels just to match the MMX version, but it's overkill
1393 * 5 would be enough (less isn't good) */
1394 const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1395 abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1396 abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1397 abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1399 const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1400 abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1401 abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1402 abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1404 const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1405 abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1406 abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1407 abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1409 if( c0 < c1 && c1 <= c2 )
1410 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1411 else if( c2 < c1 && c1 <= c0 )
1412 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1414 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1421 #ifdef CAN_COMPILE_MMXEXT
1422 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1423 uint8_t *src, int i_src )
1428 for( y = 0; y < 8; y += 2 )
1430 memcpy( dst, src, 8 );
1433 for( x = 0; x < 8; x++ )
1435 uint8_t *src2 = &src[2*i_src];
1438 movq_m2r( src[x-2], mm0 );
1439 movq_m2r( src[x-3], mm1 );
1440 movq_m2r( src[x-4], mm2 );
1442 psadbw_m2r( src2[x-4], mm0 );
1443 psadbw_m2r( src2[x-3], mm1 );
1444 psadbw_m2r( src2[x-2], mm2 );
1446 movd_r2m( mm0, c2 );
1447 movd_r2m( mm1, c1 );
1448 movd_r2m( mm2, c0 );
1450 if( c0 < c1 && c1 <= c2 )
1451 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1452 else if( c2 < c1 && c1 <= c0 )
1453 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1455 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1465 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1466 uint8_t *pix2, int i_pix2 )
1471 for( y = 0; y < 8; y++ )
1472 for( x = 0; x < 8; x++ )
1473 s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1477 #ifdef CAN_COMPILE_MMXEXT
1478 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1479 uint8_t *pix2, int i_pix2 )
1484 pxor_r2r( mm7, mm7 );
1485 pxor_r2r( mm6, mm6 );
1487 for( y = 0; y < 8; y++ )
1489 movq_m2r( pix1[0], mm0 );
1490 movq_m2r( pix2[0], mm1 );
1492 movq_r2r( mm0, mm2 );
1493 movq_r2r( mm1, mm3 );
1495 punpcklbw_r2r( mm7, mm0 );
1496 punpckhbw_r2r( mm7, mm2 );
1497 punpcklbw_r2r( mm7, mm1 );
1498 punpckhbw_r2r( mm7, mm3 );
1500 psubw_r2r( mm1, mm0 );
1501 psubw_r2r( mm3, mm2 );
1503 pmaddwd_r2r( mm0, mm0 );
1504 pmaddwd_r2r( mm2, mm2 );
1506 paddd_r2r( mm2, mm0 );
1507 paddd_r2r( mm0, mm6 );
1513 movq_r2r( mm6, mm7 );
1514 psrlq_i2r( 32, mm7 );
1515 paddd_r2r( mm6, mm7 );
1524 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1525 #ifdef CAN_COMPILE_MMXEXT
1528 * TODO: mmx version (easier in sse2)
1530 static inline void XDeintMC( uint8_t *dst, int i_dst,
1531 uint8_t *src, int i_src,
1533 int i_width, int i_height )
1535 const int d4x = mvx&0x03;
1536 const int d4y = mvy&0x03;
1538 const int cA = (4-d4x)*(4-d4y);
1539 const int cB = d4x *(4-d4y);
1540 const int cC = (4-d4x)*d4y;
1541 const int cD = d4x *d4y;
1547 src += (mvy >> 2) * i_src + (mvx >> 2);
1550 for( y = 0; y < i_height; y++ )
1552 for( x = 0; x < i_width; x++ )
1554 dst[x] = ( cA*src[x] + cB*src[x+1] +
1555 cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1563 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1564 uint8_t *pix2, int i_pix2 )
1568 movq_m2r( pix1[0*i_pix1], mm0 );
1569 movq_m2r( pix1[1*i_pix1], mm1 );
1571 psadbw_m2r( pix2[0*i_pix2], mm0 );
1572 psadbw_m2r( pix2[1*i_pix2], mm1 );
1574 movq_m2r( pix1[2*i_pix1], mm2 );
1575 movq_m2r( pix1[3*i_pix1], mm3 );
1576 psadbw_m2r( pix2[2*i_pix2], mm2 );
1577 psadbw_m2r( pix2[3*i_pix2], mm3 );
