uint8_t type;
//#define TYPE_SPLIT 1
#define BLOCK_INTRA 1
+#define BLOCK_OPT 2
//#define TYPE_NOCOLOR 4
uint8_t level; //FIXME merge into type?
}BlockNode;
+static const BlockNode null_block= { //FIXME add border maybe
+ .color= {128,128,128},
+ .mx= 0,
+ .my= 0,
+ .type= 0,
+ .level= 0,
+};
+
#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)
AVCodecContext *avctx;
RangeCoder c;
DSPContext dsp;
- AVFrame input_picture;
+ AVFrame new_picture;
+ AVFrame input_picture; ///< new_picture with the internal linesizes
AVFrame current_picture;
AVFrame last_picture;
AVFrame mconly_picture;
int block_max_depth;
Plane plane[MAX_PLANES];
BlockNode *block;
+#define ME_CACHE_SIZE 1024
+ int me_cache[ME_CACHE_SIZE];
+ int me_cache_generation;
slice_buffer sb;
MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independant of MpegEncContext, so this will be removed then (FIXME/XXX)
#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))
//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))
+static void iterative_me(SnowContext *s);
+
static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, DWTELEM * base_buffer)
{
int i;
-
+
buf->base_buffer = base_buffer;
buf->line_count = line_count;
buf->line_width = line_width;
buf->data_count = max_allocated_lines;
buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count);
buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines);
-
+
for (i = 0; i < max_allocated_lines; i++)
{
buf->data_stack[i] = (DWTELEM *) av_malloc (sizeof(DWTELEM) * line_width);
}
-
+
buf->data_stack_top = max_allocated_lines - 1;
}
static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
{
- int i;
int offset;
DWTELEM * buffer;
-
-// av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);
-
+
+// av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);
+
assert(buf->data_stack_top >= 0);
// assert(!buf->line[line]);
if (buf->line[line])
return buf->line[line];
-
+
offset = buf->line_width * line;
buffer = buf->data_stack[buf->data_stack_top];
buf->data_stack_top--;
buf->line[line] = buffer;
-
+
// av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1);
-
+
return buffer;
}
static void slice_buffer_release(slice_buffer * buf, int line)
{
- int i;
int offset;
DWTELEM * buffer;
buf->data_stack_top++;
buf->data_stack[buf->data_stack_top] = buffer;
buf->line[line] = NULL;
-
+
// av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1);
}
{
int i;
slice_buffer_flush(buf);
-
+
for (i = buf->data_count - 1; i >= 0; i--)
{
assert(buf->data_stack[i]);
av_free(buf->line);
}
-#ifdef __sgi
+#ifdef __sgi
// Avoid a name clash on SGI IRIX
-#undef qexp
+#undef qexp
#endif
#define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
static uint8_t qexp[QROOT];
static inline int mirror(int v, int m){
- if (v<0) return -v;
- else if(v>m) return 2*m-v;
- else return v;
+ while((unsigned)v > (unsigned)m){
+ v=-v;
+ if(v<0) v+= 2*m;
+ }
+ return v;
}
static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
const int a= ABS(v);
const int e= av_log2(a);
#if 1
- const int el= FFMIN(e, 10);
+ const int el= FFMIN(e, 10);
put_rac(c, state+0, 0);
for(i=0; i<el; i++){
if(is_signed)
put_rac(c, state+11 + el, v < 0); //11..21
#else
-
+
put_rac(c, state+0, 0);
if(e<=9){
for(i=0; i<e; i++){
if(log2>0) r+=r;
}
put_rac(c, state+4+log2, 0);
-
+
for(i=log2-1; i>=0; i--){
put_rac(c, state+31-i, (v>>i)&1);
}
log2++;
if(log2>0) r+=r;
}
-
+
for(i=log2-1; i>=0; i--){
v+= get_rac(c, state+31-i)<<i;
}
dst += dst_step;
src += src_step;
}
-
+
for(i=0; i<w; i++){
dst[i*dst_step] = LIFT(src[i*src_step], ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), inverse);
}
-
+
if(mirror_right){
dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse);
}
dst += dst_step;
src += src_step;
}
-
+
for(i=0; i<w; i++){
int r= 3*(ref[i*ref_step] + ref[(i+1)*ref_step]);
r += r>>4;
r += r>>8;
dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse);
}
-
+
if(mirror_right){
int r= 3*2*ref[w*ref_step];
r += r>>4;
dst += dst_step;
src += src_step;
}
-
+
for(i=0; i<w; i++){
dst[i*dst_step] = LIFTS(src[i*src_step], mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, inverse);
}
-
+
if(mirror_right){
dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
}
static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){
int x, i;
-
+
for(x=start; x<width; x+=2){
int64_t sum=0;
for(y=start; y<height; y+=2){
for(x=0; x<width; x++){
int64_t sum=0;
-
+
for(i=0; i<n; i++){
int y2= y + 2*i - n + 1;
if (y2< 0) y2= -y2;
#define N4 0
#define SHIFT4 0
#define COEFFS4 NULL
-#elif 1 // 11/5
+#elif 1 // 11/5
#define N1 0
#define SHIFT1 1
#define COEFFS1 NULL
DWTELEM temp[width];
const int width2= width>>1;
const int w2= (width+1)>>1;
- int A1,A2,A3,A4, x;
+ int x;
inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0);
inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0);
inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0);
inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0);
-
+
for(x=0; x<width2; x++){
temp[x ]= b[2*x ];
temp[x+w2]= b[2*x + 1];
static void horizontal_composeX(DWTELEM *b, int width){
DWTELEM temp[width];
const int width2= width>>1;
- int A1,A2,A3,A4, x;
+ int x;
const int w2= (width+1)>>1;
memcpy(temp, b, width*sizeof(int));
static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){
int x, y;
-
+
for(y=0; y<height; y++){
for(x=0; x<width; x++){
buffer[y*stride + x] *= SCALEX;
for(y=0; y<height; y++){
horizontal_decomposeX(buffer + y*stride, width);
}
-
+
inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 0);
inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 0);
inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 0);
- inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0);
+ inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0);
}
static void spatial_composeX(DWTELEM *buffer, int width, int height, int stride){
int x, y;
-
+
inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 1);
inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 1);
inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 1);
