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     ret = ioctl(video_fd,VIDIOCGMBUF,&s->gb_buffers);
 188     if (ret < 0) {
 189         /* try to use read based access */
 190         struct video_window win;
 191         int val;
 192
 193         win.x = 0;
 194         win.y = 0;
 195         win.width = width;
 196         win.height = height;
 197         win.chromakey = -1;
 198         win.flags = 0;
 199
 200         ioctl(video_fd, VIDIOCSWIN, &win);
 201
 202         s->frame_format = pict.palette;
 203
 204         val = 1;
 205         ioctl(video_fd, VIDIOCCAPTURE, &val);
 206
 207         s->time_frame = av_gettime() * s->frame_rate / s->frame_rate_base;
 208         s->use_mmap = 0;
 209
 210         /* ATI All In Wonder automatic activation */
 211         if (!strcmp(s->video_cap.name, "Km")) {
 212             if (aiw_init(s) < 0)
 213                 goto fail;
 214             s->aiw_enabled = 1;
 215             /* force 420P format because convertion from YUV422 to YUV420P
 216                is done in this driver (ugly) */
 217             s->frame_format = VIDEO_PALETTE_YUV420P;
 218         }
 219     } else {
 220         s->video_buf = mmap(0,s->gb_buffers.size,PROT_READ|PROT_WRITE,MAP_SHARED,video_fd,0);
 221         if ((unsigned char*)-1 == s->video_buf) {
 222             s->video_buf = mmap(0,s->gb_buffers.size,PROT_READ|PROT_WRITE,MAP_PRIVATE,video_fd,0);
 223             if ((unsigned char*)-1 == s->video_buf) {
 224                 perror("mmap");
 225                 goto fail;
 226             }
 227         }
 228         s->gb_frame = 0;
 229         s->time_frame = av_gettime() * s->frame_rate / s->frame_rate_base;
 230
 231         /* start to grab the first frame */
 232         s->gb_buf.frame = s->gb_frame % s->gb_buffers.frames;
 233         s->gb_buf.height = height;
 234         s->gb_buf.width = width;
 235         s->gb_buf.format = pict.palette;
 236
 237         ret = ioctl(video_fd, VIDIOCMCAPTURE, &s->gb_buf);
 238         if (ret < 0) {
 239             if (errno != EAGAIN) {
 240             fail1:
 241                 av_log(s1, AV_LOG_ERROR, "Fatal: grab device does not support suitable format\n");
 242             } else {
 243                 av_log(s1, AV_LOG_ERROR,"Fatal: grab device does not receive any video signal\n");
 244             }
 245             goto fail;
 246         }
 247         for (j = 1; j < s->gb_buffers.frames; j++) {
 248           s->gb_buf.frame = j;
 249           ioctl(video_fd, VIDIOCMCAPTURE, &s->gb_buf);
 250         }
 251         s->frame_format = s->gb_buf.format;
 252         s->use_mmap = 1;
 253     }
 254
 255     switch(s->frame_format) {
 256     case VIDEO_PALETTE_YUV420P:
 257         frame_size = (width * height * 3) / 2;
 258         st->codec->pix_fmt = PIX_FMT_YUV420P;
 259         break;
 260     case VIDEO_PALETTE_YUV422:
 261         frame_size = width * height * 2;
 262         st->codec->pix_fmt = PIX_FMT_YUV422;
 263         break;
 264     case VIDEO_PALETTE_RGB24:
 265         frame_size = width * height * 3;
 266         st->codec->pix_fmt = PIX_FMT_BGR24; /* NOTE: v4l uses BGR24, not RGB24 ! */
 267         break;
 268     case VIDEO_PALETTE_GREY:
 269         frame_size = width * height * 1;
 270         st->codec->pix_fmt = PIX_FMT_GRAY8;
 271         break;
 272     default:
 273         goto fail;
 274     }
 275     s->fd = video_fd;
 276     s->frame_size = frame_size;
 277
 278     st->codec->codec_type = CODEC_TYPE_VIDEO;
 279     st->codec->codec_id = CODEC_ID_RAWVIDEO;
 280     st->codec->width = width;
 281     st->codec->height = height;
 282     st->codec->time_base.den      = frame_rate;
 283     st->codec->time_base.num = frame_rate_base;
 284     st->codec->bit_rate = frame_size * 1/av_q2d(st->codec->time_base) * 8;
 285
 286     return 0;
