omg, major typo (fails if card supports rgb24)
[ffmpeg.git] / libavformat / grab.c
1 /*
2  * Linux video grab interface
3  * Copyright (c) 2000,2001 Fabrice Bellard.
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 #include "avformat.h"
22 #include <unistd.h>
23 #include <fcntl.h>
24 #include <sys/ioctl.h>
25 #include <sys/mman.h>
26 #include <sys/time.h>
27 #define _LINUX_TIME_H 1
28 #include <linux/videodev.h>
29 #include <time.h>
30
31 typedef struct {
32     int fd;
33     int frame_format; /* see VIDEO_PALETTE_xxx */
34     int use_mmap;
35     int width, height;
36     int frame_rate;
37     int frame_rate_base;
38     int64_t time_frame;
39     int frame_size;
40     struct video_capability video_cap;
41     struct video_audio audio_saved;
42     uint8_t *video_buf;
43     struct video_mbuf gb_buffers;
44     struct video_mmap gb_buf;
45     int gb_frame;
46
47     /* ATI All In Wonder specific stuff */
48     /* XXX: remove and merge in libavcodec/imgconvert.c */
49     int aiw_enabled;
50     int deint;
51     int halfw;
52     uint8_t *src_mem;
53     uint8_t *lum_m4_mem;
54 } VideoData;
55
56 static int aiw_init(VideoData *s);
57 static int aiw_read_picture(VideoData *s, uint8_t *data);
58 static int aiw_close(VideoData *s);
59
60 static int grab_read_header(AVFormatContext *s1, AVFormatParameters *ap)
61 {
62     VideoData *s = s1->priv_data;
63     AVStream *st;
64     int width, height;
65     int video_fd, frame_size;
66     int ret, frame_rate, frame_rate_base;
67     int desired_palette, desired_depth;
68     struct video_tuner tuner;
69     struct video_audio audio;
70     struct video_picture pict;
71     const char *video_device;
72     int j;
73
74     if (ap->width <= 0 || ap->height <= 0 || ap->time_base.den <= 0) {
75         av_log(s1, AV_LOG_ERROR, "Bad capture size (%dx%d) or wrong time base (%d)\n",
76             ap->width, ap->height, ap->time_base.den);
77
78         return -1;
79     }
80
81     width = ap->width;
82     height = ap->height;
83     frame_rate      = ap->time_base.den;
84     frame_rate_base = ap->time_base.num;
85
86     if((unsigned)width > 32767 || (unsigned)height > 32767) {
87         av_log(s1, AV_LOG_ERROR, "Capture size is out of range: %dx%d\n",
88             width, height);
89
90         return -1;
91     }
92
93     st = av_new_stream(s1, 0);
94     if (!st)
95         return -ENOMEM;
96     av_set_pts_info(st, 64, 1, 1000000); /* 64 bits pts in us */
97
98     s->width = width;
99     s->height = height;
100     s->frame_rate      = frame_rate;
101     s->frame_rate_base = frame_rate_base;
102
103     video_device = ap->device;
104     if (!video_device)
105         video_device = "/dev/video";
106     video_fd = open(video_device, O_RDWR);
107     if (video_fd < 0) {
108         perror(video_device);
109         goto fail;
110     }
111
112     if (ioctl(video_fd,VIDIOCGCAP, &s->video_cap) < 0) {
113         perror("VIDIOCGCAP");
114         goto fail;
115     }
116
117     if (!(s->video_cap.type & VID_TYPE_CAPTURE)) {
118         av_log(s1, AV_LOG_ERROR, "Fatal: grab device does not handle capture\n");
119         goto fail;
120     }
121
122     desired_palette = -1;
123     desired_depth = -1;
124     if (ap->pix_fmt == PIX_FMT_YUV420P) {
125         desired_palette = VIDEO_PALETTE_YUV420P;
126         desired_depth = 12;
127     } else if (ap->pix_fmt == PIX_FMT_YUV422) {
128         desired_palette = VIDEO_PALETTE_YUV422;
129         desired_depth = 16;
130     } else if (ap->pix_fmt == PIX_FMT_BGR24) {
131         desired_palette = VIDEO_PALETTE_RGB24;
132         desired_depth = 24;
133     }
134
135     /* set tv standard */
136     if (ap->standard && !ioctl(video_fd, VIDIOCGTUNER, &tuner)) {
137         if (!strcasecmp(ap->standard, "pal"))
138             tuner.mode = VIDEO_MODE_PAL;
139         else if (!strcasecmp(ap->standard, "secam"))
140             tuner.mode = VIDEO_MODE_SECAM;
141         else
142             tuner.mode = VIDEO_MODE_NTSC;