static void horizontal_decompose53i(DWTELEM *b, int width){
DWTELEM temp[width];
const int width2= width>>1;
- int A1,A2,A3,A4, x;
+ int x;
const int w2= (width+1)>>1;
for(x=0; x<width2; x++){
if(width&1)
temp[x ]= b[2*x ];
#if 0
+ {
+ int A1,A2,A3,A4;
A2= temp[1 ];
A4= temp[0 ];
A1= temp[0+width2];
A2 += (A1 + A3 + 2)>>2;
b[width -1] = A3;
b[width2-1] = A2;
-#else
+ }
+#else
lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0);
#endif
static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] -= (b0[i] + b2[i])>>1;
}
static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] += (b0[i] + b2[i] + 2)>>2;
}
int y;
DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
DWTELEM *b1= buffer + mirror(-2 , height-1)*stride;
-
+
for(y=-2; y<height; y+=2){
DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
{START_TIMER
- if(b1 <= b3) horizontal_decompose53i(b2, width);
- if(y+2 < height) horizontal_decompose53i(b3, width);
+ if(y+1<(unsigned)height) horizontal_decompose53i(b2, width);
+ if(y+2<(unsigned)height) horizontal_decompose53i(b3, width);
STOP_TIMER("horizontal_decompose53i")}
-
+
{START_TIMER
- if(b1 <= b3) vertical_decompose53iH0(b1, b2, b3, width);
- if(b0 <= b2) vertical_decompose53iL0(b0, b1, b2, width);
+ if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width);
STOP_TIMER("vertical_decompose53i*")}
-
+
b0=b2;
b1=b3;
}
static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
}
static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
#ifdef lift5
b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
#ifdef liftS
b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
}
DWTELEM *b1= buffer + mirror(-4 , height-1)*stride;
DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
-
+
for(y=-4; y<height; y+=2){
DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
{START_TIMER
- if(b3 <= b5) horizontal_decompose97i(b4, width);
- if(y+4 < height) horizontal_decompose97i(b5, width);
+ if(y+3<(unsigned)height) horizontal_decompose97i(b4, width);
+ if(y+4<(unsigned)height) horizontal_decompose97i(b5, width);
if(width>400){
STOP_TIMER("horizontal_decompose97i")
}}
-
+
{START_TIMER
- if(b3 <= b5) vertical_decompose97iH0(b3, b4, b5, width);
- if(b2 <= b4) vertical_decompose97iL0(b2, b3, b4, width);
- if(b1 <= b3) vertical_decompose97iH1(b1, b2, b3, width);
- if(b0 <= b2) vertical_decompose97iL1(b0, b1, b2, width);
+ if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width);
+ if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width);
+ if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width);
if(width>400){
STOP_TIMER("vertical_decompose97i")
}}
-
+
b0=b2;
b1=b3;
b2=b4;
void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
int level;
-
+
for(level=0; level<decomposition_count; level++){
switch(type){
case 0: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
DWTELEM temp[width];
const int width2= width>>1;
const int w2= (width+1)>>1;
- int A1,A2,A3,A4, x;
+ int x;
#if 0
+ int A1,A2,A3,A4;
A2= temp[1 ];
A4= temp[0 ];
A1= temp[0+width2];
A2 += (A1 + A3 + 2)>>2;
b[width -1] = A3;
b[width2-1] = A2;
-#else
+#else
lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1);
lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
#endif
static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] += (b0[i] + b2[i])>>1;
}
static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] -= (b0[i] + b2[i] + 2)>>2;
}
static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
int y= cs->y;
- int mirror0 = mirror(y-1, height-1);
- int mirror1 = mirror(y , height-1);
- int mirror2 = mirror(y+1, height-1);
- int mirror3 = mirror(y+2, height-1);
-
+
DWTELEM *b0= cs->b0;
DWTELEM *b1= cs->b1;
- DWTELEM *b2= slice_buffer_get_line(sb, mirror2 * stride_line);
- DWTELEM *b3= slice_buffer_get_line(sb, mirror3 * stride_line);
+ DWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
+ DWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
{START_TIMER
- if(mirror1 <= mirror3) vertical_compose53iL0(b1, b2, b3, width);
- if(mirror0 <= mirror2) vertical_compose53iH0(b0, b1, b2, width);
+ if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
STOP_TIMER("vertical_compose53i*")}
{START_TIMER
- if(y-1 >= 0) horizontal_compose53i(b0, width);
- if(mirror0 <= mirror2) horizontal_compose53i(b1, width);
+ if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
+ if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
STOP_TIMER("horizontal_compose53i")}
cs->b0 = b2;
DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
{START_TIMER
- if(b1 <= b3) vertical_compose53iL0(b1, b2, b3, width);
- if(b0 <= b2) vertical_compose53iH0(b0, b1, b2, width);
+ if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
STOP_TIMER("vertical_compose53i*")}
{START_TIMER
- if(y-1 >= 0) horizontal_compose53i(b0, width);
- if(b0 <= b2) horizontal_compose53i(b1, width);
+ if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
+ if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
STOP_TIMER("horizontal_compose53i")}
cs->b0 = b2;
spatial_compose53i_init(&cs, buffer, height, stride);
while(cs.y <= height)
spatial_compose53i_dy(&cs, buffer, width, height, stride);
-}
+}
+
-
static void horizontal_compose97i(DWTELEM *b, int width){
DWTELEM temp[width];
const int w2= (width+1)>>1;
static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
}
static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
#ifdef lift5
b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
#ifdef liftS
b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
int i;
-
+
for(i=0; i<width; i++){
b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
}
static void vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
int i;
-
+
for(i=0; i<width; i++){
+#ifndef lift5
int r;
+#endif
b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
#ifdef lift5
b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
static void spatial_compose97i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
int y = cs->y;
-
- int mirror0 = mirror(y - 1, height - 1);
