Windows CE

开发平台：

C/C++

snow.c：源码内容

s->block[index + i + j*w]= block;
}
}
}
static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
const int offset[3]= {
y*c-> stride + x,
((y*c->uvstride + x)>>1),
((y*c->uvstride + x)>>1),
};
int i;
for(i=0; i<3; i++){
c->src[0][i]= src [i];
c->ref[0][i]= ref [i] + offset[i];
}
assert(!ref_index);
}
//FIXME copy&paste
#define P_LEFT P[1]
#define P_TOP P[2]
#define P_TOPRIGHT P[3]
#define P_MEDIAN P[4]
#define P_MV1 P[9]
#define FLAG_QPEL 1 //must be 1
static int encode_q_branch(SnowContext *s, int level, int x, int y){
uint8_t p_buffer[1024];
uint8_t i_buffer[1024];
uint8_t p_state[sizeof(s->block_state)];
uint8_t i_state[sizeof(s->block_state)];
RangeCoder pc, ic;
uint8_t *pbbak= s->c.bytestream;
uint8_t *pbbak_start= s->c.bytestream_start;
int score, score2, iscore, i_len, p_len, block_s, sum;
const int w= s->b_width << s->block_max_depth;
const int h= s->b_height << s->block_max_depth;
const int rem_depth= s->block_max_depth - level;
const int index= (x + y*w) << rem_depth;
const int block_w= 1<<(LOG2_MB_SIZE - level);
static BlockNode null_block= { //FIXME add border maybe
.color= {128,128,128},
.mx= 0,
.my= 0,
.type= 0,
.level= 0,
};
int trx= (x+1)<<rem_depth;
int try= (y+1)<<rem_depth;
BlockNode *left = x ? &s->block[index-1] : &null_block;
BlockNode *top = y ? &s->block[index-w] : &null_block;
BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
BlockNode *tl = y && x ? &s->block[index-w-1] : left;
BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
int pl = left->color[0];
int pcb= left->color[1];
int pcr= left->color[2];
int pmx= mid_pred(left->mx, top->mx, tr->mx);
int pmy= mid_pred(left->my, top->my, tr->my);
int mx=0, my=0;
int l,cr,cb, i;
const int stride= s->current_picture.linesize[0];
const int uvstride= s->current_picture.linesize[1];
const int instride= s->input_picture.linesize[0];
const int uvinstride= s->input_picture.linesize[1];
uint8_t *new_l = s->input_picture.data[0] + (x + y* instride)*block_w;
uint8_t *new_cb= s->input_picture.data[1] + (x + y*uvinstride)*block_w/2;
uint8_t *new_cr= s->input_picture.data[2] + (x + y*uvinstride)*block_w/2;
uint8_t current_mb[3][stride*block_w];
uint8_t *current_data[3]= {&current_mb[0][0], &current_mb[1][0], &current_mb[2][0]};
int P[10][2];
int16_t last_mv[3][2];
int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
const int shift= 1+qpel;
MotionEstContext *c= &s->m.me;
int mx_context= av_log2(2*ABS(left->mx - top->mx));
int my_context= av_log2(2*ABS(left->my - top->my));
int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
assert(sizeof(s->block_state) >= 256);
if(s->keyframe){
set_blocks(s, level, x, y, pl, pcb, pcr, pmx, pmy, BLOCK_INTRA);
return 0;
}
//FIXME optimize
for(i=0; i<block_w; i++)
memcpy(&current_mb[0][0] + stride*i, new_l + instride*i, block_w);
for(i=0; i<block_w>>1; i++)
memcpy(&current_mb[1][0] + uvstride*i, new_cb + uvinstride*i, block_w>>1);
for(i=0; i<block_w>>1; i++)
memcpy(&current_mb[2][0] + uvstride*i, new_cr + uvinstride*i, block_w>>1);
// clip predictors / edge ?
P_LEFT[0]= left->mx;
P_LEFT[1]= left->my;
P_TOP [0]= top->mx;
P_TOP [1]= top->my;
P_TOPRIGHT[0]= tr->mx;
P_TOPRIGHT[1]= tr->my;
last_mv[0][0]= s->block[index].mx;
last_mv[0][1]= s->block[index].my;
last_mv[1][0]= right->mx;
last_mv[1][1]= right->my;
last_mv[2][0]= bottom->mx;
last_mv[2][1]= bottom->my;
s->m.mb_stride=2;
s->m.mb_x=
s->m.mb_y= 0;
s->m.me.skip= 0;
init_ref(c, current_data, s->last_picture.data, NULL, block_w*x, block_w*y, 0);
assert(s->m.me. stride == stride);
assert(s->m.me.uvstride == uvstride);
c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
c->xmin = - x*block_w - 16+2;
c->ymin = - y*block_w - 16+2;
c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
if(P_LEFT[1] > (c->ymax<<shift)) P_LEFT[1] = (c->ymax<<shift);
if(P_TOP[0] > (c->xmax<<shift)) P_TOP[0] = (c->xmax<<shift);
if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
if (!y) {
c->pred_x= P_LEFT[0];
c->pred_y= P_LEFT[1];
} else {
c->pred_x = P_MEDIAN[0];
c->pred_y = P_MEDIAN[1];
}
score= ff_epzs_motion_search(&s->m, &mx, &my, P, 0, /*ref_index*/ 0, last_mv,
(1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
assert(mx >= c->xmin);
assert(mx <= c->xmax);
assert(my >= c->ymin);
assert(my <= c->ymax);
score= s->m.me.sub_motion_search(&s->m, &mx, &my, score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
score= ff_get_mb_score(&s->m, mx, my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
//FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2
// subpel search
pc= s->c;
pc.bytestream_start=
pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
memcpy(p_state, s->block_state, sizeof(s->block_state));
if(level!=s->block_max_depth)
put_rac(&pc, &p_state[4 + s_context], 1);
put_rac(&pc, &p_state[1 + left->type + top->type], 0);
put_symbol(&pc, &p_state[128 + 32*mx_context], mx - pmx, 1);
put_symbol(&pc, &p_state[128 + 32*my_context], my - pmy, 1);
p_len= pc.bytestream - pc.bytestream_start;
score += (s->lambda2*(p_len*8
+ (pc.outstanding_count - s->c.outstanding_count)*8
+ (-av_log2(pc.range) + av_log2(s->c.range))
))>>FF_LAMBDA_SHIFT;
block_s= block_w*block_w;
sum = pix_sum(&current_mb[0][0], stride, block_w);
l= (sum + block_s/2)/block_s;
iscore = pix_norm1(&current_mb[0][0], stride, block_w) - 2*l*sum + l*l*block_s;
block_s= block_w*block_w>>2;
sum = pix_sum(&current_mb[1][0], uvstride, block_w>>1);
cb= (sum + block_s/2)/block_s;
// iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
