moved from mkvideo.c
authorkramm <kramm>
Sun, 4 May 2003 15:38:00 +0000 (15:38 +0000)
committerkramm <kramm>
Sun, 4 May 2003 15:38:00 +0000 (15:38 +0000)
lib/h.263/swfvideo.c [new file with mode: 0644]

diff --git a/lib/h.263/swfvideo.c b/lib/h.263/swfvideo.c
new file mode 100644 (file)
index 0000000..94b3d25
--- /dev/null
@@ -0,0 +1,1510 @@
+/* swfvideo.c
+   Routines for handling h.263 video tags
+
+   Part of the swftools package.
+
+   Copyright (c) 2003 Matthias Kramm <kramm@quiss.org>
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <assert.h>
+#include <math.h>
+#include "../rfxswf.h"
+#include "h263tables.c"
+
+#ifdef MAIN
+U16 totalframes = 0;
+#endif
+void swf_SetVideoStreamDefine(TAG*tag, VIDEOSTREAM*stream, U16 frames, U16 width, U16 height)
+{
+    swf_SetU16(tag, frames);
+    swf_SetU16(tag, width);
+    swf_SetU16(tag, height);
+    //swf_SetU8(tag, 1); /* smoothing on */
+    swf_SetU8(tag, 0); /* smoothing off */
+    swf_SetU8(tag, 2); /* codec = h.263 sorenson spark */
+
+    totalframes = frames;
+
+    memset(stream, 0, sizeof(VIDEOSTREAM));
+    stream->olinex = width;
+    width+=15;width&=~15;
+    height+=15;height&=~15;
+    stream->linex = width;
+    stream->width = width;
+    stream->height = height;
+    stream->bbx = width/16;
+    stream->bby = height/16;
+    stream->current = (YUV*)malloc(width*height*sizeof(YUV));
+    stream->oldpic = (YUV*)malloc(width*height*sizeof(YUV));
+    stream->mvdx = (int*)malloc(stream->bbx*stream->bby*sizeof(int));
+    stream->mvdy = (int*)malloc(stream->bbx*stream->bby*sizeof(int));
+    stream->do_motion = 0;
+
+    memset(stream->oldpic, 0, width*height*sizeof(YUV));
+    memset(stream->current, 0, width*height*sizeof(YUV));
+}
+void swf_VideoStreamClear(VIDEOSTREAM*stream)
+{
+    free(stream->oldpic);stream->oldpic = 0;
+    free(stream->current);stream->current = 0;
+    free(stream->mvdx);stream->mvdx=0;
+    free(stream->mvdy);stream->mvdy=0;
+}
+
+typedef struct _block_t
+{
+    int y1[64];
+    int y2[64];
+    int y3[64];
+    int y4[64];
+    int u[64];
+    int v[64];
+} block_t;
+
+static int zigzagtable[64] = {
+    0, 1, 5, 6, 14, 15, 27, 28,
+    2, 4, 7, 13, 16, 26, 29, 42,
+    3, 8, 12, 17, 25, 30, 41, 43,
+    9, 11, 18, 24, 31, 40, 44, 53,
+    10, 19, 23, 32, 39, 45, 52, 54,
+    20, 22, 33, 38, 46, 51, 55, 60,
+    21, 34, 37, 47, 50, 56, 59, 61,
+    35, 36, 48, 49, 57, 58, 62, 63};
+
+static void zigzag(int*src)
+{
+    int tmp[64];
+    int t;
+    for(t=0;t<64;t++) {
+       tmp[zigzagtable[t]] = src[t];
+    }
+    memcpy(src, tmp, sizeof(int)*64);
+}
+
+#define PI 3.14159265358979
+#define SQRT2 1.414214
+#define RSQRT2 (1.0/1.414214)
+
+static double table[8][8] =
+{
+{0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548},
+{0.980785280403230,0.831469612302545,0.555570233019602,0.195090322016128,-0.195090322016128,-0.555570233019602,-0.831469612302545,-0.980785280403230},
+{0.923879532511287,0.382683432365090,-0.382683432365090,-0.923879532511287,-0.923879532511287,-0.382683432365090,0.382683432365090,0.923879532511287},
+{0.831469612302545,-0.195090322016128,-0.980785280403230,-0.555570233019602,0.555570233019602,0.980785280403230,0.195090322016129,-0.831469612302545},
+{0.707106781186548,-0.707106781186547,-0.707106781186548,0.707106781186547,0.707106781186548,-0.707106781186547,-0.707106781186547,0.707106781186547},
+{0.555570233019602,-0.980785280403230,0.195090322016128,0.831469612302545,-0.831469612302545,-0.195090322016128,0.980785280403231,-0.555570233019602},
+{0.382683432365090,-0.923879532511287,0.923879532511287,-0.382683432365090,-0.382683432365091,0.923879532511287,-0.923879532511286,0.382683432365090},
+{0.195090322016128,-0.555570233019602,0.831469612302545,-0.980785280403231,0.980785280403230,-0.831469612302545,0.555570233019602,-0.195090322016129}
+};
+
+static void dct(int*src)
+{
+    double tmp[64];
+    int x,y,u,v,t;
+
+    for(v=0;v<8;v++)
+    for(u=0;u<8;u++)
+    {
+       double c = 0;
+       for(x=0;x<8;x++)
+       {
+           c+=table[u][x]*src[v*8+x];
+       }
+       tmp[v*8+u] = c;
+    }
+    for(u=0;u<8;u++)
+    for(v=0;v<8;v++)
+    {
+       double c = 0;
+       for(y=0;y<8;y++)
+       {
+           c+=table[v][y]*tmp[y*8+u];
+       }
+       src[v*8+u] = (int)(c*0.25+0.5);
+    }
+}
+
+static void idct(int*src)
+{
+    double tmp[64];
+    int x,y,u,v;
+    for(y=0;y<8;y++)
+    for(x=0;x<8;x++)
+    {
+       double c = 0;
+       for(u=0;u<8;u++)
+       {
+           c+=table[u][x]*src[y*8+u];
+       }
+       tmp[y*8+x] = c;
+    }
+    for(y=0;y<8;y++)
+    for(x=0;x<8;x++)
+    {
+       double c = 0;
+       for(v=0;v<8;v++)
+       {
+           c+=table[v][y]*tmp[v*8+x];
+       }
+       src[y*8+x] = (int)(c*0.25+0.5);
+    }
+}
+
+static double c[8] = {1.0,
+0.980785280403230, // cos(Pi*1/16), sin(Pi*7/16)
+0.923879532511287, // cos(Pi*2/16), sin(Pi*6/16)
+0.831469612302545, // cos(Pi*3/16), sin(Pi*5/16)
+0.707106781186548, // cos(Pi*4/16), sin(Pi*4/16), 1/sqrt(2)
+0.555570233019602, // cos(Pi*5/16), sin(Pi*3/16)
+0.382683432365090, // cos(Pi*6/16), sin(Pi*2/16)
+0.195090322016128 // cos(Pi*7/16), sin(Pi*1/16)