- int mirror1 = mirror(y + 0, height - 1);
- int mirror2 = mirror(y + 1, height - 1);
- int mirror3 = mirror(y + 2, height - 1);
- int mirror4 = mirror(y + 3, height - 1);
- int mirror5 = mirror(y + 4, height - 1);
+
DWTELEM *b0= cs->b0;
DWTELEM *b1= cs->b1;
DWTELEM *b2= cs->b2;
DWTELEM *b3= cs->b3;
- DWTELEM *b4= slice_buffer_get_line(sb, mirror4 * stride_line);
- DWTELEM *b5= slice_buffer_get_line(sb, mirror5 * stride_line);
-
+ DWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
+ DWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
+
{START_TIMER
if(y>0 && y+4<height){
vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
}else{
- if(mirror3 <= mirror5) vertical_compose97iL1(b3, b4, b5, width);
- if(mirror2 <= mirror4) vertical_compose97iH1(b2, b3, b4, width);
- if(mirror1 <= mirror3) vertical_compose97iL0(b1, b2, b3, width);
- if(mirror0 <= mirror2) vertical_compose97iH0(b0, b1, b2, width);
+ if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
+ if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
+ if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
}
if(width>400){
STOP_TIMER("vertical_compose97i")}}
{START_TIMER
- if(y-1>= 0) horizontal_compose97i(b0, width);
- if(mirror0 <= mirror2) horizontal_compose97i(b1, width);
-if(width>400 && mirror0 <= mirror2){
+ if(y-1<(unsigned)height) horizontal_compose97i(b0, width);
+ if(y+0<(unsigned)height) horizontal_compose97i(b1, width);
+if(width>400 && y+0<(unsigned)height){
STOP_TIMER("horizontal_compose97i")}}
cs->b0=b2;
DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
- if(stride == width && y+4 < height && 0){
- int x;
- for(x=0; x<width/2; x++)
- b5[x] += 64*2;
- for(; x<width; x++)
- b5[x] += 169*2;
- }
-
{START_TIMER
- if(b3 <= b5) vertical_compose97iL1(b3, b4, b5, width);
- if(b2 <= b4) vertical_compose97iH1(b2, b3, b4, width);
- if(b1 <= b3) vertical_compose97iL0(b1, b2, b3, width);
- if(b0 <= b2) vertical_compose97iH0(b0, b1, b2, width);
+ if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
+ if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
+ if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
+ if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
if(width>400){
STOP_TIMER("vertical_compose97i")}}
{START_TIMER
- if(y-1>= 0) horizontal_compose97i(b0, width);
- if(b0 <= b2) horizontal_compose97i(b1, width);
+ if(y-1<(unsigned)height) horizontal_compose97i(b0, width);
+ if(y+0<(unsigned)height) horizontal_compose97i(b1, width);
if(width>400 && b0 <= b2){
STOP_TIMER("horizontal_compose97i")}}
int run=0;
int runs[w*h];
int run_index=0;
-
+ int max_index;
+
for(y=0; y<h; y++){
for(x=0; x<w; x++){
int v, p=0;
if(parent){
int px= x>>1;
int py= y>>1;
- if(px<b->parent->width && py<b->parent->height)
+ if(px<b->parent->width && py<b->parent->height)
p= parent[px + py*2*stride];
}
if(!(/*ll|*/l|lt|t|rt|p)){
}
}
}
+ max_index= run_index;
runs[run_index++]= run;
run_index=0;
run= runs[run_index++];
- put_symbol2(&s->c, b->state[1], run, 3);
-
+ put_symbol2(&s->c, b->state[30], max_index, 0);
+ if(run_index <= max_index)
+ put_symbol2(&s->c, b->state[1], run, 3);
+
for(y=0; y<h; y++){
if(s->c.bytestream_end - s->c.bytestream < w*40){
av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
if(parent){
int px= x>>1;
int py= y>>1;
- if(px<b->parent->width && py<b->parent->height)
+ if(px<b->parent->width && py<b->parent->height)
p= parent[px + py*2*stride];
}
if(/*ll|*/l|lt|t|rt|p){
if(!run){
run= runs[run_index++];
- put_symbol2(&s->c, b->state[1], run, 3);
+ if(run_index <= max_index)
+ put_symbol2(&s->c, b->state[1], run, 3);
assert(v);
}else{
run--;
return 0;
}
-static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){
+static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){
// encode_subband_qtree(s, b, src, parent, stride, orientation);
// encode_subband_z0run(s, b, src, parent, stride, orientation);
return encode_subband_c0run(s, b, src, parent, stride, orientation);
const int w= b->width;
const int h= b->height;
int x,y;
-
+
if(1){
- int run;
- int index=0;
- int prev_index=-1;
- int prev2_index=0;
- int parent_index= 0;
- int prev_parent_index= 0;
-
- run= get_symbol2(&s->c, b->state[1], 3);
+ int run, runs;
+ x_and_coeff *xc= b->x_coeff;
+ x_and_coeff *prev_xc= NULL;
+ x_and_coeff *prev2_xc= xc;
+ x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
+ x_and_coeff *prev_parent_xc= parent_xc;
+
+ runs= get_symbol2(&s->c, b->state[30], 0);
+ if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
+ else run= INT_MAX;
+
for(y=0; y<h; y++){
int v=0;
int lt=0, t=0, rt=0;
- if(y && b->x_coeff[prev_index].x == 0){
- rt= b->x_coeff[prev_index].coeff;
+ if(y && prev_xc->x == 0){
+ rt= prev_xc->coeff;
}
for(x=0; x<w; x++){
int p=0;
const int l= v;
-
+
lt= t; t= rt;
if(y){
- if(b->x_coeff[prev_index].x <= x)
- prev_index++;
- if(b->x_coeff[prev_index].x == x + 1)
- rt= b->x_coeff[prev_index].coeff;
+ if(prev_xc->x <= x)
+ prev_xc++;
+ if(prev_xc->x == x + 1)
+ rt= prev_xc->coeff;
else
rt=0;
}
- if(parent){
- if(x>>1 > parent->x_coeff[parent_index].x){
- parent_index++;
+ if(parent_xc){
+ if(x>>1 > parent_xc->x){
+ parent_xc++;
}
- if(x>>1 == parent->x_coeff[parent_index].x){
- p= parent->x_coeff[parent_index].coeff;
+ if(x>>1 == parent_xc->x){
+ p= parent_xc->coeff;
}
}
if(/*ll|*/l|lt|t|rt|p){
int context= av_log2(/*ABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
v=get_rac(&s->c, &b->state[0][context]);
+ if(v){
+ v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
+ v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
+
+ xc->x=x;
+ (xc++)->coeff= v;
+ }
}else{
if(!run){
- run= get_symbol2(&s->c, b->state[1], 3);
- v=1;
+ if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
+ else run= INT_MAX;
+ v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
+ v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
+
+ xc->x=x;
+ (xc++)->coeff= v;
}else{
+ int max_run;
run--;
v=0;
- if(y && parent){
- int max_run;
-
- max_run= FFMIN(run, b->x_coeff[prev_index].x - x - 2);
- max_run= FFMIN(max_run, 2*parent->x_coeff[parent_index].x - x - 1);
