4 Part of the swftools package.
6 Copyright (c) 2003 Matthias Kramm <kramm@quiss.org> */
12 #include "../rfxswf.h"
13 #include "h263tables.c"
19 void swf_SetVideoStreamDefine(TAG*tag, VIDEOSTREAM*stream, U16 frames, U16 width, U16 height)
21 swf_SetU16(tag, frames);
22 swf_SetU16(tag, width);
23 swf_SetU16(tag, height);
24 //swf_SetU8(tag, 1); /* smoothing on */
25 swf_SetU8(tag, 0); /* smoothing off */
26 swf_SetU8(tag, 2); /* codec = h.263 sorenson spark */
30 memset(stream, 0, sizeof(VIDEOSTREAM));
31 stream->olinex = width;
33 height+=15;height&=~15;
34 stream->linex = width;
35 stream->width = width;
36 stream->height = height;
37 stream->bbx = width/16;
38 stream->bby = height/16;
39 stream->current = (YUV*)malloc(width*height*sizeof(YUV));
40 stream->oldpic = (YUV*)malloc(width*height*sizeof(YUV));
41 stream->mvdx = (int*)malloc(stream->bbx*stream->bby*sizeof(int));
42 stream->mvdy = (int*)malloc(stream->bbx*stream->bby*sizeof(int));
43 stream->do_motion = 0;
45 memset(stream->oldpic, 0, width*height*sizeof(YUV));
46 memset(stream->current, 0, width*height*sizeof(YUV));
48 void swf_VideoStreamClear(VIDEOSTREAM*stream)
50 free(stream->oldpic);stream->oldpic = 0;
51 free(stream->current);stream->current = 0;
52 free(stream->mvdx);stream->mvdx=0;
53 free(stream->mvdy);stream->mvdy=0;
56 typedef struct _block_t
66 static int zigzagtable[64] = {
67 0, 1, 5, 6, 14, 15, 27, 28,
68 2, 4, 7, 13, 16, 26, 29, 42,
69 3, 8, 12, 17, 25, 30, 41, 43,
70 9, 11, 18, 24, 31, 40, 44, 53,
71 10, 19, 23, 32, 39, 45, 52, 54,
72 20, 22, 33, 38, 46, 51, 55, 60,
73 21, 34, 37, 47, 50, 56, 59, 61,
74 35, 36, 48, 49, 57, 58, 62, 63};
76 static void zigzag(int*src)
81 tmp[zigzagtable[t]] = src[t];
83 memcpy(src, tmp, sizeof(int)*64);
86 #define PI 3.14159265358979
87 #define SQRT2 1.414214
88 #define RSQRT2 (1.0/1.414214)
90 static double table[8][8] =
92 {0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548,0.707106781186548},
93 {0.980785280403230,0.831469612302545,0.555570233019602,0.195090322016128,-0.195090322016128,-0.555570233019602,-0.831469612302545,-0.980785280403230},
94 {0.923879532511287,0.382683432365090,-0.382683432365090,-0.923879532511287,-0.923879532511287,-0.382683432365090,0.382683432365090,0.923879532511287},
95 {0.831469612302545,-0.195090322016128,-0.980785280403230,-0.555570233019602,0.555570233019602,0.980785280403230,0.195090322016129,-0.831469612302545},
96 {0.707106781186548,-0.707106781186547,-0.707106781186548,0.707106781186547,0.707106781186548,-0.707106781186547,-0.707106781186547,0.707106781186547},
97 {0.555570233019602,-0.980785280403230,0.195090322016128,0.831469612302545,-0.831469612302545,-0.195090322016128,0.980785280403231,-0.555570233019602},
98 {0.382683432365090,-0.923879532511287,0.923879532511287,-0.382683432365090,-0.382683432365091,0.923879532511287,-0.923879532511286,0.382683432365090},
99 {0.195090322016128,-0.555570233019602,0.831469612302545,-0.980785280403231,0.980785280403230,-0.831469612302545,0.555570233019602,-0.195090322016129}
102 static void dct(int*src)
113 c+=table[u][x]*src[v*8+x];
123 c+=table[v][y]*tmp[y*8+u];
125 src[v*8+u] = (int)(c*0.25+0.5);
129 static void idct(int*src)
139 c+=table[u][x]*src[y*8+u];
149 c+=table[v][y]*tmp[v*8+x];
151 src[y*8+x] = (int)(c*0.25+0.5);
155 static double c[8] = {1.0,
156 0.980785280403230, // cos(Pi*1/16), sin(Pi*7/16)
157 0.923879532511287, // cos(Pi*2/16), sin(Pi*6/16)
158 0.831469612302545, // cos(Pi*3/16), sin(Pi*5/16)
159 0.707106781186548, // cos(Pi*4/16), sin(Pi*4/16), 1/sqrt(2)
160 0.555570233019602, // cos(Pi*5/16), sin(Pi*3/16)
