new video grab device for ATI All-In-Wonder cards based on the GATOS km driver patch...
authorMichael Niedermayer <michaelni@gmx.at>
Sun, 20 Oct 2002 21:45:38 +0000 (21:45 +0000)
committerMichael Niedermayer <michaelni@gmx.at>
Sun, 20 Oct 2002 21:45:38 +0000 (21:45 +0000)
Originally committed as revision 1053 to svn://svn.ffmpeg.org/ffmpeg/trunk

libav/grab.c

index 20cade3..c19d2c4 100644 (file)
@@ -323,8 +323,658 @@ AVInputFormat video_grab_device_format = {
     .flags = AVFMT_NOFILE,
 };
 
+/*
+ * Done below so we can register the aiw grabber
+ * /
+int video_grab_init(void)
+{
+    av_register_input_format(&video_grab_device_format);
+    return 0;
+}
+*/
+
+typedef struct {
+    int fd;
+    int frame_format; /* see VIDEO_PALETTE_xxx */
+    int width, height;
+    int frame_rate;
+    INT64 time_frame;
+    int frame_size;
+    int deint;
+    int halfw;
+    UINT8 *src_mem;
+    UINT8 *lum_m4_mem;
+} AIWVideoData;
+
+static int aiw_grab_read_header(AVFormatContext *s1, AVFormatParameters *ap)
+{
+    AIWVideoData *s = s1->priv_data;
+    AVStream *st;
+    int width, height;
+    int video_fd, frame_size;
+    int ret, frame_rate;
+    int desired_palette;
+
+    if (!ap || ap->width <= 0 || ap->height <= 0 || ap->frame_rate <= 0)
+        return -1;
+    
+    width = ap->width;
+    height = ap->height;
+    frame_rate = ap->frame_rate;
+
+    st = av_new_stream(s1, 0);
+    if (!st)
+        return -ENOMEM;
+
+    s->width = width;
+    s->height = height;
+    s->frame_rate = frame_rate;
+
+    video_fd = open(v4l_device, O_RDONLY | O_NONBLOCK);
+    if (video_fd < 0) {
+        perror(v4l_device);
+        goto fail;
+    }
+    
+    if (ioctl(video_fd,VIDIOCGCAP,&video_cap) < 0) {
+        perror("VIDIOCGCAP");
+        goto fail;
+    }
+
+    if (!(video_cap.type & VID_TYPE_CAPTURE)) {
+        fprintf(stderr, "Fatal: grab device does not handle capture\n");
+        goto fail;
+    }
+
+    desired_palette = -1;
+    if (st->codec.pix_fmt == PIX_FMT_YUV420P) {
+        desired_palette = VIDEO_PALETTE_YUV420P;
+    } else if (st->codec.pix_fmt == PIX_FMT_YUV422) {
+        desired_palette = VIDEO_PALETTE_YUV422;
+    } else if (st->codec.pix_fmt == PIX_FMT_BGR24) {
+        desired_palette = VIDEO_PALETTE_RGB24;
+    }    
+    
+    /* unmute audio */
+
+    ret = ioctl(video_fd,VIDIOCGMBUF,&gb_buffers);
+    if (ret < 0) {
+        /* try to use read based access */
+        struct video_window win;
+        struct video_picture pict;
+        int val;
+
+        win.x = 0;
+        win.y = 0;
+        win.width = width;
+        win.height = height;
+        win.chromakey = -1;
+        win.flags = 0;
+
+        ioctl(video_fd, VIDIOCSWIN, &win);
+
+        ioctl(video_fd, VIDIOCGPICT, &pict);
+#if 0
+        printf("v4l: colour=%d hue=%d brightness=%d constrast=%d whiteness=%d\n",
+               pict.colour,
+               pict.hue,
+               pict.brightness,
+               pict.contrast,
+               pict.whiteness);
+#endif        
+        /* try to choose a suitable video format */
+        pict.palette=VIDEO_PALETTE_YUV422;
+        ret = ioctl(video_fd, VIDIOCSPICT, &pict);
+        if (ret < 0) {
+            fprintf(stderr,"Could Not Find YUY2 capture window.\n");
+            goto fail;
+        }