+};
+
+static double cc[8];
+int ccquant = -1;
+
+static void preparequant(int quant)
+{
+    if(ccquant == quant)
+       return;
+    cc[0] = c[0]/(quant*2*4);
+    cc[1] = c[1]/(quant*2*4);
+    cc[2] = c[2]/(quant*2*4);
+    cc[3] = c[3]/(quant*2*4);
+    cc[4] = c[4]/(quant*2*4);
+    cc[5] = c[5]/(quant*2*4);
+    cc[6] = c[6]/(quant*2*4);
+    cc[7] = c[7]/(quant*2*4);
+    ccquant = quant;
+}
+
+inline static void innerdct(double*a,double*b, double*c)
+{
+    // c1*c7*2 = c6
+    // c2*c6*2 = c4
+    // c3*c5*2 = c2
+    // c4*c4*2 = 1
+
+     //{  1,  3,  5,  7, -7, -5, -3, -1},
+     //{  3, -7, -1, -5,  5,  1,  7, -3},
+     //{  5, -1,  7,  3, -3, -7,  1, -5},
+     //{  7, -5,  3, -1,  1, -3,  5, -7}
+    double b0,b1,b2,b3,b4,b5;
+    b2 = (a[0]+a[7]);
+    b3 = (a[1]+a[6]);
+    b4 = (a[2]+a[5]);
+    b5 = (a[3]+a[4]);
+
+    b0 = (b2+b5)*c[4];
+    b1 = (b3+b4)*c[4];
+    b[0*8] = b0 + b1;
+    b[4*8] = b0 - b1;
+    b[2*8] = (b2-b5)*c[2] + (b3-b4)*c[6];
+    b[6*8] = (b2-b5)*c[6] + (b4-b3)*c[2];
+
+    b0 = (a[0]-a[7]);
+    b1 = (a[1]-a[6]);
+    b2 = (a[2]-a[5]);
+    b3 = (a[3]-a[4]);
+
+    b[1*8] = b0*c[1] + b1*c[3] + b2*c[5] + b3*c[7];
+    b[3*8] = b0*c[3] - b1*c[7] - b2*c[1] - b3*c[5];
+    b[5*8] = b0*c[5] - b1*c[1] + b2*c[7] + b3*c[3];
+    b[7*8] = b0*c[7] - b1*c[5] + b2*c[3] - b3*c[1];
+}
+
+static void dct2(int*src, int*dest)
+{
+    double tmp[64], tmp2[64];
+    double*p;
+    int u,x,v,t;
+
+    for(t=0;t<64;t++)
+       tmp2[t] = src[t];
+
+    for(v=0;v<8;v++)
+    {
+       double* a=&tmp2[v*8];
+       double* b=&tmp[v];
+       innerdct(a,b,c);
+    }
+    for(v=0;v<8;v++)
+    {
+       double* a=&tmp[v*8];
+       double* b=&tmp2[v];
+       innerdct(a,b,cc);
+    }
+    for(t=0;t<64;t++) {
+       dest[zigzagtable[t]] = (int)(tmp2[t]);
+    }
+}
+
+
+static inline int truncate256(int a)
+{
+    if(a>255) return 255;
+    if(a<0) return 0;
+    return a;
+}
+
+static void getregion(block_t* bb, YUV*pic, int posx, int posy, int linex)
+{
+    YUV*p1;
+    YUV*p2;
+    int y1=0, y2=0, y3=0, y4=0;
+    int u=0,v=0;
+    int x,y;
+    posx*=16;
+    posy*=16;
+    p1 = &pic[posy*linex+posx];
+    p2 = p1;
+    for(y=0;y<8;y++) {
+       for(x=0;x<8;x++) {
+           bb->u[u++] = (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
+           bb->v[v++] = (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
+           bb->y1[y1++] = p1[x].y;
+           bb->y2[y2++] = p1[x+8].y;
+           bb->y3[y3++] = p1[linex*8+x].y;
+           bb->y4[y4++] = p1[linex*8+x+8].y;
+       }
+       p1+=linex;
+       p2+=linex*2;
+    }
+}
+
+static void getmvdregion(block_t* bb, YUV*pic, int posx, int posy, int mvdx, int mvdy, int linex)
+{
+    YUV*p1;
+    YUV*p2;
+    int yy=0,uv=0;
+    int x,y;
+    int yhp = 0, uvhp=0;
+    int uvposx, uvposy;
+    posx = posx*16 + ((mvdx&~1)/2);
+    posy = posy*16 + ((mvdy&~1)/2);
+    p1 = &pic[posy*linex+posx];
+    p2 = &pic[(posy&~1)*linex+(posx&~1)];
+    uvhp = ((mvdx&1)|((mvdx>>1)&1))|((mvdy&2)|((mvdy&1)<<1));
+    yhp = ((mvdy&1)<<1|(mvdx&1));
+
+    /* y */
+    if(yhp==0 || yhp==2) {
+       for(y=0;y<8;y++) {
+           for(x=0;x<8;x++) {
+               bb->y1[yy] = p1[x].y;
+               bb->y2[yy] = p1[x+8].y;
+               bb->y3[yy] = p1[linex*8+x].y;
+               bb->y4[yy] = p1[linex*8+x+8].y;
+               yy++;
+           }
+           p1+=linex;
+
+           if(yhp==2) {
+               yy-=8;
+               for(x=0;x<8;x++) {
+                   bb->y1[yy] += p1[x].y; bb->y1[yy] /= 2;
+                   bb->y2[yy] += p1[x+8].y; bb->y2[yy] /= 2;
+                   bb->y3[yy] += p1[linex*8+x].y; bb->y3[yy] /= 2;
+                   bb->y4[yy] += p1[linex*8+x+8].y; bb->y4[yy] /= 2;
+                   yy++;
+               }
+           }
+       }
+    } else if(yhp==1 || yhp==3) {
+       for(y=0;y<8;y++) {
+           for(x=0;x<8;x++) {
+               bb->y1[yy] = (p1[x].y + p1[x+1].y);
+               bb->y2[yy] = (p1[x+8].y + p1[x+8+1].y);
+               bb->y3[yy] = (p1[linex*8+x].y + p1[linex*8+x+1].y);
+               bb->y4[yy] = (p1[linex*8+x+8].y + p1[linex*8+x+8+1].y);
+               yy++;
+           }
+           yy-=8;
+           p1+=linex;
+           if(yhp==3) {
+               for(x=0;x<8;x++) {
+                   bb->y1[yy] += (p1[x].y + p1[x+1].y); bb->y1[yy]/=4;
+                   bb->y2[yy] += (p1[x+8].y + p1[x+8+1].y); bb->y2[yy]/=4;
+                   bb->y3[yy] += (p1[linex*8+x].y + p1[linex*8+x+1].y); bb->y3[yy]/=4;
+                   bb->y4[yy] += (p1[linex*8+x+8].y + p1[linex*8+x+8+1].y); bb->y4[yy]/=4;
+                   yy++;
+               }
+           } else {
+               for(x=0;x<8;x++) {
+                   bb->y1[yy]/=2; bb->y2[yy]/=2; bb->y3[yy]/=2; bb->y4[yy]/=2;
+                   yy++;
+               }
+           }
+       }
+    }
+
+    /* u,v */
+    if(uvhp==0 || uvhp==2) {
+       for(y=0;y<8;y++) {
+           for(x=0;x<8;x++) {
+               bb->u[uv] = (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
+               bb->v[uv] = (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