sum = pix_sum(&current_mb[2][0], uvstride, block_w>>1);
cr= (sum + block_s/2)/block_s;
// iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
ic= s->c;
ic.bytestream_start=
ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
memcpy(i_state, s->block_state, sizeof(s->block_state));
if(level!=s->block_max_depth)
put_rac(&ic, &i_state[4 + s_context], 1);
put_rac(&ic, &i_state[1 + left->type + top->type], 1);
put_symbol(&ic, &i_state[32], l-pl , 1);
put_symbol(&ic, &i_state[64], cb-pcb, 1);
put_symbol(&ic, &i_state[96], cr-pcr, 1);
i_len= ic.bytestream - ic.bytestream_start;
iscore += (s->lambda2*(i_len*8
+ (ic.outstanding_count - s->c.outstanding_count)*8
+ (-av_log2(ic.range) + av_log2(s->c.range))
))>>FF_LAMBDA_SHIFT;
// assert(score==256*256*256*64-1);
assert(iscore < 255*255*256 + s->lambda2*10);
assert(iscore >= 0);
assert(l>=0 && l<=255);
assert(pl>=0 && pl<=255);
if(level==0){
int varc= iscore >> 8;
int vard= score >> 8;
if (vard <= 64 || vard < varc)
c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
else
c->scene_change_score+= s->m.qscale;
}
if(level!=s->block_max_depth){
put_rac(&s->c, &s->block_state[4 + s_context], 0);
score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
if(score2 < score && score2 < iscore)
return score2;
}
if(iscore < score){
memcpy(pbbak, i_buffer, i_len);
s->c= ic;
s->c.bytestream_start= pbbak_start;
s->c.bytestream= pbbak + i_len;
set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, BLOCK_INTRA);
memcpy(s->block_state, i_state, sizeof(s->block_state));
return iscore;
}else{
memcpy(pbbak, p_buffer, p_len);
s->c= pc;
s->c.bytestream_start= pbbak_start;
s->c.bytestream= pbbak + p_len;
set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, 0);
memcpy(s->block_state, p_state, sizeof(s->block_state));
return score;
}
}
static void decode_q_branch(SnowContext *s, int level, int x, int y){
const int w= s->b_width << s->block_max_depth;
const int rem_depth= s->block_max_depth - level;
const int index= (x + y*w) << rem_depth;
static BlockNode null_block= { //FIXME add border maybe
.color= {128,128,128},
.mx= 0,
.my= 0,
.type= 0,
.level= 0,
};
int trx= (x+1)<<rem_depth;
BlockNode *left = x ? &s->block[index-1] : &null_block;
BlockNode *top = y ? &s->block[index-w] : &null_block;
BlockNode *tl = y && x ? &s->block[index-w-1] : left;
BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
if(s->keyframe){
set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, BLOCK_INTRA);
return;
}
if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
int type;
int l = left->color[0];
int cb= left->color[1];
int cr= left->color[2];
int mx= mid_pred(left->mx, top->mx, tr->mx);
int my= mid_pred(left->my, top->my, tr->my);
int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 0*av_log2(2*ABS(tr->mx - top->mx));
int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my));
type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
if(type){
l += get_symbol(&s->c, &s->block_state[32], 1);
cb+= get_symbol(&s->c, &s->block_state[64], 1);
cr+= get_symbol(&s->c, &s->block_state[96], 1);
}else{
mx+= get_symbol(&s->c, &s->block_state[128 + 32*mx_context], 1);
my+= get_symbol(&s->c, &s->block_state[128 + 32*my_context], 1);
}
set_blocks(s, level, x, y, l, cb, cr, mx, my, type);
}else{
decode_q_branch(s, level+1, 2*x+0, 2*y+0);
decode_q_branch(s, level+1, 2*x+1, 2*y+0);
decode_q_branch(s, level+1, 2*x+0, 2*y+1);
decode_q_branch(s, level+1, 2*x+1, 2*y+1);
}
}
static void encode_blocks(SnowContext *s){
int x, y;
int w= s->b_width;
int h= s->b_height;
for(y=0; y<h; y++){
if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
av_log(s->avctx, AV_LOG_ERROR, "encoded frame too largen");
return;
}
for(x=0; x<w; x++){
encode_q_branch(s, 0, x, y);
}
}
}
static void decode_blocks(SnowContext *s){
int x, y;
int w= s->b_width;
int h= s->b_height;
for(y=0; y<h; y++){
for(x=0; x<w; x++){
decode_q_branch(s, 0, x, y);
}
}
}
static void mc_block(uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
int x, y;
START_TIMER
for(y=0; y < b_h+5; y++){
for(x=0; x < b_w; x++){
int a0= src[x ];
int a1= src[x + 1];
int a2= src[x + 2];
int a3= src[x + 3];
int a4= src[x + 4];
int a5= src[x + 5];
// int am= 9*(a1+a2) - (a0+a3);
int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
// int am= 18*(a2+a3) - 2*(a1+a4);
// int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3;
// int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;
// if(b_w==16) am= 8*(a1+a2);
if(dx<8) am = (32*a2*( 8-dx) + am* dx + 128)>>8;
else am = ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8;
/* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/
if(am&(~255)) am= ~(am>>31);
tmp[x] = am;
/* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6;
else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6;
else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6;
else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/
}
tmp += stride;
src += stride;
}
tmp -= (b_h+5)*stride;
for(y=0; y < b_h; y++){
for(x=0; x < b_w; x++){
int a0= tmp[x + 0*stride];
int a1= tmp[x + 1*stride];
int a2= tmp[x + 2*stride];
int a3= tmp[x + 3*stride];
int a4= tmp[x + 4*stride];
int a5= tmp[x + 5*stride];
int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
// int am= 18*(a2+a3) - 2*(a1+a4);
/* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3;
int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/
// if(b_w==16) am= 8*(a1+a2);
if(dy<8) am = (32*a2*( 8-dy) + am* dy + 128)>>8;
else am = ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8;
