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;
325 return VLC_EGENERIC; /* unknown chroma */
329 /* Try to open the real video output */
330 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
332 if( p_vout->p_sys->p_vout == NULL )
334 /* Everything failed */
335 msg_Err( p_vout, "cannot open vout, aborting" );
340 var_AddCallback( p_vout, "deinterlace-mode", FilterCallback, NULL );
342 ALLOCATE_DIRECTBUFFERS( VOUT_MAX_PICTURES );
344 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
346 ADD_PARENT_CALLBACKS( SendEventsToChild );
351 /*****************************************************************************
352 * SpawnRealVout: spawn the real video output.
353 *****************************************************************************/
354 static vout_thread_t *SpawnRealVout( vout_thread_t *p_vout )
356 vout_thread_t *p_real_vout = NULL;
357 video_format_t fmt = {0};
359 msg_Dbg( p_vout, "spawning the real video output" );
361 fmt.i_width = fmt.i_visible_width = p_vout->output.i_width;
362 fmt.i_height = fmt.i_visible_height = p_vout->output.i_height;
363 fmt.i_x_offset = fmt.i_y_offset = 0;
364 fmt.i_chroma = p_vout->output.i_chroma;
365 fmt.i_aspect = p_vout->output.i_aspect;
366 fmt.i_sar_num = p_vout->output.i_aspect * fmt.i_height / fmt.i_width;
367 fmt.i_sar_den = VOUT_ASPECT_FACTOR;
369 switch( p_vout->render.i_chroma )
371 case VLC_FOURCC('I','4','2','0'):
372 case VLC_FOURCC('I','Y','U','V'):
373 case VLC_FOURCC('Y','V','1','2'):
374 switch( p_vout->p_sys->i_mode )
376 case DEINTERLACE_MEAN:
377 case DEINTERLACE_DISCARD:
378 fmt.i_height = fmt.i_visible_height = p_vout->output.i_height / 2;
379 p_real_vout = vout_Create( p_vout, &fmt );
382 case DEINTERLACE_BOB:
383 case DEINTERLACE_BLEND:
384 case DEINTERLACE_LINEAR:
386 p_real_vout = vout_Create( p_vout, &fmt );
391 case VLC_FOURCC('I','4','2','2'):
392 fmt.i_chroma = VLC_FOURCC('I','4','2','0');
393 p_real_vout = vout_Create( p_vout, &fmt );
403 /*****************************************************************************
404 * End: terminate Deinterlace video thread output method
405 *****************************************************************************/
406 static void End( vout_thread_t *p_vout )
410 /* Free the fake output buffers we allocated */
411 for( i_index = I_OUTPUTPICTURES ; i_index ; )
414 free( PP_OUTPUTPICTURE[ i_index ]->p_data_orig );
417 if( p_vout->p_sys->p_vout )
419 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
420 vlc_object_detach( p_vout->p_sys->p_vout );
421 vout_Destroy( p_vout->p_sys->p_vout );
424 DEL_PARENT_CALLBACKS( SendEventsToChild );
427 /*****************************************************************************
428 * Destroy: destroy Deinterlace video thread output method
429 *****************************************************************************
430 * Terminate an output method created by DeinterlaceCreateOutputMethod
431 *****************************************************************************/
432 static void Destroy( vlc_object_t *p_this )
434 vout_thread_t *p_vout = (vout_thread_t *)p_this;
435 vlc_mutex_destroy( &p_vout->p_sys->filter_lock );
436 free( p_vout->p_sys );
439 /*****************************************************************************
440 * Render: displays previously rendered output
441 *****************************************************************************
442 * This function send the currently rendered image to Deinterlace image,
443 * waits until it is displayed and switch the two rendering buffers, preparing
445 *****************************************************************************/
446 static void Render ( vout_thread_t *p_vout, picture_t *p_pic )
448 picture_t *pp_outpic[2];
450 pp_outpic[0] = pp_outpic[1] = NULL;
452 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
454 /* Get a new picture */
455 while( ( pp_outpic[0] = vout_CreatePicture( p_vout->p_sys->p_vout,
459 if( p_vout->b_die || p_vout->b_error )
461 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
464 msleep( VOUT_OUTMEM_SLEEP );
467 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[0], p_pic->date );
469 /* If we are using double rate, get an additional new picture */
470 if( p_vout->p_sys->b_double_rate )
472 while( ( pp_outpic[1] = vout_CreatePicture( p_vout->p_sys->p_vout,
476 if( p_vout->b_die || p_vout->b_error )
478 vout_DestroyPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
479 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
482 msleep( VOUT_OUTMEM_SLEEP );
485 /* 20ms is a bit arbitrary, but it's only for the first image we get */
486 if( !p_vout->p_sys->last_date )
488 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
489 p_pic->date + 20000 );
493 vout_DatePicture( p_vout->p_sys->p_vout, pp_outpic[1],
494 (3 * p_pic->date - p_vout->p_sys->last_date) / 2 );
496 p_vout->p_sys->last_date = p_pic->date;
499 switch( p_vout->p_sys->i_mode )
501 case DEINTERLACE_DISCARD:
502 RenderDiscard( p_vout, pp_outpic[0], p_pic, 0 );
503 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
506 case DEINTERLACE_BOB:
507 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
508 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
509 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
510 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
513 case DEINTERLACE_LINEAR:
514 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
515 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
516 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
517 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[1] );
520 case DEINTERLACE_MEAN:
521 RenderMean( p_vout, pp_outpic[0], p_pic );
522 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
525 case DEINTERLACE_BLEND:
526 RenderBlend( p_vout, pp_outpic[0], p_pic );
527 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
531 RenderX( p_vout, pp_outpic[0], p_pic );
532 vout_DisplayPicture( p_vout->p_sys->p_vout, pp_outpic[0] );
535 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
538 /*****************************************************************************
539 * RenderDiscard: only keep TOP or BOTTOM field, discard the other.