+               uv++;
+           }
+           p2+=linex*2;
+           if(uvhp==2) {
+               uv-=8;
+               for(x=0;x<8;x++) {
+                   bb->u[uv] += (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
+                   bb->v[uv] += (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
+                   bb->u[uv] /= 2;
+                   bb->v[uv] /= 2;
+                   uv++;
+               }
+           }
+       }
+    } else /* uvhp==1 || uvhp==3 */ {
+       for(y=0;y<8;y++) {
+           for(x=0;x<8;x++) {
+               bb->u[uv] = ((p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4+
+                            (p2[x*2+2].u + p2[x*2+1+2].u + p2[linex+x*2+2].u + p2[linex+x*2+1+2].u)/4);
+               bb->v[uv] = ((p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4+
+                            (p2[x*2+2].v + p2[x*2+1+2].v + p2[linex+x*2+2].v + p2[linex+x*2+1+2].v)/4);
+               uv++;
+           }
+           uv-=8;
+           p2+=linex*2;
+           if(uvhp==3) {
+               for(x=0;x<8;x++) {
+                   bb->u[uv] += ((p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4+
+                                 (p2[x*2+2].u + p2[x*2+1+2].u + p2[linex+x*2+2].u + p2[linex+x*2+1+2].u)/4);
+                   bb->v[uv] += ((p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4+
+                                 (p2[x*2+2].v + p2[x*2+1+2].v + p2[linex+x*2+2].v + p2[linex+x*2+1+2].v)/4);
+                   bb->u[uv] /= 4;
+                   bb->v[uv] /= 4;
+                   uv++;
+               }
+           } else {
+               for(x=0;x<8;x++) {
+                   bb->u[uv] /= 2;
+                   bb->v[uv] /= 2;
+                   uv++;
+               }
+           }
+       }
+    }
+}
+
+static void rgb2yuv(YUV*dest, RGBA*src, int dlinex, int slinex, int width, int height)
+{
+    int x,y;
+    for(y=0;y<height;y++) {
+       for(x=0;x<width;x++) {
+           int r,g,b;
+           r = src[y*slinex+x].r;
+           g = src[y*slinex+x].g;
+           b = src[y*slinex+x].b;
+           /*dest[y*dlinex+x].y = (r*0.299 + g*0.587 + b*0.114);
+           dest[y*dlinex+x].u = (r*-0.169 + g*-0.332 + b*0.500 + 128.0);
+           dest[y*dlinex+x].v = (r*0.500 + g*-0.419 + b*-0.0813 + 128.0);*/
+
+           //dest[y*dlinex+x].y = 128;//(r*((int)( 0.299*256)) + g*((int)( 0.587*256)) + b*((int)( 0.114 *256)))>>8;
+           dest[y*dlinex+x].y = (r*((int)( 0.299*256)) + g*((int)( 0.587*256)) + b*((int)( 0.114 *256)))>>8;
+           dest[y*dlinex+x].u = (r*((int)(-0.169*256)) + g*((int)(-0.332*256)) + b*((int)( 0.500 *256))+ 128*256)>>8;
+           dest[y*dlinex+x].v = (r*((int)( 0.500*256)) + g*((int)(-0.419*256)) + b*((int)(-0.0813*256))+ 128*256)>>8;
+       }
+    }
+}
+static void copyregion(VIDEOSTREAM*s, YUV*dest, YUV*src, int bx, int by)
+{
+    YUV*p1 = &src[by*s->linex*16+bx*16];
+    YUV*p2 = &dest[by*s->linex*16+bx*16];
+    int y;
+    for(y=0;y<16;y++) {
+       memcpy(p1, p2, 16*sizeof(YUV));
+       p1+=s->linex;p2+=s->linex;
+    }
+}
+
+static void yuv2rgb(RGBA*dest, YUV*src, int linex, int width, int height)
+{
+    int x,y;
+    for(y=0;y<height;y++) {
+       for(x=0;x<width;x++) {
+           int u,v,yy;
+           u = src[y*linex+x].u;
+           v = src[y*linex+x].v;
+           yy = src[y*linex+x].y;
+           dest[y*linex+x].r = truncate256(yy + ((360*(v-128))>>8));
+           dest[y*linex+x].g = truncate256(yy - ((88*(u-128)+183*(v-128))>>8));
+           dest[y*linex+x].b = truncate256(yy + ((455 * (u-128))>>8));
+       }
+    }
+}
+static void copyblock(VIDEOSTREAM*s, YUV*dest, block_t*b, int bx, int by)
+{
+    YUV*p1 = &dest[(by*16)*s->linex+bx*16];
+    YUV*p2 = &dest[(by*16+8)*s->linex+bx*16];
+    int x,y;
+    for(y=0;y<8;y++) {
+       for(x=0;x<8;x++) {
+           int u,v,yy;
+           p1[x+0].u = b->u[(y/2)*8+(x/2)];
+           p1[x+0].v = b->v[(y/2)*8+(x/2)];
+           p1[x+0].y = b->y1[y*8+x];
+           p1[x+8].u = b->u[(y/2)*8+(x/2)+4];
+           p1[x+8].v = b->v[(y/2)*8+(x/2)+4];
+           p1[x+8].y = b->y2[y*8+x];
+           p2[x+0].u = b->u[(y/2+4)*8+(x/2)];
+           p2[x+0].v = b->v[(y/2+4)*8+(x/2)];
+           p2[x+0].y = b->y3[y*8+x];
+           p2[x+8].u = b->u[(y/2+4)*8+(x/2)+4];
+           p2[x+8].v = b->v[(y/2+4)*8+(x/2)+4];
+           p2[x+8].y = b->y4[y*8+x];
+       }
+       p1+=s->linex;
+       p2+=s->linex;
+    }
+}
+
+static int compareregions(VIDEOSTREAM*s, int bx, int by)
+{
+    int linex = s->width;
+    YUV*p1 = &s->current[by*linex*16+bx*16];
+    YUV*p2 = &s->oldpic[by*linex*16+bx*16];
+    int diff = 0;
+    int x,y;
+    for(y=0;y<16;y++) {
+       for(x=0;x<16;x++) {
+           YUV*m = &p1[x];
+           YUV*n = &p2[x];
+           int y = m->y - n->y;
+           int u = m->u - n->u;
+           int v = m->v - n->v;
+           diff += y*y+(u*u+v*v)/4;
+       }
+       p1+=linex;
+       p2+=linex;
+    }
+    return diff/256;
+}
+
+static inline int valtodc(int val)
+{
+    assert(val>=0);
+
+    /* table 12/h.263 */
+
+    //val+=4; //round
+    val/=8;