- x+= max_run;
- run-= max_run;
- }
+ if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
+ else max_run= FFMIN(run, w-x-1);
+ if(parent_xc)
+ max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
+ x+= max_run;
+ run-= max_run;
}
}
- if(v){
- int context= av_log2(/*ABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
- v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
- v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
-
- b->x_coeff[index].x=x;
- b->x_coeff[index++].coeff= v;
- }
}
- b->x_coeff[index++].x= w+1; //end marker
- prev_index= prev2_index;
- prev2_index= index;
-
- if(parent){
+ (xc++)->x= w+1; //end marker
+ prev_xc= prev2_xc;
+ prev2_xc= xc;
+
+ if(parent_xc){
if(y&1){
- while(parent->x_coeff[parent_index].x != parent->width+1)
- parent_index++;
- parent_index++;
- prev_parent_index= parent_index;
+ while(parent_xc->x != parent->width+1)
+ parent_xc++;
+ parent_xc++;
+ prev_parent_xc= parent_xc;
}else{
- parent_index= prev_parent_index;
+ parent_xc= prev_parent_xc;
}
}
}
- b->x_coeff[index++].x= w+1; //end marker
+ (xc++)->x= w+1; //end marker
}
}
static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
const int w= b->width;
- int x,y;
+ int y;
const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
int new_index = 0;
-
+
START_TIMER
if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
if (start_y != 0)
new_index = save_state[0];
-
+
for(y=start_y; y<h; y++){
int x = 0;
int v;
if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){
STOP_TIMER("decode_subband")
}
-
+
/* Save our variables for the next slice. */
save_state[0] = new_index;
-
+
return;
}
static int alloc_blocks(SnowContext *s){
int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
int h= -((-s->avctx->height)>>LOG2_MB_SIZE);
-
+
s->b_width = w;
s->b_height= h;
-
+
s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
return 0;
}
const int block_w= 1<<rem_depth;
BlockNode block;
int i,j;
-
+
block.color[0]= l;
block.color[1]= cb;
block.color[2]= cr;
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;
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;
+ int l,cr,cb;
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]= {¤t_mb[0][0], ¤t_mb[1][0], ¤t_mb[2][0]};
+ uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w,
+ s->input_picture.data[1] + (x + y*uvstride)*block_w/2,
+ s->input_picture.data[2] + (x + y*uvstride)*block_w/2};
int P[10][2];
int16_t last_mv[3][2];
int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
return 0;
}
- //FIXME optimize
- for(i=0; i<block_w; i++)
- memcpy(¤t_mb[0][0] + stride*i, new_l + instride*i, block_w);
- for(i=0; i<block_w>>1; i++)
- memcpy(¤t_mb[1][0] + uvstride*i, new_cb + uvinstride*i, block_w>>1);
- for(i=0; i<block_w>>1; i++)
- memcpy(¤t_mb[2][0] + uvstride*i, new_cr + uvinstride*i, block_w>>1);
-
// clip predictors / edge ?
P_LEFT[0]= left->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_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_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);
c->pred_y = P_MEDIAN[1];
}
- score= ff_epzs_motion_search(&s->m, &mx, &my, P, 0, /*ref_index*/ 0, last_mv,
+ 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=
))>>FF_LAMBDA_SHIFT;
block_s= block_w*block_w;
- sum = pix_sum(¤t_mb[0][0], stride, block_w);
+ sum = pix_sum(current_data[0], stride, block_w);
l= (sum + block_s/2)/block_s;
- iscore = pix_norm1(¤t_mb[0][0], stride, block_w) - 2*l*sum + l*l*block_s;
-
+ iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
+
block_s= block_w*block_w>>2;
- sum = pix_sum(¤t_mb[1][0], uvstride, block_w>>1);
+ sum = pix_sum(current_data[1], uvstride, block_w>>1);
cb= (sum + block_s/2)/block_s;
// iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
- sum = pix_sum(¤t_mb[2][0], uvstride, block_w>>1);
+ sum = pix_sum(current_data[2], uvstride, block_w>>1);
cr= (sum + block_s/2)/block_s;
// iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
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+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;
}
}
+static always_inline int same_block(BlockNode *a, BlockNode *b){
+ if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
+ return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
+ }else{
+ return !((a->mx - b->mx) | (a->my - b->my) | ((a->type ^ b->type)&BLOCK_INTRA));
+ }
+}
+
+static void encode_q_branch2(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;
+ int trx= (x+1)<<rem_depth;
+ BlockNode *b= &s->block[index];
+ 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 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_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;
+
+ if(s->keyframe){
+ set_blocks(s, level, x, y, pl, pcb, pcr, pmx, pmy, BLOCK_INTRA);
+ return;
+ }
+
+ if(level!=s->block_max_depth){
+ if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
+ put_rac(&s->c, &s->block_state[4 + s_context], 1);
+ }else{
+ put_rac(&s->c, &s->block_state[4 + s_context], 0);
+ encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
+ encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
+ encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
+ encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
+ return;
+ }
+ }
+ if(b->type & BLOCK_INTRA){
+ put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
+ put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
+ put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
+ put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
+ set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, BLOCK_INTRA);
+ }else{
+ put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
+ put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
+ put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
+ set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, 0);
+ }
+}
+
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;
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){
int w= s->b_width;
int h= s->b_height;
+ if(s->avctx->me_method == ME_ITER && !s->keyframe)