540 *****************************************************************************/
541 static void RenderDiscard( vout_thread_t *p_vout,
542 picture_t *p_outpic, picture_t *p_pic, int i_field )
546 /* Copy image and skip lines */
547 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
549 uint8_t *p_in, *p_out_end, *p_out;
552 p_in = p_pic->p[i_plane].p_pixels
553 + i_field * p_pic->p[i_plane].i_pitch;
555 p_out = p_outpic->p[i_plane].p_pixels;
556 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
557 * p_outpic->p[i_plane].i_visible_lines;
559 switch( p_vout->render.i_chroma )
561 case VLC_FOURCC('I','4','2','0'):
562 case VLC_FOURCC('I','Y','U','V'):
563 case VLC_FOURCC('Y','V','1','2'):
565 for( ; p_out < p_out_end ; )
567 p_vout->p_vlc->pf_memcpy( p_out, p_in,
568 p_pic->p[i_plane].i_pitch );
570 p_out += p_pic->p[i_plane].i_pitch;
571 p_in += 2 * p_pic->p[i_plane].i_pitch;
575 case VLC_FOURCC('I','4','2','2'):
577 i_increment = 2 * p_pic->p[i_plane].i_pitch;
579 if( i_plane == Y_PLANE )
581 for( ; p_out < p_out_end ; )
583 p_vout->p_vlc->pf_memcpy( p_out, p_in,
584 p_pic->p[i_plane].i_pitch );
585 p_out += p_pic->p[i_plane].i_pitch;
586 p_vout->p_vlc->pf_memcpy( p_out, p_in,
587 p_pic->p[i_plane].i_pitch );
588 p_out += p_pic->p[i_plane].i_pitch;
594 for( ; p_out < p_out_end ; )
596 p_vout->p_vlc->pf_memcpy( p_out, p_in,
597 p_pic->p[i_plane].i_pitch );
598 p_out += p_pic->p[i_plane].i_pitch;
610 /*****************************************************************************
611 * RenderBob: renders a BOB picture - simple copy
612 *****************************************************************************/
613 static void RenderBob( vout_thread_t *p_vout,
614 picture_t *p_outpic, picture_t *p_pic, int i_field )
618 /* Copy image and skip lines */
619 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
621 uint8_t *p_in, *p_out_end, *p_out;
623 p_in = p_pic->p[i_plane].p_pixels;
624 p_out = p_outpic->p[i_plane].p_pixels;
625 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
626 * p_outpic->p[i_plane].i_visible_lines;
628 switch( p_vout->render.i_chroma )
630 case VLC_FOURCC('I','4','2','0'):
631 case VLC_FOURCC('I','Y','U','V'):
632 case VLC_FOURCC('Y','V','1','2'):
633 /* For BOTTOM field we need to add the first line */
636 p_vout->p_vlc->pf_memcpy( p_out, p_in,
637 p_pic->p[i_plane].i_pitch );
638 p_in += p_pic->p[i_plane].i_pitch;
639 p_out += p_pic->p[i_plane].i_pitch;
642 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
644 for( ; p_out < p_out_end ; )
646 p_vout->p_vlc->pf_memcpy( p_out, p_in,
647 p_pic->p[i_plane].i_pitch );
649 p_out += p_pic->p[i_plane].i_pitch;
651 p_vout->p_vlc->pf_memcpy( p_out, p_in,
652 p_pic->p[i_plane].i_pitch );
654 p_in += 2 * p_pic->p[i_plane].i_pitch;
655 p_out += p_pic->p[i_plane].i_pitch;
658 p_vout->p_vlc->pf_memcpy( p_out, p_in,
659 p_pic->p[i_plane].i_pitch );
661 /* For TOP field we need to add the last line */
664 p_in += p_pic->p[i_plane].i_pitch;
665 p_out += p_pic->p[i_plane].i_pitch;
666 p_vout->p_vlc->pf_memcpy( p_out, p_in,
667 p_pic->p[i_plane].i_pitch );
671 case VLC_FOURCC('I','4','2','2'):
672 /* For BOTTOM field we need to add the first line */
675 p_vout->p_vlc->pf_memcpy( p_out, p_in,
676 p_pic->p[i_plane].i_pitch );
677 p_in += p_pic->p[i_plane].i_pitch;
678 p_out += p_pic->p[i_plane].i_pitch;
681 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
683 if( i_plane == Y_PLANE )
685 for( ; p_out < p_out_end ; )
687 p_vout->p_vlc->pf_memcpy( p_out, p_in,
688 p_pic->p[i_plane].i_pitch );
690 p_out += p_pic->p[i_plane].i_pitch;
692 p_vout->p_vlc->pf_memcpy( p_out, p_in,
693 p_pic->p[i_plane].i_pitch );
695 p_in += 2 * p_pic->p[i_plane].i_pitch;
696 p_out += p_pic->p[i_plane].i_pitch;
701 for( ; p_out < p_out_end ; )
703 p_vout->p_vlc->pf_memcpy( p_out, p_in,
704 p_pic->p[i_plane].i_pitch );
706 p_out += p_pic->p[i_plane].i_pitch;
707 p_in += 2 * p_pic->p[i_plane].i_pitch;
711 p_vout->p_vlc->pf_memcpy( p_out, p_in,
712 p_pic->p[i_plane].i_pitch );
714 /* For TOP field we need to add the last line */
717 p_in += p_pic->p[i_plane].i_pitch;
718 p_out += p_pic->p[i_plane].i_pitch;
719 p_vout->p_vlc->pf_memcpy( p_out, p_in,
720 p_pic->p[i_plane].i_pitch );
727 #define Merge p_vout->p_sys->pf_merge
728 #define EndMerge if(p_vout->p_sys->pf_end_merge) p_vout->p_sys->pf_end_merge
730 /*****************************************************************************
731 * RenderLinear: BOB with linear interpolation
732 *****************************************************************************/
733 static void RenderLinear( vout_thread_t *p_vout,
734 picture_t *p_outpic, picture_t *p_pic, int i_field )
738 /* Copy image and skip lines */
739 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
741 uint8_t *p_in, *p_out_end, *p_out;
743 p_in = p_pic->p[i_plane].p_pixels;
744 p_out = p_outpic->p[i_plane].p_pixels;
745 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
746 * p_outpic->p[i_plane].i_visible_lines;
748 /* For BOTTOM field we need to add the first line */
751 p_vout->p_vlc->pf_memcpy( p_out, p_in,
752 p_pic->p[i_plane].i_pitch );
753 p_in += p_pic->p[i_plane].i_pitch;
754 p_out += p_pic->p[i_plane].i_pitch;
757 p_out_end -= 2 * p_outpic->p[i_plane].i_pitch;
759 for( ; p_out < p_out_end ; )
761 p_vout->p_vlc->pf_memcpy( p_out, p_in,
762 p_pic->p[i_plane].i_pitch );
764 p_out += p_pic->p[i_plane].i_pitch;
766 Merge( p_out, p_in, p_in + 2 * p_pic->p[i_plane].i_pitch,
767 p_pic->p[i_plane].i_pitch );
769 p_in += 2 * p_pic->p[i_plane].i_pitch;
770 p_out += p_pic->p[i_plane].i_pitch;
773 p_vout->p_vlc->pf_memcpy( p_out, p_in,
774 p_pic->p[i_plane].i_pitch );
776 /* For TOP field we need to add the last line */
779 p_in += p_pic->p[i_plane].i_pitch;
780 p_out += p_pic->p[i_plane].i_pitch;
781 p_vout->p_vlc->pf_memcpy( p_out, p_in,
782 p_pic->p[i_plane].i_pitch );
788 static void RenderMean( vout_thread_t *p_vout,
789 picture_t *p_outpic, picture_t *p_pic )
793 /* Copy image and skip lines */
794 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
796 uint8_t *p_in, *p_out_end, *p_out;
798 p_in = p_pic->p[i_plane].p_pixels;
800 p_out = p_outpic->p[i_plane].p_pixels;