+        if ((width == video_cap.maxwidth && height == video_cap.maxheight) ||
+            (width == video_cap.maxwidth && height == video_cap.maxheight*2) ||
+            (width == video_cap.maxwidth/2 && height == video_cap.maxheight)) {
+
+            s->deint=0;
+            s->halfw=0;
+            if (height == video_cap.maxheight*2) s->deint=1;
+            if (width == video_cap.maxwidth/2) s->halfw=1;
+        } else {
+            fprintf(stderr,"\nIncorrect Grab Size Supplied - Supported Sizes Are:\n");
+            fprintf(stderr," %dx%d  %dx%d %dx%d\n\n",
+                video_cap.maxwidth,video_cap.maxheight,
+                video_cap.maxwidth,video_cap.maxheight*2,
+                video_cap.maxwidth/2,video_cap.maxheight);
+            goto fail;
+        }
+
+        s->frame_format = pict.palette;
+
+        val = 1;
+        ioctl(video_fd, VIDIOCCAPTURE, &val);
+
+        s->time_frame = av_gettime();
+    } else {
+        fprintf(stderr,"mmap-based capture will not work with this grab.\n");
+        goto fail;
+    }
+
+    frame_size = (width * height * 3) / 2;
+    st->codec.pix_fmt = PIX_FMT_YUV420P;
+    s->fd = video_fd;
+    s->frame_size = frame_size;
+    
+    st->codec.codec_type = CODEC_TYPE_VIDEO;
+    st->codec.codec_id = CODEC_ID_RAWVIDEO;
+    st->codec.width = width;
+    st->codec.height = height;
+    st->codec.frame_rate = frame_rate;
+
+    if (s->halfw == 0) {
+        s->src_mem = av_malloc(s->width*2);
+    } else {
+        s->src_mem = av_malloc(s->width*4);
+    }
+    if (!s->src_mem) goto fail;
+
+    s->lum_m4_mem = av_malloc(s->width);
+    if (!s->lum_m4_mem) {
+        av_free(s->src_mem);
+        goto fail;
+    }
+
+    return 0;
+ fail:
+    if (video_fd >= 0)
+        close(video_fd);
+    av_free(st);
+    return -EIO;
+}
+
+//#ifdef HAVE_MMX
+//#undef HAVE_MMX
+//#endif
+
+#ifdef HAVE_MMX
+#include "../libavcodec/i386/mmx.h"
+
+#define LINE_WITH_UV \
+                    movq_m2r(ptr[0],mm0); \
+                    movq_m2r(ptr[8],mm1);  \
+                    movq_r2r(mm0, mm4); \
+                    punpcklbw_r2r(mm1,mm0); \
+                    punpckhbw_r2r(mm1,mm4); \
+                    movq_r2r(mm0,mm5); \
+                    punpcklbw_r2r(mm4,mm0); \
+                    punpckhbw_r2r(mm4,mm5); \
+                    movq_r2r(mm0,mm1); \
+                    punpcklbw_r2r(mm5,mm1); \
+                    movq_r2m(mm1,lum[0]); \
+                    movq_m2r(ptr[16],mm2); \
+                    movq_m2r(ptr[24],mm1); \
+                    movq_r2r(mm2,mm4); \
+                    punpcklbw_r2r(mm1,mm2); \
+                    punpckhbw_r2r(mm1,mm4); \
+                    movq_r2r(mm2,mm3); \
+                    punpcklbw_r2r(mm4,mm2); \
+                    punpckhbw_r2r(mm4,mm3); \
+                    movq_r2r(mm2,mm1); \
+                    punpcklbw_r2r(mm3,mm1); \
+                    movq_r2m(mm1,lum[8]); \
+                    punpckhdq_r2r(mm2,mm0); \
+                    punpckhdq_r2r(mm3,mm5); \
+                    movq_r2m(mm0,cb[0]); \
+                    movq_r2m(mm5,cr[0]);
+
+#define LINE_NO_UV \
+                    movq_m2r(ptr[0],mm0);\
+                    movq_m2r(ptr[8],mm1);\
+                    movq_r2r(mm0, mm4);\
+                    punpcklbw_r2r(mm1,mm0); \
+                    punpckhbw_r2r(mm1,mm4);\