161 0.382683432365090, // cos(Pi*6/16), sin(Pi*2/16)
162 0.195090322016128 // cos(Pi*7/16), sin(Pi*1/16)
168 static void preparequant(int quant)
172 cc[0] = c[0]/(quant*2*4);
173 cc[1] = c[1]/(quant*2*4);
174 cc[2] = c[2]/(quant*2*4);
175 cc[3] = c[3]/(quant*2*4);
176 cc[4] = c[4]/(quant*2*4);
177 cc[5] = c[5]/(quant*2*4);
178 cc[6] = c[6]/(quant*2*4);
179 cc[7] = c[7]/(quant*2*4);
183 inline static void innerdct(double*a,double*b, double*c)
190 //{ 1, 3, 5, 7, -7, -5, -3, -1},
191 //{ 3, -7, -1, -5, 5, 1, 7, -3},
192 //{ 5, -1, 7, 3, -3, -7, 1, -5},
193 //{ 7, -5, 3, -1, 1, -3, 5, -7}
194 double b0,b1,b2,b3,b4,b5;
204 b[2*8] = (b2-b5)*c[2] + (b3-b4)*c[6];
205 b[6*8] = (b2-b5)*c[6] + (b4-b3)*c[2];
212 b[1*8] = b0*c[1] + b1*c[3] + b2*c[5] + b3*c[7];
213 b[3*8] = b0*c[3] - b1*c[7] - b2*c[1] - b3*c[5];
214 b[5*8] = b0*c[5] - b1*c[1] + b2*c[7] + b3*c[3];
215 b[7*8] = b0*c[7] - b1*c[5] + b2*c[3] - b3*c[1];
218 static void dct2(int*src, int*dest)
220 double tmp[64], tmp2[64];
229 double* a=&tmp2[v*8];
240 dest[zigzagtable[t]] = (int)(tmp2[t]);
245 static inline int truncate256(int a)
247 if(a>255) return 255;
252 static void getregion(block_t* bb, YUV*pic, int posx, int posy, int linex)
256 int y1=0, y2=0, y3=0, y4=0;
261 p1 = &pic[posy*linex+posx];
265 bb->u[u++] = (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
266 bb->v[v++] = (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
267 bb->y1[y1++] = p1[x].y;
268 bb->y2[y2++] = p1[x+8].y;
269 bb->y3[y3++] = p1[linex*8+x].y;
270 bb->y4[y4++] = p1[linex*8+x+8].y;
277 static void getmvdregion(block_t* bb, YUV*pic, int posx, int posy, int mvdx, int mvdy, int linex)
285 posx = posx*16 + ((mvdx&~1)/2);
286 posy = posy*16 + ((mvdy&~1)/2);
287 p1 = &pic[posy*linex+posx];
288 p2 = &pic[(posy&~1)*linex+(posx&~1)];
289 uvhp = ((mvdx&1)|((mvdx>>1)&1))|((mvdy&2)|((mvdy&1)<<1));
290 yhp = ((mvdy&1)<<1|(mvdx&1));
293 if(yhp==0 || yhp==2) {
296 bb->y1[yy] = p1[x].y;
297 bb->y2[yy] = p1[x+8].y;
298 bb->y3[yy] = p1[linex*8+x].y;
299 bb->y4[yy] = p1[linex*8+x+8].y;
307 bb->y1[yy] += p1[x].y; bb->y1[yy] /= 2;
308 bb->y2[yy] += p1[x+8].y; bb->y2[yy] /= 2;
309 bb->y3[yy] += p1[linex*8+x].y; bb->y3[yy] /= 2;
310 bb->y4[yy] += p1[linex*8+x+8].y; bb->y4[yy] /= 2;
315 } else if(yhp==1 || yhp==3) {
318 bb->y1[yy] = (p1[x].y + p1[x+1].y);
319 bb->y2[yy] = (p1[x+8].y + p1[x+8+1].y);
320 bb->y3[yy] = (p1[linex*8+x].y + p1[linex*8+x+1].y);
321 bb->y4[yy] = (p1[linex*8+x+8].y + p1[linex*8+x+8+1].y);
328 bb->y1[yy] += (p1[x].y + p1[x+1].y); bb->y1[yy]/=4;
329 bb->y2[yy] += (p1[x+8].y + p1[x+8+1].y); bb->y2[yy]/=4;
330 bb->y3[yy] += (p1[linex*8+x].y + p1[linex*8+x+1].y); bb->y3[yy]/=4;
331 bb->y4[yy] += (p1[linex*8+x+8].y + p1[linex*8+x+8+1].y); bb->y4[yy]/=4;
336 bb->y1[yy]/=2; bb->y2[yy]/=2; bb->y3[yy]/=2; bb->y4[yy]/=2;
344 if(uvhp==0 || uvhp==2) {
347 bb->u[uv] = (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
348 bb->v[uv] = (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
355 bb->u[uv] += (p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4;
356 bb->v[uv] += (p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4;
363 } else /* uvhp==1 || uvhp==3 */ {
366 bb->u[uv] = ((p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4+
367 (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);
368 bb->v[uv] = ((p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4+
369 (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);
376 bb->u[uv] += ((p2[x*2].u + p2[x*2+1].u + p2[linex+x*2].u + p2[linex+x*2+1].u)/4+
377 (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);
378 bb->v[uv] += ((p2[x*2].v + p2[x*2+1].v + p2[linex+x*2].v + p2[linex+x*2+1].v)/4+