801 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
802 * p_outpic->p[i_plane].i_visible_lines;
804 /* All lines: mean value */
805 for( ; p_out < p_out_end ; )
807 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
808 p_pic->p[i_plane].i_pitch );
810 p_out += p_pic->p[i_plane].i_pitch;
811 p_in += 2 * p_pic->p[i_plane].i_pitch;
817 static void RenderBlend( vout_thread_t *p_vout,
818 picture_t *p_outpic, picture_t *p_pic )
822 /* Copy image and skip lines */
823 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
825 uint8_t *p_in, *p_out_end, *p_out;
827 p_in = p_pic->p[i_plane].p_pixels;
829 p_out = p_outpic->p[i_plane].p_pixels;
830 p_out_end = p_out + p_outpic->p[i_plane].i_pitch
831 * p_outpic->p[i_plane].i_visible_lines;
833 switch( p_vout->render.i_chroma )
835 case VLC_FOURCC('I','4','2','0'):
836 case VLC_FOURCC('I','Y','U','V'):
837 case VLC_FOURCC('Y','V','1','2'):
838 /* First line: simple copy */
839 p_vout->p_vlc->pf_memcpy( p_out, p_in,
840 p_pic->p[i_plane].i_pitch );
841 p_out += p_pic->p[i_plane].i_pitch;
843 /* Remaining lines: mean value */
844 for( ; p_out < p_out_end ; )
846 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
847 p_pic->p[i_plane].i_pitch );
849 p_out += p_pic->p[i_plane].i_pitch;
850 p_in += p_pic->p[i_plane].i_pitch;
854 case VLC_FOURCC('I','4','2','2'):
855 /* First line: simple copy */
856 p_vout->p_vlc->pf_memcpy( p_out, p_in,
857 p_pic->p[i_plane].i_pitch );
858 p_out += p_pic->p[i_plane].i_pitch;
860 /* Remaining lines: mean value */
861 if( i_plane == Y_PLANE )
863 for( ; p_out < p_out_end ; )
865 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
866 p_pic->p[i_plane].i_pitch );
868 p_out += p_pic->p[i_plane].i_pitch;
869 p_in += p_pic->p[i_plane].i_pitch;
875 for( ; p_out < p_out_end ; )
877 Merge( p_out, p_in, p_in + p_pic->p[i_plane].i_pitch,
878 p_pic->p[i_plane].i_pitch );
880 p_out += p_pic->p[i_plane].i_pitch;
881 p_in += 2*p_pic->p[i_plane].i_pitch;
892 static void MergeGeneric( void *_p_dest, const void *_p_s1,
893 const void *_p_s2, size_t i_bytes )
895 uint8_t* p_dest = (uint8_t*)_p_dest;
896 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
897 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
898 uint8_t* p_end = p_dest + i_bytes - 8;
900 while( p_dest < p_end )
902 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
903 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
904 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
905 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
906 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
907 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
908 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
909 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
914 while( p_dest < p_end )
916 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
920 #if defined(CAN_COMPILE_MMXEXT)
921 static void MergeMMX( void *_p_dest, const void *_p_s1, const void *_p_s2,
924 uint8_t* p_dest = (uint8_t*)_p_dest;
925 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
926 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
927 uint8_t* p_end = p_dest + i_bytes - 8;
928 while( p_dest < p_end )
930 __asm__ __volatile__( "movq %2,%%mm1;"
932 "movq %%mm1, %0" :"=m" (*p_dest):
942 while( p_dest < p_end )
944 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
949 #if defined(CAN_COMPILE_SSE)
950 static void MergeSSE2( void *_p_dest, const void *_p_s1, const void *_p_s2,
953 uint8_t* p_dest = (uint8_t*)_p_dest;
954 const uint8_t *p_s1 = (const uint8_t *)_p_s1;
955 const uint8_t *p_s2 = (const uint8_t *)_p_s2;
957 while( (ptrdiff_t)p_s1 % 16 )
959 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
961 p_end = p_dest + i_bytes - 16;
962 while( p_dest < p_end )
964 __asm__ __volatile__( "movdqu %2,%%xmm1;"
966 "movdqu %%xmm1, %0" :"=m" (*p_dest):
976 while( p_dest < p_end )
978 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
983 #if defined(CAN_COMPILE_MMXEXT) || defined(CAN_COMPILE_SSE)
984 static void EndMMX( void )
986 __asm__ __volatile__( "emms" :: );
990 #ifdef CAN_COMPILE_C_ALTIVEC
991 static void MergeAltivec( void *_p_dest, const void *_p_s1,
992 const void *_p_s2, size_t i_bytes )
994 uint8_t *p_dest = (uint8_t *)_p_dest;
995 uint8_t *p_s1 = (uint8_t *)_p_s1;
996 uint8_t *p_s2 = (uint8_t *)_p_s2;
997 uint8_t *p_end = p_dest + i_bytes - 15;
999 /* Use C until the first 16-bytes aligned destination pixel */
1000 while( (int)p_dest & 0xF )
1002 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1005 if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) )
1007 /* Unaligned source */
1008 vector unsigned char s1v, s2v, destv;
1009 vector unsigned char s1oldv, s2oldv, s1newv, s2newv;
1010 vector unsigned char perm1v, perm2v;
1012 perm1v = vec_lvsl( 0, p_s1 );
1013 perm2v = vec_lvsl( 0, p_s2 );
1014 s1oldv = vec_ld( 0, p_s1 );
1015 s2oldv = vec_ld( 0, p_s2 );
1017 while( p_dest < p_end )
1019 s1newv = vec_ld( 16, p_s1 );
1020 s2newv = vec_ld( 16, p_s2 );
1021 s1v = vec_perm( s1oldv, s1newv, perm1v );
1022 s2v = vec_perm( s2oldv, s2newv, perm2v );
1025 destv = vec_avg( s1v, s2v );
1026 vec_st( destv, 0, p_dest );
1035 /* Aligned source */
1036 vector unsigned char s1v, s2v, destv;
1038 while( p_dest < p_end )
1040 s1v = vec_ld( 0, p_s1 );
1041 s2v = vec_ld( 0, p_s2 );
1042 destv = vec_avg( s1v, s2v );
1043 vec_st( destv, 0, p_dest );
1053 while( p_dest < p_end )
1055 *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
1060 /*****************************************************************************
1061 * RenderX: This algo works on a 8x8 block basic, it copies the top field
1062 * and apply a process to recreate the bottom field :
1063 * If a 8x8 block is classified as :
1064 * - progressive: it applies a small blend (1,6,1)
1066 * * in the MMX version: we do a ME between the 2 fields, if there is a
1067 * good match we use MC to recreate the bottom field (with a small
1069 * * otherwise: it recreates the bottom field by an edge oriented
1071 *****************************************************************************/
1073 /* XDeint8x8Detect: detect if a 8x8 block is interlaced.