+                    movq_r2r(mm0,mm5);\
+                    punpcklbw_r2r(mm4,mm0);\
+                    punpckhbw_r2r(mm4,mm5);\
+                    movq_r2r(mm0,mm1);\
+                    punpcklbw_r2r(mm5,mm1);\
+                    movq_r2m(mm1,lum[0]);\
+                    movq_m2r(ptr[16],mm2);\
+                    movq_m2r(ptr[24],mm1);\
+                    movq_r2r(mm2,mm4);\
+                    punpcklbw_r2r(mm1,mm2);\
+                    punpckhbw_r2r(mm1,mm4);\
+                    movq_r2r(mm2,mm3);\
+                    punpcklbw_r2r(mm4,mm2);\
+                    punpckhbw_r2r(mm4,mm3);\
+                    movq_r2r(mm2,mm1);\
+                    punpcklbw_r2r(mm3,mm1);\
+                    movq_r2m(mm1,lum[8]);
+
+#define LINE_WITHUV_AVG \
+                    movq_m2r(ptr[0], mm0);\
+                    movq_m2r(ptr[8], mm1);\
+                    movq_r2r(mm0, mm4);\
+                    punpcklbw_r2r(mm1,mm0);\
+                    punpckhbw_r2r(mm1,mm4);\
+                    movq_r2r(mm0,mm5);\
+                    punpcklbw_r2r(mm4,mm0);\
+                    punpckhbw_r2r(mm4,mm5);\
+                    movq_r2r(mm0,mm1);\
+                    movq_r2r(mm5,mm2);\
+                    punpcklbw_r2r(mm7,mm1);\
+                    punpcklbw_r2r(mm7,mm2);\
+                    paddw_r2r(mm6,mm1);\
+                    paddw_r2r(mm2,mm1);\
+                    psraw_i2r(1,mm1);\
+                    packuswb_r2r(mm7,mm1);\
+                    movd_r2m(mm1,lum[0]);\
+                    movq_m2r(ptr[16],mm2);\
+                    movq_m2r(ptr[24],mm1);\
+                    movq_r2r(mm2,mm4);\
+                    punpcklbw_r2r(mm1,mm2);\
+                    punpckhbw_r2r(mm1,mm4);\
+                    movq_r2r(mm2,mm3);\
+                    punpcklbw_r2r(mm4,mm2);\
+                    punpckhbw_r2r(mm4,mm3);\
+                    movq_r2r(mm2,mm1);\
+                    movq_r2r(mm3,mm4);\
+                    punpcklbw_r2r(mm7,mm1);\
+                    punpcklbw_r2r(mm7,mm4);\
+                    paddw_r2r(mm6,mm1);\
+                    paddw_r2r(mm4,mm1);\
+                    psraw_i2r(1,mm1);\
+                    packuswb_r2r(mm7,mm1);\
+                    movd_r2m(mm1,lum[4]);\
+                    punpckhbw_r2r(mm7,mm0);\
+                    punpckhbw_r2r(mm7,mm2);\
+                    paddw_r2r(mm6,mm0);\
+                    paddw_r2r(mm2,mm0);\
+                    psraw_i2r(1,mm0);\
+                    packuswb_r2r(mm7,mm0);\
+                    punpckhbw_r2r(mm7,mm5);\
+                    punpckhbw_r2r(mm7,mm3);\
+                    paddw_r2r(mm6,mm5);\
+                    paddw_r2r(mm3,mm5);\
+                    psraw_i2r(1,mm5);\
+                    packuswb_r2r(mm7,mm5);\
+                    movd_r2m(mm0,cb[0]);\
+                    movd_r2m(mm5,cr[0]);
+
+#define LINE_NOUV_AVG \
+                    movq_m2r(ptr[0],mm0);\
+                    movq_m2r(ptr[8],mm1);\
+                    pand_r2r(mm5,mm0);\
+                    pand_r2r(mm5,mm1);\
+                    pmaddwd_r2r(mm6,mm0);\
+                    pmaddwd_r2r(mm6,mm1);\
+                    packssdw_r2r(mm1,mm0);\
+                    paddw_r2r(mm6,mm0);\
+                    psraw_i2r(1,mm0);\
+                    movq_m2r(ptr[16],mm2);\