if(am&(~255)) am= ~(am>>31);
dst[x] = am;
/* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6;
else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6;
else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6;
else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/
}
dst += stride;
tmp += stride;
}
STOP_TIMER("mc_block")
}
#define mca(dx,dy,b_w)
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, uint8_t *src, int stride, int h){
uint8_t tmp[stride*(b_w+5)];
assert(h==b_w);
mc_block(dst, src-2-2*stride, tmp, stride, b_w, b_w, dx, dy);
}
mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)
mca( 0, 0,8)
mca( 8, 0,8)
mca( 0, 8,8)
mca( 8, 8,8)
static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
if(block->type){
int x, y;
const int color= block->color[plane_index];
for(y=0; y < b_h; y++){
for(x=0; x < b_w; x++){
dst[x + y*stride]= color;
}
}
}else{
const int scale= plane_index ? s->mv_scale : 2*s->mv_scale;
int mx= block->mx*scale;
int my= block->my*scale;
const int dx= mx&15;
const int dy= my&15;
sx += (mx>>4) - 2;
sy += (my>>4) - 2;
src += sx + sy*stride;
if( (unsigned)sx >= w - b_w - 4
|| (unsigned)sy >= h - b_h - 4){
ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+5, b_h+5, sx, sy, w, h);
src= tmp + MB_SIZE;
}
if((dx&3) || (dy&3) || b_w!=b_h || (b_w!=4 && b_w!=8 && b_w!=16))
mc_block(dst, src, tmp, stride, b_w, b_h, dx, dy);
else
s->dsp.put_h264_qpel_pixels_tab[2-(b_w>>3)][dy+(dx>>2)](dst,src + 2 + 2*stride,stride);
}
}
static always_inline int same_block(BlockNode *a, BlockNode *b){
return !((a->mx - b->mx) | (a->my - b->my) | a->type | b->type);
}
//FIXME name clenup (b_w, block_w, b_width stuff)
static always_inline void add_yblock_buffered(SnowContext *s, slice_buffer * sb, DWTELEM *old_dst, uint8_t *dst8, uint8_t *src, uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int plane_index){
DWTELEM * dst = NULL;
const int b_width = s->b_width << s->block_max_depth;
const int b_height= s->b_height << s->block_max_depth;
const int b_stride= b_width;
BlockNode *lt= &s->block[b_x + b_y*b_stride];
BlockNode *rt= lt+1;
BlockNode *lb= lt+b_stride;
BlockNode *rb= lb+1;
uint8_t *block[4];
int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align
uint8_t *ptmp;
int x,y;
if(b_x<0){
lt= rt;
lb= rb;
}else if(b_x + 1 >= b_width){
rt= lt;
rb= lb;
}
if(b_y<0){
lt= lb;
rt= rb;
}else if(b_y + 1 >= b_height){
lb= lt;
rb= rt;
}
if(src_x<0){ //FIXME merge with prev & always round internal width upto *16
obmc -= src_x;
b_w += src_x;
src_x=0;
}else if(src_x + b_w > w){
b_w = w - src_x;
}
if(src_y<0){
obmc -= src_y*obmc_stride;
b_h += src_y;
src_y=0;
}else if(src_y + b_h> h){
b_h = h - src_y;
}
if(b_w<=0 || b_h<=0) return;
assert(src_stride > 2*MB_SIZE + 5);
// old_dst += src_x + src_y*dst_stride;
dst8+= src_x + src_y*src_stride;
// src += src_x + src_y*src_stride;
ptmp= tmp + 3*tmp_step;
block[0]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
if(same_block(lt, rt)){
block[1]= block[0];
}else{
block[1]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[1], src, tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
}
if(same_block(lt, lb)){
block[2]= block[0];
}else if(same_block(rt, lb)){
block[2]= block[1];
}else{
block[2]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[2], src, tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
}
if(same_block(lt, rb) ){
block[3]= block[0];
}else if(same_block(rt, rb)){
block[3]= block[1];
}else if(same_block(lb, rb)){
block[3]= block[2];
}else{
block[3]= ptmp;
pred_block(s, block[3], src, tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
}
#if 0
for(y=0; y<b_h; y++){
for(x=0; x<b_w; x++){
int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc2= obmc + (obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
uint8_t *obmc4= obmc3+ (obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
#else
{
START_TIMER
int block_index = 0;
for(y=0; y<b_h; y++){
//FIXME ugly missue of obmc_stride
uint8_t *obmc1= obmc + y*obmc_stride;
uint8_t *obmc2= obmc1+ (obmc_stride>>1);
uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
uint8_t *obmc4= obmc3+ (obmc_stride>>1);
dst = slice_buffer_get_line(sb, src_y + y);
for(x=0; x<b_w; x++){
int v= obmc1[x] * block[3][x + y*src_stride]
+obmc2[x] * block[2][x + y*src_stride]
+obmc3[x] * block[1][x + y*src_stride]
+obmc4[x] * block[0][x + y*src_stride];
v <<= 8 - LOG2_OBMC_MAX;
if(FRAC_BITS != 8){
v += 1<<(7 - FRAC_BITS);
v >>= 8 - FRAC_BITS;
}
if(add){
// v += old_dst[x + y*dst_stride];
v += dst[x + src_x];
v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
if(v&(~255)) v= ~(v>>31);
dst8[x + y*src_stride] = v;
}else{
// old_dst[x + y*dst_stride] -= v;
dst[x + src_x] -= v;
}
}
}
STOP_TIMER("Inner add y block")
}
#endif
}
//FIXME name clenup (b_w, block_w, b_width stuff)
static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8, uint8_t *src, uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int plane_index){
const int b_width = s->b_width << s->block_max_depth;
const int b_height= s->b_height << s->block_max_depth;
const int b_stride= b_width;
BlockNode *lt= &s->block[b_x + b_y*b_stride];
BlockNode *rt= lt+1;
BlockNode *lb= lt+b_stride;
BlockNode *rb= lb+1;
uint8_t *block[4];
int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align
uint8_t *ptmp;
int x,y;
if(b_x<0){
lt= rt;
lb= rb;
}else if(b_x + 1 >= b_width){
rt= lt;
rb= lb;
}
if(b_y<0){
lt= lb;
rt= rb;
}else if(b_y + 1 >= b_height){
lb= lt;
rb= rt;
}
if(src_x<0){ //FIXME merge with prev & always round internal width upto *16
obmc -= src_x;