1074 * XXX: It need to access to 8x10
1075 * We use more than 8 lines to help with scrolling (text)
1076 * (and because XDeint8x8Frame use line 9)
1077 * XXX: smooth/uniform area with noise detection doesn't works well
1078 * but it's not really a problem because they don't have much details anyway
1080 static inline int ssd( int a ) { return a*a; }
1081 static inline int XDeint8x8DetectC( uint8_t *src, int i_src )
1087 /* Detect interlacing */
1089 for( y = 0; y < 7; y += 2 )
1092 for( x = 0; x < 8; x++ )
1094 fr += ssd(src[ x] - src[1*i_src+x]) +
1095 ssd(src[i_src+x] - src[2*i_src+x]);
1096 ff += ssd(src[ x] - src[2*i_src+x]) +
1097 ssd(src[i_src+x] - src[3*i_src+x]);
1099 if( ff < 6*fr/8 && fr > 32 )
1105 return fc < 1 ? VLC_FALSE : VLC_TRUE;
1107 #ifdef CAN_COMPILE_MMXEXT
1108 static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
1115 /* Detect interlacing */
1117 pxor_r2r( mm7, mm7 );
1118 for( y = 0; y < 9; y += 2 )
1121 pxor_r2r( mm5, mm5 );
1122 pxor_r2r( mm6, mm6 );
1123 for( x = 0; x < 8; x+=4 )
1125 movd_m2r( src[ x], mm0 );
1126 movd_m2r( src[1*i_src+x], mm1 );
1127 movd_m2r( src[2*i_src+x], mm2 );
1128 movd_m2r( src[3*i_src+x], mm3 );
1130 punpcklbw_r2r( mm7, mm0 );
1131 punpcklbw_r2r( mm7, mm1 );
1132 punpcklbw_r2r( mm7, mm2 );
1133 punpcklbw_r2r( mm7, mm3 );
1135 movq_r2r( mm0, mm4 );
1137 psubw_r2r( mm1, mm0 );
1138 psubw_r2r( mm2, mm4 );
1140 psubw_r2r( mm1, mm2 );
1141 psubw_r2r( mm1, mm3 );
1143 pmaddwd_r2r( mm0, mm0 );
1144 pmaddwd_r2r( mm4, mm4 );
1145 pmaddwd_r2r( mm2, mm2 );
1146 pmaddwd_r2r( mm3, mm3 );
1147 paddd_r2r( mm0, mm2 );
1148 paddd_r2r( mm4, mm3 );
1149 paddd_r2r( mm2, mm5 );
1150 paddd_r2r( mm3, mm6 );
1153 movq_r2r( mm5, mm0 );
1154 psrlq_i2r( 32, mm0 );
1155 paddd_r2r( mm0, mm5 );
1156 movd_r2m( mm5, fr );
1158 movq_r2r( mm6, mm0 );
1159 psrlq_i2r( 32, mm0 );
1160 paddd_r2r( mm0, mm6 );
1161 movd_r2m( mm6, ff );
1163 if( ff < 6*fr/8 && fr > 32 )
1172 /* XDeint8x8Frame: apply a small blend between field (1,6,1).
1173 * This won't destroy details, and help if there is a bit of interlacing.
1174 * (It helps with paning to avoid flickers)
1178 static inline void XDeint8x8FrameC( uint8_t *dst, int i_dst,
1179 uint8_t *src, int i_src )
1184 for( y = 0; y < 8; y += 2 )
1186 memcpy( dst, src, 8 );
1189 for( x = 0; x < 8; x++ )
1190 dst[x] = (src[x] + 6*src[1*i_src+x] + src[2*i_src+x] + 4 ) >> 3;
1196 static inline void XDeint8x8MergeC( uint8_t *dst, int i_dst,
1197 uint8_t *src1, int i_src1,
1198 uint8_t *src2, int i_src2 )
1203 for( y = 0; y < 8; y += 2 )
1205 memcpy( dst, src1, 8 );
1208 for( x = 0; x < 8; x++ )
1209 dst[x] = (src1[x] + 6*src2[x] + src1[i_src1+x] + 4 ) >> 3;
1217 #ifdef CAN_COMPILE_MMXEXT
1218 static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
1219 uint8_t *src1, int i_src1,
1220 uint8_t *src2, int i_src2 )
1222 static const uint64_t m_4 = I64C(0x0004000400040004);
1226 pxor_r2r( mm7, mm7 );
1227 for( y = 0; y < 8; y += 2 )
1229 for( x = 0; x < 8; x +=4 )
1231 movd_m2r( src1[x], mm0 );
1232 movd_r2m( mm0, dst[x] );
1234 movd_m2r( src2[x], mm1 );
1235 movd_m2r( src1[i_src1+x], mm2 );
1237 punpcklbw_r2r( mm7, mm0 );
1238 punpcklbw_r2r( mm7, mm1 );
1239 punpcklbw_r2r( mm7, mm2 );
1240 paddw_r2r( mm1, mm1 );
1241 movq_r2r( mm1, mm3 );
1242 paddw_r2r( mm3, mm3 );
1243 paddw_r2r( mm2, mm0 );
1244 paddw_r2r( mm3, mm1 );
1245 paddw_m2r( m_4, mm1 );
1246 paddw_r2r( mm1, mm0 );
1247 psraw_i2r( 3, mm0 );
1248 packuswb_r2r( mm7, mm0 );
1249 movd_r2m( mm0, dst[i_dst+x] );
1260 static inline void XDeint8x8Set( uint8_t *dst, int i_dst, uint8_t v )
1263 for( y = 0; y < 8; y++ )
1264 memset( &dst[y*i_dst], v, 8 );
1267 /* XDeint8x8FieldE: Stupid deinterlacing (1,0,1) for block that miss a
1270 * TODO: a better one for the inner part.
