X-Git-Url: http://git.asbjorn.biz/?a=blobdiff_plain;f=lib%2Fh.263%2Fmkvideo.c;h=20e58339491766f2e708caf51099d5499a5cf166;hb=d6ae22a11779055f9b4cb417b5f93562da99aaec;hp=ebe859ab24a2569e9480109842e021f219bcbe0c;hpb=bac2af381e6c670485d9792d893f39a3df429ab9;p=swftools.git

diff --git a/lib/h.263/mkvideo.c b/lib/h.263/mkvideo.c
index ebe859a..20e5833 100644
--- a/lib/h.263/mkvideo.c
+++ b/lib/h.263/mkvideo.c
@@ -9,24 +9,9 @@
 #include <stdio.h>
 #include <assert.h>
 #include <math.h>
-#include "../lib/rfxswf.h"
-#include "png.h"
+#include "../rfxswf.h"
 #include "h263tables.c"
-
-typedef struct _YUV
-{
-    unsigned char y,u,v;
-} YUV;
-
-typedef struct _VIDEOSTREAM
-{
-    int width;
-    int height;
-    int frame;
-    int linex;
-    YUV*oldpic;
-    YUV*current;
-} VIDEOSTREAM;
+#include "swfvideo.h"
 
 void swf_SetVideoStreamDefine(TAG*tag, VIDEOSTREAM*stream, U16 frames, U16 width, U16 height)
 {
@@ -38,15 +23,29 @@ void swf_SetVideoStreamDefine(TAG*tag, VIDEOSTREAM*stream, U16 frames, U16 width
     swf_SetU8(tag, 2); /* codec = h.263 sorenson spark */
 
     memset(stream, 0, sizeof(VIDEOSTREAM));
+    stream->olinex = width;
+    width+=15;width&=~15;
+    height+=15;height&=~15;
     stream->linex = width;
-    width&=~15;
-    height&=~15;
     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
@@ -59,16 +58,6 @@ typedef struct _block_t
     int v[64];
 } block_t;
 
-typedef struct _fblock_t
-{
-    double y1[64];
-    double y2[64];
-    double y3[64];
-    double y4[64];
-    double u[64];
-    double v[64];
-} fblock_t;
-
 static int zigzagtable[64] = {
     0, 1, 5, 6, 14, 15, 27, 28, 
     2, 4, 7, 13, 16, 26, 29, 42, 
@@ -79,15 +68,14 @@ static int zigzagtable[64] = {
     21, 34, 37, 47, 50, 56, 59, 61, 
     35, 36, 48, 49, 57, 58, 62, 63};
 
-static void fzigzag(double*src) 
+static void zigzag(int*src) 
 {
-    double tmp[64];
+    int tmp[64];
     int t;
     for(t=0;t<64;t++) {
-	((int*)&tmp[zigzagtable[t]])[0] = ((int*)&src[t])[0];
-	((int*)&tmp[zigzagtable[t]])[1] = ((int*)&src[t])[1];
+	tmp[zigzagtable[t]] = src[t];
     }
-    memcpy(src, tmp, sizeof(double)*64);
+    memcpy(src, tmp, sizeof(int)*64);
 }
 
 #define PI 3.14159265358979
@@ -106,7 +94,7 @@ static double table[8][8] =
 {0.195090322016128,-0.555570233019602,0.831469612302545,-0.980785280403231,0.980785280403230,-0.831469612302545,0.555570233019602,-0.195090322016129}
 };
 
-static void dct(double*src)
+static void dct(int*src)
 {
     double tmp[64];
     int x,y,u,v,t;
@@ -129,11 +117,11 @@ static void dct(double*src)
 	{
 	    c+=table[v][y]*tmp[y*8+u];
 	}
-	src[v*8+u] = c*0.25;
+	src[v*8+u] = (int)(c*0.25);
     }
 }
 