+ iterative_me(s);
+
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 large\n");
return;
}
for(x=0; x<w; x++){
- encode_q_branch(s, 0, x, y);
+ if(s->avctx->me_method == ME_ITER)
+ encode_q_branch2(s, 0, x, y);
+ else
+ encode_q_branch (s, 0, x, y);
}
}
}
// if(b_w==16) am= 8*(a1+a2);
- if(dx<8) tmp[x]= (32*a2*( 8-dx) + am* dx + 128)>>8;
- else tmp[x]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8;
+ 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;
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 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) dst[x]= (32*a2*( 8-dy) + am* dy + 128)>>8;
- else dst[x]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8;
+ 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;
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){
+ if(block->type & BLOCK_INTRA){
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;
+ const int color = block->color[plane_index];
+ const int color4= color*0x01010101;
+ if(b_w==16){
+ for(y=0; y < b_h; y++){
+ *(uint32_t*)&dst[0 + y*stride]= color4;
+ *(uint32_t*)&dst[4 + y*stride]= color4;
+ *(uint32_t*)&dst[8 + y*stride]= color4;
+ *(uint32_t*)&dst[12+ y*stride]= color4;
+ }
+ }else if(b_w==8){
+ for(y=0; y < b_h; y++){
+ *(uint32_t*)&dst[0 + y*stride]= color4;
+ *(uint32_t*)&dst[4 + y*stride]= color4;
+ }
+ }else if(b_w==4){
+ for(y=0; y < b_h; y++){
+ *(uint32_t*)&dst[0 + y*stride]= color4;
+ }
+ }else{
+ for(y=0; y < b_h; y++){
+ for(x=0; x < b_w; x++){
+ dst[x + y*stride]= color;
+ }
}
}
}else{
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))
+ assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
+ assert(!(b_w&(b_w-1)));
+ assert(b_w>1 && b_h>1);
+ if((dx&3) || (dy&3) || b_w==2 || b_h==2)
mc_block(dst, src, tmp, stride, b_w, b_h, dx, dy);
- else
+ else if(b_w==b_h)
s->dsp.put_h264_qpel_pixels_tab[2-(b_w>>3)][dy+(dx>>2)](dst,src + 2 + 2*stride,stride);
+ else if(b_w==2*b_h){
+ s->dsp.put_h264_qpel_pixels_tab[2-(b_h>>3)][dy+(dx>>2)](dst ,src + 2 + 2*stride,stride);
+ s->dsp.put_h264_qpel_pixels_tab[2-(b_h>>3)][dy+(dx>>2)](dst+b_h,src + 2 + b_h + 2*stride,stride);
+ }else{
+ assert(2*b_w==b_h);
+ s->dsp.put_h264_qpel_pixels_tab[2-(b_w>>3)][dy+(dx>>2)](dst ,src + 2 + 2*stride ,stride);
+ s->dsp.put_h264_qpel_pixels_tab[2-(b_w>>3)][dy+(dx>>2)](dst+b_w*stride,src + 2 + 2*stride+b_w*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;
BlockNode *rt= lt+1;
BlockNode *lb= lt+b_stride;
BlockNode *rb= lb+1;
- uint8_t *block[4];
- uint8_t tmp[src_stride*(b_h+5)]; //FIXME align
+ 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){
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;
}else if(src_y + b_h> h){
b_h = h - src_y;
}
-
+
if(b_w<=0 || b_h<=0) return;
-assert(src_stride > 7*MB_SIZE);
+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;
- block[0]= tmp+3*MB_SIZE;
- pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
+ 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]= tmp + 4*MB_SIZE;
+ 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]= tmp+5*MB_SIZE;
+ 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);
}
}else if(same_block(lb, rb)){
block[3]= block[2];
}else{
- block[3]= tmp+6*MB_SIZE;
+ 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
{
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;
BlockNode *rt= lt+1;
BlockNode *lb= lt+b_stride;
BlockNode *rb= lb+1;
- uint8_t *block[4];
- uint8_t tmp[src_stride*(b_h+5)]; //FIXME align
+ 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){
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;
}else if(src_y + b_h> h){
b_h = h - src_y;
}
-
+
if(b_w<=0 || b_h<=0) return;
-assert(src_stride > 7*MB_SIZE);
+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;
- block[0]= tmp+3*MB_SIZE;
- pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
+ 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]= tmp + 4*MB_SIZE;
+ 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]= tmp+5*MB_SIZE;
+ 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);
}
}else if(same_block(lb, rb)){
block[3]= block[2];
}else{
- block[3]= tmp+6*MB_SIZE;
+ 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
+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);
int w= p->width;
int h= p->height;
START_TIMER
-
+
if(s->keyframe || (s->avctx->debug&512)){
if(mb_y==mb_h)
return;
return;
}
-
+
for(mb_x=0; mb_x<=mb_w; mb_x++){
START_TIMER
- add_yblock_buffered(s, sb, old_buffer, dst8, ref, obmc,
+ 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, ref_stride, obmc_stride,
mb_x - 1, mb_y - 1,
add, plane_index);
-
+
STOP_TIMER("add_yblock")
}
-
+
STOP_TIMER("predict_slice")
}
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;
+ const 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;
return;
}
-
+
for(mb_x=0; mb_x<=mb_w; mb_x++){
START_TIMER
- add_yblock(s, buf, dst8, ref, obmc,
+ 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, ref_stride, obmc_stride,
mb_x - 1, mb_y - 1,
add, plane_index);
-
+
STOP_TIMER("add_yblock")
}
-
+
STOP_TIMER("predict_slice")
}
predict_slice(s, buf, plane_index, add, mb_y);
}
+static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
+ int i, x2, y2;
+ Plane *p= &s->plane[plane_index];
+ const int block_size = MB_SIZE >> s->block_max_depth;
+ const 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];
+ const int obmc_stride= plane_index ? block_size : 2*block_size;
+ const int ref_stride= s->current_picture.linesize[plane_index];
+ uint8_t *ref= s-> last_picture.data[plane_index];
+ uint8_t *dst= s->current_picture.data[plane_index];
+ uint8_t *src= s-> input_picture.data[plane_index];
+ const static DWTELEM zero_dst[4096]; //FIXME
+ const int b_stride = s->b_width << s->block_max_depth;
+ const int w= p->width;
+ const int h= p->height;
+ int index= mb_x + mb_y*b_stride;
+ BlockNode *b= &s->block[index];
+ BlockNode backup= *b;
+ int ab=0;
+ int aa=0;
+
+ b->type|= BLOCK_INTRA;
+ b->color[plane_index]= 0;
+
+ for(i=0; i<4; i++){
+ int mb_x2= mb_x + (i &1) - 1;
+ int mb_y2= mb_y + (i>>1) - 1;