+    /* TODO: what to do for zero values? skip the block? */
+    if(val==0)
+       return 1;
+    if(val==128)
+       return 255;
+    if(val>254)
+       return 254;
+    return val;
+}
+static int dctoval(int dc)
+{
+    int val;
+    assert(dc>0);
+    assert(dc!=128);
+    assert(dc<256);
+    /* table 12/h.263 */
+    val = dc*8;
+    if(val == 255*8)
+       val = 128*8;
+    return val;
+}
+
+static int codehuffman(TAG*tag, struct huffcode*table, int index)
+{
+    /* TODO: !optimize! */
+    int i=0;
+    while(table[index].code[i]) {
+       if(table[index].code[i]=='0')
+           swf_SetBits(tag, 0, 1);
+       else
+           swf_SetBits(tag, 1, 1);
+       i++;
+    }
+    return i;
+}
+
+static void quantize8x8(int*src, int*dest, int has_dc, int quant)
+{
+    int t,pos=0;
+    double q = 1.0/(quant*2);
+    if(has_dc) {
+       dest[0] = valtodc((int)src[0]); /*DC*/
+       pos++;
+    }
+    for(t=pos;t<64;t++)
+    {
+       //dest[t] = (int)src[t];
+    /* exact: if(quant&1){dest[t] = (dest[t]/quant - 1)/2;}else{dest[t] = ((dest[t]+1)/quant - 1)/2;} */
+        //if(quant&1){dest[t] = (dest[t]/quant - 1)/2;}else{dest[t] = ((dest[t]+1)/quant - 1)/2;}
+       //dest[t] = dest[t]/(quant*2);
+       dest[t] = (int)(src[t]*q);
+       /* TODO: warn if this happens- the video will be buggy */
+       if(dest[t]>127) dest[t]=127;
+       if(dest[t]<-127) dest[t]=-127;
+    }
+}
+
+static void dequantize8x8(int*b, int has_dc, int quant)
+{
+    int t,pos=0;
+    if(has_dc) {
+       b[0] = dctoval(b[0]); //DC
+       pos++;
+    }
+    for(t=pos;t<64;t++) {
+       if(b[t]) {
+           int sign = 0;
+           if(b[t]<0) {
+               b[t] = -b[t];
+               sign = 1;
+           }
+
+           if(quant&1) {
+               b[t] = quant*(2*b[t]+1); //-7,8,24,40
+           } else {
+               b[t] = quant*(2*b[t]+1)-1; //-8,7,23,39
+           }
+
+           if(sign)
+               b[t] = -b[t];
+       }
+
+       /* paragraph 6.2.2, "clipping of reconstruction levels": */
+       if(b[t]>2047) b[t]=2047;
+       if(b[t]<-2048) b[t]=-2048;
+    }
+}
+
+static int hascoef(int*b, int has_dc)
+{
+    int t;
+    int pos=0;
+    if(has_dc)
+       pos++;
+    for(t=pos;t<64;t++) {
+       if(b[t])
+           return 1;
+    }
+    return 0;
+}
+
+static int coefbits8x8(int*bb, int has_dc)
+{
+    int t;
+    int pos=0;
+    int bits=0;
+    int last;
+
+    if(has_dc) {
+       bits+=8;
+       pos++;
+    }
+    for(last=63;last>=pos;last--) {
+       if(bb[last])
+           break;
+    }
+    if(last < pos)
+       return bits;
+    while(1) {
+       int run=0, level=0, islast=0,t;
+       while(!bb[pos] && pos<last) {
+           pos++;
+           run++;
+       }
+       if(pos==last)
+           islast=1;
+       level=bb[pos];
+       if(level<0) level=-level;
+       assert(level);
+       for(t=0;t<RLE_ESCAPE;t++) {
+           if(rle_params[t].run == run &&
+              rle_params[t].level == level &&
+              rle_params[t].last == islast) {
+               bits += rle[t].len + 1;
+               break;
+           }
+       }
+       if(t==RLE_ESCAPE) {
+           bits += rle[RLE_ESCAPE].len + 1 + 6 + 8;
+       }
+       if(islast)
+           break;
+       pos++;
+    }
+    return bits;
+}
+
+static void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
+{
+    int t;
+    int pos=0;
+    int bits=0;
+
+    if(has_dc) {
+       swf_SetBits(tag, bb[0], 8);
+       pos++;
+    }
+
+    if(has_tcoef) {
+       int last;
+       /* determine last non-null coefficient */
+       for(last=63;last>=pos;last--) {
+           /* TODO: we could leave out small coefficients
+                    after a certain point (32?) */
+           if(bb[last])
+               break;
+       }
+       /* blocks without coefficients should not be included
+          in the cbpy/cbpc patterns: */
+       assert(bb[last]);
+
+       while(1) {
+           int run=0;
+           int level=0;
+           int islast=0;
+           int sign=0;
+           int t;
+           while(!bb[pos] && pos<last) {
+               pos++;
+               run++;
+           }
+           if(pos==last)
+               islast=1;
+           level=bb[pos];
+           assert(level);
+           if(level<0) {
+               level = -level;
+               sign = 1;
+           }
+           for(t=0;t<RLE_ESCAPE;t++) {
+               /* TODO: lookup table */
+               if(rle_params[t].run == run &&
+                  rle_params[t].level == level &&
+                  rle_params[t].last == islast) {
+                   codehuffman(tag, rle, t);
+                   swf_SetBits(tag, sign, 1);
+                   break;
+               }
+           }
+           if(t==RLE_ESCAPE) {
+               codehuffman(tag, rle, RLE_ESCAPE);
+               level=bb[pos];
+               /* table 14/h.263 */
+               assert(level);
+               assert(level>=-127);
+               assert(level<=127);
+
+               swf_SetBits(tag, islast, 1);
+               swf_SetBits(tag, run, 6);
+               swf_SetBits(tag, level, 8); //FIXME: fixme??