1272 static inline void XDeint8x8FieldEC( uint8_t *dst, int i_dst,
1273 uint8_t *src, int i_src )
1278 for( y = 0; y < 8; y += 2 )
1280 memcpy( dst, src, 8 );
1283 for( x = 0; x < 8; x++ )
1284 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1289 #ifdef CAN_COMPILE_MMXEXT
1290 static inline void XDeint8x8FieldEMMXEXT( uint8_t *dst, int i_dst,
1291 uint8_t *src, int i_src )
1296 for( y = 0; y < 8; y += 2 )
1298 movq_m2r( src[0], mm0 );
1299 movq_r2m( mm0, dst[0] );
1302 movq_m2r( src[2*i_src], mm1 );
1303 pavgb_r2r( mm1, mm0 );
1305 movq_r2m( mm0, dst[0] );
1313 /* XDeint8x8Field: Edge oriented interpolation
1314 * (Need -4 and +5 pixels H, +1 line)
1316 static inline void XDeint8x8FieldC( uint8_t *dst, int i_dst,
1317 uint8_t *src, int i_src )
1322 for( y = 0; y < 8; y += 2 )
1324 memcpy( dst, src, 8 );
1327 for( x = 0; x < 8; x++ )
1329 uint8_t *src2 = &src[2*i_src];
1330 /* I use 8 pixels just to match the MMX version, but it's overkill
1331 * 5 would be enough (less isn't good) */
1332 const int c0 = abs(src[x-4]-src2[x-2]) + abs(src[x-3]-src2[x-1]) +
1333 abs(src[x-2]-src2[x+0]) + abs(src[x-1]-src2[x+1]) +
1334 abs(src[x+0]-src2[x+2]) + abs(src[x+1]-src2[x+3]) +
1335 abs(src[x+2]-src2[x+4]) + abs(src[x+3]-src2[x+5]);
1337 const int c1 = abs(src[x-3]-src2[x-3]) + abs(src[x-2]-src2[x-2]) +
1338 abs(src[x-1]-src2[x-1]) + abs(src[x+0]-src2[x+0]) +
1339 abs(src[x+1]-src2[x+1]) + abs(src[x+2]-src2[x+2]) +
1340 abs(src[x+3]-src2[x+3]) + abs(src[x+4]-src2[x+4]);
1342 const int c2 = abs(src[x-2]-src2[x-4]) + abs(src[x-1]-src2[x-3]) +
1343 abs(src[x+0]-src2[x-2]) + abs(src[x+1]-src2[x-1]) +
1344 abs(src[x+2]-src2[x+0]) + abs(src[x+3]-src2[x+1]) +
1345 abs(src[x+4]-src2[x+2]) + abs(src[x+5]-src2[x+3]);
1347 if( c0 < c1 && c1 <= c2 )
1348 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1349 else if( c2 < c1 && c1 <= c0 )
1350 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1352 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1359 #ifdef CAN_COMPILE_MMXEXT
1360 static inline void XDeint8x8FieldMMXEXT( uint8_t *dst, int i_dst,
1361 uint8_t *src, int i_src )
1366 for( y = 0; y < 8; y += 2 )
1368 memcpy( dst, src, 8 );
1371 for( x = 0; x < 8; x++ )
1373 uint8_t *src2 = &src[2*i_src];
1376 movq_m2r( src[x-2], mm0 );
1377 movq_m2r( src[x-3], mm1 );
1378 movq_m2r( src[x-4], mm2 );
1380 psadbw_m2r( src2[x-4], mm0 );
1381 psadbw_m2r( src2[x-3], mm1 );
1382 psadbw_m2r( src2[x-2], mm2 );
1384 movd_r2m( mm0, c2 );
1385 movd_r2m( mm1, c1 );
1386 movd_r2m( mm2, c0 );
1388 if( c0 < c1 && c1 <= c2 )
1389 dst[x] = (src[x-1] + src2[x+1]) >> 1;
1390 else if( c2 < c1 && c1 <= c0 )
1391 dst[x] = (src[x+1] + src2[x-1]) >> 1;
1393 dst[x] = (src[x+0] + src2[x+0]) >> 1;
1403 static inline int XDeint8x8SsdC( uint8_t *pix1, int i_pix1,
1404 uint8_t *pix2, int i_pix2 )
1409 for( y = 0; y < 8; y++ )
1410 for( x = 0; x < 8; x++ )
1411 s += ssd( pix1[y*i_pix1+x] - pix2[y*i_pix2+x] );
1415 #ifdef CAN_COMPILE_MMXEXT
1416 static inline int XDeint8x8SsdMMXEXT( uint8_t *pix1, int i_pix1,
1417 uint8_t *pix2, int i_pix2 )
1422 pxor_r2r( mm7, mm7 );
1423 pxor_r2r( mm6, mm6 );
1425 for( y = 0; y < 8; y++ )
1427 movq_m2r( pix1[0], mm0 );
1428 movq_m2r( pix2[0], mm1 );
1430 movq_r2r( mm0, mm2 );
1431 movq_r2r( mm1, mm3 );
1433 punpcklbw_r2r( mm7, mm0 );
1434 punpckhbw_r2r( mm7, mm2 );
1435 punpcklbw_r2r( mm7, mm1 );
1436 punpckhbw_r2r( mm7, mm3 );
1438 psubw_r2r( mm1, mm0 );
1439 psubw_r2r( mm3, mm2 );
1441 pmaddwd_r2r( mm0, mm0 );
1442 pmaddwd_r2r( mm2, mm2 );
1444 paddd_r2r( mm2, mm0 );
1445 paddd_r2r( mm0, mm6 );
1451 movq_r2r( mm6, mm7 );
1452 psrlq_i2r( 32, mm7 );
1453 paddd_r2r( mm6, mm7 );
1462 /* A little try with motion, but doesn't work better that pure intra (and slow) */
1463 #ifdef CAN_COMPILE_MMXEXT
1466 * TODO: mmx version (easier in sse2)
1468 static inline void XDeintMC( uint8_t *dst, int i_dst,
1469 uint8_t *src, int i_src,
1471 int i_width, int i_height )
1473 const int d4x = mvx&0x03;
1474 const int d4y = mvy&0x03;
1476 const int cA = (4-d4x)*(4-d4y);
1477 const int cB = d4x *(4-d4y);
1478 const int cC = (4-d4x)*d4y;
1479 const int cD = d4x *d4y;
1485 src += (mvy >> 2) * i_src + (mvx >> 2);
1488 for( y = 0; y < i_height; y++ )