b_w += src_x;
src_x=0;
}else if(src_x + b_w > w){
b_w = w - src_x;
}
if(src_y<0){
obmc -= src_y*obmc_stride;
b_h += src_y;
src_y=0;
}else if(src_y + b_h> h){
b_h = h - src_y;
}
if(b_w<=0 || b_h<=0) return;
assert(src_stride > 2*MB_SIZE + 5);
dst += src_x + src_y*dst_stride;
dst8+= src_x + src_y*src_stride;
// src += src_x + src_y*src_stride;
ptmp= tmp + 3*tmp_step;
block[0]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
if(same_block(lt, rt)){
block[1]= block[0];
}else{
block[1]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[1], src, tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
}
if(same_block(lt, lb)){
block[2]= block[0];
}else if(same_block(rt, lb)){
block[2]= block[1];
}else{
block[2]= ptmp;
ptmp+=tmp_step;
pred_block(s, block[2], src, tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
}
if(same_block(lt, rb) ){
block[3]= block[0];
}else if(same_block(rt, rb)){
block[3]= block[1];
}else if(same_block(lb, rb)){
block[3]= block[2];
}else{
block[3]= ptmp;
pred_block(s, block[3], src, tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
}
#if 0
for(y=0; y<b_h; y++){
for(x=0; x<b_w; x++){
int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc2= obmc + (obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
for(y=0; y<b_h; y++){
uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
uint8_t *obmc4= obmc3+ (obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX);
if(add) dst[x + y*dst_stride] += v;
else dst[x + y*dst_stride] -= v;
}
}
#else
for(y=0; y<b_h; y++){
//FIXME ugly missue of obmc_stride
uint8_t *obmc1= obmc + y*obmc_stride;
uint8_t *obmc2= obmc1+ (obmc_stride>>1);
uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
uint8_t *obmc4= obmc3+ (obmc_stride>>1);
for(x=0; x<b_w; x++){
int v= obmc1[x] * block[3][x + y*src_stride]
+obmc2[x] * block[2][x + y*src_stride]
+obmc3[x] * block[1][x + y*src_stride]
+obmc4[x] * block[0][x + y*src_stride];
v <<= 8 - LOG2_OBMC_MAX;
if(FRAC_BITS != 8){
v += 1<<(7 - FRAC_BITS);
v >>= 8 - FRAC_BITS;
}
if(add){
v += dst[x + y*dst_stride];
v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
if(v&(~255)) v= ~(v>>31);
dst8[x + y*src_stride] = v;
}else{
dst[x + y*dst_stride] -= v;
}
}
}
#endif
}
static always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, DWTELEM * old_buffer, int plane_index, int add, int mb_y){
Plane *p= &s->plane[plane_index];
const int mb_w= s->b_width << s->block_max_depth;
const int mb_h= s->b_height << s->block_max_depth;
int x, y, mb_x;
int block_size = MB_SIZE >> s->block_max_depth;
int block_w = plane_index ? block_size/2 : block_size;
const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
int obmc_stride= plane_index ? block_size : 2*block_size;
int ref_stride= s->current_picture.linesize[plane_index];
uint8_t *ref = s->last_picture.data[plane_index];
uint8_t *dst8= s->current_picture.data[plane_index];
int w= p->width;
int h= p->height;
START_TIMER
if(s->keyframe || (s->avctx->debug&512)){
if(mb_y==mb_h)
return;
if(add){
for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++)
{
// DWTELEM * line = slice_buffer_get_line(sb, y);
DWTELEM * line = sb->line[y];
for(x=0; x<w; x++)
{
// int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
v >>= FRAC_BITS;
if(v&(~255)) v= ~(v>>31);
dst8[x + y*ref_stride]= v;
}
}
}else{
for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++)
{
// DWTELEM * line = slice_buffer_get_line(sb, y);
DWTELEM * line = sb->line[y];
for(x=0; x<w; x++)
{
line[x] -= 128 << FRAC_BITS;
// buf[x + y*w]-= 128<<FRAC_BITS;
}
}
}
return;
}
for(mb_x=0; mb_x<=mb_w; mb_x++){
START_TIMER
add_yblock_buffered(s, sb, old_buffer, dst8, ref, obmc,
block_w*mb_x - block_w/2,
block_w*mb_y - block_w/2,
block_w, block_w,
w, h,
w, ref_stride, obmc_stride,
mb_x - 1, mb_y - 1,
add, plane_index);
STOP_TIMER("add_yblock")
}
STOP_TIMER("predict_slice")
}
static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){
Plane *p= &s->plane[plane_index];
const int mb_w= s->b_width << s->block_max_depth;
const int mb_h= s->b_height << s->block_max_depth;
int x, y, mb_x;
int block_size = MB_SIZE >> s->block_max_depth;
int block_w = plane_index ? block_size/2 : block_size;
const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
int obmc_stride= plane_index ? block_size : 2*block_size;
int ref_stride= s->current_picture.linesize[plane_index];
uint8_t *ref = s->last_picture.data[plane_index];
uint8_t *dst8= s->current_picture.data[plane_index];
int w= p->width;
int h= p->height;
START_TIMER
if(s->keyframe || (s->avctx->debug&512)){
if(mb_y==mb_h)
return;
if(add){
for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
for(x=0; x<w; x++){
int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
v >>= FRAC_BITS;
if(v&(~255)) v= ~(v>>31);
dst8[x + y*ref_stride]= v;
}
}
}else{
for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
for(x=0; x<w; x++){
buf[x + y*w]-= 128<<FRAC_BITS;
}
}
}
return;
}
for(mb_x=0; mb_x<=mb_w; mb_x++){
START_TIMER
add_yblock(s, buf, dst8, ref, obmc,
block_w*mb_x - block_w/2,
block_w*mb_y - block_w/2,
block_w, block_w,
w, h,
w, ref_stride, obmc_stride,
mb_x - 1, mb_y - 1,
add, plane_index);
STOP_TIMER("add_yblock")
}
STOP_TIMER("predict_slice")
}
static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){
const int mb_h= s->b_height << s->block_max_depth;
int mb_y;
for(mb_y=0; mb_y<=mb_h; mb_y++)
predict_slice(s, buf, plane_index, add, mb_y);
}
static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){
const int level= b->level;
const int w= b->width;
const int h= b->height;