+           }
+
+           if(islast)
+               break;
+           pos++;
+       }
+    }
+}
+
+static void quantize(block_t*fb, block_t*b, int has_dc, int quant)
+{
+    quantize8x8(fb->y1, b->y1, has_dc, quant);
+    quantize8x8(fb->y2, b->y2, has_dc, quant);
+    quantize8x8(fb->y3, b->y3, has_dc, quant);
+    quantize8x8(fb->y4, b->y4, has_dc, quant);
+    quantize8x8(fb->u, b->u, has_dc, quant);
+    quantize8x8(fb->v, b->v, has_dc, quant);
+}
+
+static void dodct(block_t*fb)
+{
+    dct(fb->y1); dct(fb->y2); dct(fb->y3); dct(fb->y4);
+    dct(fb->u);  dct(fb->v);
+    zigzag(fb->y1);
+    zigzag(fb->y2);
+    zigzag(fb->y3);
+    zigzag(fb->y4);
+    zigzag(fb->u);
+    zigzag(fb->v);
+}
+static void dodctandquant(block_t*fb, block_t*b, int has_dc, int quant)
+{
+    int t;
+    if(has_dc) {
+       dodct(fb);
+       quantize(fb,b,has_dc,quant);
+       return;
+    }
+    preparequant(quant);
+    dct2(fb->y1,b->y1); dct2(fb->y2,b->y2); dct2(fb->y3,b->y3); dct2(fb->y4,b->y4);
+    dct2(fb->u,b->u);  dct2(fb->v,b->v);
+}
+
+static void doidct(block_t*b)
+{
+    block_t fb;
+    int t;
+    for(t=0;t<64;t++) {
+       fb.y1[t] = b->y1[zigzagtable[t]];
+       fb.y2[t] = b->y2[zigzagtable[t]];
+       fb.y3[t] = b->y3[zigzagtable[t]];
+       fb.y4[t] = b->y4[zigzagtable[t]];
+       fb.u[t] = b->u[zigzagtable[t]];
+       fb.v[t] = b->v[zigzagtable[t]];
+    }
+    idct(fb.y1); idct(fb.y2); idct(fb.y3); idct(fb.y4);
+    idct(fb.u);  idct(fb.v);
+    for(t=0;t<64;t++) {
+       b->y1[t] = fb.y1[t];
+       b->y2[t] = fb.y2[t];
+       b->y3[t] = fb.y3[t];
+       b->y4[t] = fb.y4[t];
+       b->u[t] = fb.u[t];
+       b->v[t] = fb.v[t];
+    }
+}
+static void truncateblock(block_t*b)
+{
+    int t;
+    for(t=0;t<64;t++) {
+       b->y1[t] = truncate256(b->y1[t]);
+       b->y2[t] = truncate256(b->y2[t]);
+       b->y3[t] = truncate256(b->y3[t]);
+       b->y4[t] = truncate256(b->y4[t]);
+       b->u[t] = truncate256(b->u[t]);
+       b->v[t] = truncate256(b->v[t]);
+    }
+}
+
+static void dequantize(block_t*b, int has_dc, int quant)
+{
+    dequantize8x8(b->y1, has_dc, quant);
+    dequantize8x8(b->y2, has_dc, quant);
+    dequantize8x8(b->y3, has_dc, quant);
+    dequantize8x8(b->y4, has_dc, quant);
+    dequantize8x8(b->u, has_dc, quant);
+    dequantize8x8(b->v, has_dc, quant);
+}
+
+static void getblockpatterns(block_t*b, int*cbpybits,int*cbpcbits, int has_dc)
+{
+    *cbpybits = 0;
+    *cbpcbits = 0;
+
+    *cbpybits|=hascoef(b->y1, has_dc)*8;
+    *cbpybits|=hascoef(b->y2, has_dc)*4;
+    *cbpybits|=hascoef(b->y3, has_dc)*2;
+    *cbpybits|=hascoef(b->y4, has_dc)*1;
+
+    *cbpcbits|=hascoef(b->u, has_dc)*2;
+    *cbpcbits|=hascoef(b->v, has_dc)*1;
+}
+
+static void setQuant(TAG*tag, int dquant)
+{
+    int code = 0;
+    /* 00 01 10 11
+       -1 -2 +1 +2
+    */
+    if(dquant == -1) {
+       swf_SetBits(tag, 0x0, 2);
+    } else if(dquant == -2) {
+       swf_SetBits(tag, 0x1, 2);
+    } else if(dquant == +1) {
+       swf_SetBits(tag, 0x2, 2);
+    } else if(dquant == +2) {
+       swf_SetBits(tag, 0x3, 2);
+    } else {
+       assert(0*strlen("invalid dquant"));
+    }
+}
+
+static void change_quant(int quant, int*dquant)
+{
+    /* TODO */
+    *dquant = 0;
+}
+
+static void encode_blockI(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
+{
+    block_t fb;
+    block_t b;
+    int dquant=0;
+    int cbpcbits = 0, cbpybits=0;
+
+    getregion(&fb, s->current, bx, by, s->width);
+
+    change_quant(*quant, &dquant);
+    *quant+=dquant;
+
+    dodctandquant(&fb, &b, 1, *quant);
+    //quantize(&fb, &b, 1, *quant);
+
+    //decode_blockI(s, &b, bx, by);
+
+    getblockpatterns(&b, &cbpybits, &cbpcbits, 1);
+
+    if(dquant) {
+       codehuffman(tag, mcbpc_intra, 4+cbpcbits);
+    } else {
+       codehuffman(tag, mcbpc_intra, 0+cbpcbits);
+    }
+
+    codehuffman(tag, cbpy, cbpybits);
+
+    if(dquant) {
+       setQuant(tag, dquant);
+    }
+
+    /* luminance */
+    encode8x8(tag, b.y1, 1, cbpybits&8);
+    encode8x8(tag, b.y2, 1, cbpybits&4);
+    encode8x8(tag, b.y3, 1, cbpybits&2);
+    encode8x8(tag, b.y4, 1, cbpybits&1);
+
+    /* chrominance */
+    encode8x8(tag, b.u, 1, cbpcbits&2);
+    encode8x8(tag, b.v, 1, cbpcbits&1);
+
+    /* reconstruct */
+    dequantize(&b, 1, *quant);
+    doidct(&b);
+    truncateblock(&b);
+    copyblock(s, s->current, &b, bx, by);
+}
+
+static void yuvdiff(block_t*a, block_t*b)
+{
+    int t;
+    for(t=0;t<64;t++) {
+       a->y1[t] = (a->y1[t] - b->y1[t]);
+       a->y2[t] = (a->y2[t] - b->y2[t]);
+       a->y3[t] = (a->y3[t] - b->y3[t]);
+       a->y4[t] = (a->y4[t] - b->y4[t]);
+       a->u[t]  = (a->u[t] - b->u[t]);
+       a->v[t]  = (a->v[t] - b->v[t]);
+    }
+}
+
+static void predictmvd(VIDEOSTREAM*s, int bx, int by, int*px, int*py)
+{
+    int i1,i2;
+    int x1,y1,x2,y2,x3,y3;
+    int x4,y4,p;
+    if(bx) {x1=s->mvdx[by*s->bbx+bx-1];
+           y1=s->mvdy[by*s->bbx+bx-1];
+    } else {x1=y1=0;}
+
+    if(by) {x2=s->mvdx[(by-1)*s->bbx+bx];
+           y2=s->mvdy[(by-1)*s->bbx+bx];
+           if(bx<s->bbx-1) {
+               x3=s->mvdx[(by-1)*s->bbx+bx+1];
+               y3=s->mvdy[(by-1)*s->bbx+bx+1];
+           } else {
+               x3=y3=0;
+           }
+           }
+    else   {x2=x3=x1;y2=y3=y1;}
+
+          if((x1 <= x2 && x2 <= x3) ||
+             (x3 <= x2 && x2 <= x1)) {
+       x4=x2;
+    } else if((x2 <= x1 && x1 <= x3) ||
+             (x3 <= x1 && x1 <= x2)) {
+       x4=x1;
+    } else if((x1 <= x3 && x3 <= x2) ||
+             (x2 <= x3 && x3 <= x1)) {
+       x4=x3;
+    }
+
+          if((y1 <= y2 && y2 <= y3) ||
+             (y3 <= y2 && y2 <= y1)) {
+       y4=y2;
+    } else if((y2 <= y1 && y1 <= y3) ||