 287  fail:
 288     if (video_fd >= 0)
 289         close(video_fd);
 290     av_free(st);
 291     return AVERROR_IO;
 292 }
 293
 294 static int v4l_mm_read_picture(VideoData *s, uint8_t *buf)
 295 {
 296     uint8_t *ptr;
 297
 298     while (ioctl(s->fd, VIDIOCSYNC, &s->gb_frame) < 0 &&
 299            (errno == EAGAIN || errno == EINTR));
 300
 301     ptr = s->video_buf + s->gb_buffers.offsets[s->gb_frame];
 302     memcpy(buf, ptr, s->frame_size);
 303
 304     /* Setup to capture the next frame */
 305     s->gb_buf.frame = s->gb_frame;
 306     if (ioctl(s->fd, VIDIOCMCAPTURE, &s->gb_buf) < 0) {
 307         if (errno == EAGAIN)
 308             av_log(NULL, AV_LOG_ERROR, "Cannot Sync\n");
 309         else
 310             perror("VIDIOCMCAPTURE");
 311         return AVERROR_IO;
 312     }
 313
 314     /* This is now the grabbing frame */
 315     s->gb_frame = (s->gb_frame + 1) % s->gb_buffers.frames;
 316
 317     return s->frame_size;
 318 }
 319
 320 static int grab_read_packet(AVFormatContext *s1, AVPacket *pkt)
 321 {
 322     VideoData *s = s1->priv_data;
 323     int64_t curtime, delay;
 324     struct timespec ts;
 325
 326     /* Calculate the time of the next frame */
 327     s->time_frame += INT64_C(1000000);
 328
 329     /* wait based on the frame rate */
 330     for(;;) {
 331         curtime = av_gettime();
 332         delay = s->time_frame  * s->frame_rate_base / s->frame_rate - curtime;
 333         if (delay <= 0) {
 334             if (delay < INT64_C(-1000000) * s->frame_rate_base / s->frame_rate) {
 335                 /* printf("grabbing is %d frames late (dropping)\n", (int) -(delay / 16666)); */
 336                 s->time_frame += INT64_C(1000000);
 337             }
 338             break;
 339         }
 340         ts.tv_sec = delay / 1000000;
 341         ts.tv_nsec = (delay % 1000000) * 1000;
 342         nanosleep(&ts, NULL);
 343     }
 344
 345     if (av_new_packet(pkt, s->frame_size) < 0)
 346         return AVERROR_IO;
 347
 348     pkt->pts = curtime;
 349
 350     /* read one frame */
 351     if (s->aiw_enabled) {
 352         return aiw_read_picture(s, pkt->data);
 353     } else if (s->use_mmap) {
 354         return v4l_mm_read_picture(s, pkt->data);
 355     } else {
 356         if (read(s->fd, pkt->data, pkt->size) != pkt->size)
 357             return AVERROR_IO;
 358         return s->frame_size;
 359     }
 360 }
 361
 362 static int grab_read_close(AVFormatContext *s1)
 363 {
 364     VideoData *s = s1->priv_data;
 365
 366     if (s->aiw_enabled)
 367         aiw_close(s);
 368
 369     if (s->use_mmap)
 370         munmap(s->video_buf, s->gb_buffers.size);
 371
 372     /* mute audio. we must force it because the BTTV driver does not
 373        return its state correctly */
 374     s->audio_saved.flags |= VIDEO_AUDIO_MUTE;
 375     ioctl(s->fd, VIDIOCSAUDIO, &s->audio_saved);
 376
 377     close(s->fd);
 378     return 0;
 379 }
 380
 381 AVInputFormat video_grab_device_demuxer = {
 382     "video4linux",
 383     "video grab",
 384     sizeof(VideoData),
 385     NULL,
 386     grab_read_header,
 387     grab_read_packet,
 388     grab_read_close,
 389     .flags = AVFMT_NOFILE,
 390 };
 391
 392 /* All in Wonder specific stuff */
 393 /* XXX: remove and merge in libavcodec/imgconvert.c */
 394
 395 static int aiw_init(VideoData *s)
 396 {
 397     int width, height;
 398
 399     width = s->width;
 400     height = s->height;
 401
 402     if ((width == s->video_cap.maxwidth && height == s->video_cap.maxheight) ||
 403         (width == s->video_cap.maxwidth && height == s->video_cap.maxheight*2) ||
 404         (width == s->video_cap.maxwidth/2 && height == s->video_cap.maxheight)) {
 405
 406         s->deint=0;
 407         s->halfw=0;
 408         if (height == s->video_cap.maxheight*2) s->deint=1;