379 (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);
395 static void rgb2yuv(YUV*dest, RGBA*src, int dlinex, int slinex, int width, int height)
398 for(y=0;y<height;y++) {
399 for(x=0;x<width;x++) {
401 r = src[y*slinex+x].r;
402 g = src[y*slinex+x].g;
403 b = src[y*slinex+x].b;
404 /*dest[y*dlinex+x].y = (r*0.299 + g*0.587 + b*0.114);
405 dest[y*dlinex+x].u = (r*-0.169 + g*-0.332 + b*0.500 + 128.0);
406 dest[y*dlinex+x].v = (r*0.500 + g*-0.419 + b*-0.0813 + 128.0);*/
408 //dest[y*dlinex+x].y = 128;//(r*((int)( 0.299*256)) + g*((int)( 0.587*256)) + b*((int)( 0.114 *256)))>>8;
409 dest[y*dlinex+x].y = (r*((int)( 0.299*256)) + g*((int)( 0.587*256)) + b*((int)( 0.114 *256)))>>8;
410 dest[y*dlinex+x].u = (r*((int)(-0.169*256)) + g*((int)(-0.332*256)) + b*((int)( 0.500 *256))+ 128*256)>>8;
411 dest[y*dlinex+x].v = (r*((int)( 0.500*256)) + g*((int)(-0.419*256)) + b*((int)(-0.0813*256))+ 128*256)>>8;
415 static void copyregion(VIDEOSTREAM*s, YUV*dest, YUV*src, int bx, int by)
417 YUV*p1 = &src[by*s->linex*16+bx*16];
418 YUV*p2 = &dest[by*s->linex*16+bx*16];
421 memcpy(p1, p2, 16*sizeof(YUV));
422 p1+=s->linex;p2+=s->linex;
426 static void yuv2rgb(RGBA*dest, YUV*src, int linex, int width, int height)
429 for(y=0;y<height;y++) {
430 for(x=0;x<width;x++) {
432 u = src[y*linex+x].u;
433 v = src[y*linex+x].v;
434 yy = src[y*linex+x].y;
435 dest[y*linex+x].r = truncate256(yy + ((360*(v-128))>>8));
436 dest[y*linex+x].g = truncate256(yy - ((88*(u-128)+183*(v-128))>>8));
437 dest[y*linex+x].b = truncate256(yy + ((455 * (u-128))>>8));
441 static void copyblock(VIDEOSTREAM*s, YUV*dest, block_t*b, int bx, int by)
443 YUV*p1 = &dest[(by*16)*s->linex+bx*16];
444 YUV*p2 = &dest[(by*16+8)*s->linex+bx*16];
449 p1[x+0].u = b->u[(y/2)*8+(x/2)];
450 p1[x+0].v = b->v[(y/2)*8+(x/2)];
451 p1[x+0].y = b->y1[y*8+x];
452 p1[x+8].u = b->u[(y/2)*8+(x/2)+4];
453 p1[x+8].v = b->v[(y/2)*8+(x/2)+4];
454 p1[x+8].y = b->y2[y*8+x];
455 p2[x+0].u = b->u[(y/2+4)*8+(x/2)];
456 p2[x+0].v = b->v[(y/2+4)*8+(x/2)];
457 p2[x+0].y = b->y3[y*8+x];
458 p2[x+8].u = b->u[(y/2+4)*8+(x/2)+4];
459 p2[x+8].v = b->v[(y/2+4)*8+(x/2)+4];
460 p2[x+8].y = b->y4[y*8+x];
467 static int compareregions(VIDEOSTREAM*s, int bx, int by)
469 int linex = s->width;
470 YUV*p1 = &s->current[by*linex*16+bx*16];
471 YUV*p2 = &s->oldpic[by*linex*16+bx*16];
481 diff += y*y+(u*u+v*v)/4;
489 static inline int valtodc(int val)
497 /* TODO: what to do for zero values? skip the block? */
506 static int dctoval(int dc)
519 static int codehuffman(TAG*tag, struct huffcode*table, int index)
521 /* TODO: !optimize! */
523 while(table[index].code[i]) {
524 if(table[index].code[i]=='0')
525 swf_SetBits(tag, 0, 1);
527 swf_SetBits(tag, 1, 1);
533 static void quantize8x8(int*src, int*dest, int has_dc, int quant)
536 double q = 1.0/(quant*2);
538 dest[0] = valtodc((int)src[0]); /*DC*/
543 //dest[t] = (int)src[t];
544 /* exact: if(quant&1){dest[t] = (dest[t]/quant - 1)/2;}else{dest[t] = ((dest[t]+1)/quant - 1)/2;} */
545 //if(quant&1){dest[t] = (dest[t]/quant - 1)/2;}else{dest[t] = ((dest[t]+1)/quant - 1)/2;}
546 //dest[t] = dest[t]/(quant*2);
547 dest[t] = (int)(src[t]*q);
548 /* TODO: warn if this happens- the video will be buggy */
549 if(dest[t]>127) dest[t]=127;
550 if(dest[t]<-127) dest[t]=-127;
554 static void dequantize8x8(int*b, int has_dc, int quant)
558 b[0] = dctoval(b[0]); //DC
561 for(t=pos;t<64;t++) {
570 b[t] = quant*(2*b[t]+1); //-7,8,24,40
572 b[t] = quant*(2*b[t]+1)-1; //-8,7,23,39
579 /* paragraph 6.2.2, "clipping of reconstruction levels": */
580 if(b[t]>2047) b[t]=2047;