1490 for( x = 0; x < i_width; x++ )
1492 dst[x] = ( cA*src[x] + cB*src[x+1] +
1493 cC*srcp[x] + cD*srcp[x+1] + 8 ) >> 4;
1501 static int XDeint8x4SadMMXEXT( uint8_t *pix1, int i_pix1,
1502 uint8_t *pix2, int i_pix2 )
1506 movq_m2r( pix1[0*i_pix1], mm0 );
1507 movq_m2r( pix1[1*i_pix1], mm1 );
1509 psadbw_m2r( pix2[0*i_pix2], mm0 );
1510 psadbw_m2r( pix2[1*i_pix2], mm1 );
1512 movq_m2r( pix1[2*i_pix1], mm2 );
1513 movq_m2r( pix1[3*i_pix1], mm3 );
1514 psadbw_m2r( pix2[2*i_pix2], mm2 );
1515 psadbw_m2r( pix2[3*i_pix2], mm3 );
1517 paddd_r2r( mm1, mm0 );
1518 paddd_r2r( mm3, mm2 );
1519 paddd_r2r( mm2, mm0 );
1525 static inline int XDeint8x4TestQpel( uint8_t *src, int i_src,
1526 uint8_t *ref, int i_stride,
1528 int xmax, int ymax )
1530 uint8_t buffer[8*4];
1532 if( abs(mx) >= 4*xmax || abs(my) >= 4*ymax )
1535 XDeintMC( buffer, 8, ref, i_stride, mx, my, 8, 4 );
1536 return XDeint8x4SadMMXEXT( src, i_src, buffer, 8 );
1538 static inline int XDeint8x4TestInt( uint8_t *src, int i_src,
1539 uint8_t *ref, int i_stride,
1541 int xmax, int ymax )
1543 if( abs(mx) >= xmax || abs(my) >= ymax )
1546 return XDeint8x4SadMMXEXT( src, i_src, &ref[my*i_stride+mx], i_stride );
1549 static inline void XDeint8x8FieldMotion( uint8_t *dst, int i_dst,
1550 uint8_t *src, int i_src,
1552 int xmax, int ymax )
1554 static const int dx[8] = { 0, 0, -1, 1, -1, -1, 1, 1 };
1555 static const int dy[8] = {-1, 1, 0, 0, -1, 1, -1, 1 };
1556 uint8_t *next = &src[i_src];
1557 const int i_src2 = 2*i_src;
1562 uint8_t *rec = &dst[i_dst];
1564 /* We construct with intra method the missing field */
1565 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1567 /* Now we will try to find a match with ME with the other field */
1569 /* ME: A small/partial EPZS
1570 * We search only for small MV (with high motion intra will be perfect */
1571 if( xmax > 4 ) xmax = 4;
1572 if( ymax > 4 ) ymax = 4;
1574 /* Init with NULL Mv */
1576 mvs = XDeint8x4SadMMXEXT( rec, i_src2, next, i_src2 );
1578 /* Try predicted Mv */
1579 if( (s=XDeint8x4TestInt( rec, i_src2, next, i_src2, *mpx, *mpy, xmax, ymax)) < mvs )
1585 /* Search interger pel (small mv) */
1586 for( i_step = 0; i_step < 4; i_step++ )
1592 for( i = 0; i < 4; i++ )
1594 s = XDeint8x4TestInt( rec, i_src2,
1595 next, i_src2, mvx+dx[i], mvy+dy[i],
1615 if( mvs > 4 && mvs < 256 )
1618 /* XXX: for now only HPEL (too slow) */
1619 for( i_step = 0; i_step < 4; i_step++ )
1625 for( i = 0; i < 8; i++ )
1627 s = XDeint8x4TestQpel( rec, i_src2, next, i_src2,
1628 mvx+dx[i], mvy+dy[i],
1646 uint8_t buffer[8*4];
1647 XDeintMC( buffer, 8, next, i_src2, mvx, mvy, 8, 4 );
1648 XDeint8x8MergeMMXEXT( dst, i_dst, src, 2*i_src, buffer, 8 );
1650 //XDeint8x8Set( dst, i_dst, 0 );
1657 /* Kernel interpolation (1,-5,20,20,-5,1)
1658 * Loose a bit more details+add aliasing than edge interpol but avoid
1661 static inline uint8_t clip1( int a )
1670 static inline void XDeint8x8Field( uint8_t *dst, int i_dst,
1671 uint8_t *src, int i_src )
1676 for( y = 0; y < 8; y += 2 )
1678 const int i_src2 = i_src*2;
1680 memcpy( dst, src, 8 );
1683 for( x = 0; x < 8; x++ )
1687 pix = 1*(src[-2*i_src2+x]+src[3*i_src2+x]) +
1688 -5*(src[-1*i_src2+x]+src[2*i_src2+x])
1689 +20*(src[ 0*i_src2+x]+src[1*i_src2+x]);
1691 dst[x] = clip1( ( pix + 16 ) >> 5 );
1701 /* NxN arbitray size (and then only use pixel in the NxN block)
1703 static inline int XDeintNxNDetect( uint8_t *src, int i_src,
1704 int i_height, int i_width )
1711 /* Detect interlacing */
1712 /* FIXME way too simple, need to be more like XDeint8x8Detect */
1715 for( y = 0; y < i_height - 2; y += 2 )
1717 const uint8_t *s = &src[y*i_src];
1718 for( x = 0; x < i_width; x++ )
1720 fr += ssd(s[ x] - s[1*i_src+x]);
1721 ff += ssd(s[ x] - s[2*i_src+x]);
1723 if( ff < fr && fr > i_width / 2 )
1727 return fc < 2 ? VLC_FALSE : VLC_TRUE;
1730 static inline void XDeintNxNFrame( uint8_t *dst, int i_dst,
1731 uint8_t *src, int i_src,
1732 int i_width, int i_height )
1737 for( y = 0; y < i_height; y += 2 )
1739 memcpy( dst, src, i_width );
1742 if( y < i_height - 2 )
1744 for( x = 0; x < i_width; x++ )
1745 dst[x] = (src[x] + 2*src[1*i_src+x] + src[2*i_src+x] + 2 ) >> 2;