 409         if (width == s->video_cap.maxwidth/2) s->halfw=1;
 410     } else {
 411         av_log(NULL, AV_LOG_ERROR, "\nIncorrect Grab Size Supplied - Supported Sizes Are:\n");
 412         av_log(NULL, AV_LOG_ERROR, " %dx%d  %dx%d %dx%d\n\n",
 413                 s->video_cap.maxwidth,s->video_cap.maxheight,
 414                 s->video_cap.maxwidth,s->video_cap.maxheight*2,
 415                 s->video_cap.maxwidth/2,s->video_cap.maxheight);
 416         goto fail;
 417     }
 418
 419     if (s->halfw == 0) {
 420         s->src_mem = av_malloc(s->width*2);
 421     } else {
 422         s->src_mem = av_malloc(s->width*4);
 423     }
 424     if (!s->src_mem) goto fail;
 425
 426     s->lum_m4_mem = av_malloc(s->width);
 427     if (!s->lum_m4_mem)
 428         goto fail;
 429     return 0;
 430  fail:
 431     av_freep(&s->src_mem);
 432     av_freep(&s->lum_m4_mem);
 433     return -1;
 434 }
 435
 436 #ifdef HAVE_MMX
 437 #include "libavcodec/i386/mmx.h"
 438
 439 #define LINE_WITH_UV \
 440                     movq_m2r(ptr[0],mm0); \
 441                     movq_m2r(ptr[8],mm1);  \
 442                     movq_r2r(mm0, mm4); \
 443                     punpcklbw_r2r(mm1,mm0); \
 444                     punpckhbw_r2r(mm1,mm4); \
 445                     movq_r2r(mm0,mm5); \
 446                     punpcklbw_r2r(mm4,mm0); \
 447                     punpckhbw_r2r(mm4,mm5); \
 448                     movq_r2r(mm0,mm1); \
 449                     punpcklbw_r2r(mm5,mm1); \
 450                     movq_r2m(mm1,lum[0]); \
 451                     movq_m2r(ptr[16],mm2); \
 452                     movq_m2r(ptr[24],mm1); \
 453                     movq_r2r(mm2,mm4); \
 454                     punpcklbw_r2r(mm1,mm2); \
 455                     punpckhbw_r2r(mm1,mm4); \
 456                     movq_r2r(mm2,mm3); \
 457                     punpcklbw_r2r(mm4,mm2); \
 458                     punpckhbw_r2r(mm4,mm3); \
 459                     movq_r2r(mm2,mm1); \
 460                     punpcklbw_r2r(mm3,mm1); \
 461                     movq_r2m(mm1,lum[8]); \
 462                     punpckhdq_r2r(mm2,mm0); \
 463                     punpckhdq_r2r(mm3,mm5); \
 464                     movq_r2m(mm0,cb[0]); \
 465                     movq_r2m(mm5,cr[0]);
 466
 467 #define LINE_NO_UV \
 468                     movq_m2r(ptr[0],mm0);\
 469                     movq_m2r(ptr[8],mm1);\
 470                     movq_r2r(mm0, mm4);\
 471                     punpcklbw_r2r(mm1,mm0); \
 472                     punpckhbw_r2r(mm1,mm4);\
 473                     movq_r2r(mm0,mm5);\
 474                     punpcklbw_r2r(mm4,mm0);\
 475                     punpckhbw_r2r(mm4,mm5);\
 476                     movq_r2r(mm0,mm1);\
 477                     punpcklbw_r2r(mm5,mm1);\
 478                     movq_r2m(mm1,lum[0]);\
 479                     movq_m2r(ptr[16],mm2);\
 480                     movq_m2r(ptr[24],mm1);\
 481                     movq_r2r(mm2,mm4);\
 482                     punpcklbw_r2r(mm1,mm2);\
 483                     punpckhbw_r2r(mm1,mm4);\
 484                     movq_r2r(mm2,mm3);\
 485                     punpcklbw_r2r(mm4,mm2);\
 486                     punpckhbw_r2r(mm4,mm3);\
 487                     movq_r2r(mm2,mm1);\
 488                     punpcklbw_r2r(mm3,mm1);\
 489                     movq_r2m(mm1,lum[8]);
 490
 491 #define LINE_WITHUV_AVG \
 492                     movq_m2r(ptr[0], mm0);\
 493                     movq_m2r(ptr[8], mm1);\
 494                     movq_r2r(mm0, mm4);\
 495                     punpcklbw_r2r(mm1,mm0);\
 496                     punpckhbw_r2r(mm1,mm4);\
 497                     movq_r2r(mm0,mm5);\
 498                     punpcklbw_r2r(mm4,mm0);\
 499                     punpckhbw_r2r(mm4,mm5);\
 500                     movq_r2r(mm0,mm1);\