143         ioctl(video_fd, VIDIOCSTUNER, &tuner);
144     }
145
146     /* unmute audio */
147     audio.audio = 0;
148     ioctl(video_fd, VIDIOCGAUDIO, &audio);
149     memcpy(&s->audio_saved, &audio, sizeof(audio));
150     audio.flags &= ~VIDEO_AUDIO_MUTE;
151     ioctl(video_fd, VIDIOCSAUDIO, &audio);
152
153     ioctl(video_fd, VIDIOCGPICT, &pict);
154 #if 0
155     printf("v4l: colour=%d hue=%d brightness=%d constrast=%d whiteness=%d\n",
156            pict.colour,
157            pict.hue,
158            pict.brightness,
159            pict.contrast,
160            pict.whiteness);
161 #endif
162     /* try to choose a suitable video format */
163     pict.palette = desired_palette;
164     pict.depth= desired_depth;
165     if (desired_palette == -1 || (ret = ioctl(video_fd, VIDIOCSPICT, &pict)) < 0) {
166         pict.palette=VIDEO_PALETTE_YUV420P;
167         pict.depth=12;
168         ret = ioctl(video_fd, VIDIOCSPICT, &pict);
169         if (ret < 0) {
170             pict.palette=VIDEO_PALETTE_YUV422;
171             pict.depth=16;
172             ret = ioctl(video_fd, VIDIOCSPICT, &pict);
173             if (ret < 0) {
174                 pict.palette=VIDEO_PALETTE_RGB24;
175                 pict.depth=24;
176                 ret = ioctl(video_fd, VIDIOCSPICT, &pict);
177                 if (ret < 0) {
178                     pict.palette=VIDEO_PALETTE_GREY;
179                     pict.depth=8;
180                     ret = ioctl(video_fd, VIDIOCSPICT, &pict);
181                     if (ret < 0)
182                         goto fail1;
183                 }
184             }
185         }
186     }
187
188     ret = ioctl(video_fd,VIDIOCGMBUF,&s->gb_buffers);
189     if (ret < 0) {
190         /* try to use read based access */
191         struct video_window win;
192         int val;
193
194         win.x = 0;
195         win.y = 0;
196         win.width = width;
197         win.height = height;
198         win.chromakey = -1;
199         win.flags = 0;
200
201         ioctl(video_fd, VIDIOCSWIN, &win);
202
203         s->frame_format = pict.palette;
204
205         val = 1;
206         ioctl(video_fd, VIDIOCCAPTURE, &val);
207
208         s->time_frame = av_gettime() * s->frame_rate / s->frame_rate_base;
209         s->use_mmap = 0;
210
211         /* ATI All In Wonder automatic activation */
212         if (!strcmp(s->video_cap.name, "Km")) {
213             if (aiw_init(s) < 0)
214                 goto fail;
215             s->aiw_enabled = 1;
216             /* force 420P format because convertion from YUV422 to YUV420P
217                is done in this driver (ugly) */
218             s->frame_format = VIDEO_PALETTE_YUV420P;
219         }
220     } else {
221         s->video_buf = mmap(0,s->gb_buffers.size,PROT_READ|PROT_WRITE,MAP_SHARED,video_fd,0);
222         if ((unsigned char*)-1 == s->video_buf) {
223             s->video_buf = mmap(0,s->gb_buffers.size,PROT_READ|PROT_WRITE,MAP_PRIVATE,video_fd,0);
224             if ((unsigned char*)-1 == s->video_buf) {
225                 perror("mmap");
226                 goto fail;
227             }
228         }
229         s->gb_frame = 0;
230         s->time_frame = av_gettime() * s->frame_rate / s->frame_rate_base;
231
232         /* start to grab the first frame */
233         s->gb_buf.frame = s->gb_frame % s->gb_buffers.frames;
234         s->gb_buf.height = height;
235         s->gb_buf.width = width;
236         s->gb_buf.format = pict.palette;
237
238         ret = ioctl(video_fd, VIDIOCMCAPTURE, &s->gb_buf);
239         if (ret < 0) {
240             if (errno != EAGAIN) {
241             fail1:
242                 av_log(s1, AV_LOG_ERROR, "Fatal: grab device does not support suitable format\n");
243             } else {
244                 av_log(s1, AV_LOG_ERROR,"Fatal: grab device does not receive any video signal\n");
245             }
246             goto fail;
247         }