+ int x= block_w*mb_x2 + block_w/2;
+ int y= block_w*mb_y2 + block_w/2;
+
+ add_yblock(s, zero_dst, dst, ref, obmc,
+ x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, plane_index);
+
+ for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_w); y2++){
+ for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
+ int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_w*mb_y - block_w/2))*obmc_stride;
+ int obmc_v= obmc[index];
+ if(y<0) obmc_v += obmc[index + block_w*obmc_stride];
+ if(x<0) obmc_v += obmc[index + block_w];
+ if(y+block_w>h) obmc_v += obmc[index - block_w*obmc_stride];
+ if(x+block_w>w) obmc_v += obmc[index - block_w];
+ //FIXME precalc this or simplify it somehow else
+
+ ab += (src[x2 + y2*ref_stride] - dst[x2 + y2*ref_stride]) * obmc_v;
+ aa += obmc_v * obmc_v; //FIXME precalclate this
+ }
+ }
+ }
+ *b= backup;
+
+ return clip(((ab<<6) + aa/2)/aa, 0, 255); //FIXME we shouldnt need cliping
+}
+
+static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
+ int i, y2;
+ Plane *p= &s->plane[plane_index];
+ const int block_size = MB_SIZE >> s->block_max_depth;
+ const 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];
+ const int obmc_stride= plane_index ? block_size : 2*block_size;
+ const int ref_stride= s->current_picture.linesize[plane_index];
+ uint8_t *ref= s-> last_picture.data[plane_index];
+ uint8_t *dst= s->current_picture.data[plane_index];
+ uint8_t *src= s-> input_picture.data[plane_index];
+ const static DWTELEM zero_dst[4096]; //FIXME
+ const int b_stride = s->b_width << s->block_max_depth;
+ const int b_height = s->b_height<< s->block_max_depth;
+ const int w= p->width;
+ const int h= p->height;
+ int distortion= 0;
+ int rate= 0;
+ const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
+
+ for(i=0; i<4; i++){
+ int mb_x2= mb_x + (i &1) - 1;
+ int mb_y2= mb_y + (i>>1) - 1;
+ int x= block_w*mb_x2 + block_w/2;
+ int y= block_w*mb_y2 + block_w/2;
+
+ add_yblock(s, zero_dst, dst, ref, obmc,
+ x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, plane_index);
+
+ //FIXME find a cleaner/simpler way to skip the outside stuff
+ for(y2= y; y2<0; y2++)
+ memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
+ for(y2= h; y2<y+block_w; y2++)
+ memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
+ if(x<0){
+ for(y2= y; y2<y+block_w; y2++)
+ memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
+ }
+ if(x+block_w > w){
+ for(y2= y; y2<y+block_w; y2++)
+ memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
+ }
+
+ assert(block_w== 8 || block_w==16);
+ distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w);
+ }
+
+ if(plane_index==0){
+ for(i=0; i<4; i++){
+/* ..RRr
+ * .RXx.
+ * rxx..
+ */
+ int x= mb_x + (i&1) - (i>>1);
+ int y= mb_y + (i>>1);
+ int index= x + y*b_stride;
+ BlockNode *b = &s->block[index];
+ BlockNode *left = x ? &s->block[index-1] : &null_block;
+ BlockNode *top = y ? &s->block[index-b_stride] : &null_block;
+ BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left;
+ BlockNode *tr = y && x+1<b_stride ? &s->block[index-b_stride+1] : tl;
+ int dmx= b->mx - mid_pred(left->mx, top->mx, tr->mx);
+ int dmy= b->my - mid_pred(left->my, top->my, tr->my);
+// int mx_context= av_log2(2*ABS(left->mx - top->mx));
+// int my_context= av_log2(2*ABS(left->my - top->my));
+
+ if(x<0 || x>=b_stride || y>=b_height)
+ continue;
+/*
+1 0 0
+01X 1-2 1
+001XX 3-6 2-3
+0001XXX 7-14 4-7
+00001XXXX 15-30 8-15
+*/
+//FIXME try accurate rate
+//FIXME intra and inter predictors if surrounding blocks arent the same type
+ if(b->type & BLOCK_INTRA){
+ rate += 3+2*( av_log2(2*ABS(left->color[0] - b->color[0]))
+ + av_log2(2*ABS(left->color[1] - b->color[1]))
+ + av_log2(2*ABS(left->color[2] - b->color[2])));
+ }else
+ rate += 2*(1 + av_log2(2*ABS(dmx))
+ + av_log2(2*ABS(dmy))); //FIXME kill the 2* can be merged in lambda
+ }
+ }
+
+ return distortion + rate*penalty_factor;
+}
+
+static always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, int *best_rd){
+ const int b_stride= s->b_width << s->block_max_depth;
+ BlockNode *block= &s->block[mb_x + mb_y * b_stride];
+ BlockNode backup= *block;
+ int rd, index, value;
+
+ assert(mb_x>=0 && mb_y>=0);
+ assert(mb_x<b_stride);
+
+ if(intra){
+ block->color[0] = p[0];
+ block->color[1] = p[1];
+ block->color[2] = p[2];
+ block->type |= BLOCK_INTRA;
+ }else{
+ index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
+ value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6);
+ if(s->me_cache[index] == value)
+ return 0;
+ s->me_cache[index]= value;
+
+ block->mx= p[0];
+ block->my= p[1];
+ block->type &= ~BLOCK_INTRA;
+ }
+
+ rd= get_block_rd(s, mb_x, mb_y, 0);
+
+//FIXME chroma
+ if(rd < *best_rd){
+ *best_rd= rd;
+ return 1;
+ }else{
+ *block= backup;
+ return 0;
+ }
+}
+
+/* special case for int[2] args we discard afterward, fixes compilation prob with gcc 2.95 */
+static always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int intra, int *best_rd){
+ int p[2] = {p0, p1};
+ return check_block(s, mb_x, mb_y, p, intra, best_rd);
+}
+
+static void iterative_me(SnowContext *s){
+ int pass, mb_x, mb_y;
+ 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;
+ int color[3];
+
+ for(pass=0; pass<50; pass++){
+ int change= 0;
+
+ for(mb_y= 0; mb_y<b_height; mb_y++){
+ for(mb_x= 0; mb_x<b_width; mb_x++){
+ int dia_change, i, j;
+ int best_rd= INT_MAX;
+ BlockNode backup;
+ const int index= mb_x + mb_y * b_stride;
+ BlockNode *block= &s->block[index];
+ BlockNode *tb = mb_y ? &s->block[index-b_stride ] : &null_block;
+ BlockNode *lb = mb_x ? &s->block[index -1] : &null_block;
+ BlockNode *rb = mb_x<b_width ? &s->block[index +1] : &null_block;
+ BlockNode *bb = mb_y<b_height ? &s->block[index+b_stride ] : &null_block;
+ BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : &null_block;
+ BlockNode *trb= mb_x<b_width && mb_y ? &s->block[index-b_stride+1] : &null_block;
+ BlockNode *blb= mb_x && mb_y<b_height ? &s->block[index+b_stride-1] : &null_block;
+ BlockNode *brb= mb_x<b_width && mb_y<b_height ? &s->block[index+b_stride+1] : &null_block;