 501                     movq_r2r(mm5,mm2);\
 502                     punpcklbw_r2r(mm7,mm1);\
 503                     punpcklbw_r2r(mm7,mm2);\
 504                     paddw_r2r(mm6,mm1);\
 505                     paddw_r2r(mm2,mm1);\
 506                     psraw_i2r(1,mm1);\
 507                     packuswb_r2r(mm7,mm1);\
 508                     movd_r2m(mm1,lum[0]);\
 509                     movq_m2r(ptr[16],mm2);\
 510                     movq_m2r(ptr[24],mm1);\
 511                     movq_r2r(mm2,mm4);\
 512                     punpcklbw_r2r(mm1,mm2);\
 513                     punpckhbw_r2r(mm1,mm4);\
 514                     movq_r2r(mm2,mm3);\
 515                     punpcklbw_r2r(mm4,mm2);\
 516                     punpckhbw_r2r(mm4,mm3);\
 517                     movq_r2r(mm2,mm1);\
 518                     movq_r2r(mm3,mm4);\
 519                     punpcklbw_r2r(mm7,mm1);\
 520                     punpcklbw_r2r(mm7,mm4);\
 521                     paddw_r2r(mm6,mm1);\
 522                     paddw_r2r(mm4,mm1);\
 523                     psraw_i2r(1,mm1);\
 524                     packuswb_r2r(mm7,mm1);\
 525                     movd_r2m(mm1,lum[4]);\
 526                     punpckhbw_r2r(mm7,mm0);\
 527                     punpckhbw_r2r(mm7,mm2);\
 528                     paddw_r2r(mm6,mm0);\
 529                     paddw_r2r(mm2,mm0);\
 530                     psraw_i2r(1,mm0);\
 531                     packuswb_r2r(mm7,mm0);\
 532                     punpckhbw_r2r(mm7,mm5);\
 533                     punpckhbw_r2r(mm7,mm3);\
 534                     paddw_r2r(mm6,mm5);\
 535                     paddw_r2r(mm3,mm5);\
 536                     psraw_i2r(1,mm5);\
 537                     packuswb_r2r(mm7,mm5);\
 538                     movd_r2m(mm0,cb[0]);\
 539                     movd_r2m(mm5,cr[0]);
 540
 541 #define LINE_NOUV_AVG \
 542                     movq_m2r(ptr[0],mm0);\
 543                     movq_m2r(ptr[8],mm1);\
 544                     pand_r2r(mm5,mm0);\
 545                     pand_r2r(mm5,mm1);\
 546                     pmaddwd_r2r(mm6,mm0);\
 547                     pmaddwd_r2r(mm6,mm1);\
 548                     packssdw_r2r(mm1,mm0);\
 549                     paddw_r2r(mm6,mm0);\
 550                     psraw_i2r(1,mm0);\
 551                     movq_m2r(ptr[16],mm2);\
 552                     movq_m2r(ptr[24],mm3);\
 553                     pand_r2r(mm5,mm2);\
 554                     pand_r2r(mm5,mm3);\
 555                     pmaddwd_r2r(mm6,mm2);\
 556                     pmaddwd_r2r(mm6,mm3);\
 557                     packssdw_r2r(mm3,mm2);\
 558                     paddw_r2r(mm6,mm2);\
 559                     psraw_i2r(1,mm2);\
 560                     packuswb_r2r(mm2,mm0);\
 561                     movq_r2m(mm0,lum[0]);
 562
 563 #define DEINT_LINE_LUM(ptroff) \
 564                     movd_m2r(lum_m4[(ptroff)],mm0);\
 565                     movd_m2r(lum_m3[(ptroff)],mm1);\
 566                     movd_m2r(lum_m2[(ptroff)],mm2);\
 567                     movd_m2r(lum_m1[(ptroff)],mm3);\
 568                     movd_m2r(lum[(ptroff)],mm4);\
 569                     punpcklbw_r2r(mm7,mm0);\
 570                     movd_r2m(mm2,lum_m4[(ptroff)]);\
 571                     punpcklbw_r2r(mm7,mm1);\
 572                     punpcklbw_r2r(mm7,mm2);\
 573                     punpcklbw_r2r(mm7,mm3);\
 574                     punpcklbw_r2r(mm7,mm4);\
 575                     psllw_i2r(2,mm1);\
 576                     psllw_i2r(1,mm2);\
 577                     paddw_r2r(mm6,mm1);\
 578                     psllw_i2r(2,mm3);\
 579                     paddw_r2r(mm2,mm1);\
 580                     paddw_r2r(mm4,mm0);\
 581                     paddw_r2r(mm3,mm1);\
 582                     psubusw_r2r(mm0,mm1);\