248         for (j = 1; j < s->gb_buffers.frames; j++) {
249           s->gb_buf.frame = j;
250           ioctl(video_fd, VIDIOCMCAPTURE, &s->gb_buf);
251         }
252         s->frame_format = s->gb_buf.format;
253         s->use_mmap = 1;
254     }
255
256     switch(s->frame_format) {
257     case VIDEO_PALETTE_YUV420P:
258         frame_size = (width * height * 3) / 2;
259         st->codec->pix_fmt = PIX_FMT_YUV420P;
260         break;
261     case VIDEO_PALETTE_YUV422:
262         frame_size = width * height * 2;
263         st->codec->pix_fmt = PIX_FMT_YUV422;
264         break;
265     case VIDEO_PALETTE_RGB24:
266         frame_size = width * height * 3;
267         st->codec->pix_fmt = PIX_FMT_BGR24; /* NOTE: v4l uses BGR24, not RGB24 ! */
268         break;
269     case VIDEO_PALETTE_GREY:
270         frame_size = width * height * 1;
271         st->codec->pix_fmt = PIX_FMT_GRAY8;
272         break;
273     default:
274         goto fail;
275     }
276     s->fd = video_fd;
277     s->frame_size = frame_size;
278
279     st->codec->codec_type = CODEC_TYPE_VIDEO;
280     st->codec->codec_id = CODEC_ID_RAWVIDEO;
281     st->codec->width = width;
282     st->codec->height = height;
283     st->codec->time_base.den      = frame_rate;
284     st->codec->time_base.num = frame_rate_base;
285     st->codec->bit_rate = frame_size * 1/av_q2d(st->codec->time_base) * 8;
286
287     return 0;
288  fail:
289     if (video_fd >= 0)
290         close(video_fd);
291     av_free(st);
292     return AVERROR_IO;
293 }
294
295 static int v4l_mm_read_picture(VideoData *s, uint8_t *buf)
296 {
297     uint8_t *ptr;
298
299     while (ioctl(s->fd, VIDIOCSYNC, &s->gb_frame) < 0 &&
300            (errno == EAGAIN || errno == EINTR));
301
302     ptr = s->video_buf + s->gb_buffers.offsets[s->gb_frame];
303     memcpy(buf, ptr, s->frame_size);
304
305     /* Setup to capture the next frame */
306     s->gb_buf.frame = s->gb_frame;
307     if (ioctl(s->fd, VIDIOCMCAPTURE, &s->gb_buf) < 0) {
308         if (errno == EAGAIN)
309             av_log(NULL, AV_LOG_ERROR, "Cannot Sync\n");
310         else
311             perror("VIDIOCMCAPTURE");
312         return AVERROR_IO;
313     }
314
315     /* This is now the grabbing frame */
316     s->gb_frame = (s->gb_frame + 1) % s->gb_buffers.frames;
317
318     return s->frame_size;
319 }
320
321 static int grab_read_packet(AVFormatContext *s1, AVPacket *pkt)
322 {
323     VideoData *s = s1->priv_data;
324     int64_t curtime, delay;
325     struct timespec ts;
326
327     /* Calculate the time of the next frame */
328     s->time_frame += INT64_C(1000000);
329
330     /* wait based on the frame rate */
331     for(;;) {
332         curtime = av_gettime();
333         delay = s->time_frame  * s->frame_rate_base / s->frame_rate - curtime;
334         if (delay <= 0) {
335             if (delay < INT64_C(-1000000) * s->frame_rate_base / s->frame_rate) {
336                 /* printf("grabbing is %d frames late (dropping)\n", (int) -(delay / 16666)); */
337                 s->time_frame += INT64_C(1000000);
338             }
339             break;
340         }
341         ts.tv_sec = delay / 1000000;
342         ts.tv_nsec = (delay % 1000000) * 1000;
343         nanosleep(&ts, NULL);
344     }
345
346     if (av_new_packet(pkt, s->frame_size) < 0)
347         return AVERROR_IO;
348
349     pkt->pts = curtime;
350
351     /* read one frame */
352     if (s->aiw_enabled) {
353         return aiw_read_picture(s, pkt->data);
354     } else if (s->use_mmap) {
355         return v4l_mm_read_picture(s, pkt->data);
356     } else {
357         if (read(s->fd, pkt->data, pkt->size) != pkt->size)
358             return AVERROR_IO;
359         return s->frame_size;
360     }
361 }
362
363 static int grab_read_close(AVFormatContext *s1)
364 {
365     VideoData *s = s1->priv_data;
366
367     if (s->aiw_enabled)
368         aiw_close(s);
369
370     if (s->use_mmap)
371         munmap(s->video_buf, s->gb_buffers.size);