-static void idct(double*src)
+static void idct(int*src)
 {
     double tmp[64];
     int x,y,u,v;
@@ -155,10 +143,100 @@ static void idct(double*src)
 	{
 	    c+=table[v][y]*tmp[v*8+x];
 	}
-	src[y*8+x] = c*0.25;
+	src[y*8+x] = (int)(c*0.25);
+    }
+}
+
+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;
@@ -166,38 +244,63 @@ static inline int truncate256(int a)
     return a;
 }
 
-static void getregion(fblock_t* bb, YUV*pic, int bx, int by, int linex)
+static void getregion(block_t* bb, YUV*pic, int posx, int posy, int linex)
 {
-    YUV*p1 = &pic[by*linex*16+bx*16];
-    YUV*p2 = p1;
+    YUV*p1;
+    YUV*p2;
     int y1=0, y2=0, y3=0, y4=0;
     int u=0,v=0;
     int x,y;
-    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;
+    int hp = (posy&1)<<1|(posx&1);
+    posx>>=1;
+    posy>>=1;
+    p1 = &pic[posy*linex+posx];
+    p2 = p1;
+    if(hp==0x0) {
+	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;
+	}
+    } else if(hp==0x1) {
+	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 + p1[x+1].y)/2;
+		bb->y2[y2++] = (p1[x+8].y + p1[x+8+1].y)/2;
+		bb->y3[y3++] = (p1[linex*8+x].y + p1[linex*8+x+1].y)/2;
+		bb->y4[y4++] = (p1[linex*8+x+8].y + p1[linex*8+x+8+1].y)/2;
+	    }
+	    p1+=linex;
+	    p2+=linex*2;
 	}
-	p1+=linex;
-	p2+=linex*2;
     }
+    else
+	assert(0);
 }
-static void rgb2yuv(YUV*dest, RGBA*src, int linex, int width, int height)
+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++) {
-	    double r,g,b;
-	    r = src[y*linex+x].r;
-	    g = src[y*linex+x].g;
-	    b = src[y*linex+x].b;
-	    dest[y*linex+x].y = (r*0.299 + g*0.587 + b*0.114);
-	    dest[y*linex+x].u = (r*-0.169 + g*-0.332 + b*0.500 + 128.0);
-	    dest[y*linex+x].v = (r*0.500 + g*-0.419 + b*-0.0813 + 128.0);
+	    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 = (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;
 	}
     }
 }
@@ -275,7 +378,7 @@ static int compareregions(VIDEOSTREAM*s, int bx, int by)
     return diff/256;
 }
 
-static int valtodc(int val)
+static inline int valtodc(int val)
 {
     assert(val>=0);
 
@@ -319,19 +422,24 @@ static int codehuffman(TAG*tag, struct huffcode*table, int index)
     return i;
 }
 
-static void quantize8x8(double*src, int*dest, int has_dc, int quant)
+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];
+	//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] = 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;
     }
 }
 
@@ -379,10 +487,57 @@ static int hascoef(int*b, int has_dc)
     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);
@@ -421,6 +576,7 @@ static void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
 		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) {
@@ -449,22 +605,43 @@ static void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
     }
 }
 
-static void dodct(fblock_t*fb)
+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)
 {
-    int t;
     dct(fb->y1); dct(fb->y2); dct(fb->y3); dct(fb->y4); 
     dct(fb->u);  dct(fb->v);  
-    fzigzag(fb->y1);
-    fzigzag(fb->y2);
-    fzigzag(fb->y3);
-    fzigzag(fb->y4);
-    fzigzag(fb->u);
-    fzigzag(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)
 {
-    fblock_t fb;
+    block_t fb;
     int t;
     for(t=0;t<64;t++) {
 	fb.y1[t] = b->y1[zigzagtable[t]];
@@ -498,15 +675,6 @@ static void truncateblock(block_t*b)
     }
 }
 
-static void quantize(fblock_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 dequantize(block_t*b, int has_dc, int quant)
 {
     dequantize8x8(b->y1, has_dc, quant); 
@@ -558,17 +726,18 @@ static void change_quant(int quant, int*dquant)
 
 static void encode_blockI(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
 {
-    fblock_t fb;
+    block_t fb;
     block_t b;
     int dquant=0;
     int cbpcbits = 0, cbpybits=0;
 