581 if(b[t]<-2048) b[t]=-2048;
585 static int hascoef(int*b, int has_dc)
591 for(t=pos;t<64;t++) {
598 static int coefbits8x8(int*bb, int has_dc)
609 for(last=63;last>=pos;last--) {
616 int run=0, level=0, islast=0,t;
617 while(!bb[pos] && pos<last) {
624 if(level<0) level=-level;
626 for(t=0;t<RLE_ESCAPE;t++) {
627 if(rle_params[t].run == run &&
628 rle_params[t].level == level &&
629 rle_params[t].last == islast) {
630 bits += rle[t].len + 1;
635 bits += rle[RLE_ESCAPE].len + 1 + 6 + 8;
644 static void encode8x8(TAG*tag, int*bb, int has_dc, int has_tcoef)
651 swf_SetBits(tag, bb[0], 8);
657 /* determine last non-null coefficient */
658 for(last=63;last>=pos;last--) {
659 /* TODO: we could leave out small coefficients
660 after a certain point (32?) */
664 /* blocks without coefficients should not be included
665 in the cbpy/cbpc patterns: */
674 while(!bb[pos] && pos<last) {
686 for(t=0;t<RLE_ESCAPE;t++) {
687 /* TODO: lookup table */
688 if(rle_params[t].run == run &&
689 rle_params[t].level == level &&
690 rle_params[t].last == islast) {
691 codehuffman(tag, rle, t);
692 swf_SetBits(tag, sign, 1);
697 codehuffman(tag, rle, RLE_ESCAPE);
704 swf_SetBits(tag, islast, 1);
705 swf_SetBits(tag, run, 6);
706 swf_SetBits(tag, level, 8); //FIXME: fixme??
716 static void quantize(block_t*fb, block_t*b, int has_dc, int quant)
718 quantize8x8(fb->y1, b->y1, has_dc, quant);
719 quantize8x8(fb->y2, b->y2, has_dc, quant);
720 quantize8x8(fb->y3, b->y3, has_dc, quant);
721 quantize8x8(fb->y4, b->y4, has_dc, quant);
722 quantize8x8(fb->u, b->u, has_dc, quant);
723 quantize8x8(fb->v, b->v, has_dc, quant);
726 static void dodct(block_t*fb)
728 dct(fb->y1); dct(fb->y2); dct(fb->y3); dct(fb->y4);
729 dct(fb->u); dct(fb->v);
737 static void dodctandquant(block_t*fb, block_t*b, int has_dc, int quant)
742 quantize(fb,b,has_dc,quant);
746 dct2(fb->y1,b->y1); dct2(fb->y2,b->y2); dct2(fb->y3,b->y3); dct2(fb->y4,b->y4);
747 dct2(fb->u,b->u); dct2(fb->v,b->v);
750 static void doidct(block_t*b)
755 fb.y1[t] = b->y1[zigzagtable[t]];
756 fb.y2[t] = b->y2[zigzagtable[t]];
757 fb.y3[t] = b->y3[zigzagtable[t]];
758 fb.y4[t] = b->y4[zigzagtable[t]];
759 fb.u[t] = b->u[zigzagtable[t]];
760 fb.v[t] = b->v[zigzagtable[t]];
762 idct(fb.y1); idct(fb.y2); idct(fb.y3); idct(fb.y4);
763 idct(fb.u); idct(fb.v);
773 static void truncateblock(block_t*b)
777 b->y1[t] = truncate256(b->y1[t]);
778 b->y2[t] = truncate256(b->y2[t]);
779 b->y3[t] = truncate256(b->y3[t]);
780 b->y4[t] = truncate256(b->y4[t]);
781 b->u[t] = truncate256(b->u[t]);
782 b->v[t] = truncate256(b->v[t]);
786 static void dequantize(block_t*b, int has_dc, int quant)
788 dequantize8x8(b->y1, has_dc, quant);
789 dequantize8x8(b->y2, has_dc, quant);
790 dequantize8x8(b->y3, has_dc, quant);
791 dequantize8x8(b->y4, has_dc, quant);
792 dequantize8x8(b->u, has_dc, quant);
793 dequantize8x8(b->v, has_dc, quant);
796 static void getblockpatterns(block_t*b, int*cbpybits,int*cbpcbits, int has_dc)
801 *cbpybits|=hascoef(b->y1, has_dc)*8;
802 *cbpybits|=hascoef(b->y2, has_dc)*4;
803 *cbpybits|=hascoef(b->y3, has_dc)*2;
804 *cbpybits|=hascoef(b->y4, has_dc)*1;
806 *cbpcbits|=hascoef(b->u, has_dc)*2;
807 *cbpcbits|=hascoef(b->v, has_dc)*1;
810 static void setQuant(TAG*tag, int dquant)
817 swf_SetBits(tag, 0x0, 2);
818 } else if(dquant == -2) {
819 swf_SetBits(tag, 0x1, 2);
820 } else if(dquant == +1) {
821 swf_SetBits(tag, 0x2, 2);
822 } else if(dquant == +2) {
823 swf_SetBits(tag, 0x3, 2);
825 assert(0*strlen("invalid dquant"));
829 static void change_quant(int quant, int*dquant)