372
373     /* mute audio. we must force it because the BTTV driver does not
374        return its state correctly */
375     s->audio_saved.flags |= VIDEO_AUDIO_MUTE;
376     ioctl(s->fd, VIDIOCSAUDIO, &s->audio_saved);
377
378     close(s->fd);
379     return 0;
380 }
381
382 AVInputFormat video_grab_device_demuxer = {
383     "video4linux",
384     "video grab",
385     sizeof(VideoData),
386     NULL,
387     grab_read_header,
388     grab_read_packet,
389     grab_read_close,
390     .flags = AVFMT_NOFILE,
391 };
392
393 /* All in Wonder specific stuff */
394 /* XXX: remove and merge in libavcodec/imgconvert.c */
395
396 static int aiw_init(VideoData *s)
397 {
398     int width, height;
399
400     width = s->width;
401     height = s->height;
402
403     if ((width == s->video_cap.maxwidth && height == s->video_cap.maxheight) ||
404         (width == s->video_cap.maxwidth && height == s->video_cap.maxheight*2) ||
405         (width == s->video_cap.maxwidth/2 && height == s->video_cap.maxheight)) {
406
407         s->deint=0;
408         s->halfw=0;
409         if (height == s->video_cap.maxheight*2) s->deint=1;
410         if (width == s->video_cap.maxwidth/2) s->halfw=1;
411     } else {
412         av_log(NULL, AV_LOG_ERROR, "\nIncorrect Grab Size Supplied - Supported Sizes Are:\n");
413         av_log(NULL, AV_LOG_ERROR, " %dx%d  %dx%d %dx%d\n\n",
414                 s->video_cap.maxwidth,s->video_cap.maxheight,
415                 s->video_cap.maxwidth,s->video_cap.maxheight*2,
416                 s->video_cap.maxwidth/2,s->video_cap.maxheight);
417         goto fail;
418     }
419
420     if (s->halfw == 0) {
421         s->src_mem = av_malloc(s->width*2);
422     } else {
423         s->src_mem = av_malloc(s->width*4);
424     }
425     if (!s->src_mem) goto fail;
426
427     s->lum_m4_mem = av_malloc(s->width);
428     if (!s->lum_m4_mem)
429         goto fail;
430     return 0;
431  fail:
432     av_freep(&s->src_mem);
433     av_freep(&s->lum_m4_mem);
434     return -1;
435 }
436
437 #ifdef HAVE_MMX
438 #include "libavcodec/i386/mmx.h"
439
440 #define LINE_WITH_UV \
441                     movq_m2r(ptr[0],mm0); \
442                     movq_m2r(ptr[8],mm1);  \
443                     movq_r2r(mm0, mm4); \
444                     punpcklbw_r2r(mm1,mm0); \
445                     punpckhbw_r2r(mm1,mm4); \
446                     movq_r2r(mm0,mm5); \
447                     punpcklbw_r2r(mm4,mm0); \
448                     punpckhbw_r2r(mm4,mm5); \
449                     movq_r2r(mm0,mm1); \
450                     punpcklbw_r2r(mm5,mm1); \
451                     movq_r2m(mm1,lum[0]); \
452                     movq_m2r(ptr[16],mm2); \
453                     movq_m2r(ptr[24],mm1); \
454                     movq_r2r(mm2,mm4); \
455                     punpcklbw_r2r(mm1,mm2); \
456                     punpckhbw_r2r(mm1,mm4); \
457                     movq_r2r(mm2,mm3); \
458                     punpcklbw_r2r(mm4,mm2); \
459                     punpckhbw_r2r(mm4,mm3); \
460                     movq_r2r(mm2,mm1); \
461                     punpcklbw_r2r(mm3,mm1); \
462                     movq_r2m(mm1,lum[8]); \
463                     punpckhdq_r2r(mm2,mm0); \
464                     punpckhdq_r2r(mm3,mm5); \
465                     movq_r2m(mm0,cb[0]); \
466                     movq_r2m(mm5,cr[0]);
467
468 #define LINE_NO_UV \
469                     movq_m2r(ptr[0],mm0);\
470                     movq_m2r(ptr[8],mm1);\
471                     movq_r2r(mm0, mm4);\
472                     punpcklbw_r2r(mm1,mm0); \
473                     punpckhbw_r2r(mm1,mm4);\
474                     movq_r2r(mm0,mm5);\
475                     punpcklbw_r2r(mm4,mm0);\
476                     punpckhbw_r2r(mm4,mm5);\
477                     movq_r2r(mm0,mm1);\
478                     punpcklbw_r2r(mm5,mm1);\
479                     movq_r2m(mm1,lum[0]);\
480                     movq_m2r(ptr[16],mm2);\