const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
int x,y, thres1, thres2;
START_TIMER
if(s->qlog == LOSSLESS_QLOG) return;
bias= bias ? 0 : (3*qmul)>>3;
thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
thres2= 2*thres1;
if(!bias){
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int i= src[x + y*stride];
if((unsigned)(i+thres1) > thres2){
if(i>=0){
i<<= QEXPSHIFT;
i/= qmul; //FIXME optimize
src[x + y*stride]= i;
}else{
i= -i;
i<<= QEXPSHIFT;
i/= qmul; //FIXME optimize
src[x + y*stride]= -i;
}
}else
src[x + y*stride]= 0;
}
}
}else{
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int i= src[x + y*stride];
if((unsigned)(i+thres1) > thres2){
if(i>=0){
i<<= QEXPSHIFT;
i= (i + bias) / qmul; //FIXME optimize
src[x + y*stride]= i;
}else{
i= -i;
i<<= QEXPSHIFT;
i= (i + bias) / qmul; //FIXME optimize
src[x + y*stride]= -i;
}
}else
src[x + y*stride]= 0;
}
}
}
if(level+1 == s->spatial_decomposition_count){
// STOP_TIMER("quantize")
}
}
static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int start_y, int end_y){
const int w= b->width;
const int h= b->height;
const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
int x,y;
START_TIMER
if(s->qlog == LOSSLESS_QLOG) return;
for(y=start_y; y<end_y; y++){
// DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
DWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
for(x=0; x<w; x++){
int i= line[x];
if(i<0){
line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
}else if(i>0){
line[x]= (( i*qmul + qadd)>>(QEXPSHIFT));
}
}
}
if(w > 200 /*level+1 == s->spatial_decomposition_count*/){
STOP_TIMER("dquant")
}
}
static void dequantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride){
const int w= b->width;
const int h= b->height;
const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
int x,y;
START_TIMER
if(s->qlog == LOSSLESS_QLOG) return;
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int i= src[x + y*stride];
if(i<0){
src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
}else if(i>0){
src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT));
}
}
}
if(w > 200 /*level+1 == s->spatial_decomposition_count*/){
STOP_TIMER("dquant")
}
}
static void decorrelate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
const int w= b->width;
const int h= b->height;
int x,y;
for(y=h-1; y>=0; y--){
for(x=w-1; x>=0; x--){
int i= x + y*stride;
if(x){
if(use_median){
if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
else src[i] -= src[i - 1];
}else{
if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
else src[i] -= src[i - 1];
}
}else{
if(y) src[i] -= src[i - stride];
}
}
}
}
static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
const int w= b->width;
const int h= b->height;
int x,y;
// START_TIMER
DWTELEM * line;
DWTELEM * prev;
if (start_y != 0)
line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
for(y=start_y; y<end_y; y++){
prev = line;
// line = slice_buffer_get_line_from_address(sb, src + (y * stride));
line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
for(x=0; x<w; x++){
if(x){
if(use_median){
if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
else line[x] += line[x - 1];
}else{
if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
else line[x] += line[x - 1];
}
}else{
if(y) line[x] += prev[x];
}
}
}
// STOP_TIMER("correlate")
}
static void correlate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
const int w= b->width;
const int h= b->height;
int x,y;
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int i= x + y*stride;
if(x){
if(use_median){
if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
else src[i] += src[i - 1];
}else{
if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
else src[i] += src[i - 1];
}
}else{
if(y) src[i] += src[i - stride];
}
}
}
}
static void encode_header(SnowContext *s){
int plane_index, level, orientation;
uint8_t kstate[32];
memset(kstate, MID_STATE, sizeof(kstate));
put_rac(&s->c, kstate, s->keyframe);
if(s->keyframe || s->always_reset)
reset_contexts(s);
if(s->keyframe){
put_symbol(&s->c, s->header_state, s->version, 0);
put_rac(&s->c, s->header_state, s->always_reset);
put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
put_rac(&s->c, s->header_state, s->spatial_scalability);
// put_rac(&s->c, s->header_state, s->rate_scalability);
for(plane_index=0; plane_index<2; plane_index++){
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1:0; orientation<4; orientation++){
if(orientation==2) continue;
put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
}
}
}
}
put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0);
put_symbol(&s->c, s->header_state, s->qlog, 1);
put_symbol(&s->c, s->header_state, s->mv_scale, 0);
put_symbol(&s->c, s->header_state, s->qbias, 1);
put_symbol(&s->c, s->header_state, s->block_max_depth, 0);
}
static int decode_header(SnowContext *s){
int plane_index, level, orientation;
uint8_t kstate[32];
memset(kstate, MID_STATE, sizeof(kstate));
s->keyframe= get_rac(&s->c, kstate);
if(s->keyframe || s->always_reset)
reset_contexts(s);
if(s->keyframe){
s->version= get_symbol(&s->c, s->header_state, 0);
if(s->version>0){
av_log(s->avctx, AV_LOG_ERROR, "version %d not supported", s->version);
return -1;
}
s->always_reset= get_rac(&s->c, s->header_state);
s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0);
s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
s->spatial_scalability= get_rac(&s->c, s->header_state);
// s->rate_scalability= get_rac(&s->c, s->header_state);
for(plane_index=0; plane_index<3; plane_index++){