+
+ if(pass && (block->type & BLOCK_OPT))
+ continue;
+ block->type |= BLOCK_OPT;
+
+ backup= *block;
+
+ if(!s->me_cache_generation)
+ memset(s->me_cache, 0, sizeof(s->me_cache));
+ s->me_cache_generation += 1<<22;
+
+ // get previous score (cant be cached due to OBMC)
+ check_block_inter(s, mb_x, mb_y, block->mx, block->my, 0, &best_rd);
+ check_block_inter(s, mb_x, mb_y, 0, 0, 0, &best_rd);
+ check_block_inter(s, mb_x, mb_y, tb->mx, tb->my, 0, &best_rd);
+ check_block_inter(s, mb_x, mb_y, lb->mx, lb->my, 0, &best_rd);
+ check_block_inter(s, mb_x, mb_y, rb->mx, rb->my, 0, &best_rd);
+ check_block_inter(s, mb_x, mb_y, bb->mx, bb->my, 0, &best_rd);
+
+ /* fullpel ME */
+ //FIXME avoid subpel interpol / round to nearest integer
+ do{
+ dia_change=0;
+ for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
+ for(j=0; j<i; j++){
+ dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), 0, &best_rd);
+ dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), 0, &best_rd);
+ dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), 0, &best_rd);
+ dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), 0, &best_rd);
+ }
+ }
+ }while(dia_change);
+ /* subpel ME */
+ do{
+ static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
+ dia_change=0;
+ for(i=0; i<8; i++)
+ dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], 0, &best_rd);
+ }while(dia_change);
+ //FIXME or try the standard 2 pass qpel or similar
+#if 1
+ for(i=0; i<3; i++){
+ color[i]= get_dc(s, mb_x, mb_y, i);
+ }
+ check_block(s, mb_x, mb_y, color, 1, &best_rd);
+ //FIXME RD style color selection
+#endif
+ if(!same_block(block, &backup)){
+ if(tb != &null_block) tb ->type &= ~BLOCK_OPT;
+ if(lb != &null_block) lb ->type &= ~BLOCK_OPT;
+ if(rb != &null_block) rb ->type &= ~BLOCK_OPT;
+ if(bb != &null_block) bb ->type &= ~BLOCK_OPT;
+ if(tlb!= &null_block) tlb->type &= ~BLOCK_OPT;
+ if(trb!= &null_block) trb->type &= ~BLOCK_OPT;
+ if(blb!= &null_block) blb->type &= ~BLOCK_OPT;
+ if(brb!= &null_block) brb->type &= ~BLOCK_OPT;
+ change ++;
+ }
+ }
+ }
+ av_log(NULL, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
+ if(!change)
+ break;
+ }
+}
+
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 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
+// 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;
}else{
for(y=0; y<h; y++){
for(x=0; x<w; x++){
- int i= src[x + y*stride];
-
+ int i= src[x + y*stride];
+
if((unsigned)(i+thres1) > thres2){
if(i>=0){
i<<= QEXPSHIFT;
}
}
-static void dequantize_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride){
+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=0; y<h; y++){
+
+ 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++){
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];
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]);
}
}
-static void correlate_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
+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;
-
- for(y=0; y<h; y++){
+
+ 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;
}
}
}
-
+
// STOP_TIMER("correlate")
}
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]);
static void encode_header(SnowContext *s){
int plane_index, level, orientation;
- uint8_t kstate[32];
-
- memset(kstate, MID_STATE, sizeof(kstate));
+ uint8_t kstate[32];
+
+ memset(kstate, MID_STATE, sizeof(kstate));
put_rac(&s->c, kstate, s->keyframe);
if(s->keyframe || s->always_reset)
}
}
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->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);
}
int plane_index, level, orientation;
uint8_t kstate[32];
- memset(kstate, MID_STATE, sizeof(kstate));
+ memset(kstate, MID_STATE, sizeof(kstate));
s->keyframe= get_rac(&s->c, kstate);
if(s->keyframe || s->always_reset)
}
}
}
-
+
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);
+ if(s->block_max_depth > 1){
+ av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
+ s->block_max_depth= 0;
+ return -1;
+ }
return 0;
}
for(i=0; i<QROOT; i++){
qexp[i]= lrintf(v);
- v *= pow(2, 1.0 / QROOT);
+ v *= pow(2, 1.0 / QROOT);
}
}
int level, orientation, plane_index, dec;
s->avctx= avctx;
-
+
dsputil_init(&s->dsp, avctx);
#define mcf(dx,dy)\
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++){
+
+ for(plane_index=0; plane_index<3; plane_index++){
int w= s->avctx->width;
int h= s->avctx->height;
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;
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));
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;
}
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);
SnowContext *s = avctx->priv_data;
int plane_index;
- if(avctx->strict_std_compliance >= 0){
- av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it wont be decodeable with future versions!!!\n"
- "use vstrict=-1 / -strict -1 to use it anyway\n");
+ 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 anyway\n");
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:
// 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;
+
+ ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
+ ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
+
+ s->avctx->get_buffer(s->avctx, &s->input_picture);
+
return 0;
}
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() failed\n");
return -1;
}
-
+
return 0;
}
AVFrame *pict = data;
const int width= s->avctx->width;
const int height= s->avctx->height;
- int level, orientation, plane_index;
+ int level, orientation, plane_index, i, y;
ff_init_range_encoder(c, buf, buf_size);
ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
-
- s->input_picture = *pict;
- s->keyframe=avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
- pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE;
-
+ for(i=0; i<3; i++){
+ int shift= !!i;
+ for(y=0; y<(height>>shift); y++)
+ memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]],
+ &pict->data[i][y * pict->linesize[i]],
+ width>>shift);
+ }
+ s->new_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, 0);