481                     movq_m2r(ptr[24],mm1);\
482                     movq_r2r(mm2,mm4);\
483                     punpcklbw_r2r(mm1,mm2);\
484                     punpckhbw_r2r(mm1,mm4);\
485                     movq_r2r(mm2,mm3);\
486                     punpcklbw_r2r(mm4,mm2);\
487                     punpckhbw_r2r(mm4,mm3);\
488                     movq_r2r(mm2,mm1);\
489                     punpcklbw_r2r(mm3,mm1);\
490                     movq_r2m(mm1,lum[8]);
491
492 #define LINE_WITHUV_AVG \
493                     movq_m2r(ptr[0], mm0);\
494                     movq_m2r(ptr[8], mm1);\
495                     movq_r2r(mm0, mm4);\
496                     punpcklbw_r2r(mm1,mm0);\
497                     punpckhbw_r2r(mm1,mm4);\
498                     movq_r2r(mm0,mm5);\
499                     punpcklbw_r2r(mm4,mm0);\
500                     punpckhbw_r2r(mm4,mm5);\
501                     movq_r2r(mm0,mm1);\
502                     movq_r2r(mm5,mm2);\
503                     punpcklbw_r2r(mm7,mm1);\
504                     punpcklbw_r2r(mm7,mm2);\
505                     paddw_r2r(mm6,mm1);\
506                     paddw_r2r(mm2,mm1);\
507                     psraw_i2r(1,mm1);\
508                     packuswb_r2r(mm7,mm1);\
509                     movd_r2m(mm1,lum[0]);\
510                     movq_m2r(ptr[16],mm2);\
511                     movq_m2r(ptr[24],mm1);\
512                     movq_r2r(mm2,mm4);\
513                     punpcklbw_r2r(mm1,mm2);\
514                     punpckhbw_r2r(mm1,mm4);\
515                     movq_r2r(mm2,mm3);\
516                     punpcklbw_r2r(mm4,mm2);\
517                     punpckhbw_r2r(mm4,mm3);\
518                     movq_r2r(mm2,mm1);\
519                     movq_r2r(mm3,mm4);\
520                     punpcklbw_r2r(mm7,mm1);\
521                     punpcklbw_r2r(mm7,mm4);\
522                     paddw_r2r(mm6,mm1);\
523                     paddw_r2r(mm4,mm1);\
524                     psraw_i2r(1,mm1);\
525                     packuswb_r2r(mm7,mm1);\
526                     movd_r2m(mm1,lum[4]);\
527                     punpckhbw_r2r(mm7,mm0);\
528                     punpckhbw_r2r(mm7,mm2);\
529                     paddw_r2r(mm6,mm0);\
530                     paddw_r2r(mm2,mm0);\
531                     psraw_i2r(1,mm0);\
532                     packuswb_r2r(mm7,mm0);\
533                     punpckhbw_r2r(mm7,mm5);\
534                     punpckhbw_r2r(mm7,mm3);\
535                     paddw_r2r(mm6,mm5);\
536                     paddw_r2r(mm3,mm5);\
537                     psraw_i2r(1,mm5);\
538                     packuswb_r2r(mm7,mm5);\
539                     movd_r2m(mm0,cb[0]);\
540                     movd_r2m(mm5,cr[0]);
541
542 #define LINE_NOUV_AVG \
543                     movq_m2r(ptr[0],mm0);\
544                     movq_m2r(ptr[8],mm1);\
545                     pand_r2r(mm5,mm0);\
546                     pand_r2r(mm5,mm1);\
547                     pmaddwd_r2r(mm6,mm0);\
548                     pmaddwd_r2r(mm6,mm1);\
549                     packssdw_r2r(mm1,mm0);\
550                     paddw_r2r(mm6,mm0);\
551                     psraw_i2r(1,mm0);\
552                     movq_m2r(ptr[16],mm2);\
553                     movq_m2r(ptr[24],mm3);\
554                     pand_r2r(mm5,mm2);\
555                     pand_r2r(mm5,mm3);\
556                     pmaddwd_r2r(mm6,mm2);\
557                     pmaddwd_r2r(mm6,mm3);\
558                     packssdw_r2r(mm3,mm2);\
559                     paddw_r2r(mm6,mm2);\
560                     psraw_i2r(1,mm2);\
561                     packuswb_r2r(mm2,mm0);\
562                     movq_r2m(mm0,lum[0]);
563
564 #define DEINT_LINE_LUM(ptroff) \
565                     movd_m2r(lum_m4[(ptroff)],mm0);\
566                     movd_m2r(lum_m3[(ptroff)],mm1);\
567                     movd_m2r(lum_m2[(ptroff)],mm2);\
568                     movd_m2r(lum_m1[(ptroff)],mm3);\