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1:0; orientation<4; orientation++){
int q;
if (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
else q= get_symbol(&s->c, s->header_state, 1);
s->plane[plane_index].band[level][orientation].qlog= q;
}
}
}
}
s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0);
if(s->spatial_decomposition_type > 2){
av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
return -1;
}
s->qlog= get_symbol(&s->c, s->header_state, 1);
s->mv_scale= get_symbol(&s->c, s->header_state, 0);
s->qbias= get_symbol(&s->c, s->header_state, 1);
s->block_max_depth= get_symbol(&s->c, s->header_state, 0);
return 0;
}
static void init_qexp(){
int i;
double v=128;
for(i=0; i<QROOT; i++){
qexp[i]= lrintf(v);
v *= pow(2, 1.0 / QROOT);
}
}
static int common_init(AVCodecContext *avctx){
SnowContext *s = avctx->priv_data;
int width, height;
int level, orientation, plane_index, dec;
s->avctx= avctx;
dsputil_init(&s->dsp, avctx);
#define mcf(dx,dy)
s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=
s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=
s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];
s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=
s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=
s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4];
mcf( 0, 0)
mcf( 4, 0)
mcf( 8, 0)
mcf(12, 0)
mcf( 0, 4)
mcf( 4, 4)
mcf( 8, 4)
mcf(12, 4)
mcf( 0, 8)
mcf( 4, 8)
mcf( 8, 8)
mcf(12, 8)
mcf( 0,12)
mcf( 4,12)
mcf( 8,12)
mcf(12,12)
#define mcfh(dx,dy)
s->dsp.put_pixels_tab [0][dy/4+dx/8]=
s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=
mc_block_hpel ## dx ## dy ## 16;
s->dsp.put_pixels_tab [1][dy/4+dx/8]=
s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=
mc_block_hpel ## dx ## dy ## 8;
mcfh(0, 0)
mcfh(8, 0)
mcfh(0, 8)
mcfh(8, 8)
if(!qexp[0])
init_qexp();
dec= s->spatial_decomposition_count= 5;
s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
s->chroma_h_shift= 1; //FIXME XXX
s->chroma_v_shift= 1;
// dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
width= s->avctx->width;
height= s->avctx->height;
s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM));
s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0;
for(plane_index=0; plane_index<3; plane_index++){
int w= s->avctx->width;
int h= s->avctx->height;
if(plane_index){
w>>= s->chroma_h_shift;
h>>= s->chroma_v_shift;
}
s->plane[plane_index].width = w;
s->plane[plane_index].height= h;
//av_log(NULL, AV_LOG_DEBUG, "%d %dn", w, h);
for(level=s->spatial_decomposition_count-1; level>=0; level--){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &s->plane[plane_index].band[level][orientation];
b->buf= s->spatial_dwt_buffer;
b->level= level;
b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
b->width = (w + !(orientation&1))>>1;
b->height= (h + !(orientation>1))>>1;
b->stride_line = 1 << (s->spatial_decomposition_count - level);
b->buf_x_offset = 0;
b->buf_y_offset = 0;
if(orientation&1){
b->buf += (w+1)>>1;
b->buf_x_offset = (w+1)>>1;
}
if(orientation>1){
b->buf += b->stride>>1;
b->buf_y_offset = b->stride_line >> 1;
}
if(level)
b->parent= &s->plane[plane_index].band[level-1][orientation];
b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
}
w= (w+1)>>1;
h= (h+1)>>1;
}
}
reset_contexts(s);
/*
width= s->width= avctx->width;
height= s->height= avctx->height;
assert(width && height);
*/
s->avctx->get_buffer(s->avctx, &s->mconly_picture);
return 0;
}
static void calculate_vissual_weight(SnowContext *s, Plane *p){
int width = p->width;
int height= p->height;
int level, orientation, x, y;
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
DWTELEM *buf= b->buf;
int64_t error=0;
memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height);
buf[b->width/2 + b->height/2*b->stride]= 256*256;
ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(y=0; y<height; y++){
for(x=0; x<width; x++){
int64_t d= s->spatial_dwt_buffer[x + y*width];
error += d*d;
}
}
b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
// av_log(NULL, AV_LOG_DEBUG, "%d %d %dn", level, orientation, b->qlog/*, sqrt(error)*/);
}
}
}
static int encode_init(AVCodecContext *avctx)
{
SnowContext *s = avctx->priv_data;
int plane_index;
if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){
av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it may not be decodable with future versions!!!n"
"use vstrict=-2 / -strict -2 to use it anywayn");
return -1;
}
common_init(avctx);
alloc_blocks(s);
s->version=0;
s->m.avctx = avctx;
s->m.flags = avctx->flags;
s->m.bit_rate= avctx->bit_rate;
s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
h263_encode_init(&s->m); //mv_penalty
if(avctx->flags&CODEC_FLAG_PASS1){
if(!avctx->stats_out)
avctx->stats_out = av_mallocz(256);
}
if(avctx->flags&CODEC_FLAG_PASS2){
if(ff_rate_control_init(&s->m) < 0)
return -1;
}
for(plane_index=0; plane_index<3; plane_index++){
calculate_vissual_weight(s, &s->plane[plane_index]);
}
avctx->coded_frame= &s->current_picture;
switch(avctx->pix_fmt){
// case PIX_FMT_YUV444P:
// case PIX_FMT_YUV422P:
case PIX_FMT_YUV420P:
case PIX_FMT_GRAY8:
// case PIX_FMT_YUV411P:
// case PIX_FMT_YUV410P:
s->colorspace_type= 0;
break;
/* case PIX_FMT_RGBA32:
s->colorspace= 1;
break;*/
default:
av_log(avctx, AV_LOG_ERROR, "format not supportedn");
return -1;
}
// avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
s->chroma_h_shift= 1;
s->chroma_v_shift= 1;
return 0;
}
static int frame_start(SnowContext *s){
AVFrame tmp;
int w= s->avctx->width; //FIXME round up to x16 ?