 583                     psrlw_i2r(3,mm1);\
 584                     packuswb_r2r(mm7,mm1);\
 585                     movd_r2m(mm1,lum_m2[(ptroff)]);
 586
 587 #else
 588 #include "libavcodec/dsputil.h"
 589
 590 #define LINE_WITH_UV \
 591                     lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
 592                     cb[0]=ptr[1];cb[1]=ptr[5];\
 593                     cr[0]=ptr[3];cr[1]=ptr[7];\
 594                     lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
 595                     cb[2]=ptr[9];cb[3]=ptr[13];\
 596                     cr[2]=ptr[11];cr[3]=ptr[15];\
 597                     lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
 598                     cb[4]=ptr[17];cb[5]=ptr[21];\
 599                     cr[4]=ptr[19];cr[5]=ptr[23];\
 600                     lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];\
 601                     cb[6]=ptr[25];cb[7]=ptr[29];\
 602                     cr[6]=ptr[27];cr[7]=ptr[31];
 603
 604 #define LINE_NO_UV \
 605                     lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
 606                     lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
 607                     lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
 608                     lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];
 609
 610 #define LINE_WITHUV_AVG \
 611                     sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
 612                     sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
 613                     sum=(ptr[1]+ptr[5]+1) >> 1;cb[0]=sum; \
 614                     sum=(ptr[3]+ptr[7]+1) >> 1;cr[0]=sum; \
 615                     sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
 616                     sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
 617                     sum=(ptr[9]+ptr[13]+1) >> 1;cb[1]=sum; \
 618                     sum=(ptr[11]+ptr[15]+1) >> 1;cr[1]=sum; \
 619                     sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
 620                     sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
 621                     sum=(ptr[17]+ptr[21]+1) >> 1;cb[2]=sum; \
 622                     sum=(ptr[19]+ptr[23]+1) >> 1;cr[2]=sum; \
 623                     sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
 624                     sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum; \
 625                     sum=(ptr[25]+ptr[29]+1) >> 1;cb[3]=sum; \
 626                     sum=(ptr[27]+ptr[31]+1) >> 1;cr[3]=sum;
 627
 628 #define LINE_NOUV_AVG \
 629                     sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
 630                     sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
 631                     sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
 632                     sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
 633                     sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
 634                     sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
 635                     sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
 636                     sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum;
 637
 638 #define DEINT_LINE_LUM(ptroff) \
 639                     sum=(-lum_m4[(ptroff)]+(lum_m3[(ptroff)]<<2)+(lum_m2[(ptroff)]<<1)+(lum_m1[(ptroff)]<<2)-lum[(ptroff)]); \
 640                     lum_m4[(ptroff)]=lum_m2[(ptroff)];\
 641                     lum_m2[(ptroff)]=cm[(sum+4)>>3];\
 642                     sum=(-lum_m4[(ptroff)+1]+(lum_m3[(ptroff)+1]<<2)+(lum_m2[(ptroff)+1]<<1)+(lum_m1[(ptroff)+1]<<2)-lum[(ptroff)+1]); \
 643                     lum_m4[(ptroff)+1]=lum_m2[(ptroff)+1];\
 644                     lum_m2[(ptroff)+1]=cm[(sum+4)>>3];\