569                     movd_m2r(lum[(ptroff)],mm4);\
570                     punpcklbw_r2r(mm7,mm0);\
571                     movd_r2m(mm2,lum_m4[(ptroff)]);\
572                     punpcklbw_r2r(mm7,mm1);\
573                     punpcklbw_r2r(mm7,mm2);\
574                     punpcklbw_r2r(mm7,mm3);\
575                     punpcklbw_r2r(mm7,mm4);\
576                     psllw_i2r(2,mm1);\
577                     psllw_i2r(1,mm2);\
578                     paddw_r2r(mm6,mm1);\
579                     psllw_i2r(2,mm3);\
580                     paddw_r2r(mm2,mm1);\
581                     paddw_r2r(mm4,mm0);\
582                     paddw_r2r(mm3,mm1);\
583                     psubusw_r2r(mm0,mm1);\
584                     psrlw_i2r(3,mm1);\
585                     packuswb_r2r(mm7,mm1);\
586                     movd_r2m(mm1,lum_m2[(ptroff)]);
587
588 #else
589 #include "libavcodec/dsputil.h"
590
591 #define LINE_WITH_UV \
592                     lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
593                     cb[0]=ptr[1];cb[1]=ptr[5];\
594                     cr[0]=ptr[3];cr[1]=ptr[7];\
595                     lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
596                     cb[2]=ptr[9];cb[3]=ptr[13];\
597                     cr[2]=ptr[11];cr[3]=ptr[15];\
598                     lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
599                     cb[4]=ptr[17];cb[5]=ptr[21];\
600                     cr[4]=ptr[19];cr[5]=ptr[23];\
601                     lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];\
602                     cb[6]=ptr[25];cb[7]=ptr[29];\
603                     cr[6]=ptr[27];cr[7]=ptr[31];
604
605 #define LINE_NO_UV \
606                     lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
607                     lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
608                     lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
609                     lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];
610
611 #define LINE_WITHUV_AVG \
612                     sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
613                     sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
614                     sum=(ptr[1]+ptr[5]+1) >> 1;cb[0]=sum; \
615                     sum=(ptr[3]+ptr[7]+1) >> 1;cr[0]=sum; \
616                     sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
617                     sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
618                     sum=(ptr[9]+ptr[13]+1) >> 1;cb[1]=sum; \
619                     sum=(ptr[11]+ptr[15]+1) >> 1;cr[1]=sum; \
620                     sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
621                     sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
622                     sum=(ptr[17]+ptr[21]+1) >> 1;cb[2]=sum; \
623                     sum=(ptr[19]+ptr[23]+1) >> 1;cr[2]=sum; \
624                     sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
625                     sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum; \
626                     sum=(ptr[25]+ptr[29]+1) >> 1;cb[3]=sum; \
627                     sum=(ptr[27]+ptr[31]+1) >> 1;cr[3]=sum;
628
629 #define LINE_NOUV_AVG \
630                     sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
631                     sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
632                     sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
633                     sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
634                     sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
635                     sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
636                     sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
637                     sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum;
638
639 #define DEINT_LINE_LUM(ptroff) \
640                     sum=(-lum_m4[(ptroff)]+(lum_m3[(ptroff)]<<2)+(lum_m2[(ptroff)]<<1)+(lum_m1[(ptroff)]<<2)-lum[(ptroff)]); \
