/*
* Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
*
- * This library is free software; you can redistribute it and/or
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
*
- * This library is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "avcodec.h"
-#include "common.h"
#include "dsputil.h"
#include "snow.h"
#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)
+#define ENCODER_EXTRA_BITS 4
typedef struct x_and_coeff{
int16_t x;
int height;
int qlog; ///< log(qscale)/log[2^(1/6)]
DWTELEM *buf;
+ IDWTELEM *ibuf;
int buf_x_offset;
int buf_y_offset;
int stride_line; ///< Stride measured in lines, not pixels.
}Plane;
typedef struct SnowContext{
-// 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)
+// MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX)
AVCodecContext *avctx;
RangeCoder c;
int always_reset;
int version;
int spatial_decomposition_type;
+ int last_spatial_decomposition_type;
int temporal_decomposition_type;
int spatial_decomposition_count;
int temporal_decomposition_count;
int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
uint32_t *ref_scores[MAX_REF_FRAMES];
DWTELEM *spatial_dwt_buffer;
+ IDWTELEM *spatial_idwt_buffer;
int colorspace_type;
int chroma_h_shift;
int chroma_v_shift;
int spatial_scalability;
int qlog;
+ int last_qlog;
int lambda;
int lambda2;
int pass1_rc;
int mv_scale;
+ int last_mv_scale;
int qbias;
+ int last_qbias;
#define QBIAS_SHIFT 3
int b_width;
int b_height;
int block_max_depth;
+ int last_block_max_depth;
Plane plane[MAX_PLANES];
BlockNode *block;
#define ME_CACHE_SIZE 1024
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)
+ MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX)
}SnowContext;
typedef struct {
- DWTELEM *b0;
- DWTELEM *b1;
- DWTELEM *b2;
- DWTELEM *b3;
+ IDWTELEM *b0;
+ IDWTELEM *b1;
+ IDWTELEM *b2;
+ IDWTELEM *b3;
int y;
} dwt_compose_t;
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)
+static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
{
int i;
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);
+ buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count);
+ buf->data_stack = av_malloc (sizeof(IDWTELEM *) * 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[i] = av_malloc (sizeof(IDWTELEM) * line_width);
}
buf->data_stack_top = max_allocated_lines - 1;
}
-static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
+static IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
{
int offset;
- DWTELEM * buffer;
+ IDWTELEM * buffer;
// av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);
static void slice_buffer_release(slice_buffer * buf, int line)
{
int offset;
- DWTELEM * buffer;
+ IDWTELEM * buffer;
assert(line >= 0 && line < buf->line_count);
assert(buf->line[line]);
int i;
if(v){
- const int a= ABS(v);
+ const int a= FFABS(v);
const int e= av_log2(a);
#if 1
const int el= FFMIN(e, 10);
return v;
}
-static always_inline void lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
+static av_always_inline void lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
const int mirror_left= !highpass;
const int mirror_right= (width&1) ^ highpass;
const int w= (width>>1) - 1 + (highpass & width);
}
}
-#ifndef lift5
-static always_inline void lift5(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
+static av_always_inline void inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
const int mirror_left= !highpass;
const int mirror_right= (width&1) ^ highpass;
const int w= (width>>1) - 1 + (highpass & width);
int i;
+#define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
if(mirror_left){
- int r= 3*2*ref[0];
- r += r>>4;
- r += r>>8;
- dst[0] = LIFT(src[0], ((r+add)>>shift), inverse);
+ dst[0] = LIFT(src[0], ((mul*2*ref[0]+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);
+ 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){
- int r= 3*2*ref[w*ref_step];
- r += r>>4;
- r += r>>8;
- dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse);
+ dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse);
}
}
-#endif
#ifndef liftS
-static always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
+static av_always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
const int mirror_left= !highpass;
const int mirror_right= (width&1) ^ highpass;
const int w= (width>>1) - 1 + (highpass & width);
int i;
assert(shift == 4);
-#define LIFTS(src, ref, inv) ((inv) ? (src) - (((ref) - 4*(src))>>shift): (16*4*(src) + 4*(ref) + 8 + (5<<27))/(5*16) - (1<<23))
+#define LIFTS(src, ref, inv) ((inv) ? (src) + (((ref) + 4*(src))>>shift): -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
if(mirror_left){
dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
dst += dst_step;
dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
}
}
-#endif
-
-
-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(i=0; i<n; i++){
- int x2= x + 2*i - n + 1;
- if (x2< 0) x2= -x2;
- else if(x2>=width) x2= 2*width-x2-2;
- sum += coeffs[i]*(int64_t)dst[x2];
- }
- if(inverse) dst[x] -= (sum + (1<<shift)/2)>>shift;
- else dst[x] += (sum + (1<<shift)/2)>>shift;
- }
-}
-
-static void inplace_liftV(DWTELEM *dst, int width, int height, int stride, int *coeffs, int n, int shift, int start, int inverse){
