X-Git-Url: http://git.asbjorn.biz/?a=blobdiff_plain;f=lib%2Fh.263%2Fmkvideo.c;h=250f4c738a5385e1c2ff9144b3386bf0dcbf1021;hb=95ab8f4e51eac64a2cb420a9c309e0f3cbdf0048;hp=88a482a5486bcdcaca373cf64ca5b7ce1434ea64;hpb=78235556f27a7dc3928cc88aef6fba0a5f4b03d9;p=swftools.git

diff --git a/lib/h.263/mkvideo.c b/lib/h.263/mkvideo.c
index 88a482a..250f4c7 100644
--- a/lib/h.263/mkvideo.c
+++ b/lib/h.263/mkvideo.c
@@ -13,8 +13,22 @@
 #include "png.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;
 
-void swf_SetVideoStreamDefine(TAG*tag, U16 frames, U16 width, U16 height)
+void swf_SetVideoStreamDefine(TAG*tag, VIDEOSTREAM*stream, U16 frames, U16 width, U16 height)
 {
     width=width&~15; height=height&~15;
     swf_SetU16(tag, frames);
@@ -22,9 +36,20 @@ void swf_SetVideoStreamDefine(TAG*tag, U16 frames, U16 width, U16 height)
     swf_SetU16(tag, height);
     swf_SetU8(tag, 1); /* smoothing on */
     swf_SetU8(tag, 2); /* codec = h.263 sorenson spark */
+
+    memset(stream, 0, sizeof(VIDEOSTREAM));
+    stream->linex = width;
+    width&=~15;
+    height&=~15;
+    stream->width = width;
+    stream->height = height;
+    stream->current = (YUV*)malloc(width*height*sizeof(YUV));
+    stream->oldpic = (YUV*)malloc(width*height*sizeof(YUV));
+
+    memset(stream->oldpic, 0, width*height*sizeof(YUV));
 }
 
-struct block_t
+typedef struct _block_t
 {
     int y1[64];
     int y2[64];
@@ -32,9 +57,9 @@ struct block_t
     int y4[64];
     int u[64];
     int v[64];
-};
+} block_t;
 
-struct fblock_t
+typedef struct _fblock_t
 {
     double y1[64];
     double y2[64];
@@ -42,109 +67,221 @@ struct fblock_t
     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, 
+    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};
 
-void zigzag(int*src) 
-{
-    int table[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};
-    int tmp[64];
+static void fzigzag(double*src) 
+{
+    double tmp[64];
     int t;
     for(t=0;t<64;t++) {
-	tmp[table[t]] = src[t];
+	((int*)&tmp[zigzagtable[t]])[0] = ((int*)&src[t])[0];
+	((int*)&tmp[zigzagtable[t]])[1] = ((int*)&src[t])[1];
     }
-    memcpy(src, tmp, sizeof(int)*64);
+    memcpy(src, tmp, sizeof(double)*64);
 }
 
 #define PI 3.14159265358979
 #define SQRT2 1.414214
 #define RSQRT2 (1.0/1.414214)
 
-void dct(double*src)
+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(double*src)
 {
     double tmp[64];
-    int x,y,u,v;
-    for(y=0;y<8;y++)
-    for(x=0;x<8;x++)
+    int x,y,u,v,t;
+
+    for(v=0;v<8;v++)
+    for(u=0;u<8;u++)
     {
 	double c = 0;
-	for(v=0;v<8;v++)
-	for(u=0;u<8;u++)
+	for(x=0;x<8;x++)
 	{
-	    double f = 0.25*cos(PI*(2.0*x+1.0)*u/16.0)*cos(PI*(2.0*y+1.0)*v/16.0);
-	    if(!u) f *= RSQRT2;
-	    if(!v) f *= RSQRT2;
-	    c+=f*src[v*8+u];
+	    c+=table[u][x]*src[v*8+x];
 	}
-	tmp[y*8+x] = c;
+	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] = c*0.25;
     }
-    memcpy(src, tmp, sizeof(double)*64);
 }
 
-void idct(double*src)
+static void idct(double*src)
 {
     double tmp[64];
     int x,y,u,v;
-    for(v=0;v<8;v++)
-    for(u=0;u<8;u++)
+    for(y=0;y<8;y++)
+    for(x=0;x<8;x++)
     {
 	double c = 0;
-	for(y=0;y<8;y++)
-	for(x=0;x<8;x++)
+	for(u=0;u<8;u++)
 	{
-	    double f = 0.25*cos(PI*(2.0*x+1.0)*u/16.0)*cos(PI*(2.0*y+1.0)*v/16.0);
-	    if(!u) f *= RSQRT2;
-	    if(!v) f *= RSQRT2;
-	    c+=f*src[y*8+x];
+	    c+=table[u][x]*src[y*8+u];
 	}
-	tmp[v*8+u] = c;
+	tmp[y*8+x] = c;
     }
-    memcpy(src, tmp, sizeof(double)*64);
+    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] = c*0.25;
+    }
+}
+
+static inline int truncate256(int a)
+{
+    if(a>255) return 255;
+    if(a<0) return 0;
+    return a;
 }
 