+                    movq_m2r(ptr[24],mm3);\
+                    pand_r2r(mm5,mm2);\
+                    pand_r2r(mm5,mm3);\
+                    pmaddwd_r2r(mm6,mm2);\
+                    pmaddwd_r2r(mm6,mm3);\
+                    packssdw_r2r(mm3,mm2);\
+                    paddw_r2r(mm6,mm2);\
+                    psraw_i2r(1,mm2);\
+                    packuswb_r2r(mm2,mm0);\
+                    movq_r2m(mm0,lum[0]);
+
+#define DEINT_LINE_LUM(ptroff) \
+                    movd_m2r(lum_m4[(ptroff)],mm0);\
+                    movd_m2r(lum_m3[(ptroff)],mm1);\
+                    movd_m2r(lum_m2[(ptroff)],mm2);\
+                    movd_m2r(lum_m1[(ptroff)],mm3);\
+                    movd_m2r(lum[(ptroff)],mm4);\
+                    punpcklbw_r2r(mm7,mm0);\
+                    movd_r2m(mm2,lum_m4[(ptroff)]);\
+                    punpcklbw_r2r(mm7,mm1);\
+                    punpcklbw_r2r(mm7,mm2);\
+                    punpcklbw_r2r(mm7,mm3);\
+                    punpcklbw_r2r(mm7,mm4);\
+                    psllw_i2r(2,mm1);\
+                    psllw_i2r(1,mm2);\
+                    paddw_r2r(mm6,mm1);\
+                    psllw_i2r(2,mm3);\
+                    paddw_r2r(mm2,mm1);\
+                    paddw_r2r(mm4,mm0);\
+                    paddw_r2r(mm3,mm1);\
+                    psubusw_r2r(mm0,mm1);\
+                    psrlw_i2r(3,mm1);\
+                    packuswb_r2r(mm7,mm1);\
+                    movd_r2m(mm1,lum_m2[(ptroff)]);
+
+#else
+#include "../libavcodec/dsputil.h"
+
+#define LINE_WITH_UV \
+                    lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
+                    cb[0]=ptr[1];cb[1]=ptr[5];\
+                    cr[0]=ptr[3];cr[1]=ptr[7];\
+                    lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
+                    cb[2]=ptr[9];cb[3]=ptr[13];\
+                    cr[2]=ptr[11];cr[3]=ptr[15];\
+                    lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
+                    cb[4]=ptr[17];cb[5]=ptr[21];\
+                    cr[4]=ptr[19];cr[5]=ptr[23];\
+                    lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];\
+                    cb[6]=ptr[25];cb[7]=ptr[29];\
+                    cr[6]=ptr[27];cr[7]=ptr[31];
+
+#define LINE_NO_UV \
+                    lum[0]=ptr[0];lum[1]=ptr[2];lum[2]=ptr[4];lum[3]=ptr[6];\
+                    lum[4]=ptr[8];lum[5]=ptr[10];lum[6]=ptr[12];lum[7]=ptr[14];\
+                    lum[8]=ptr[16];lum[9]=ptr[18];lum[10]=ptr[20];lum[11]=ptr[22];\
+                    lum[12]=ptr[24];lum[13]=ptr[26];lum[14]=ptr[28];lum[15]=ptr[30];
+
+#define LINE_WITHUV_AVG \
+                    sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
+                    sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
+                    sum=(ptr[1]+ptr[5]+1) >> 1;cb[0]=sum; \
+                    sum=(ptr[3]+ptr[7]+1) >> 1;cr[0]=sum; \
+                    sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
+                    sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
+                    sum=(ptr[9]+ptr[13]+1) >> 1;cb[1]=sum; \
+                    sum=(ptr[11]+ptr[15]+1) >> 1;cr[1]=sum; \
+                    sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
+                    sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