641                     lum_m4[(ptroff)]=lum_m2[(ptroff)];\
642                     lum_m2[(ptroff)]=cm[(sum+4)>>3];\
643                     sum=(-lum_m4[(ptroff)+1]+(lum_m3[(ptroff)+1]<<2)+(lum_m2[(ptroff)+1]<<1)+(lum_m1[(ptroff)+1]<<2)-lum[(ptroff)+1]); \
644                     lum_m4[(ptroff)+1]=lum_m2[(ptroff)+1];\
645                     lum_m2[(ptroff)+1]=cm[(sum+4)>>3];\
646                     sum=(-lum_m4[(ptroff)+2]+(lum_m3[(ptroff)+2]<<2)+(lum_m2[(ptroff)+2]<<1)+(lum_m1[(ptroff)+2]<<2)-lum[(ptroff)+2]); \
647                     lum_m4[(ptroff)+2]=lum_m2[(ptroff)+2];\
648                     lum_m2[(ptroff)+2]=cm[(sum+4)>>3];\
649                     sum=(-lum_m4[(ptroff)+3]+(lum_m3[(ptroff)+3]<<2)+(lum_m2[(ptroff)+3]<<1)+(lum_m1[(ptroff)+3]<<2)-lum[(ptroff)+3]); \
650                     lum_m4[(ptroff)+3]=lum_m2[(ptroff)+3];\
651                     lum_m2[(ptroff)+3]=cm[(sum+4)>>3];
652
653 #endif
654
655
656 /* Read two fields separately. */
657 static int aiw_read_picture(VideoData *s, uint8_t *data)
658 {
659     uint8_t *ptr, *lum, *cb, *cr;
660     int h;
661 #ifndef HAVE_MMX
662     int sum;
663 #endif
664     uint8_t* src = s->src_mem;
665     uint8_t *ptrend = &src[s->width*2];
666     lum=data;
667     cb=&lum[s->width*s->height];
668     cr=&cb[(s->width*s->height)/4];
669     if (s->deint == 0 && s->halfw == 0) {
670         while (read(s->fd,src,s->width*2) < 0) {
671             usleep(100);
672         }
673         for (h = 0; h < s->height-2; h+=2) {
674             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
675                 LINE_WITH_UV
676                     }
677             read(s->fd,src,s->width*2);
678             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
679                 LINE_NO_UV
680                     }
681             read(s->fd,src,s->width*2);
682         }
683         /*
684          * Do last two lines
685          */
686         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
687             LINE_WITH_UV
688                 }
689         read(s->fd,src,s->width*2);
690         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
691             LINE_NO_UV
692                 }
693         /* drop second field */
694         while (read(s->fd,src,s->width*2) < 0) {
695             usleep(100);
696         }
697         for (h = 0; h < s->height - 1; h++) {
698             read(s->fd,src,s->width*2);
699         }
700     } else if (s->halfw == 1) {
701 #ifdef HAVE_MMX
702         mmx_t rounder;
703         mmx_t masker;
704         rounder.uw[0]=1;
705         rounder.uw[1]=1;
706         rounder.uw[2]=1;
707         rounder.uw[3]=1;
708         masker.ub[0]=0xff;
709         masker.ub[1]=0;
710         masker.ub[2]=0xff;
711         masker.ub[3]=0;
712         masker.ub[4]=0xff;
713         masker.ub[5]=0;
714         masker.ub[6]=0xff;
715         masker.ub[7]=0;
716         pxor_r2r(mm7,mm7);
717         movq_m2r(rounder,mm6);
718 #endif
719         while (read(s->fd,src,s->width*4) < 0) {
720             usleep(100);
721         }
722         ptrend = &src[s->width*4];
723         for (h = 0; h < s->height-2; h+=2) {
724             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
725                 LINE_WITHUV_AVG
726                     }
727             read(s->fd,src,s->width*4);
728 #ifdef HAVE_MMX
729             movq_m2r(masker,mm5);
730 #endif
731             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
732                 LINE_NOUV_AVG
733                     }
734             read(s->fd,src,s->width*4);
735         }
736         /*
737  * Do last two lines
738  */
739         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
740             LINE_WITHUV_AVG
741                 }
742         read(s->fd,src,s->width*4);
743 #ifdef HAVE_MMX
744         movq_m2r(masker,mm5);
745 #endif
746         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