-void getregion(struct fblock_t* bb, RGBA*pic, int bx, int by, int width, int height)
+static void getregion(fblock_t* bb, YUV*pic, int bx, int by, int linex)
 {
-    RGBA*p1 = &pic[by*width*16+bx*16];
-    RGBA*p2 = p1;
-    int linex = width;
+    YUV*p1 = &pic[by*linex*16+bx*16];
+    YUV*p2 = p1;
     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++) {
-	    double r,g,b;
-	    r = (p2[x*2].r + p2[x*2+1].r + p2[linex+x*2].r + p2[linex+x*2+1].r)/4.0;
-	    g = (p2[x*2].g + p2[x*2+1].g + p2[linex+x*2].g + p2[linex+x*2+1].g)/4.0;
-	    b = (p2[x*2].b + p2[x*2+1].b + p2[linex+x*2].b + p2[linex+x*2+1].b)/4.0;
-	    bb->u[u++] = (r*-0.169 + g*-0.332 + b*0.500 + 128.0);
-	    bb->v[v++] = (r*0.500 + g*-0.419 + b*-0.0813 + 128.0);
-
-	    r = p1[x].r; g = p1[x].g; b = p1[x].b;
-	    bb->y1[y1++] = (r*0.299 + g*0.587 + b*0.114);
-	    r = p1[x+8].r; g = p1[x+8].g; b = p1[x+8].b;
-	    bb->y2[y2++] = (r*0.299 + g*0.587 + b*0.114);
-	    r = p1[linex*8+x].r; g = p1[linex*8+x].g; b = p1[linex*8+x].b;
-	    bb->y3[y3++] = (r*0.299 + g*0.587 + b*0.114);
-	    r = p1[linex*8+x+8].r; g = p1[linex*8+x+8].g; b = p1[linex*8+x+8].b;
-	    bb->y4[y4++] = (r*0.299 + g*0.587 + b*0.114);
+	    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 rgb2yuv(YUV*dest, RGBA*src, int linex, 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);
+	}
+    }
+}
+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;
+    }
+}
 
-int valtodc(int val)
+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 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)
@@ -155,8 +292,20 @@ int valtodc(int val)
 	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;
+}
 
-void codehuffman(TAG*tag, struct huffcode*table, int index)
+static int codehuffman(TAG*tag, struct huffcode*table, int index)
 {
     /* TODO: !optimize! */
     int i=0;
@@ -167,48 +316,120 @@ void codehuffman(TAG*tag, struct huffcode*table, int index)
 	    swf_SetBits(tag, 1, 1);
 	i++;
     }
+    return i;
 }
 
-int see=0;
-
-void quantize(double*src, int*dest, int quant, int has_dc)
+static void quantize8x8(double*src, int*dest, int has_dc, int quant)
 {
     int t,pos=0;
     if(has_dc) {
-	dest[0] = (int)src[0]; /*DC*/
-	dest[0] = valtodc(dest[0]);
+	dest[0] = valtodc((int)src[0]); /*DC*/
 	pos++;
     }
     for(t=pos;t<64;t++)
     {
 	dest[t] = (int)src[t];
-	//val = (quant*(2*level+1)-1)+quant&1
-	if(quant&1) {
-	    dest[t] = (dest[t]/quant - 1)/2;
-	} else {
-	    dest[t] = ((dest[t]+1)/quant - 1)/2;
+    /* 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);
+    }
+}
+
+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;
     }
 }
 
-int hascoef(int*b, int has_dc)
+static int hascoef(int*b, int has_dc)
 {
     int t;
     int pos=0;
-    int range=2;
     if(has_dc)
 	pos++;
     for(t=pos;t<64;t++) {
-	if(b[t]<=-range || b[t]>=range)
+	if(b[t])
 	    return 1;
     }
     return 0;
 }
 