+                    sum=(ptr[17]+ptr[21]+1) >> 1;cb[2]=sum; \
+                    sum=(ptr[19]+ptr[23]+1) >> 1;cr[2]=sum; \
+                    sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
+                    sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum; \
+                    sum=(ptr[25]+ptr[29]+1) >> 1;cb[3]=sum; \
+                    sum=(ptr[27]+ptr[31]+1) >> 1;cr[3]=sum; 
+
+#define LINE_NOUV_AVG \
+                    sum=(ptr[0]+ptr[2]+1) >> 1;lum[0]=sum; \
+                    sum=(ptr[4]+ptr[6]+1) >> 1;lum[1]=sum; \
+                    sum=(ptr[8]+ptr[10]+1) >> 1;lum[2]=sum; \
+                    sum=(ptr[12]+ptr[14]+1) >> 1;lum[3]=sum; \
+                    sum=(ptr[16]+ptr[18]+1) >> 1;lum[4]=sum; \
+                    sum=(ptr[20]+ptr[22]+1) >> 1;lum[5]=sum; \
+                    sum=(ptr[24]+ptr[26]+1) >> 1;lum[6]=sum; \
+                    sum=(ptr[28]+ptr[30]+1) >> 1;lum[7]=sum; 
+
+#define DEINT_LINE_LUM(ptroff) \
+                    sum=(-lum_m4[(ptroff)]+(lum_m3[(ptroff)]<<2)+(lum_m2[(ptroff)]<<1)+(lum_m1[(ptroff)]<<2)-lum[(ptroff)]); \
+                    lum_m4[(ptroff)]=lum_m2[(ptroff)];\
+                    lum_m2[(ptroff)]=cm[(sum+4)>>3];\
+                    sum=(-lum_m4[(ptroff)+1]+(lum_m3[(ptroff)+1]<<2)+(lum_m2[(ptroff)+1]<<1)+(lum_m1[(ptroff)+1]<<2)-lum[(ptroff)+1]); \
+                    lum_m4[(ptroff)+1]=lum_m2[(ptroff)+1];\
+                    lum_m2[(ptroff)+1]=cm[(sum+4)>>3];\
+                    sum=(-lum_m4[(ptroff)+2]+(lum_m3[(ptroff)+2]<<2)+(lum_m2[(ptroff)+2]<<1)+(lum_m1[(ptroff)+2]<<2)-lum[(ptroff)+2]); \
+                    lum_m4[(ptroff)+2]=lum_m2[(ptroff)+2];\
+                    lum_m2[(ptroff)+2]=cm[(sum+4)>>3];\
+                    sum=(-lum_m4[(ptroff)+3]+(lum_m3[(ptroff)+3]<<2)+(lum_m2[(ptroff)+3]<<1)+(lum_m1[(ptroff)+3]<<2)-lum[(ptroff)+3]); \
+                    lum_m4[(ptroff)+3]=lum_m2[(ptroff)+3];\
+                    lum_m2[(ptroff)+3]=cm[(sum+4)>>3];
+
+#endif
+
+
+static int aiw_grab_read_packet(AVFormatContext *s1, AVPacket *pkt)
+{
+    AIWVideoData *s = s1->priv_data;
+    INT64 curtime, delay;
+    struct timespec ts;
+    int first;
+    INT64 per_frame = (INT64_C(1000000) * FRAME_RATE_BASE) / s->frame_rate;
+    int dropped = 0;
+
+    /* Calculate the time of the next frame */
+    s->time_frame += per_frame;
+
+    /* wait based on the frame rate */
+    for(first = 1;; first = 0) {
+        curtime = av_gettime();
+        delay = s->time_frame - curtime;
+        if (delay <= 0) {
+            if (delay < -per_frame) {
+                /* printf("grabbing is %d frames late (dropping)\n", (int) -(delay / 16666)); */
+                dropped = 1;
+                s->time_frame += per_frame;
+            }
+            break;
+        }    
+        ts.tv_sec = delay / 1000000;
+        ts.tv_nsec = (delay % 1000000) * 1000;
+        nanosleep(&ts, NULL);
+    }
+
+    if (av_new_packet(pkt, s->frame_size) < 0)
+        return -EIO;
+
+    if (dropped)
+        pkt->flags |= PKT_FLAG_DROPPED_FRAME;
+
+    /* read fields */
+    {
+        UINT8 *ptr, *lum, *cb, *cr;
+        int h;
+#ifndef HAVE_MMX
+        int sum;
+#endif
+        UINT8* src = s->src_mem;
+        UINT8 *ptrend = &src[s->width*2];
+        lum=&pkt->data[0];
+        cb=&lum[s->width*s->height];