 645                     sum=(-lum_m4[(ptroff)+2]+(lum_m3[(ptroff)+2]<<2)+(lum_m2[(ptroff)+2]<<1)+(lum_m1[(ptroff)+2]<<2)-lum[(ptroff)+2]); \
 646                     lum_m4[(ptroff)+2]=lum_m2[(ptroff)+2];\
 647                     lum_m2[(ptroff)+2]=cm[(sum+4)>>3];\
 648                     sum=(-lum_m4[(ptroff)+3]+(lum_m3[(ptroff)+3]<<2)+(lum_m2[(ptroff)+3]<<1)+(lum_m1[(ptroff)+3]<<2)-lum[(ptroff)+3]); \
 649                     lum_m4[(ptroff)+3]=lum_m2[(ptroff)+3];\
 650                     lum_m2[(ptroff)+3]=cm[(sum+4)>>3];
 651
 652 #endif
 653
 654
 655 /* Read two fields separately. */
 656 static int aiw_read_picture(VideoData *s, uint8_t *data)
 657 {
 658     uint8_t *ptr, *lum, *cb, *cr;
 659     int h;
 660 #ifndef HAVE_MMX
 661     int sum;
 662 #endif
 663     uint8_t* src = s->src_mem;
 664     uint8_t *ptrend = &src[s->width*2];
 665     lum=data;
 666     cb=&lum[s->width*s->height];
 667     cr=&cb[(s->width*s->height)/4];
 668     if (s->deint == 0 && s->halfw == 0) {
 669         while (read(s->fd,src,s->width*2) < 0) {
 670             usleep(100);
 671         }
 672         for (h = 0; h < s->height-2; h+=2) {
 673             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 674                 LINE_WITH_UV
 675                     }
 676             read(s->fd,src,s->width*2);
 677             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
 678                 LINE_NO_UV
 679                     }
 680             read(s->fd,src,s->width*2);
 681         }
 682         /*
 683          * Do last two lines
 684          */
 685         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 686             LINE_WITH_UV
 687                 }
 688         read(s->fd,src,s->width*2);
 689         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
 690             LINE_NO_UV
 691                 }
 692         /* drop second field */
 693         while (read(s->fd,src,s->width*2) < 0) {
 694             usleep(100);
 695         }
 696         for (h = 0; h < s->height - 1; h++) {
 697             read(s->fd,src,s->width*2);
 698         }
 699     } else if (s->halfw == 1) {
 700 #ifdef HAVE_MMX
 701         mmx_t rounder;
 702         mmx_t masker;
 703         rounder.uw[0]=1;
 704         rounder.uw[1]=1;
 705         rounder.uw[2]=1;
 706         rounder.uw[3]=1;
 707         masker.ub[0]=0xff;
 708         masker.ub[1]=0;
 709         masker.ub[2]=0xff;
 710         masker.ub[3]=0;
 711         masker.ub[4]=0xff;
 712         masker.ub[5]=0;
 713         masker.ub[6]=0xff;
 714         masker.ub[7]=0;
 715         pxor_r2r(mm7,mm7);
 716         movq_m2r(rounder,mm6);
 717 #endif
 718         while (read(s->fd,src,s->width*4) < 0) {
 719             usleep(100);
 720         }
 721         ptrend = &src[s->width*4];
 722         for (h = 0; h < s->height-2; h+=2) {
 723             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
 724                 LINE_WITHUV_AVG
 725                     }
 726             read(s->fd,src,s->width*4);
 727 #ifdef HAVE_MMX
 728             movq_m2r(masker,mm5);
 729 #endif
 730             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
 731                 LINE_NOUV_AVG
 732                     }
 733             read(s->fd,src,s->width*4);
 734         }
 735         /*
 736  * Do last two lines
 737  */
 738         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
 739             LINE_WITHUV_AVG
 740                 }
 741         read(s->fd,src,s->width*4);
 742 #ifdef HAVE_MMX
 743         movq_m2r(masker,mm5);
 744 #endif
 745         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
 746             LINE_NOUV_AVG
 747                 }
 748         /* drop second field */
 749         while (read(s->fd,src,s->width*4) < 0) {
 750             usleep(100);
 751         }
 752         for (h = 0; h < s->height - 1; h++) {