-void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
+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);
@@ -219,12 +440,15 @@ void encode8x8(TAG*tag, int*bb, int has_dc, int 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;
 	}
-	assert(bb[last]);
 	/* blocks without coefficients should not be included
-	   in the cbpy/cbpc patterns */
+	   in the cbpy/cbpc patterns: */
+	assert(bb[last]);
+
 	while(1) {
 	    int run=0;
 	    int level=0;
@@ -238,11 +462,13 @@ void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
 	    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) {
@@ -253,87 +479,377 @@ void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
 	    }
 	    if(t==RLE_ESCAPE) {
 		codehuffman(tag, rle, RLE_ESCAPE);
+		level=bb[pos];
 		/* table 14/h.263 */
 		assert(level);
-		if(level<-127) level = -127;
-		if(level>127) level = 127;
+		assert(level>=-127);
+		assert(level<=127);
 
 		swf_SetBits(tag, islast, 1);
 		swf_SetBits(tag, run, 6);
-		swf_SetBits(tag, level, 8); //fixme
+		swf_SetBits(tag, level, 8); //FIXME: fixme??
 	    }
 
 	    if(islast)
 		break;
 	    pos++;
 	}
+    }
+}
 
-	//codehuffman(tag, rle, 58);
-	//swf_SetBits(tag, 1, 1); //sign
+static void dodct(fblock_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); 
+}
+
+static void doidct(block_t*b)
+{
+    fblock_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]);
     }
-    see++;
 }
 
-void encode_blockI(TAG*tag, RGBA*pic, int bx, int by, int width, int height, int*quant)
+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)
 {
-    struct fblock_t b;
-    int dquant=0; /* TODO: should we take advantage of the dquant feature?*/
-    //int cbpcbits = 3;
-    //int cbpybits = 15;
-    int cbpcbits = 0;
-    int cbpybits = 0;
-    int y1[64],y2[64],y3[64],y4[64],u[64],v[64];
+    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;
 
-    getregion(&b, pic, bx, by, width, height);
-    dct(b.y1); quantize(b.y1,y1,1,*quant); zigzag(y1); cbpybits|=hascoef(y1, 1)*8;
-    dct(b.y2); quantize(b.y2,y2,1,*quant); zigzag(y2); cbpybits|=hascoef(y2, 1)*4;
-    dct(b.y3); quantize(b.y3,y3,1,*quant); zigzag(y3); cbpybits|=hascoef(y3, 1)*2;
-    dct(b.y4); quantize(b.y4,y4,1,*quant); zigzag(y4); cbpybits|=hascoef(y4, 1)*1;
-    dct(b.u); quantize(b.u,u,1,*quant); zigzag(u); cbpcbits|=hascoef(u, 1)*2;
-    dct(b.v); quantize(b.v,v,1,*quant); zigzag(v); cbpcbits|=hascoef(v, 1)*1;
+    *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)
+{
+    fblock_t fb;
+    block_t b;
+    int dquant=0;
+    int cbpcbits = 0, cbpybits=0;
+
+    getregion(&fb, s->current, bx, by, s->width);
+    dodct(&fb);
+    
+    change_quant(*quant, &dquant);
+    *quant+=dquant;
+    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);
     }
-    //swf_SetBits(tag, 1, 1); /*cbpc-00 mb_type=3 (no bquant), cbc=0*/
 
-    /* if this was an intra frame, we'd need to code (cbpybits^15) */
     codehuffman(tag, cbpy, cbpybits);
 
     if(dquant) {
-	/* 00 01 10 11
-	   -1 -2 +1 +2
-	*/
-	swf_SetBits(tag, 0x3, 2);
+	setQuant(tag, dquant);
     }
 
     /* luminance */
-    encode8x8(tag, y1, 1, cbpybits&8);
-    encode8x8(tag, y2, 1, cbpybits&4);
-    encode8x8(tag, y3, 1, cbpybits&2);
-    encode8x8(tag, y4, 1, cbpybits&1);
-    /*swf_SetBits(tag, 0x1, 8);
-    swf_SetBits(tag, 0xfe, 8);
-    swf_SetBits(tag, 0x1, 8);
-    swf_SetBits(tag, 0x1, 8);*/
+    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, u, 1, cbpcbits&2);
-    encode8x8(tag, v, 1, cbpcbits&1);
-    /*swf_SetBits(tag, 0xfe, 8);
-    swf_SetBits(tag, 0xfe, 8);*/
+    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);
 }
 
-void swf_SetVideoStreamIFrame(TAG*tag, RGBA*pic, U16 width, U16 height, int frame)
+static void yuvdiff(fblock_t*a, fblock_t*b)
 {
-    U32 i32;
-    int bx, by, bbx, bby;
-    int quant = 9;
+    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]);
+    }
+}
 