1749 /* Blend last line */
1750 for( x = 0; x < i_width; x++ )
1751 dst[x] = (src[x] + src[1*i_src+x] ) >> 1;
1758 static inline void XDeintNxNField( uint8_t *dst, int i_dst,
1759 uint8_t *src, int i_src,
1760 int i_width, int i_height )
1765 for( y = 0; y < i_height; y += 2 )
1767 memcpy( dst, src, i_width );
1770 if( y < i_height - 2 )
1772 for( x = 0; x < i_width; x++ )
1773 dst[x] = (src[x] + src[2*i_src+x] ) >> 1;
1777 /* Blend last line */
1778 for( x = 0; x < i_width; x++ )
1779 dst[x] = (src[x] + src[i_src+x]) >> 1;
1786 static inline void XDeintNxN( uint8_t *dst, int i_dst, uint8_t *src, int i_src,
1787 int i_width, int i_height )
1789 if( XDeintNxNDetect( src, i_src, i_width, i_height ) )
1790 XDeintNxNField( dst, i_dst, src, i_src, i_width, i_height );
1792 XDeintNxNFrame( dst, i_dst, src, i_src, i_width, i_height );
1796 static inline int median( int a, int b, int c )
1798 int min = a, max =a;
1809 return a + b + c - min - max;
1815 static inline void XDeintBand8x8C( uint8_t *dst, int i_dst,
1816 uint8_t *src, int i_src,
1817 const int i_mbx, int i_modx )
1821 for( x = 0; x < i_mbx; x++ )
1824 if( ( s = XDeint8x8DetectC( src, i_src ) ) )
1826 if( x == 0 || x == i_mbx - 1 )
1827 XDeint8x8FieldEC( dst, i_dst, src, i_src );
1829 XDeint8x8FieldC( dst, i_dst, src, i_src );
1833 XDeint8x8MergeC( dst, i_dst,
1834 &src[0*i_src], 2*i_src,
1835 &src[1*i_src], 2*i_src );
1843 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1845 #ifdef CAN_COMPILE_MMXEXT
1846 static inline void XDeintBand8x8MMXEXT( uint8_t *dst, int i_dst,
1847 uint8_t *src, int i_src,
1848 const int i_mbx, int i_modx )
1852 /* Reset current line */
1853 for( x = 0; x < i_mbx; x++ )
1856 if( ( s = XDeint8x8DetectMMXEXT( src, i_src ) ) )
1858 if( x == 0 || x == i_mbx - 1 )
1859 XDeint8x8FieldEMMXEXT( dst, i_dst, src, i_src );
1861 XDeint8x8FieldMMXEXT( dst, i_dst, src, i_src );
1865 XDeint8x8MergeMMXEXT( dst, i_dst,
1866 &src[0*i_src], 2*i_src,
1867 &src[1*i_src], 2*i_src );
1875 XDeintNxN( dst, i_dst, src, i_src, i_modx, 8 );
1879 static void RenderX( vout_thread_t *p_vout,
1880 picture_t *p_outpic, picture_t *p_pic )
1884 /* Copy image and skip lines */
1885 for( i_plane = 0 ; i_plane < p_pic->i_planes ; i_plane++ )
1887 const int i_mby = ( p_outpic->p[i_plane].i_visible_lines + 7 )/8 - 1;
1888 const int i_mbx = p_outpic->p[i_plane].i_visible_pitch/8;
1890 const int i_mody = p_outpic->p[i_plane].i_visible_lines - 8*i_mby;
1891 const int i_modx = p_outpic->p[i_plane].i_visible_pitch - 8*i_mbx;
1893 const int i_dst = p_outpic->p[i_plane].i_pitch;
1894 const int i_src = p_pic->p[i_plane].i_pitch;
1898 for( y = 0; y < i_mby; y++ )
1900 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1901 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1903 #ifdef CAN_COMPILE_MMXEXT
1904 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1905 XDeintBand8x8MMXEXT( dst, i_dst, src, i_src, i_mbx, i_modx );
1908 XDeintBand8x8C( dst, i_dst, src, i_src, i_mbx, i_modx );
1911 /* Last line (C only)*/
1914 uint8_t *dst = &p_outpic->p[i_plane].p_pixels[8*y*i_dst];
1915 uint8_t *src = &p_pic->p[i_plane].p_pixels[8*y*i_src];
1917 for( x = 0; x < i_mbx; x++ )
1919 XDeintNxN( dst, i_dst, src, i_src, 8, i_mody );
1926 XDeintNxN( dst, i_dst, src, i_src, i_modx, i_mody );
1930 #ifdef CAN_COMPILE_MMXEXT
1931 if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT )
1936 /*****************************************************************************
1937 * SendEvents: forward mouse and keyboard events to the parent p_vout
1938 *****************************************************************************/
1939 static int SendEvents( vlc_object_t *p_this, char const *psz_var,
1940 vlc_value_t oldval, vlc_value_t newval, void *_p_vout )
1942 vout_thread_t *p_vout = (vout_thread_t *)_p_vout;
1943 vlc_value_t sentval = newval;
1945 if( !strcmp( psz_var, "mouse-y" ) )
1947 switch( p_vout->p_sys->i_mode )
1949 case DEINTERLACE_MEAN:
1950 case DEINTERLACE_DISCARD:
1956 var_Set( p_vout, psz_var, sentval );
1961 /*****************************************************************************
1962 * FilterCallback: called when changing the deinterlace method on the fly.