+        cr=&cb[(s->width*s->height)/4];
+        if (s->deint == 0 && s->halfw == 0) {
+            while (read(s->fd,src,s->width*2) < 0) {
+                usleep(100);
+            }
+            for (h = 0; h < s->height-2; h+=2) {
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                    LINE_WITH_UV
+                }
+                read(s->fd,src,s->width*2);
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
+                    LINE_NO_UV
+                }
+                read(s->fd,src,s->width*2);
+            }
+/*
+ * Do last two lines
+ */
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                LINE_WITH_UV
+            }
+            read(s->fd,src,s->width*2);
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
+                LINE_NO_UV
+            }
+            /* drop second field */
+            while (read(s->fd,src,s->width*2) < 0) {
+                usleep(100);
+            }
+            for (h = 0; h < s->height - 1; h++) {
+                read(s->fd,src,s->width*2);
+            }
+        } else if (s->halfw == 1) {
+#ifdef HAVE_MMX
+            mmx_t rounder;
+            mmx_t masker;
+            rounder.uw[0]=1;
+            rounder.uw[1]=1;
+            rounder.uw[2]=1;
+            rounder.uw[3]=1;
+            masker.ub[0]=0xff;
+            masker.ub[1]=0;
+            masker.ub[2]=0xff;
+            masker.ub[3]=0;
+            masker.ub[4]=0xff;
+            masker.ub[5]=0;
+            masker.ub[6]=0xff;
+            masker.ub[7]=0;
+            pxor_r2r(mm7,mm7);
+            movq_m2r(rounder,mm6);
+#endif
+            while (read(s->fd,src,s->width*4) < 0) {
+                usleep(100);
+            }
+            ptrend = &src[s->width*4];
+            for (h = 0; h < s->height-2; h+=2) {
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
+                    LINE_WITHUV_AVG
+                }
+                read(s->fd,src,s->width*4);
+#ifdef HAVE_MMX
+                movq_m2r(masker,mm5);
+#endif
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
+                    LINE_NOUV_AVG
+                }
+                read(s->fd,src,s->width*4);
+            }
+/*
+ * Do last two lines
+ */
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8, cb+=4, cr+=4) {
+                LINE_WITHUV_AVG
+            }
+            read(s->fd,src,s->width*4);
+#ifdef HAVE_MMX
+            movq_m2r(masker,mm5);
+#endif
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=8) {
+                LINE_NOUV_AVG
+            }
+            /* drop second field */
+            while (read(s->fd,src,s->width*4) < 0) {
+                usleep(100);
+            }
+            for (h = 0; h < s->height - 1; h++) {
+                read(s->fd,src,s->width*4);
+            }
+        } else {
+            UINT8 *lum_m1, *lum_m2, *lum_m3, *lum_m4;
+#ifdef HAVE_MMX
+            mmx_t rounder;
+            rounder.uw[0]=4;
+            rounder.uw[1]=4;
+            rounder.uw[2]=4;
+            rounder.uw[3]=4;
+            movq_m2r(rounder,mm6);
+            pxor_r2r(mm7,mm7);
+#else
+            UINT8 *cm = cropTbl + MAX_NEG_CROP;
+#endif
+
+            /* read two fields and deinterlace them */
+            while (read(s->fd,src,s->width*2) < 0) {
+                usleep(100);
+            }