835 static void encode_blockI(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
840 int cbpcbits = 0, cbpybits=0;
842 getregion(&fb, s->current, bx, by, s->width);
844 change_quant(*quant, &dquant);
847 dodctandquant(&fb, &b, 1, *quant);
848 //quantize(&fb, &b, 1, *quant);
850 //decode_blockI(s, &b, bx, by);
852 getblockpatterns(&b, &cbpybits, &cbpcbits, 1);
855 codehuffman(tag, mcbpc_intra, 4+cbpcbits);
857 codehuffman(tag, mcbpc_intra, 0+cbpcbits);
860 codehuffman(tag, cbpy, cbpybits);
863 setQuant(tag, dquant);
867 encode8x8(tag, b.y1, 1, cbpybits&8);
868 encode8x8(tag, b.y2, 1, cbpybits&4);
869 encode8x8(tag, b.y3, 1, cbpybits&2);
870 encode8x8(tag, b.y4, 1, cbpybits&1);
873 encode8x8(tag, b.u, 1, cbpcbits&2);
874 encode8x8(tag, b.v, 1, cbpcbits&1);
877 dequantize(&b, 1, *quant);
880 copyblock(s, s->current, &b, bx, by);
883 static void yuvdiff(block_t*a, block_t*b)
887 a->y1[t] = (a->y1[t] - b->y1[t]);
888 a->y2[t] = (a->y2[t] - b->y2[t]);
889 a->y3[t] = (a->y3[t] - b->y3[t]);
890 a->y4[t] = (a->y4[t] - b->y4[t]);
891 a->u[t] = (a->u[t] - b->u[t]);
892 a->v[t] = (a->v[t] - b->v[t]);
896 static void predictmvd(VIDEOSTREAM*s, int bx, int by, int*px, int*py)
899 int x1,y1,x2,y2,x3,y3;
901 if(bx) {x1=s->mvdx[by*s->bbx+bx-1];
902 y1=s->mvdy[by*s->bbx+bx-1];
905 if(by) {x2=s->mvdx[(by-1)*s->bbx+bx];
906 y2=s->mvdy[(by-1)*s->bbx+bx];
908 x3=s->mvdx[(by-1)*s->bbx+bx+1];
909 y3=s->mvdy[(by-1)*s->bbx+bx+1];
914 else {x2=x3=x1;y2=y3=y1;}
916 if((x1 <= x2 && x2 <= x3) ||
917 (x3 <= x2 && x2 <= x1)) {
919 } else if((x2 <= x1 && x1 <= x3) ||
920 (x3 <= x1 && x1 <= x2)) {
922 } else if((x1 <= x3 && x3 <= x2) ||
923 (x2 <= x3 && x3 <= x1)) {
927 if((y1 <= y2 && y2 <= y3) ||
928 (y3 <= y2 && y2 <= y1)) {
930 } else if((y2 <= y1 && y1 <= y3) ||
931 (y3 <= y1 && y1 <= y2)) {
933 } else if((y1 <= y3 && y3 <= y2) ||
934 (y2 <= y3 && y3 <= y1)) {
940 assert((x4>=-32 && x4<=31) && (y4>=-32 && y4<=31));
943 static inline int mvd2index(int px, int py, int x, int y, int xy)
945 assert((x>=-32 && x<=31) && (y>=-32 && y<=31));
946 //assert((x&1)==0 && (y&1)==0);//for now
947 //assert((x&2)==0 && (y&2)==0);//for now(2)
962 assert(x>=0 && x<64);
966 static int encode_blockP(TAG*tag, VIDEOSTREAM*s, int bx, int by, int*quant)
975 int cbpcbits = 0, cbpybits=0;
985 int bits_v00 = 65535;
989 diff = compareregions(s, bx, by);
991 if(diff < 20 /*TODO: should be a parameter- good values are between 32 and 48 */) {
992 /* TODO: measure the error an I-block encoding would do, and base the decision
994 swf_SetBits(tag, 1,1); /* cod=1, block skipped */
995 /* copy the region from the last frame so that we have a complete reconstruction */
996 copyregion(s, s->current, s->oldpic, bx, by);
1000 predictmvd(s,bx,by,&predictmvdx,&predictmvdy);
1001 getregion(&fb, s->current, bx, by, s->width);
1003 { /* consider I-block */
1006 memcpy(&fb_i, &fb, sizeof(block_t));
1007 dodctandquant(&fb_i, &b_i, 1, *quant);
1008 //quantize(&fb_i, &b_i, 1, *quant);
1009 getblockpatterns(&b_i, &y, &c, 1);
1011 bits_i += mcbpc_inter[3*4+c].len;
1012 bits_i += cbpy[y].len;
1013 bits_i += coefbits8x8(b_i.y1, 1);
1014 bits_i += coefbits8x8(b_i.y2, 1);
1015 bits_i += coefbits8x8(b_i.y3, 1);
1016 bits_i += coefbits8x8(b_i.y4, 1);
1017 bits_i += coefbits8x8(b_i.u, 1);
1018 bits_i += coefbits8x8(b_i.v, 1);
1021 { /* consider mvd(x,y)-block */
1030 int bestx=0,besty=0,bestbits=65536;
1031 int startx=-2,endx=2;
1032 int starty=-2,endy=2;
1036 if(bx==s->bbx-1) endx=0;
1037 if(by==s->bby-1) endy=0;
1039 for(hx=startx;hx<=endx;hx+=1)
1040 for(hy=starty;hy<=endy;hy+=1)
1045 memcpy(&fbdiff, &fb, sizeof(block_t));