+             (y3 <= y1 && y1 <= y2)) {
+       y4=y1;
+    } else if((y1 <= y3 && y3 <= y2) ||
+             (y2 <= y3 && y3 <= y1)) {
+       y4=y3;
+    }
+
+    *px = x4;
+    *py = y4;
+    assert((x4>=-32 && x4<=31) && (y4>=-32 && y4<=31));
+}
+
+static inline int mvd2index(int px, int py, int x, int y, int xy)
+{
+    assert((x>=-32 && x<=31) && (y>=-32 && y<=31));
+    //assert((x&1)==0 && (y&1)==0);//for now
+    //assert((x&2)==0 && (y&2)==0);//for now(2)
+
+    x-=px;
+    y-=py;
+
+    if(xy)
+       x=y;
+    x+=32;
+
+    /* (x&63) */
+    if(x>63)
+       x-=64;
+    if(x<0)
+       x+=64;
+
+    assert(x>=0 && x<64);
+    return x;
+}
+
+static int encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
+{
+    block_t fb;
+    block_t b;
+    int dquant=0;
+    int has_mvd=0;
+    int has_mvd24=0;
+    int has_dc=1;
+    int mode = 0;
+    int cbpcbits = 0, cbpybits=0;
+    int diff;
+    int predictmvdx;
+    int predictmvdy;
+
+    block_t b_i;
+    int bits_i;
+
+    block_t fbold_v00;
+    block_t b_v00;
+    int bits_v00 = 65535;
+    int x_v00=0;
+    int y_v00=0;
+
+    diff = compareregions(s, bx, by);
+
+    if(diff < *quant) {
+       /* TODO: measure the error an I-block encoding would do, and base the decision
+          on that */
+       swf_SetBits(tag, 1,1); /* cod=1, block skipped */
+       /* copy the region from the last frame so that we have a complete reconstruction */
+       copyregion(s, s->current, s->oldpic, bx, by);
+       return 1;
+    }
+
+    predictmvd(s,bx,by,&predictmvdx,&predictmvdy);
+    getregion(&fb, s->current, bx, by, s->width);
+
+    { /* consider I-block */
+       block_t fb_i;
+       int y,c;
+       memcpy(&fb_i, &fb, sizeof(block_t));
+       dodctandquant(&fb_i, &b_i, 1, *quant);
+       //quantize(&fb_i, &b_i, 1, *quant);
+       getblockpatterns(&b_i, &y, &c, 1);
+       bits_i = 1; //cod
+       bits_i += mcbpc_inter[3*4+c].len;
+       bits_i += cbpy[y].len;
+       bits_i += coefbits8x8(b_i.y1, 1);
+       bits_i += coefbits8x8(b_i.y2, 1);
+       bits_i += coefbits8x8(b_i.y3, 1);
+       bits_i += coefbits8x8(b_i.y4, 1);
+       bits_i += coefbits8x8(b_i.u, 1);
+       bits_i += coefbits8x8(b_i.v, 1);
+    }
+
+    { /* consider mvd(x,y)-block */
+       block_t fbdiff;
+       int y,c;
+
+       x_v00=0;
+       y_v00=0;
+
+       if(s->do_motion) {
+           int hx,hy;
+           int bestx=0,besty=0,bestbits=65536;
+           int startx=-2,endx=2;
+           int starty=-2,endy=2;
+
+           if(!bx) startx=0;
+           if(!by) starty=0;
+           if(bx==s->bbx-1) endx=0;
+           if(by==s->bby-1) endy=0;
+
+           for(hx=startx;hx<=endx;hx+=1)
+           for(hy=starty;hy<=endy;hy+=1)
+           {
+               block_t b;
+               block_t fbold;
+               int bits = 0;
+               memcpy(&fbdiff, &fb, sizeof(block_t));
+               getmvdregion(&fbold, s->oldpic, bx, by, hx, hy, s->linex);
+               yuvdiff(&fbdiff, &fbold);
+               dodctandquant(&fbdiff, &b, 0, *quant);
+               //quantize(&fbdiff, &b, 0, *quant);
+               bits += coefbits8x8(b.y1, 0);
+               bits += coefbits8x8(b.y2, 0);
+               bits += coefbits8x8(b.y3, 0);
+               bits += coefbits8x8(b.y4, 0);
+               bits += coefbits8x8(b.u, 0);
+               bits += coefbits8x8(b.v, 0);
+               if(bits<bestbits) {
+                   bestbits = bits;
+                   bestx = hx;
+                   besty = hy;
+               }
+           }
+           x_v00 = bestx;
+           y_v00 = besty;
+       }
+
+       memcpy(&fbdiff, &fb, sizeof(block_t));
+       getmvdregion(&fbold_v00, s->oldpic, bx, by, x_v00, y_v00, s->linex);
+       yuvdiff(&fbdiff, &fbold_v00);
+       dodctandquant(&fbdiff, &b_v00, 0, *quant);
+       //quantize(&fbdiff, &b_v00, 0, *quant);
+       getblockpatterns(&b_v00, &y, &c, 0);
+
+       bits_v00 = 1; //cod
+       bits_v00 += mcbpc_inter[0*4+c].len;
+       bits_v00 += cbpy[y^15].len;
+       bits_v00 += mvd[mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 0)].len; // (0,0)
+       bits_v00 += mvd[mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 1)].len;
+       bits_v00 += coefbits8x8(b_v00.y1, 0);
+       bits_v00 += coefbits8x8(b_v00.y2, 0);
+       bits_v00 += coefbits8x8(b_v00.y3, 0);
+       bits_v00 += coefbits8x8(b_v00.y4, 0);
+       bits_v00 += coefbits8x8(b_v00.u, 0);
+       bits_v00 += coefbits8x8(b_v00.v, 0);
+    }
+
+    if(bits_i > bits_v00)
+    {
+       /* mvd (0,0) block (mode=0) */
+       int t;
+       mode = 0; // mvd w/o mvd24
+       has_dc = 0;
+       memcpy(&b, &b_v00, sizeof(block_t));
+
+       getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
+       swf_SetBits(tag,0,1); // COD
+       codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
+       codehuffman(tag, cbpy, cbpybits^15);
+
+       /* vector */
+       codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 0));
+       codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 1));
+       s->mvdx[by*s->bbx+bx] = x_v00;
+       s->mvdy[by*s->bbx+bx] = y_v00;
+
+       /* luminance */
+       encode8x8(tag, b.y1, has_dc, cbpybits&8);
+       encode8x8(tag, b.y2, has_dc, cbpybits&4);
+       encode8x8(tag, b.y3, has_dc, cbpybits&2);
+       encode8x8(tag, b.y4, has_dc, cbpybits&1);
+
+       /* chrominance */
+       encode8x8(tag, b.u, has_dc, cbpcbits&2);
+       encode8x8(tag, b.v, has_dc, cbpcbits&1);
+
+       /* -- reconstruction -- */
+       dequantize(&b, 0, *quant);
+       doidct(&b);
+       for(t=0;t<64;t++) {
+           b.y1[t] = truncate256(b.y1[t] + (int)fbold_v00.y1[t]);