int h= s->avctx->height;
if(s->current_picture.data[0]){
draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH );
draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2);
draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2);
}
tmp= s->last_picture;
s->last_picture= s->current_picture;
s->current_picture= tmp;
s->current_picture.reference= 1;
if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){
av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failedn");
return -1;
}
return 0;
}
static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
SnowContext *s = avctx->priv_data;
RangeCoder * const c= &s->c;
AVFrame *pict = data;
const int width= s->avctx->width;
const int height= s->avctx->height;
int level, orientation, plane_index;
ff_init_range_encoder(c, buf, buf_size);
ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
s->input_picture = *pict;
if(avctx->flags&CODEC_FLAG_PASS2){
s->m.pict_type =
pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type;
s->keyframe= pict->pict_type==FF_I_TYPE;
s->m.picture_number= avctx->frame_number;
pict->quality= ff_rate_estimate_qscale(&s->m);
}else{
s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE;
}
if(pict->quality){
s->qlog= rint(QROOT*log(pict->quality / (float)FF_QP2LAMBDA)/log(2));
//<64 >60
s->qlog += 61*QROOT/8;
}else{
s->qlog= LOSSLESS_QLOG;
}
frame_start(s);
s->current_picture.key_frame= s->keyframe;
s->m.current_picture_ptr= &s->m.current_picture;
if(pict->pict_type == P_TYPE){
int block_width = (width +15)>>4;
int block_height= (height+15)>>4;
int stride= s->current_picture.linesize[0];
assert(s->current_picture.data[0]);
assert(s->last_picture.data[0]);
s->m.avctx= s->avctx;
s->m.current_picture.data[0]= s->current_picture.data[0];
s->m. last_picture.data[0]= s-> last_picture.data[0];
s->m. new_picture.data[0]= s-> input_picture.data[0];
s->m. last_picture_ptr= &s->m. last_picture;
s->m.linesize=
s->m. last_picture.linesize[0]=
s->m. new_picture.linesize[0]=
s->m.current_picture.linesize[0]= stride;
s->m.uvlinesize= s->current_picture.linesize[1];
s->m.width = width;
s->m.height= height;
s->m.mb_width = block_width;
s->m.mb_height= block_height;
s->m.mb_stride= s->m.mb_width+1;
s->m.b8_stride= 2*s->m.mb_width+1;
s->m.f_code=1;
s->m.pict_type= pict->pict_type;
s->m.me_method= s->avctx->me_method;
s->m.me.scene_change_score=0;
s->m.flags= s->avctx->flags;
s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
s->m.out_format= FMT_H263;
s->m.unrestricted_mv= 1;
s->lambda = s->m.lambda= pict->quality * 3/2; //FIXME bug somewhere else
s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
s->m.dsp= s->dsp; //move
ff_init_me(&s->m);
}
redo_frame:
s->qbias= pict->pict_type == P_TYPE ? 2 : 0;
encode_header(s);
s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
encode_blocks(s);
s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
for(plane_index=0; plane_index<3; plane_index++){
Plane *p= &s->plane[plane_index];
int w= p->width;
int h= p->height;
int x, y;
// int bits= put_bits_count(&s->c.pb);
//FIXME optimize
if(pict->data[plane_index]) //FIXME gray hack
for(y=0; y<h; y++){
for(x=0; x<w; x++){
s->spatial_dwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
}
}
predict_plane(s, s->spatial_dwt_buffer, plane_index, 0);
if( plane_index==0
&& pict->pict_type == P_TYPE
&& s->m.me.scene_change_score > s->avctx->scenechange_threshold){
ff_init_range_encoder(c, buf, buf_size);
ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
pict->pict_type= FF_I_TYPE;
s->keyframe=1;
reset_contexts(s);
goto redo_frame;
}
if(s->qlog == LOSSLESS_QLOG){
for(y=0; y<h; y++){
for(x=0; x<w; x++){
s->spatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
}
}
}
ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
quantize(s, b, b->buf, b->stride, s->qbias);
if(orientation==0)
decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0);
encode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation);
assert(b->parent==NULL || b->parent->stride == b->stride*2);
if(orientation==0)
correlate(s, b, b->buf, b->stride, 1, 0);
}
}
// av_log(NULL, AV_LOG_DEBUG, "plane:%d bits:%dn", plane_index, put_bits_count(&s->c.pb) - bits);
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
dequantize(s, b, b->buf, b->stride);
}
}
ff_spatial_idwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
if(s->qlog == LOSSLESS_QLOG){
for(y=0; y<h; y++){
for(x=0; x<w; x++){
s->spatial_dwt_buffer[y*w + x]<<=FRAC_BITS;
}
}
}
{START_TIMER
predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
STOP_TIMER("pred-conv")}
if(s->avctx->flags&CODEC_FLAG_PSNR){
int64_t error= 0;
if(pict->data[plane_index]) //FIXME gray hack
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
error += d*d;
}
}
s->avctx->error[plane_index] += error;
s->current_picture.error[plane_index] = error;
}
}
if(s->last_picture.data[0])
avctx->release_buffer(avctx, &s->last_picture);
s->current_picture.coded_picture_number = avctx->frame_number;
s->current_picture.pict_type = pict->pict_type;
s->current_picture.quality = pict->quality;
if(avctx->flags&CODEC_FLAG_PASS1){
s->m.p_tex_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits - s->m.mv_bits;
s->m.current_picture.display_picture_number =
s->m.current_picture.coded_picture_number = avctx->frame_number;
s->m.pict_type = pict->pict_type;
s->m.current_picture.quality = pict->quality;
ff_write_pass1_stats(&s->m);
}
if(avctx->flags&CODEC_FLAG_PASS2){
s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
}
emms_c();
return ff_rac_terminate(c);
}
static void common_end(SnowContext *s){
int plane_index, level, orientation;
av_freep(&s->spatial_dwt_buffer);
av_freep(&s->m.me.scratchpad);
av_freep(&s->m.me.map);
av_freep(&s->m.me.score_map);
av_freep(&s->block);
for(plane_index=0; plane_index<3; plane_index++){
for(level=s->spatial_decomposition_count-1; level>=0; level--){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &s->plane[plane_index].band[level][orientation];
av_freep(&b->x_coeff);
}
}
}
}
static int encode_end(AVCodecContext *avctx)
{
SnowContext *s = avctx->priv_data;
common_end(s);
av_free(avctx->stats_out);
return 0;
}
static int decode_init(AVCodecContext *avctx)
{
SnowContext *s = avctx->priv_data;
int block_size;
avctx->pix_fmt= PIX_FMT_YUV420P;
common_init(avctx);
block_size = MB_SIZE >> s->block_max_depth;
slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 2)) + 1, s->plane[0].width, s->spatial_dwt_buffer);
return 0;
}
static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){
SnowContext *s = avctx->priv_data;
RangeCoder * const c= &s->c;
int bytes_read;
AVFrame *picture = data;
int level, orientation, plane_index;
ff_init_range_decoder(c, buf, buf_size);
ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P
decode_header(s);
if(!s->block) alloc_blocks(s);
frame_start(s);
//keyframe flag dupliaction mess FIXME
if(avctx->debug&FF_DEBUG_PICT_INFO)