1046 getmvdregion(&fbold, s->oldpic, bx, by, hx, hy, s->linex);
1047 yuvdiff(&fbdiff, &fbold);
1048 dodctandquant(&fbdiff, &b, 0, *quant);
1049 //quantize(&fbdiff, &b, 0, *quant);
1050 bits += coefbits8x8(b.y1, 0);
1051 bits += coefbits8x8(b.y2, 0);
1052 bits += coefbits8x8(b.y3, 0);
1053 bits += coefbits8x8(b.y4, 0);
1054 bits += coefbits8x8(b.u, 0);
1055 bits += coefbits8x8(b.v, 0);
1066 memcpy(&fbdiff, &fb, sizeof(block_t));
1067 getmvdregion(&fbold_v00, s->oldpic, bx, by, x_v00, y_v00, s->linex);
1068 yuvdiff(&fbdiff, &fbold_v00);
1069 dodctandquant(&fbdiff, &b_v00, 0, *quant);
1070 //quantize(&fbdiff, &b_v00, 0, *quant);
1071 getblockpatterns(&b_v00, &y, &c, 0);
1074 bits_v00 += mcbpc_inter[0*4+c].len;
1075 bits_v00 += cbpy[y^15].len;
1076 bits_v00 += mvd[mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 0)].len; // (0,0)
1077 bits_v00 += mvd[mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 1)].len;
1078 bits_v00 += coefbits8x8(b_v00.y1, 0);
1079 bits_v00 += coefbits8x8(b_v00.y2, 0);
1080 bits_v00 += coefbits8x8(b_v00.y3, 0);
1081 bits_v00 += coefbits8x8(b_v00.y4, 0);
1082 bits_v00 += coefbits8x8(b_v00.u, 0);
1083 bits_v00 += coefbits8x8(b_v00.v, 0);
1086 if(bits_i > bits_v00)
1088 /* mvd (0,0) block (mode=0) */
1090 mode = 0; // mvd w/o mvd24
1092 memcpy(&b, &b_v00, sizeof(block_t));
1094 getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
1095 swf_SetBits(tag,0,1); // COD
1096 codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
1097 codehuffman(tag, cbpy, cbpybits^15);
1100 codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 0));
1101 codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, x_v00, y_v00, 1));
1102 s->mvdx[by*s->bbx+bx] = x_v00;
1103 s->mvdy[by*s->bbx+bx] = y_v00;
1106 encode8x8(tag, b.y1, has_dc, cbpybits&8);
1107 encode8x8(tag, b.y2, has_dc, cbpybits&4);
1108 encode8x8(tag, b.y3, has_dc, cbpybits&2);
1109 encode8x8(tag, b.y4, has_dc, cbpybits&1);
1112 encode8x8(tag, b.u, has_dc, cbpcbits&2);
1113 encode8x8(tag, b.v, has_dc, cbpcbits&1);
1115 /* -- reconstruction -- */
1116 dequantize(&b, 0, *quant);
1119 b.y1[t] = truncate256(b.y1[t] + (int)fbold_v00.y1[t]);
1120 b.y2[t] = truncate256(b.y2[t] + (int)fbold_v00.y2[t]);
1121 b.y3[t] = truncate256(b.y3[t] + (int)fbold_v00.y3[t]);
1122 b.y4[t] = truncate256(b.y4[t] + (int)fbold_v00.y4[t]);
1123 b.u[t] = truncate256(b.u[t] + (int)fbold_v00.u[t]);
1124 b.v[t] = truncate256(b.v[t] + (int)fbold_v00.v[t]);
1126 copyblock(s, s->current, &b, bx, by);
1129 /* i block (mode=3) */
1132 memcpy(&b, &b_i, sizeof(block_t));
1133 getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
1134 swf_SetBits(tag,0,1); // COD
1135 codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
1136 codehuffman(tag, cbpy, cbpybits);
1139 encode8x8(tag, b.y1, has_dc, cbpybits&8);
1140 encode8x8(tag, b.y2, has_dc, cbpybits&4);
1141 encode8x8(tag, b.y3, has_dc, cbpybits&2);
1142 encode8x8(tag, b.y4, has_dc, cbpybits&1);
1145 encode8x8(tag, b.u, has_dc, cbpcbits&2);
1146 encode8x8(tag, b.v, has_dc, cbpcbits&1);
1148 /* -- reconstruction -- */
1149 dequantize(&b, 1, *quant);
1152 copyblock(s, s->current, &b, bx, by);
1159 quantize(&fb, &b, has_dc, *quant);
1160 getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
1162 if(!dquant && has_mvd && !has_mvd24 && !has_dc) mode = 0;
1163 else if(dquant && has_mvd && !has_mvd24 && !has_dc) mode = 1;
1164 else if(!dquant && has_mvd && has_mvd24 && !has_dc) mode = 2;
1165 else if(!dquant && !has_mvd && !has_mvd24 && has_dc) mode = 3;
1166 else if(dquant && !has_mvd && !has_mvd24 && has_dc) mode = 4;
1169 swf_SetBits(tag,0,1); /* cod - 1 if we're not going to code this block*/