-    getregion(&fb, s->current, bx, by, s->width);
-    dodct(&fb);
+    getregion(&fb, s->current, bx*2*16, by*2*16, s->width);
     
     change_quant(*quant, &dquant);
     *quant+=dquant;
-    quantize(&fb, &b, 1, *quant);
+
+    dodctandquant(&fb, &b, 1, *quant);
+    //quantize(&fb, &b, 1, *quant);
 
     //decode_blockI(s, &b, bx, by);
 
@@ -603,7 +772,7 @@ static void encode_blockI(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
     copyblock(s, s->current, &b, bx, by);
 }
 
-static void yuvdiff(fblock_t*a, fblock_t*b)
+static void yuvdiff(block_t*a, block_t*b)
 {
     int t;
     for(t=0;t<64;t++) {
@@ -616,9 +785,79 @@ static void yuvdiff(fblock_t*a, fblock_t*b)
     }
 }
 
-static void encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
+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)
 {
-    fblock_t fb,fbdiff, fbold;
+    block_t fb;
     block_t b;
     int dquant=0;
     int has_mvd=0;
@@ -627,38 +866,130 @@ static void encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
     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 < 24 /*TODO: should be a parameter- good values are between 32 and 48 */) {
+    if(diff < 20 /*TODO: should be a parameter- good values are between 32 and 48 */) {
 	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;
+	return 1;
     }
 
-    getregion(&fb, s->current, bx, by, s->width);
+    predictmvd(s,bx,by,&predictmvdx,&predictmvdy);
+    getregion(&fb, s->current, bx*2*16, by*2*16, 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=-8,endx=8;
+	    int starty=-8,endy=8;
+
+	    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+=4)
+	    for(hy=starty;hy<=endy;hy+=4)
+	    {
+		block_t b;
+		block_t fbold;
+		int bits = 0;
+		memcpy(&fbdiff, &fb, sizeof(block_t));
+		getregion(&fbold, s->oldpic, bx*2*16+hx, by*2*16+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;
+	}
 
-    //if(diff < 128)
-    if(1)
+	memcpy(&fbdiff, &fb, sizeof(block_t));
+	getregion(&fbold_v00, s->oldpic, bx*2*16+x_v00, by*2*16+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;
-    	memcpy(&fbdiff, &fb, sizeof(fblock_t));
-    	getregion(&fbold, s->oldpic, bx, by, s->linex);
-    	yuvdiff(&fbdiff, &fbold);
-	//getregiondiff(&fbdiff, s->oldpic, s->current, bx, by, s->linex);
-    	dodct(&fbdiff);
+	mode = 0; // mvd w/o mvd24
 	has_dc = 0;
+	memcpy(&b, &b_v00, sizeof(block_t));
 
-    	quantize(&fbdiff, &b, has_dc, *quant);
 	getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
-	mode = 0; // mvd w/o mvd24
 	swf_SetBits(tag,0,1); // COD
 	codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
 	codehuffman(tag, cbpy, cbpybits^15);
 
-	/* 0,0 */
-	codehuffman(tag, mvd, 32);
-	codehuffman(tag, mvd, 32);
+	/* 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);
@@ -669,27 +1000,25 @@ static void encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
 	/* 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.y1[t]);
-	    b.y2[t] = truncate256(b.y2[t] + (int)fbold.y2[t]);
-	    b.y3[t] = truncate256(b.y3[t] + (int)fbold.y3[t]);
-	    b.y4[t] = truncate256(b.y4[t] + (int)fbold.y4[t]);
-	    b.u[t] = truncate256(b.u[t] + (int)fbold.u[t]);
-	    b.v[t] = truncate256(b.v[t] + (int)fbold.v[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;
+	return bits_v00;
     } else {
 	/* i block (mode=3) */
 	mode = 3;
 	has_dc = 1;
-	dodct(&fb);
-	quantize(&fb, &b, has_dc, *quant);
+	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);
@@ -704,12 +1033,13 @@ static void encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
 	/* 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;
+	return bits_i;
     }
 