+            for (h = 0; h < (s->height/2)-2; h+=2) {
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                    LINE_WITH_UV
+                }
+                read(s->fd,src,s->width*2);
+/* skip a luminance line - will be filled in later */
+                lum += s->width;
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                    LINE_WITH_UV
+                }
+/* skip a luminance line - will be filled in later */
+                lum += s->width;
+                read(s->fd,src,s->width*2);
+            }
+/*
+ * Do last two lines
+ */
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                LINE_WITH_UV
+            }
+/* skip a luminance line - will be filled in later */
+            lum += s->width;
+            read(s->fd,src,s->width*2);
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, cb+=8, cr+=8) {
+                LINE_WITH_UV
+            }
+/*
+ *
+ * SECOND FIELD
+ *
+ */
+            lum=&pkt->data[s->width];
+            while (read(s->fd,src,s->width*2) < 0) {
+                usleep(10);
+            }
+/* First (and last) two lines not interlaced */
+            for (h = 0; h < 2; h++) {
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16) {
+                    LINE_NO_UV
+                }
+                read(s->fd,src,s->width*2);
+/* skip a luminance line */
+                lum += s->width;
+            }
+            lum_m1=&lum[-s->width];
+            lum_m2=&lum_m1[-s->width];
+            lum_m3=&lum_m2[-s->width];
+            memmove(s->lum_m4_mem,&lum_m3[-s->width],s->width);
+            for (; h < (s->height/2)-1; h++) {
+                lum_m4=s->lum_m4_mem;
+                for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16,lum_m1+=16,lum_m2+=16,lum_m3+=16,lum_m4+=16) {
+                    LINE_NO_UV
+
+                    DEINT_LINE_LUM(0)
+                    DEINT_LINE_LUM(4)
+                    DEINT_LINE_LUM(8)
+                    DEINT_LINE_LUM(12)
+                }
+                read(s->fd,src,s->width*2);
+/* skip a luminance line */
+                lum += s->width;
+                lum_m1 += s->width;
+                lum_m2 += s->width;
+                lum_m3 += s->width;
+//                lum_m4 += s->width;
+            }
+/*
+ * Do last line
+ */
+            lum_m4=s->lum_m4_mem;
+            for (ptr = &src[0]; ptr < ptrend; ptr+=32, lum+=16, lum_m1+=16, lum_m2+=16, lum_m3+=16, lum_m4+=16) {
+                LINE_NO_UV
+
+                DEINT_LINE_LUM(0)
+                DEINT_LINE_LUM(4)
+                DEINT_LINE_LUM(8)
+                DEINT_LINE_LUM(12)
+            }
+        }
+#ifdef HAVE_MMX
+        emms();
+#endif
+    }
+    return s->frame_size;
+}
+
+static int aiw_grab_read_close(AVFormatContext *s1)
+{
+    AIWVideoData *s = s1->priv_data;
+
+    close(s->fd);
+    av_free(s->lum_m4_mem);
+    av_free(s->src_mem);
+
+    return 0;
+}
+
+AVInputFormat aiw_grab_device_format = {
+    "aiw_grab_device",
+    "All-In-Wonder (km read-based) video grab",
+    sizeof(AIWVideoData),
+    NULL,
+    aiw_grab_read_header,
+    aiw_grab_read_packet,
+    aiw_grab_read_close,
+    .flags = AVFMT_NOFILE,
+};
+
 int video_grab_init(void)
 {
     av_register_input_format(&video_grab_device_format);
+    av_register_input_format(&aiw_grab_device_format);
     return 0;
 }