1579 paddd_r2r( mm1, mm0 );
1580 paddd_r2r( mm3, mm2 );
1581 paddd_r2r( mm2, mm0 );
1587 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1588 uint8_t *ref, int i_stride,
1590 int xmax, int ymax )
1592 uint8_t buffer[8*4];
1594 if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1597 XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1598 return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1600 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1601 uint8_t *ref, int i_stride,
1603 int xmax, int ymax )
1605 if( abs(mx) >= xmax || abs(my) >= ymax )
1608 return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1611 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1612 uint8_t *src, int i_src,
1614 int xmax, int ymax )
1616 static const int dx[8] = { 0, 0, -1, 1, -1, -1, 1, 1 };
1617 static const int dy[8] = {-1, 1, 0, 0, -1, 1, -1, 1 };
1618 uint8_t *next = &src[i_src];
1619 const int i_src2 = 2*i_src;
1624 uint8_t *rec = &dst[i_dst];
1626 /* We construct with intra method the missing field */
1627 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1629 /* Now we will try to find a match with ME with the other field */
1631 /* ME: A small/partial EPZS
1632 * We search only for small MV (with high motion intra will be perfect */
1633 if( xmax > 4 ) xmax = 4;
1634 if( ymax > 4 ) ymax = 4;
1636 /* Init with NULL Mv */
1638 mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1640 /* Try predicted Mv */
1641 if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1647 /* Search interger pel (small mv) */
1648 for( i_step = 0; i_step < 4; i_step++ )
1654 for( i = 0; i < 4; i++ )
1656 s = XDeint8x4TestInt( rec, i_src2,
1657 next, i_src2, mvx+dx[i], mvy+dy[i],
1677 if( mvs > 4 && mvs < 256 )
1680 /* XXX: for now only HPEL (too slow) */
1681 for( i_step = 0; i_step < 4; i_step++ )
1687 for( i = 0; i < 8; i++ )
1689 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1690 mvx+dx[i], mvy+dy[i],
1708 uint8_t buffer[8*4];
1709 XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1710 XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1712 //XDeint8x8Set( dst, i_dst, 0 );
1719 /* Kernel interpolation (1,-5,20,20,-5,1)
1720 * Lose a bit more details+add aliasing than edge interpol but avoid
1723 static inline uint8_t clip1( int a )
1732 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1733 uint8_t *src, int i_src )
1738 for( y = 0; y < 8; y += 2 )
1740 const int i_src2 = i_src*2;
1742 memcpy( dst, src, 8 );
1745 for( x = 0; x < 8; x++ )
1749 pix = 1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1750 -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1751 +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1753 dst[x] = clip1( ( pix + 16 ) >> 5 );
1763 /* NxN arbitray size (and then only use pixel in the NxN block)
1765 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1766 int i_height, int i_width )
1773 /* Detect interlacing */
1774 /* FIXME way too simple, need to be more like XDeint8x8Detect */
1777 for( y = 0; y < i_height - 2; y += 2 )
1779 const uint8_t *s = &src[y*i_src];
1780 for( x = 0; x < i_width; x++ )
1782 fr += ssd(s[ x] - s[1*i_src+x]);
1783 ff += ssd(s[ x] - s[2*i_src+x]);
1785 if( ff < fr && fr > i_width / 2 )
1789 return fc < 2 ? VLC_FALSE : VLC_TRUE;
1792 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1793 uint8_t *src, int i_src,
1794 int i_width, int i_height )
1799 for( y = 0; y < i_height; y += 2 )
1801 memcpy( dst, src, i_width );
1804 if( y < i_height - 2 )
1806 for( x = 0; x < i_width; x++ )
1807 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1811 /* Blend last line */
1812 for( x = 0; x < i_width; x++ )
1813 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1820 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1821 uint8_t *src, int i_src,
1822 int i_width, int i_height )
1827 for( y = 0; y < i_height; y += 2 )