747             LINE_NOUV_AVG
748                 }
749         /* drop second field */
750         while (read(s->fd,src,s->width*4) < 0) {
751             usleep(100);
752         }
753         for (h = 0; h < s->height - 1; h++) {
754             read(s->fd,src,s->width*4);
755         }
756     } else {
757         uint8_t *lum_m1, *lum_m2, *lum_m3, *lum_m4;
758 #ifdef HAVE_MMX
759         mmx_t rounder;
760         rounder.uw[0]=4;
761         rounder.uw[1]=4;
762         rounder.uw[2]=4;
763         rounder.uw[3]=4;
764         movq_m2r(rounder,mm6);
765         pxor_r2r(mm7,mm7);
766 #else
767         uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
768 #endif
769
770         /* read two fields and deinterlace them */
771         while (read(s->fd,src,s->width*2) < 0) {
772             usleep(100);
773         }
774         for (h = 0; h < (s->height/2)-2; h+=2) {
775             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
776                 LINE_WITH_UV
777                     }
778             read(s->fd,src,s->width*2);
779             /* skip a luminance line - will be filled in later */
780             lum += s->width;
781             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
782                 LINE_WITH_UV
783                     }
784             /* skip a luminance line - will be filled in later */
785             lum += s->width;
786             read(s->fd,src,s->width*2);
787         }
788         /*
789  * Do last two lines
790  */
791         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
792             LINE_WITH_UV
793                 }
794         /* skip a luminance line - will be filled in later */
795         lum += s->width;
796         read(s->fd,src,s->width*2);
797         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
798             LINE_WITH_UV
799                 }
800         /*
801  *
802  * SECOND FIELD
803  *
804  */
805         lum=&data[s->width];
806         while (read(s->fd,src,s->width*2) < 0) {
807             usleep(10);
808         }
809         /* First (and last) two lines not interlaced */
810         for (h = 0; h < 2; h++) {
811             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
812                 LINE_NO_UV
813                     }
814             read(s->fd,src,s->width*2);
815             /* skip a luminance line */
816             lum += s->width;
817         }
818         lum_m1=&lum[-s->width];
819         lum_m2=&lum_m1[-s->width];
820         lum_m3=&lum_m2[-s->width];
821         memmove(s->lum_m4_mem,&lum_m3[-s->width],s->width);
822         for (; h < (s->height/2)-1; h++) {
823             lum_m4=s->lum_m4_mem;
824             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16,lum_m1+=16,lum_m2+=16,lum_m3+=16,lum_m4+=16) {
825                 LINE_NO_UV
826
827                     DEINT_LINE_LUM(0)
828                     DEINT_LINE_LUM(4)
829                     DEINT_LINE_LUM(8)
830                     DEINT_LINE_LUM(12)
831                     }
832             read(s->fd,src,s->width*2);
833             /* skip a luminance line */
834             lum += s->width;
835             lum_m1 += s->width;
836             lum_m2 += s->width;
837             lum_m3 += s->width;
838             //                lum_m4 += s->width;
839         }
840         /*
841  * Do last line
842  */
843         lum_m4=s->lum_m4_mem;
844         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, lum_m1+=16, lum_m2+=16, lum_m3+=16, lum_m4+=16) {
845             LINE_NO_UV
846
847                 DEINT_LINE_LUM(0)
848                 DEINT_LINE_LUM(4)
849                 DEINT_LINE_LUM(8)
850                 DEINT_LINE_LUM(12)
851                 }
852     }
853 #ifdef HAVE_MMX
854     emms();
855 #endif
856     return s->frame_size;
857 }
858
859 static int aiw_close(VideoData *s)
860 {
861     av_freep(&s->lum_m4_mem);
862     av_freep(&s->src_mem);
863     return 0;
864 }