1171 codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
1172 codehuffman(tag, cbpy, (mode==3 || mode==4)?cbpybits:cbpybits^15);
1175 setQuant(tag, dquant);
1180 codehuffman(tag, mvd, 32);
1181 codehuffman(tag, mvd, 32);
1187 encode8x8(tag, b.y1, has_dc, cbpybits&8);
1188 encode8x8(tag, b.y2, has_dc, cbpybits&4);
1189 encode8x8(tag, b.y3, has_dc, cbpybits&2);
1190 encode8x8(tag, b.y4, has_dc, cbpybits&1);
1193 encode8x8(tag, b.u, has_dc, cbpcbits&2);
1194 encode8x8(tag, b.v, has_dc, cbpcbits&1);
1198 static int bmid = 0;
1199 void setdbgpic(TAG*tag, RGBA*pic, int width, int height)
1204 tag = swf_InsertTag(tag,ST_REMOVEOBJECT2);
1205 swf_SetU16(tag, 133);
1207 tag = swf_InsertTag(tag, ST_DEFINEBITSLOSSLESS);
1208 swf_SetU16(tag, 1000+bmid);
1209 swf_SetLosslessBits(tag, width, height, (void*)pic, BMF_32BIT);
1211 tag = swf_InsertTag(tag, ST_DEFINESHAPE);
1212 swf_SetU16(tag, 2000+bmid);
1213 swf_ShapeSetBitmapRect(tag, 1000+bmid, width, height);
1215 tag = swf_InsertTag(tag,ST_PLACEOBJECT2);
1216 swf_GetMatrix(0,&m);
1218 swf_ObjectPlace(tag, 2000+bmid, 133, &m, 0, 0);
1223 #define TYPE_IFRAME 0
1224 #define TYPE_PFRAME 1
1226 static void writeHeader(TAG*tag, int width, int height, int frame, int quant, int type)
1229 swf_SetU16(tag, frame);
1230 swf_SetBits(tag, 1, 17); /* picture start code*/
1231 swf_SetBits(tag, 0, 5); /* version=0, version 1 would optimize rle behaviour*/
1232 swf_SetBits(tag, frame, 8); /* time reference */
1234 /* write dimensions, taking advantage of some predefined sizes
1235 if the opportunity presents itself */
1236 i32 = width<<16|height;
1239 case 352<<16|288: swf_SetBits(tag, 2, 3);break;
1240 case 176<<16|144: swf_SetBits(tag, 3, 3);break;
1241 case 128<<16|96: swf_SetBits(tag, 4, 3);break;
1242 case 320<<16|240: swf_SetBits(tag, 5, 3);break;
1243 case 160<<16|120: swf_SetBits(tag, 6, 3);break;
1245 if(width>255 || height>255) {
1246 swf_SetBits(tag, 1, 3);
1247 swf_SetBits(tag, width, 16);
1248 swf_SetBits(tag, height, 16);
1250 swf_SetBits(tag, 0, 3);
1251 swf_SetBits(tag, width, 8);
1252 swf_SetBits(tag, height, 8);
1256 swf_SetBits(tag, type, 2); /* I-Frame or P-Frame */
1257 swf_SetBits(tag, 0, 1); /* No deblock filter */
1259 swf_SetBits(tag, quant, 5); /* quantizer (1-31), may be updated later on*/
1260 swf_SetBits(tag, 0, 1); /* No extra info */
1263 void swf_SetVideoStreamIFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
1267 if(quant<1) quant=1;
1268 if(quant>31) quant=31;
1270 writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_IFRAME);
1272 rgb2yuv(s->current, pic, s->linex, s->olinex, s->width, s->height);
1276 for(by=0;by<s->bby;by++)
1278 for(bx=0;bx<s->bbx;bx++)
1280 encode_blockI(tag, s, bx, by, &quant);
1284 memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
1287 void swf_SetVideoStreamPFrame(TAG*tag, VIDEOSTREAM*s, RGBA*pic, int quant)
1291 if(quant<1) quant=1;
1292 if(quant>31) quant=31;
1294 writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
1296 rgb2yuv(s->current, pic, s->linex, s->olinex, s->width, s->height);
1297 memset(s->mvdx, 0, s->bbx*s->bby*sizeof(int));
1298 memset(s->mvdy, 0, s->bbx*s->bby*sizeof(int));
1300 for(by=0;by<s->bby;by++)
1302 for(bx=0;bx<s->bbx;bx++)
1304 encode_blockP(tag, s, bx, by, &quant);
1308 memcpy(s->oldpic, s->current, s->width*s->height*sizeof(YUV));
1313 yuv2rgb(pic, s->current, s->linex, s->width, s->height);
1314 setdbgpic(tag, pic, s->width, s->height);
1316 if(s->frame == (int)totalframes-1)
1319 FILE*fi = fopen("test.ppm", "wb");
1320 fprintf(fi, "P6\n%d %d\n255\n", s->width, s->height);