 753             read(s->fd,src,s->width*4);
 754         }
 755     } else {
 756         uint8_t *lum_m1, *lum_m2, *lum_m3, *lum_m4;
 757 #ifdef HAVE_MMX
 758         mmx_t rounder;
 759         rounder.uw[0]=4;
 760         rounder.uw[1]=4;
 761         rounder.uw[2]=4;
 762         rounder.uw[3]=4;
 763         movq_m2r(rounder,mm6);
 764         pxor_r2r(mm7,mm7);
 765 #else
 766         uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 767 #endif
 768
 769         /* read two fields and deinterlace them */
 770         while (read(s->fd,src,s->width*2) < 0) {
 771             usleep(100);
 772         }
 773         for (h = 0; h < (s->height/2)-2; h+=2) {
 774             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 775                 LINE_WITH_UV
 776                     }
 777             read(s->fd,src,s->width*2);
 778             /* skip a luminance line - will be filled in later */
 779             lum += s->width;
 780             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 781                 LINE_WITH_UV
 782                     }
 783             /* skip a luminance line - will be filled in later */
 784             lum += s->width;
 785             read(s->fd,src,s->width*2);
 786         }
 787         /*
 788  * Do last two lines
 789  */
 790         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 791             LINE_WITH_UV
 792                 }
 793         /* skip a luminance line - will be filled in later */
 794         lum += s->width;
 795         read(s->fd,src,s->width*2);
 796         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
 797             LINE_WITH_UV
 798                 }
 799         /*
 800  *
 801  * SECOND FIELD
 802  *
 803  */
 804         lum=&data[s->width];
 805         while (read(s->fd,src,s->width*2) < 0) {
 806             usleep(10);
 807         }
 808         /* First (and last) two lines not interlaced */
 809         for (h = 0; h < 2; h++) {
 810             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
 811                 LINE_NO_UV
 812                     }
 813             read(s->fd,src,s->width*2);
 814             /* skip a luminance line */
 815             lum += s->width;
 816         }
 817         lum_m1=&lum[-s->width];
 818         lum_m2=&lum_m1[-s->width];
 819         lum_m3=&lum_m2[-s->width];
 820         memmove(s->lum_m4_mem,&lum_m3[-s->width],s->width);
 821         for (; h < (s->height/2)-1; h++) {
 822             lum_m4=s->lum_m4_mem;
 823             for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16,lum_m1+=16,lum_m2+=16,lum_m3+=16,lum_m4+=16) {
 824                 LINE_NO_UV
 825
 826                     DEINT_LINE_LUM(0)
 827                     DEINT_LINE_LUM(4)
 828                     DEINT_LINE_LUM(8)
 829                     DEINT_LINE_LUM(12)
 830                     }
 831             read(s->fd,src,s->width*2);
 832             /* skip a luminance line */
 833             lum += s->width;
 834             lum_m1 += s->width;
 835             lum_m2 += s->width;
 836             lum_m3 += s->width;
 837             //                lum_m4 += s->width;
 838         }
 839         /*
 840  * Do last line
 841  */
 842         lum_m4=s->lum_m4_mem;
 843         for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, lum_m1+=16, lum_m2+=16, lum_m3+=16, lum_m4+=16) {
 844             LINE_NO_UV
 845
 846                 DEINT_LINE_LUM(0)
 847                 DEINT_LINE_LUM(4)
 848                 DEINT_LINE_LUM(8)
 849                 DEINT_LINE_LUM(12)
 850                 }
 851     }
 852 #ifdef HAVE_MMX
 853     emms();
 854 #endif
 855     return s->frame_size;
 856 }
 857
 858 static int aiw_close(VideoData *s)
 859 {
 860     av_freep(&s->lum_m4_mem);
 861     av_freep(&s->src_mem);
 862     return 0;
 863 }