1963 *****************************************************************************/
1964 static int FilterCallback( vlc_object_t *p_this, char const *psz_cmd,
1965 vlc_value_t oldval, vlc_value_t newval,
1968 vout_thread_t * p_vout = (vout_thread_t *)p_this;
1969 int i_old_mode = p_vout->p_sys->i_mode;
1971 msg_Dbg( p_vout, "using %s deinterlace mode", newval.psz_string );
1973 vlc_mutex_lock( &p_vout->p_sys->filter_lock );
1975 SetFilterMethod( p_vout, newval.psz_string );
1977 switch( p_vout->render.i_chroma )
1979 case VLC_FOURCC('I','4','2','2'):
1980 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1984 case VLC_FOURCC('I','4','2','0'):
1985 case VLC_FOURCC('I','Y','U','V'):
1986 case VLC_FOURCC('Y','V','1','2'):
1987 switch( p_vout->p_sys->i_mode )
1989 case DEINTERLACE_MEAN:
1990 case DEINTERLACE_DISCARD:
1991 if( ( i_old_mode == DEINTERLACE_MEAN )
1992 || ( i_old_mode == DEINTERLACE_DISCARD ) )
1994 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
1999 case DEINTERLACE_BOB:
2000 case DEINTERLACE_BLEND:
2001 case DEINTERLACE_LINEAR:
2002 if( ( i_old_mode == DEINTERLACE_BOB )
2003 || ( i_old_mode == DEINTERLACE_BLEND )
2004 || ( i_old_mode == DEINTERLACE_LINEAR ) )
2006 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2017 /* We need to kill the old vout */
2019 DEL_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2021 vlc_object_detach( p_vout->p_sys->p_vout );
2022 vout_Destroy( p_vout->p_sys->p_vout );
2024 /* Try to open a new video output */
2025 p_vout->p_sys->p_vout = SpawnRealVout( p_vout );
2027 if( p_vout->p_sys->p_vout == NULL )
2029 /* Everything failed */
2030 msg_Err( p_vout, "cannot open vout, aborting" );
2032 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2033 return VLC_EGENERIC;
2036 ADD_CALLBACKS( p_vout->p_sys->p_vout, SendEvents );
2038 vlc_mutex_unlock( &p_vout->p_sys->filter_lock );
2042 /*****************************************************************************
2043 * SendEventsToChild: forward events to the child/children vout
2044 *****************************************************************************/
2045 static int SendEventsToChild( vlc_object_t *p_this, char const *psz_var,
2046 vlc_value_t oldval, vlc_value_t newval, void *p_data )
2048 vout_thread_t *p_vout = (vout_thread_t *)p_this;
2049 var_Set( p_vout->p_sys->p_vout, psz_var, newval );
2054 /*****************************************************************************
2055 * video filter2 functions
2056 *****************************************************************************/
2057 static picture_t *Deinterlace( filter_t *p_filter, picture_t *p_pic )
2059 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2060 picture_t *p_pic_dst;
2062 /* Request output picture */
2063 p_pic_dst = p_filter->pf_vout_buffer_new( p_filter );
2064 if( p_pic_dst == NULL )
2066 msg_Warn( p_filter, "can't get output picture" );
2070 switch( p_vout->p_sys->i_mode )
2072 case DEINTERLACE_DISCARD:
2074 RenderDiscard( p_vout, p_pic_dst, p_pic, 0 );
2076 msg_Err( p_vout, "discarding lines is not supported yet" );
2077 p_pic_dst->pf_release( p_pic_dst );
2081 case DEINTERLACE_BOB:
2083 RenderBob( p_vout, pp_outpic[0], p_pic, 0 );
2084 RenderBob( p_vout, pp_outpic[1], p_pic, 1 );
2088 case DEINTERLACE_LINEAR:
2090 RenderLinear( p_vout, pp_outpic[0], p_pic, 0 );
2091 RenderLinear( p_vout, pp_outpic[1], p_pic, 1 );
2093 msg_Err( p_vout, "doubling the frame rate is not supported yet" );
2094 p_pic_dst->pf_release( p_pic_dst );
2098 case DEINTERLACE_MEAN:
2099 RenderMean( p_vout, p_pic_dst, p_pic );
2102 case DEINTERLACE_BLEND:
2103 RenderBlend( p_vout, p_pic_dst, p_pic );
2107 RenderX( p_vout, p_pic_dst, p_pic );
2111 p_pic_dst->date = p_pic->date;
2112 p_pic_dst->b_force = p_pic->b_force;
2113 p_pic_dst->i_nb_fields = p_pic->i_nb_fields;
2114 p_pic_dst->b_progressive = VLC_TRUE;
2115 p_pic_dst->b_top_field_first = p_pic->b_top_field_first;
2117 p_pic->pf_release( p_pic );
2121 /*****************************************************************************
2123 *****************************************************************************/
2124 static int OpenFilter( vlc_object_t *p_this )
2126 filter_t *p_filter = (filter_t*)p_this;
2127 vout_thread_t *p_vout;
2130 if( ( p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','4','2','0') &&
2131 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('I','Y','U','V') &&
2132 p_filter->fmt_in.video.i_chroma != VLC_FOURCC('Y','V','1','2') ) ||
2133 p_filter->fmt_in.video.i_chroma != p_filter->fmt_out.video.i_chroma )
2135 return VLC_EGENERIC;
2138 /* Impossible to use VLC_OBJECT_VOUT here because it would be used
2140 p_vout = vlc_object_create( p_filter, sizeof(vout_thread_t) );
2141 vlc_object_attach( p_vout, p_filter );
2142 p_filter->p_sys = (filter_sys_t *)p_vout;
2143 p_vout->render.i_chroma = p_filter->fmt_in.video.i_chroma;
2145 sout_CfgParse( p_filter, FILTER_CFG_PREFIX, ppsz_filter_options,
2147 var_Get( p_filter, FILTER_CFG_PREFIX "mode", &val );
2148 var_Create( p_filter, "deinterlace-mode", VLC_VAR_STRING );
2149 var_Set( p_filter, "deinterlace-mode", val );
2151 if ( Create( VLC_OBJECT(p_vout) ) != VLC_SUCCESS )
2153 vlc_object_detach( p_vout );
2154 vlc_object_release( p_vout );
2155 return VLC_EGENERIC;
2158 p_filter->pf_video_filter = Deinterlace;
2160 msg_Dbg( p_filter, "deinterlacing" );
2165 /*****************************************************************************
2166 * CloseFilter: clean up the filter
2167 *****************************************************************************/
2168 static void CloseFilter( vlc_object_t *p_this )
2170 filter_t *p_filter = (filter_t*)p_this;
2171 vout_thread_t *p_vout = (vout_thread_t *)p_filter->p_sys;
2173 Destroy( VLC_OBJECT(p_vout) );
2174 vlc_object_detach( p_vout );
2175 vlc_object_release( p_vout );