+           b.y2[t] = truncate256(b.y2[t] + (int)fbold_v00.y2[t]);
+           b.y3[t] = truncate256(b.y3[t] + (int)fbold_v00.y3[t]);
+           b.y4[t] = truncate256(b.y4[t] + (int)fbold_v00.y4[t]);
+           b.u[t] = truncate256(b.u[t] + (int)fbold_v00.u[t]);
+           b.v[t] = truncate256(b.v[t] + (int)fbold_v00.v[t]);
+       }
+       copyblock(s, s->current, &b, bx, by);
+       return bits_v00;
+    } else {
+       /* i block (mode=3) */
+       mode = 3;
+       has_dc = 1;
+       memcpy(&b, &b_i, sizeof(block_t));
+       getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
+       swf_SetBits(tag,0,1); // COD
+       codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
+       codehuffman(tag, cbpy, cbpybits);
+
+       /* luminance */
+       encode8x8(tag, b.y1, has_dc, cbpybits&8);
+       encode8x8(tag, b.y2, has_dc, cbpybits&4);
+       encode8x8(tag, b.y3, has_dc, cbpybits&2);
+       encode8x8(tag, b.y4, has_dc, cbpybits&1);
+
+       /* chrominance */
+       encode8x8(tag, b.u, has_dc, cbpcbits&2);
+       encode8x8(tag, b.v, has_dc, cbpcbits&1);
+
+       /* -- reconstruction -- */
+       dequantize(&b, 1, *quant);
+       doidct(&b);
+       truncateblock(&b);
+       copyblock(s, s->current, &b, bx, by);
+       return bits_i;
+    }
+
+    exit(1);
+#if 0
+    dodct(&fb);
+    quantize(&fb, &b, has_dc, *quant);
+    getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
+
+    if(!dquant && has_mvd && !has_mvd24 && !has_dc) mode = 0;
+    else if(dquant && has_mvd && !has_mvd24 && !has_dc) mode = 1;
+    else if(!dquant && has_mvd && has_mvd24 && !has_dc) mode = 2;
+    else if(!dquant && !has_mvd && !has_mvd24 && has_dc) mode = 3;
+    else if(dquant && !has_mvd && !has_mvd24 && has_dc) mode = 4;
+    else exit(1);
+
+    swf_SetBits(tag,0,1); /* cod - 1 if we're not going to code this block*/
+
+    codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
+    codehuffman(tag, cbpy, (mode==3 || mode==4)?cbpybits:cbpybits^15);
+
+    if(dquant) {
+       setQuant(tag, dquant);
+    }
+
+    if(has_mvd) {
+       /* 0,0 */
+       codehuffman(tag, mvd, 32);
+       codehuffman(tag, mvd, 32);
+    }
+    if(has_mvd24) {
+    }
+
+    /* luminance */
+    encode8x8(tag, b.y1, has_dc, cbpybits&8);
+    encode8x8(tag, b.y2, has_dc, cbpybits&4);
+    encode8x8(tag, b.y3, has_dc, cbpybits&2);
+    encode8x8(tag, b.y4, has_dc, cbpybits&1);
+
+    /* chrominance */
+    encode8x8(tag, b.u, has_dc, cbpcbits&2);
+    encode8x8(tag, b.v, has_dc, cbpcbits&1);
+#endif
+}
+
+static int bmid = 0;
+void setdbgpic(TAG*tag, RGBA*pic, int width, int height)
+{
+    MATRIX m;
+    tag = tag->prev;
+
+    tag = swf_InsertTag(tag,ST_REMOVEOBJECT2);
+    swf_SetU16(tag, 133);
+
+    tag = swf_InsertTag(tag, ST_DEFINEBITSLOSSLESS);
+    swf_SetU16(tag, 1000+bmid);
+    swf_SetLosslessBits(tag, width, height, (void*)pic, BMF_32BIT);
+
+    tag = swf_InsertTag(tag, ST_DEFINESHAPE);
+    swf_SetU16(tag, 2000+bmid);
+    swf_ShapeSetBitmapRect(tag, 1000+bmid, width, height);
+
+    tag = swf_InsertTag(tag,ST_PLACEOBJECT2);
+    swf_GetMatrix(0,&m);
+    m.tx = width*20;
+    swf_ObjectPlace(tag, 2000+bmid, 133, &m, 0, 0);
+
+    bmid++;
+}
+
+#define TYPE_IFRAME 0
+#define TYPE_PFRAME 1
+
+static void writeHeader(TAG*tag, int width, int height, int frame, int quant, int type)
+{
+    U32 i32;
+    swf_SetU16(tag, frame);
+    swf_SetBits(tag, 1, 17); /* picture start code*/
+    swf_SetBits(tag, 0, 5); /* version=0, version 1 would optimize rle behaviour*/
+    swf_SetBits(tag, frame, 8); /* time reference */
+
+    /* write dimensions, taking advantage of some predefined sizes
+       if the opportunity presents itself */
+    i32 = width<<16|height;
+    switch(i32)
+    {
+       case 352<<16|288: swf_SetBits(tag, 2, 3);break;
+       case 176<<16|144: swf_SetBits(tag, 3, 3);break;
+       case 128<<16|96: swf_SetBits(tag, 4, 3);break;
+       case 320<<16|240: swf_SetBits(tag, 5, 3);break;
+       case 160<<16|120: swf_SetBits(tag, 6, 3);break;
+       default:
+           if(width>255 || height>255) {
+               swf_SetBits(tag, 1, 3);
+               swf_SetBits(tag, width, 16);
+               swf_SetBits(tag, height, 16);
+           } else {
+               swf_SetBits(tag, 0, 3);
+               swf_SetBits(tag, width, 8);
+               swf_SetBits(tag, height, 8);
+           }
+    }
+
+    swf_SetBits(tag, type, 2); /* I-Frame or P-Frame */
+    swf_SetBits(tag, 0, 1); /* No deblock filter */
+    assert(quant>0);
+    swf_SetBits(tag, quant, 5); /* quantizer (1-31), may be updated later on*/
+    swf_SetBits(tag, 0, 1); /* No extra info */
+}
+
+void swf_SetVideoStreamIFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
+{
+    int bx, by;
+
+    if(quant<1) quant=1;
+    if(quant>31) quant=31;
+
+    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_IFRAME);
+
+    rgb2yuv(s->current, pic, s->linex, s->olinex, s->width, s->height);
+
+    //dostat(s);
+
+    for(by=0;by<s->bby;by++)
+    {
+       for(bx=0;bx<s->bbx;bx++)
+       {
+           encode_blockI(tag, s, bx, by, &quant);
+       }
+    }
+    s->frame++;
+    memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
+}
+
+void swf_SetVideoStreamPFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
+{
+    int bx, by;
+
+    if(quant<1) quant=1;
+    if(quant>31) quant=31;
+
+    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
+
+    rgb2yuv(s->current, pic, s->linex, s->olinex, s->width, s->height);
+    memset(s->mvdx, 0, s->bbx*s->bby*sizeof(int));
+    memset(s->mvdy, 0, s->bbx*s->bby*sizeof(int));
+
+    for(by=0;by<s->bby;by++)
+    {
+       for(bx=0;bx<s->bbx;bx++)
+       {
+           encode_blockP(tag, s, bx, by, &quant);
+       }