1829 memcpy( dst, src, i_width );
1832 if( y < i_height - 2 )
1834 for( x = 0; x < i_width; x++ )
1835 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1839 /* Blend last line */
1840 for( x = 0; x < i_width; x++ )
1841 dst[x] = (src[x] + src[i_src+x]) >> 1;
1848 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1849 int i_width, int i_height )
1851 if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1852 XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1854 XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1858 static inline int median( int a, int b, int c )
1860 int min = a, max =a;
1871 return a + b + c - min - max;
1877 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1878 uint8_t *src, int i_src,
1879 const int i_mbx, int i_modx )
1883 for( x = 0; x < i_mbx; x++ )
1886 if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1888 if( x == 0 || x == i_mbx - 1 )
1889 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1891 XDeint8x8FieldC( dst, i_dst, src, i_src );
1895 XDeint8x8MergeC( dst, i_dst,
1896 &src[0*i_src], 2*i_src,
1897 &src[1*i_src], 2*i_src );
1905 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1907 #ifdef CAN_COMPILE_MMXEXT
1908 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1909 uint8_t *src, int i_src,
1910 const int i_mbx, int i_modx )
1914 /* Reset current line */
1915 for( x = 0; x < i_mbx; x++ )
1918 if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1920 if( x == 0 || x == i_mbx - 1 )
1921 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1923 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1927 XDeint8x8MergeMMXEXT( dst, i_dst,
1928 &src[0*i_src], 2*i_src,
1929 &src[1*i_src], 2*i_src );
1937 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1941 static void RenderX( vout_thread_t *p_vout,
1942 picture_t *p_outpic, picture_t *p_pic )
1946 /* Copy image and skip lines */
1947 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1949 const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1950 const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1952 const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1953 const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1955 const int i_dst = p_outpic->p[i_plane].i_pitch;
1956 const int i_src = p_pic->p[i_plane].i_pitch;
1960 for( y = 0; y < i_mby; y++ )
1962 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1963 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1965 #ifdef CAN_COMPILE_MMXEXT
1966 if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
1967 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1970 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1973 /* Last line (C only)*/
1976 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1977 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1979 for( x = 0; x < i_mbx; x++ )
1981 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1988 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1992 #ifdef CAN_COMPILE_MMXEXT
1993 if( vlc_CPU() & CPU_CAPABILITY_MMXEXT )
1998 /*****************************************************************************
1999 * SendEvents: forward mouse and keyboard events to the parent p_vout
2000 *****************************************************************************/
2001 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
2002 vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
2004 vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
2005 vlc_value_t sentval = newval;
2007 if( !strcmp( psz_var, "mouse-y" ) )
2009 switch( p_vout->p_sys->i_mode )
2011 case DEINTERLACE_MEAN:
2012 case DEINTERLACE_DISCARD:
2018 var_Set( p_vout, psz_var, sentval );
2023 /*****************************************************************************
2024 * FilterCallback: called when changing the deinterlace method on the fly.