     exit(1);
@@ -793,22 +1123,22 @@ static void writeHeader(TAG*tag, int width, int height, int frame, int quant, in
     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)
+void swf_SetVideoStreamIFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
 {
-    int bx, by, bbx, bby;
-    int quant = 3;
+    int bx, by;
+
+    if(quant<1) quant=1;
+    if(quant>31) quant=31;
 
     writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_IFRAME);
 
-    bbx = (s->width+15)/16; //TODO: move bbx,bby into VIDEOSTREAM
-    bby = (s->height+15)/16;
+    rgb2yuv(s->current, pic, s->linex, s->olinex, s->width, s->height);
 
-    rgb2yuv(s->current, pic, s->linex, s->width, s->height);
+    //dostat(s);
 
-    for(by=0;by<bby;by++)
+    for(by=0;by<s->bby;by++)
     {
-	for(bx=0;bx<bbx;bx++)
+	for(bx=0;bx<s->bbx;bx++)
 	{
 	    encode_blockI(tag, s, bx, by, &quant);
 	}
@@ -817,28 +1147,29 @@ void swf_SetVideoStreamIFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic)
     memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
 }
 
-void swf_SetVideoStreamPFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic)
+void swf_SetVideoStreamPFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
 {
-    int bx, by, bbx, bby;
-    int quant = 3;
+    int bx, by;
 
-    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
+    if(quant<1) quant=1;
+    if(quant>31) quant=31;
 
-    bbx = (s->width+15)/16;
-    bby = (s->height+15)/16;
+    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
 
-    rgb2yuv(s->current, pic, s->linex, s->width, s->height);
+    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<bby;by++)
+    for(by=0;by<s->bby;by++)
     {
-	for(bx=0;bx<bbx;bx++)
+	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));
-
+#ifdef MAIN
     {
 	int t;
 	FILE*fi = fopen("test.ppm", "wb");
@@ -852,8 +1183,11 @@ void swf_SetVideoStreamPFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic)
 	}
 	fclose(fi);
     }
+#endif
 }
 
+#ifdef MAIN
+#include "png.h"
 int main(int argn, char*argv[])
 {
     int fi;
@@ -864,7 +1198,7 @@ int main(int argn, char*argv[])
     SWFPLACEOBJECT obj;
     int width = 0;
     int height = 0;
-    int frames = 50;
+    int frames = 5;
     int framerate = 29;
     unsigned char*data;
     char* fname = "/home/kramm/pics/peppers.png";
@@ -897,6 +1231,7 @@ int main(int argn, char*argv[])
     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++)
     {
@@ -913,9 +1248,9 @@ int main(int argn, char*argv[])
 	tag = swf_InsertTag(tag, ST_VIDEOFRAME);
 	swf_SetU16(tag, 33);
 	if(t==0)
-	    swf_SetVideoStreamIFrame(tag, &stream, pic2);
+	    swf_SetVideoStreamIFrame(tag, &stream, pic2, 9);
 	else
-	    swf_SetVideoStreamPFrame(tag, &stream, pic2);
+	    swf_SetVideoStreamPFrame(tag, &stream, pic2, 9);
 
 	tag = swf_InsertTag(tag, ST_PLACEOBJECT2);
 	swf_GetPlaceObject(0, &obj);
@@ -932,6 +1267,8 @@ int main(int argn, char*argv[])
 	tag = swf_InsertTag(tag, ST_SHOWFRAME);
 	d-=0.005;
     }
+
+    swf_VideoStreamClear(&stream);
    
     tag = swf_InsertTag(tag, ST_END);
 
@@ -942,3 +1279,5 @@ int main(int argn, char*argv[])
     close(fi);
     swf_FreeTags(&swf);
 }
+#undef MAIN
+#endif