+ }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
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.current_picture_ptr= &s->m.current_picture;
s->m. last_picture_ptr= &s->m. last_picture;
s->m.linesize=
s->m. last_picture.linesize[0]=
s->m.dsp= s->dsp; //move
ff_init_me(&s->m);
+ s->dsp= s->m.dsp;
}
-
+
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;
}
}
predict_plane(s, s->spatial_dwt_buffer, plane_index, 0);
-
- if( plane_index==0
- && pict->pict_type == P_TYPE
+
+ 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);
reset_contexts(s);
goto redo_frame;
}
-
+
if(s->qlog == LOSSLESS_QLOG){
for(y=0; y<h; y++){
for(x=0; x<w; x++){
}
}
}
-
+
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);
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++){
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);
}
av_freep(&s->spatial_dwt_buffer);
- av_freep(&s->m.me.scratchpad);
+ 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(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);
}
}
SnowContext *s = avctx->priv_data;
common_end(s);
+ av_free(avctx->stats_out);
return 0;
}
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;
- /* FIXME block_size * 2 is determined empirically. block_size * 1.5 is definitely needed, but I (Robert) cannot figure out why more than that is needed. Perhaps there is a bug, or perhaps I overlooked some demands that are placed on the buffer. */
- /* FIXME The formula is WRONG. For height > 480, the buffer will overflow. */
- /* FIXME For now, I will use a full frame of lines. Fortunately, this should not materially effect cache performance because lines are allocated using a stack, so if in fact only 50 out of 496 lines are needed at a time, the other 446 will sit allocated but never accessed. */
-// slice_buffer_init(s->plane[0].sb, s->plane[0].height, (block_size * 2) + (s->spatial_decomposition_count * s->spatial_decomposition_count), s->plane[0].width, s->spatial_dwt_buffer);
- slice_buffer_init(&s->sb, s->plane[0].height, s->plane[0].height, s->plane[0].width, s->spatial_dwt_buffer);
-
+ slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 3)) + 1, s->plane[0].width, s->spatial_dwt_buffer);
+
return 0;
}
//keyframe flag dupliaction mess FIXME
if(avctx->debug&FF_DEBUG_PICT_INFO)
av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);
-
+
decode_blocks(s);
for(plane_index=0; plane_index<3; plane_index++){
int h= p->height;
int x, y;
int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
- SubBand * correlate_band;
-
+
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);
}
STOP_TIMER("unpack coeffs");
}
-
- /* Handle level 0, orientation 0 specially. It is particularly resistant to slicing but fortunately quite small, so process it in one pass. */
- correlate_band = &p->band[0][0];
- decode_subband_slice_buffered(s, correlate_band, &s->sb, 0, correlate_band->height, decode_state[0][0]);
- correlate_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, 1, 0);
- dequantize_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride);
{START_TIMER
const int mb_h= s->b_height << s->block_max_depth;
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++){
-
- const int slice_starty = block_w*mb_y;
- const int slice_h = block_w*(mb_y+1);
- {
+ 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 : 1; orientation<4; orientation++){
+ 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 = FFMIN(b->height, (mb_y ? ((block_w * our_mb_start - 4) >> (s->spatial_decomposition_count - level)) + 5 : 0));
- end_y = FFMIN(b->height, (((block_w * our_mb_end - 4) >> (s->spatial_decomposition_count - level)) + 5));
-
- if (start_y != end_y)
- decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
+ const int extra= 3;
+ start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
+ end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
+ 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);
}
predict_slice_buffered(s, &s->sb, s->spatial_dwt_buffer, plane_index, 1, mb_y);
-
- /* Nasty hack based empirically on how predict_slice_buffered() hits the buffer. */
- /* FIXME If possible, make predict_slice fit into the slice. As of now, it works on some previous lines (up to slice_height / 2) if the condition on the next line is false. */
- if (s->keyframe || (s->avctx->debug&512)){
- y = FFMIN(p->height, slice_starty);
- end_y = FFMIN(p->height, slice_h);
- }
- else{
- y = FFMAX(0, FFMIN(p->height, slice_starty - (block_w >> 1)));
- end_y = FFMAX(0, FFMIN(p->height, slice_h - (block_w >> 1)));
- }
+
+ 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))
+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 frame\n"); //FIXME
SnowContext *s = avctx->priv_data;
slice_buffer_destroy(&s->sb);
-
+
common_end(s);
return 0;
int i;
s.spatial_decomposition_count=6;
s.spatial_decomposition_type=1;
-
+
printf("testing 5/3 DWT\n");
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);
-
+
+ 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 %d\n",i, buffer[0][i], buffer[1][i]);
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);
-
+
+ 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 %d\n",i, buffer[0][i], buffer[1][i]);
-
+ if(ABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);
+
+#if 0
printf("testing AC coder\n");
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);
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
STOP_TIMER("get_symbol")
if(j!=i*i*i/3*ABS(i)) printf("fsck: %d != %d\n", i, j);
}
+#endif
{
int level, orientation, x, y;
int64_t errors[8][4];
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);
+ 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];
buf+=w;
buf+=stride>>1;
-
+
memset(buffer[0], 0, sizeof(int)*width*height);
#if 1
for(y=0; y<height; y++){
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);
+ 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-w]=64;
}
}
- ff_spatial_idwt(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);
#endif
for(y=0; y<height; y++){
for(x=0; x<width; x++){
projects / ffmpeg / blobdiff