2025 *****************************************************************************/
2026 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
2027 vlc_value_t oldval, vlc_value_t newval,
2030 vout_thread_t * p_vout = (vout_thread_t *)p_this;
2031 int i_old_mode = p_vout->p_sys->i_mode;
2033 msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
2035 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
2037 SetFilterMethod( p_vout, newval.psz_string );
2039 switch( p_vout->render.i_chroma )
2041 case VLC_FOURCC('I','4','2','2'):
2042 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2046 case VLC_FOURCC('I','4','2','0'):
2047 case VLC_FOURCC('I','Y','U','V'):
2048 case VLC_FOURCC('Y','V','1','2'):
2049 switch( p_vout->p_sys->i_mode )
2051 case DEINTERLACE_MEAN:
2052 case DEINTERLACE_DISCARD:
2053 if( ( i_old_mode == DEINTERLACE_MEAN )
2054 || ( i_old_mode == DEINTERLACE_DISCARD ) )
2056 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2061 case DEINTERLACE_BOB:
2062 case DEINTERLACE_BLEND:
2063 case DEINTERLACE_LINEAR:
2064 if( ( i_old_mode == DEINTERLACE_BOB )
2065 || ( i_old_mode == DEINTERLACE_BLEND )
2066 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2068 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2079 /* We need to kill the old vout */
2081 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2083 vlc_object_detach( p_vout->p_sys->p_vout );
2084 vout_Destroy( p_vout->p_sys->p_vout );
2086 /* Try to open a new video output */
2087 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2089 if( p_vout->p_sys->p_vout == NULL )
2091 /* Everything failed */
2092 msg_Err( p_vout, "cannot open vout, aborting" );
2094 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2095 return VLC_EGENERIC;
2098 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2100 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2104 /*****************************************************************************
2105 * SendEventsToChild: forward events to the child/children vout
2106 *****************************************************************************/
2107 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2108 vlc_value_t oldval, vlc_value_t newval, void *p_data )
2110 vout_thread_t *p_vout = (vout_thread_t *)p_this;
2111 var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2116 /*****************************************************************************
2117 * video filter2 functions
2118 *****************************************************************************/
2119 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2121 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2122 picture_t *p_pic_dst;
2124 /* Request output picture */
2125 p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2126 if( p_pic_dst == NULL )
2128 msg_Warn( p_filter, "can't get output picture" );
2132 switch( p_vout->p_sys->i_mode )
2134 case DEINTERLACE_DISCARD:
2136 RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2138 msg_Err( p_vout, "discarding lines is not supported yet" );
2139 p_pic_dst->pf_release( p_pic_dst );
2143 case DEINTERLACE_BOB:
2145 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2146 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2150 case DEINTERLACE_LINEAR:
2152 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2153 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2155 msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2156 p_pic_dst->pf_release( p_pic_dst );
2160 case DEINTERLACE_MEAN:
2161 RenderMean( p_vout, p_pic_dst, p_pic );
2164 case DEINTERLACE_BLEND:
2165 RenderBlend( p_vout, p_pic_dst, p_pic );
2169 RenderX( p_vout, p_pic_dst, p_pic );
2173 p_pic_dst->date = p_pic->date;
2174 p_pic_dst->b_force = p_pic->b_force;
2175 p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2176 p_pic_dst->b_progressive = VLC_TRUE;
2177 p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2179 p_pic->pf_release( p_pic );
2183 /*****************************************************************************
2185 *****************************************************************************/
2186 static int OpenFilter( vlc_object_t *p_this )
2188 filter_t *p_filter = (filter_t*)p_this;
2189 vout_thread_t *p_vout;
2192 if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2193 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2194 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2195 p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2197 return VLC_EGENERIC;
2200 /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2202 p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2203 vlc_object_attach( p_vout, p_filter );
2204 p_filter->p_sys = (filter_sys_t *)p_vout;
2205 p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2207 config_ChainParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2209 var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2210 var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2211 var_Set( p_filter, "deinterlace-mode", val );
2213 if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2215 vlc_object_detach( p_vout );
2216 vlc_object_release( p_vout );
2217 return VLC_EGENERIC;
2220 p_filter->pf_video_filter = Deinterlace;
2222 msg_Dbg( p_filter, "deinterlacing" );
2227 /*****************************************************************************
2228 * CloseFilter: clean up the filter
2229 *****************************************************************************/
2230 static void CloseFilter( vlc_object_t *p_this )
2232 filter_t *p_filter = (filter_t*)p_this;
2233 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2235 Destroy( VLC_OBJECT(p_vout) );
2236 vlc_object_detach( p_vout );
2237 vlc_object_release( p_vout );