1321 for(t=0;t<s->width*s->height;t++)
1323 fwrite(&pic[t].r, 1, 1, fi);
1324 fwrite(&pic[t].g, 1, 1, fi);
1325 fwrite(&pic[t].b, 1, 1, fi);
1332 int uline[64],vline[64],yline[64];
1333 void swf_SetVideoStreamMover(TAG*tag, VIDEOSTREAM*s, int quant)
1337 if(quant<1) quant=1;
1338 if(quant>31) quant=31;
1340 writeHeader(tag, s->width, s->height, s->frame, quant, TYPE_PFRAME);
1342 memset(s->mvdx, 0, s->bbx*s->bby*sizeof(int));
1343 memset(s->mvdy, 0, s->bbx*s->bby*sizeof(int));
1345 for(by=0;by<s->bby;by++)
1347 for(bx=0;bx<s->bbx;bx++)
1349 //if((lrand48()&255) || !(bx>8 && bx<24 && by>8 && by<24)) {
1351 /* mvd (0,0) block (mode=0) */
1353 int mode = 0; // mvd w/o mvd24
1355 int cbpybits=0,cbpcbits=0;
1356 int predictmvdx, predictmvdy;
1357 //int mvx=-1+(2*(s->frame&1));
1358 //int mvy=-1+((s->frame&2));
1359 int mvx=(lrand48()%4)-2;
1362 swf_SetBits(tag,0,1); // COD
1363 codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
1364 codehuffman(tag, cbpy, cbpybits^15);
1367 predictmvd(s,bx,by,&predictmvdx,&predictmvdy);
1368 codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, mvx, mvy, 0));
1369 codehuffman(tag, mvd, mvd2index(predictmvdx, predictmvdy, mvx, mvy, 1));
1370 s->mvdx[by*s->bbx+bx] = mvx;
1371 s->mvdy[by*s->bbx+bx] = mvy;
1373 /* i block (mode=3) */
1376 int cbpybits,cbpcbits;
1379 memset(&b, 0, sizeof(block_t));
1380 b.y1[0] = b.y2[0] = b.y3[0] = b.y4[0] = yline[bx];
1384 getblockpatterns(&b, &cbpybits, &cbpcbits, has_dc);
1385 swf_SetBits(tag,0,1); // COD
1386 codehuffman(tag, mcbpc_inter, mode*4+cbpcbits);
1387 codehuffman(tag, cbpy, cbpybits);
1390 encode8x8(tag, b.y1, has_dc, cbpybits&8);
1391 encode8x8(tag, b.y2, has_dc, cbpybits&4);
1392 encode8x8(tag, b.y3, has_dc, cbpybits&2);
1393 encode8x8(tag, b.y4, has_dc, cbpybits&1);
1396 encode8x8(tag, b.u, has_dc, cbpcbits&2);
1397 encode8x8(tag, b.v, has_dc, cbpcbits&1);
1405 int main(int argn, char*argv[])
1411 RGBA* pic, *pic2, rgb;
1418 char* fname = "/home/kramm/pics/peppers.png";
1422 memset(&stream, 0, sizeof(stream));
1424 getPNG(fname, &width, &height, &data);
1425 pic = (RGBA*)malloc(width*height*sizeof(RGBA));
1426 pic2 = (RGBA*)malloc(width*height*sizeof(RGBA));
1427 memcpy(pic, data, width*height*sizeof(RGBA));
1430 printf("Compressing %s, size %dx%d\n", fname, width, height);
1432 memset(&swf,0,sizeof(SWF));
1433 memset(&obj,0,sizeof(obj));
1435 swf.fileVersion = 6;
1436 swf.frameRate = framerate*256;
1437 swf.movieSize.xmax = 20*width*2;
1438 swf.movieSize.ymax = 20*height-20*64;
1440 swf.firstTag = swf_InsertTag(NULL,ST_SETBACKGROUNDCOLOR);
1442 rgb.r = 0x00;rgb.g = 0x00;rgb.b = 0x00;
1443 swf_SetRGB(tag,&rgb);
1445 tag = swf_InsertTag(tag, ST_DEFINEVIDEOSTREAM);
1446 swf_SetU16(tag, 33);
1447 swf_SetVideoStreamDefine(tag, &stream, frames, width, height);
1448 stream.do_motion = 1;
1450 for(t=0;t<frames;t++)
1454 for(y=0,yy=0;y<height;y++,yy+=d) {
1455 RGBA*line = &pic[((int)yy)*width];
1456 for(x=0,xx=0;x<width;x++,xx+=d) {
1457 pic2[y*width+x] = line[((int)xx)];
1460 printf("frame:%d\n", t);fflush(stdout);
1462 tag = swf_InsertTag(tag, ST_VIDEOFRAME);
1463 swf_SetU16(tag, 33);
1465 swf_SetVideoStreamIFrame(tag, &stream, pic2, 9);
1467 swf_SetVideoStreamPFrame(tag, &stream, pic2, 9);
1470 tag = swf_InsertTag(tag, ST_PLACEOBJECT2);
1471 swf_GetPlaceObject(0, &obj);
1480 swf_SetPlaceObject(tag,&obj);
1482 tag = swf_InsertTag(tag, ST_SHOWFRAME);
1485 swf_VideoStreamClear(&stream);
1487 tag = swf_InsertTag(tag, ST_END);
1489 fi = open("video3.swf", O_WRONLY|O_CREAT|O_TRUNC, 0644);
1490 if(swf_WriteSWC(fi,&swf)<0) {
1491 fprintf(stderr,"WriteSWF() failed.\n");