av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%dn", s->keyframe, s->qlog);
decode_blocks(s);
for(plane_index=0; plane_index<3; plane_index++){
Plane *p= &s->plane[plane_index];
int w= p->width;
int h= p->height;
int x, y;
int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
if(s->avctx->debug&2048){
memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
}
}
}
{ START_TIMER
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
unpack_coeffs(s, b, b->parent, orientation);
}
}
STOP_TIMER("unpack coeffs");
}
{START_TIMER
const int mb_h= s->b_height << s->block_max_depth;
const int block_size = MB_SIZE >> s->block_max_depth;
const int block_w = plane_index ? block_size/2 : block_size;
int mb_y;
dwt_compose_t cs[MAX_DECOMPOSITIONS];
int yd=0, yq=0;
int y;
int end_y;
ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(mb_y=0; mb_y<=mb_h; mb_y++){
int slice_starty = block_w*mb_y;
int slice_h = block_w*(mb_y+1);
if (!(s->keyframe || s->avctx->debug&512)){
slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
slice_h -= (block_w >> 1);
}
{
START_TIMER
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
int start_y;
int end_y;
int our_mb_start = mb_y;
int our_mb_end = (mb_y + 1);
start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + 2: 0);
end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + 2);
if (!(s->keyframe || s->avctx->debug&512)){
start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
}
start_y = FFMIN(b->height, start_y);
end_y = FFMIN(b->height, end_y);
if (start_y != end_y){
if (orientation == 0){
SubBand * correlate_band = &p->band[0][0];
int correlate_end_y = FFMIN(b->height, end_y + 1);
int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, start_y, end_y);
}
else
decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
}
}
}
STOP_TIMER("decode_subband_slice");
}
{ START_TIMER
for(; yd<slice_h; yd+=4){
ff_spatial_idwt_buffered_slice(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
}
STOP_TIMER("idwt slice");}
if(s->qlog == LOSSLESS_QLOG){
for(; yq<slice_h && yq<h; yq++){
DWTELEM * line = slice_buffer_get_line(&s->sb, yq);
for(x=0; x<w; x++){
line[x] <<= FRAC_BITS;
}
}
}
predict_slice_buffered(s, &s->sb, s->spatial_dwt_buffer, plane_index, 1, mb_y);
y = FFMIN(p->height, slice_starty);
end_y = FFMIN(p->height, slice_h);
while(y < end_y)
slice_buffer_release(&s->sb, y++);
}
slice_buffer_flush(&s->sb);
STOP_TIMER("idwt + predict_slices")}
}
emms_c();
if(s->last_picture.data[0])
avctx->release_buffer(avctx, &s->last_picture);
if(!(s->avctx->debug&2048))
*picture= s->current_picture;
else
*picture= s->mconly_picture;
*data_size = sizeof(AVFrame);
bytes_read= c->bytestream - c->bytestream_start;
if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of framen"); //FIXME
return bytes_read;
}
static int decode_end(AVCodecContext *avctx)
{
SnowContext *s = avctx->priv_data;
slice_buffer_destroy(&s->sb);
common_end(s);
return 0;
}
AVCodec snow_decoder = {
"snow",
CODEC_TYPE_VIDEO,
CODEC_ID_SNOW,
sizeof(SnowContext),
decode_init,
NULL,
decode_end,
decode_frame,
0 /*CODEC_CAP_DR1*/ /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
NULL
};
#ifdef CONFIG_ENCODERS
AVCodec snow_encoder = {
"snow",
CODEC_TYPE_VIDEO,
CODEC_ID_SNOW,
sizeof(SnowContext),
encode_init,
encode_frame,
encode_end,
};
#endif
#if 0
#undef malloc
#undef free
#undef printf
int main(){
int width=256;
int height=256;
int buffer[2][width*height];
SnowContext s;
int i;
s.spatial_decomposition_count=6;
s.spatial_decomposition_type=1;
printf("testing 5/3 DWTn");
for(i=0; i<width*height; i++)
buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
ff_spatial_dwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(i=0; i<width*height; i++)
if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %dn",i, buffer[0][i], buffer[1][i]);
printf("testing 9/7 DWTn");
s.spatial_decomposition_type=0;
for(i=0; i<width*height; i++)
buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
ff_spatial_dwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(i=0; i<width*height; i++)
if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %dn",i, buffer[0][i], buffer[1][i]);
printf("testing AC codern");
memset(s.header_state, 0, sizeof(s.header_state));
ff_init_range_encoder(&s.c, buffer[0], 256*256);
ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
for(i=-256; i<256; i++){
START_TIMER
put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1);
STOP_TIMER("put_symbol")
}
ff_rac_terminate(&s.c);
memset(s.header_state, 0, sizeof(s.header_state));
ff_init_range_decoder(&s.c, buffer[0], 256*256);
ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
for(i=-256; i<256; i++){
int j;
START_TIMER
j= get_symbol(&s.c, s.header_state, 1);
STOP_TIMER("get_symbol")
if(j!=i*i*i/3*ABS(i)) printf("fsck: %d != %dn", i, j);
}
{
int level, orientation, x, y;
int64_t errors[8][4];
int64_t g=0;
memset(errors, 0, sizeof(errors));
s.spatial_decomposition_count=3;
s.spatial_decomposition_type=0;
for(level=0; level<s.spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
int w= width >> (s.spatial_decomposition_count-level);
int h= height >> (s.spatial_decomposition_count-level);
int stride= width << (s.spatial_decomposition_count-level);
DWTELEM *buf= buffer[0];
int64_t error=0;
if(orientation&1) buf+=w;
if(orientation>1) buf+=stride>>1;
memset(buffer[0], 0, sizeof(int)*width*height);
buf[w/2 + h/2*stride]= 256*256;
ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
for(y=0; y<height; y++){
for(x=0; x<width; x++){
int64_t d= buffer[0][x + y*width];
error += d*d;
if(ABS(width/2-x)<9 && ABS(height/2-y)<9 && level==2) printf("%8lld ", d);
}
if(ABS(height/2-y)<9 && level==2) printf("n");
}
error= (int)(sqrt(error)+0.5);
errors[level][orientation]= error;
if(g) g=ff_gcd(g, error);
else g= error;
}
}
printf("static int const visual_weight[][4]={n");
for(level=0; level<s.spatial_decomposition_count; level++){
printf(" {");
for(orientation=0; orientation<4; orientation++){
printf("%8lld,", errors[level][orientation]/g);
}
printf("},n");
}
printf("};n");
{
int level=2;
int orientation=3;
int w= width >> (s.spatial_decomposition_count-level);
int h= height >> (s.spatial_decomposition_count-level);
int stride= width << (s.spatial_decomposition_count-level);
DWTELEM *buf= buffer[0];
int64_t error=0;
buf+=w;
buf+=stride>>1;
memset(buffer[0], 0, sizeof(int)*width*height);
#if 1
for(y=0; y<height; y++){
for(x=0; x<width; x++){
int tab[4]={0,2,3,1};
buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
}
}
ff_spatial_dwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
#else
for(y=0; y<h; y++){
for(x=0; x<w; x++){
buf[x + y*stride ]=169;
buf[x + y*stride-w]=64;
}
}
ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
#endif
for(y=0; y<height; y++){
for(x=0; x<width; x++){
int64_t d= buffer[0][x + y*width];
error += d*d;
if(ABS(width/2-x)<9 && ABS(height/2-y)<9) printf("%8lld ", d);
}
if(ABS(height/2-y)<9) printf("n");
}
}
}
return 0;
}
#endif