-    width=width&~15; height=height&~15;
+static int encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
+{
+    fblock_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;
+
+    block_t b_i;
+    int bits_i;
+
+    fblock_t fbold_v00;
+    block_t b_v00;
+    int bits_v00;
+
+    diff = compareregions(s, bx, by);
+    if(diff < 24 /*TODO: should be a parameter- good values are between 32 and 48 */) {
+	swf_SetBits(tag, 1,1); /* cod=1, block skipped */
+	copyregion(s, s->current, s->oldpic, bx, by);
+	return;
+    }
+
+    getregion(&fb, s->current, bx, by, s->width);
+
+    { /* consider I-block */
+	fblock_t fb_i;
+	int y,c;
+	memcpy(&fb_i, &fb, sizeof(fblock_t));
+	dodct(&fb_i);
+	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(0,0)-block */
+	fblock_t fbdiff;
+	int y,c;
+    	memcpy(&fbdiff, &fb, sizeof(fblock_t));
+    	getregion(&fbold_v00, s->oldpic, bx, by, s->linex);
+    	yuvdiff(&fbdiff, &fbold_v00);
+    	dodct(&fbdiff);
+    	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[32].len; // (0,0)
+	bits_v00 += mvd[32].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);
+
+	/* 0,0 */
+	codehuffman(tag, mvd, 32);
+	codehuffman(tag, mvd, 32);
+
+	/* 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));
+	//dodct(&fb);
+	//quantize(&fb, &b, has_dc, *quant);
+	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
+}
+
+#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*/
@@ -361,48 +877,105 @@ void swf_SetVideoStreamIFrame(TAG*tag, RGBA*pic, U16 width, U16 height, int fram
 	    }
     }
 
-    swf_SetBits(tag, 0, 2); /* I-Frame */
+    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 */
+}
 
-    bbx = (width+15)/16;
-    bby = (height+15)/16;
+void swf_SetVideoStreamIFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic)
+{
+    int bx, by, bbx, bby;
+    int 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->width, s->height);
+
+    for(by=0;by<bby;by++)
+    {
+	for(bx=0;bx<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 bx, by, bbx, bby;
+    int quant = 31;
+
+    writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
+
+    bbx = (s->width+15)/16;
+    bby = (s->height+15)/16;
+
+    rgb2yuv(s->current, pic, s->linex, s->width, s->height);
 
     for(by=0;by<bby;by++)
     {
 	for(bx=0;bx<bbx;bx++)
 	{
-	    encode_blockI(tag, pic, bx, by, width, height, &quant);
+	    encode_blockP(tag, s, bx, by, &quant);
 	}
     }
+    s->frame++;
+    memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
+
+    {
+	int t;
+	FILE*fi = fopen("test.ppm", "wb");
+	yuv2rgb(pic, s->current, s->linex, s->width, s->height);
+	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);
+    }
 }
 
-int main()
+int main(int argn, char*argv[])
 {
     int fi;
     int t;
     SWF swf;
     TAG * tag;
-    RGBA* pic, rgb;
+    RGBA* pic, *pic2, rgb;
     SWFPLACEOBJECT obj;
     int width = 0;
     int height = 0;
-    int frames = 1;
+    int frames = 2;
+    int framerate = 1;
     unsigned char*data;
-    char* fname = "/home/kramm/pics/lena.png";
+    char* fname = "/home/kramm/pics/peppers.png";
+    VIDEOSTREAM stream;
+    double d = 1.0;
+
+    memset(&stream, 0, sizeof(stream));
 
-    getimage(fname, &width, &height, &data);
+    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      = 29*256;
+    swf.frameRate      = framerate*256;
     swf.movieSize.xmax = 20*width;
     swf.movieSize.ymax = 20*height;
 
@@ -413,13 +986,26 @@ int main()
 
     tag = swf_InsertTag(tag, ST_DEFINEVIDEOSTREAM);
     swf_SetU16(tag, 33);
-    swf_SetVideoStreamDefine(tag, frames, width, height);
+    swf_SetVideoStreamDefine(tag, &stream, frames, width, height);
     
     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);
-	swf_SetVideoStreamIFrame(tag, pic, width, height, t);
+	if(t==0)
+	    swf_SetVideoStreamIFrame(tag, &stream, pic2);
+	else
+	    swf_SetVideoStreamPFrame(tag, &stream, pic2);
 
 	tag = swf_InsertTag(tag, ST_PLACEOBJECT2);
 	swf_GetPlaceObject(0, &obj);
@@ -434,12 +1020,13 @@ int main()
 	swf_SetPlaceObject(tag,&obj);
 
 	tag = swf_InsertTag(tag, ST_SHOWFRAME);
+	d-=0.005;
     }
    
     tag = swf_InsertTag(tag, ST_END);
 
     fi = open("video3.swf", O_WRONLY|O_CREAT|O_TRUNC, 0644);
-    if(swf_WriteSWF(fi,&swf)<0) {
+    if(swf_WriteSWC(fi,&swf)<0) {
 	fprintf(stderr,"WriteSWF() failed.\n");
     }
     close(fi);