+    }
+    s->frame++;
+    memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
+
+//#define PNG
+#ifdef MAIN
+#ifdef PNG
+    yuv2rgb(pic, s->current, s->linex, s->width, s->height);
+    setdbgpic(tag, pic, s->width, s->height);
+#endif
+    if(s->frame == (int)totalframes-1)
+    {
+       int t;
+       FILE*fi = fopen("test.ppm", "wb");
+       fprintf(fi, "P6\n%d %d\n255\n", s->width, s->height);
+       for(t=0;t<s->width*s->height;t++)
+       {
+           fwrite(&pic[t].r, 1, 1, fi);
+           fwrite(&pic[t].g, 1, 1, fi);
+           fwrite(&pic[t].b, 1, 1, fi);
+       }
+       fclose(fi);
+    }
+#endif
+}
+
+int uline[64],vline[64],yline[64];
+void swf_SetVideoStreamMover(TAG*tag, VIDEOSTREAM*s, int quant)
+{
+    int bx, by;
+
+    if(quant<1) quant=1;
+    if(quant>31) quant=31;
+
+    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
+
+    memset(s->mvdx, 0, s->bbx*s->bby*sizeof(int));
+    memset(s->mvdy, 0, s->bbx*s->bby*sizeof(int));
+
+    for(by=0;by<s->bby;by++)
+    {
+       for(bx=0;bx<s->bbx;bx++)
+       {
+           //if((lrand48()&255) || !(bx>8 && bx<24 && by>8 && by<24)) {
+           if(!(by==31)) {
+               /* mvd (0,0) block (mode=0) */
+               int t;
+               int mode = 0; // mvd w/o mvd24
+               int has_dc = 0;
+               int cbpybits=0,cbpcbits=0;
+               int predictmvdx, predictmvdy;
+               //int mvx=-1+(2*(s->frame&1));
+               //int mvy=-1+((s->frame&2));
+               int mvx=(lrand48()%4)-2;
+               int mvy=3;
+
+               swf_SetBits(tag,0,1); // COD
+               codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
+               codehuffman(tag, cbpy, cbpybits^15);
+
+               /* vector */
+               predictmvd(s,bx,by,&predictmvdx,&predictmvdy);
+               codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, mvx, mvy, 0));
+               codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, mvx, mvy, 1));
+               s->mvdx[by*s->bbx+bx] = mvx;
+               s->mvdy[by*s->bbx+bx] = mvy;
+           } else {
+               /* i block (mode=3) */
+               int mode = 3;
+               int has_dc = 1;
+               int cbpybits,cbpcbits;
+               int t;
+               block_t b;
+               memset(&b, 0, sizeof(block_t));
+               b.y1[0] = b.y2[0] = b.y3[0] = b.y4[0] = yline[bx];
+               b.u[0] = uline[bx];
+               b.v[0] = vline[bx];
+
+               getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
+               swf_SetBits(tag,0,1); // COD
+               codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
+               codehuffman(tag, cbpy, cbpybits);
+
+               /* luminance */
+               encode8x8(tag, b.y1, has_dc, cbpybits&8);
+               encode8x8(tag, b.y2, has_dc, cbpybits&4);
+               encode8x8(tag, b.y3, has_dc, cbpybits&2);
+               encode8x8(tag, b.y4, has_dc, cbpybits&1);
+
+               /* chrominance */
+               encode8x8(tag, b.u, has_dc, cbpcbits&2);
+               encode8x8(tag, b.v, has_dc, cbpcbits&1);
+           }
+       }
+    }
+}
+
+#ifdef MAIN
+#include "png.h"
+int main(int argn, char*argv[])
+{
+    int fi;
+    int t;
+    SWF swf;
+    TAG * tag;
+    RGBA* pic, *pic2, rgb;
+    SWFPLACEOBJECT obj;
+    int width = 0;
+    int height = 0;
+    int frames = 20;
+    int framerate = 29;
+    unsigned char*data;
+    char* fname = "/home/kramm/pics/peppers.png";
+    VIDEOSTREAM stream;
+    double d = 1.0;
+
+    memset(&stream, 0, sizeof(stream));
+
+    getPNG(fname, &width, &height, &data);
+    pic = (RGBA*)malloc(width*height*sizeof(RGBA));
+    pic2 = (RGBA*)malloc(width*height*sizeof(RGBA));
+    memcpy(pic, data, width*height*sizeof(RGBA));
+    free(data);
+
+    printf("Compressing %s, size %dx%d\n", fname, width, height);
+
+    memset(&swf,0,sizeof(SWF));
+    memset(&obj,0,sizeof(obj));
+
+    swf.fileVersion    = 6;
+    swf.frameRate      = framerate*256;
+    swf.movieSize.xmax = 20*width*2;
+    swf.movieSize.ymax = 20*height-20*64;
+
+    swf.firstTag = swf_InsertTag(NULL,ST_SETBACKGROUNDCOLOR);
+    tag = swf.firstTag;
+    rgb.r = 0x00;rgb.g = 0x00;rgb.b = 0x00;
+    swf_SetRGB(tag,&rgb);
+
+    tag = swf_InsertTag(tag, ST_DEFINEVIDEOSTREAM);
+    swf_SetU16(tag, 33);
+    swf_SetVideoStreamDefine(tag, &stream, frames, width, height);
+    stream.do_motion = 1;
+
+    for(t=0;t<frames;t++)
+    {
+       int x,y;
+       double xx,yy;
+       for(y=0,yy=0;y<height;y++,yy+=d)  {
+           RGBA*line = &pic[((int)yy)*width];
+           for(x=0,xx=0;x<width;x++,xx+=d) {
+               pic2[y*width+x] = line[((int)xx)];
+           }
+       }
+       printf("frame:%d\n", t);fflush(stdout);
+
+       tag = swf_InsertTag(tag, ST_VIDEOFRAME);
+       swf_SetU16(tag, 33);
+       if(t==0)
+           swf_SetVideoStreamIFrame(tag, &stream, pic2, 9);
+       else {
+           swf_SetVideoStreamPFrame(tag, &stream, pic2, 9);
+       }
+
+       tag = swf_InsertTag(tag, ST_PLACEOBJECT2);
+       swf_GetPlaceObject(0, &obj);
+       if(t==0) {
+           obj.depth = 1;
+           obj.id = 33;
+       } else {
+           obj.move = 1;
+           obj.depth = 1;
+           obj.ratio = t;
+       }
+       swf_SetPlaceObject(tag,&obj);
+
+       tag = swf_InsertTag(tag, ST_SHOWFRAME);
+       d-=0.015;
+    }
+    swf_VideoStreamClear(&stream);
+
+    tag = swf_InsertTag(tag, ST_END);
+
+    fi = open("video3.swf", O_WRONLY|O_CREAT|O_TRUNC, 0644);
+    if(swf_WriteSWC(fi,&swf)<0) {
+       fprintf(stderr,"WriteSWF() failed.\n");
+    }
+    close(fi);
+    swf_FreeTags(&swf);
+}
+#undef MAIN
+#endif