- int x, y, i;
- 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;
- else if(y2>=height) y2= 2*height-y2-2;
- sum += coeffs[i]*(int64_t)dst[x + y2*stride];
- }
- if(inverse) dst[x + y*stride] -= (sum + (1<<shift)/2)>>shift;
- else dst[x + y*stride] += (sum + (1<<shift)/2)>>shift;
- }
- }
-}
-
-#define SCALEX 1
-#define LX0 0
-#define LX1 1
-
-#if 0 // more accurate 9/7
-#define N1 2
-#define SHIFT1 14
-#define COEFFS1 (int[]){-25987,-25987}
-#define N2 2
-#define SHIFT2 19
-#define COEFFS2 (int[]){-27777,-27777}
-#define N3 2
-#define SHIFT3 15
-#define COEFFS3 (int[]){28931,28931}
-#define N4 2
-#define SHIFT4 15
-#define COEFFS4 (int[]){14533,14533}
-#elif 1 // 13/7 CRF
-#define N1 4
-#define SHIFT1 4
-#define COEFFS1 (int[]){1,-9,-9,1}
-#define N2 4
-#define SHIFT2 4
-#define COEFFS2 (int[]){-1,5,5,-1}
-#define N3 0
-#define SHIFT3 1
-#define COEFFS3 NULL
-#define N4 0
-#define SHIFT4 1
-#define COEFFS4 NULL
-#elif 1 // 3/5
-#define LX0 1
-#define LX1 0
-#define SCALEX 0.5
-#define N1 2
-#define SHIFT1 1
-#define COEFFS1 (int[]){1,1}
-#define N2 2
-#define SHIFT2 2
-#define COEFFS2 (int[]){-1,-1}
-#define N3 0
-#define SHIFT3 0
-#define COEFFS3 NULL
-#define N4 0
-#define SHIFT4 0
-#define COEFFS4 NULL
-#elif 1 // 11/5
-#define N1 0
-#define SHIFT1 1
-#define COEFFS1 NULL
-#define N2 2
-#define SHIFT2 2
-#define COEFFS2 (int[]){-1,-1}
-#define N3 2
-#define SHIFT3 0
-#define COEFFS3 (int[]){-1,-1}
-#define N4 4
-#define SHIFT4 7
-#define COEFFS4 (int[]){-5,29,29,-5}
-#define SCALEX 4
-#elif 1 // 9/7 CDF
-#define N1 2
-#define SHIFT1 7
-#define COEFFS1 (int[]){-203,-203}
-#define N2 2
-#define SHIFT2 12
-#define COEFFS2 (int[]){-217,-217}
-#define N3 2
-#define SHIFT3 7
-#define COEFFS3 (int[]){113,113}
-#define N4 2
-#define SHIFT4 9
-#define COEFFS4 (int[]){227,227}
-#define SCALEX 1
-#elif 1 // 7/5 CDF
-#define N1 0
-#define SHIFT1 1
-#define COEFFS1 NULL
-#define N2 2
-#define SHIFT2 2
-#define COEFFS2 (int[]){-1,-1}
-#define N3 2
-#define SHIFT3 0
-#define COEFFS3 (int[]){-1,-1}
-#define N4 2
-#define SHIFT4 4
-#define COEFFS4 (int[]){3,3}
-#elif 1 // 9/7 MN
-#define N1 4
-#define SHIFT1 4
-#define COEFFS1 (int[]){1,-9,-9,1}
-#define N2 2
-#define SHIFT2 2
-#define COEFFS2 (int[]){1,1}
-#define N3 0
-#define SHIFT3 1
-#define COEFFS3 NULL
-#define N4 0
-#define SHIFT4 1
-#define COEFFS4 NULL
-#else // 13/7 CRF
-#define N1 4
-#define SHIFT1 4
-#define COEFFS1 (int[]){1,-9,-9,1}
-#define N2 4
-#define SHIFT2 4
-#define COEFFS2 (int[]){-1,5,5,-1}
-#define N3 0
-#define SHIFT3 1
-#define COEFFS3 NULL
-#define N4 0
-#define SHIFT4 1
-#define COEFFS4 NULL
-#endif
-static void horizontal_decomposeX(DWTELEM *b, int width){
- DWTELEM temp[width];
- const int width2= width>>1;
- const int w2= (width+1)>>1;
- 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];
- }
- if(width&1)
- temp[x ]= b[2*x ];
- memcpy(b, temp, width*sizeof(int));
-}
-
-static void horizontal_composeX(DWTELEM *b, int width){
- DWTELEM temp[width];
- const int width2= width>>1;
- int x;
- const int w2= (width+1)>>1;
-
- memcpy(temp, b, width*sizeof(int));
- for(x=0; x<width2; x++){
- b[2*x ]= temp[x ];
- b[2*x + 1]= temp[x+w2];
- }
- if(width&1)
- b[2*x ]= temp[x ];
-
- inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 1);
- inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 1);
- inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 1);
- inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 1);
-}
-
-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;
- }
- }
+static av_always_inline void inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){
+ const int mirror_left= !highpass;
+ const int mirror_right= (width&1) ^ highpass;
+ const int w= (width>>1) - 1 + (highpass & width);
+ int i;
- for(y=0; y<height; y++){
- horizontal_decomposeX(buffer + y*stride, width);
+ assert(shift == 4);
+#define LIFTS(src, ref, inv) ((inv) ? (src) + (((ref) + 4*(src))>>shift): -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
+ if(mirror_left){
+ dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
+ dst += dst_step;
+ src += src_step;
}
- 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);
-}
-
-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);
- inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 1);
-
- for(y=0; y<height; y++){
- horizontal_composeX(buffer + y*stride, width);
+ 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);
}
- for(y=0; y<height; y++){
- for(x=0; x<width; x++){
- buffer[y*stride + x] /= SCALEX;
- }
+ if(mirror_right){
+ dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
}
}
+#endif
static void horizontal_decompose53i(DWTELEM *b, int width){
DWTELEM temp[width];
DWTELEM temp[width];
const int w2= (width+1)>>1;
- lift (temp+w2, b +1, b , 1, 2, 2, width, -W_AM, W_AO, W_AS, 1, 0);
- liftS(temp , b , temp+w2, 1, 2, 1, width, -W_BM, W_BO, W_BS, 0, 0);
- lift5(b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
+ lift (temp+w2, b +1, b , 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1);
+ liftS(temp , b , temp+w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0);
+ lift (b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0);
}
int i;
for(i=0; i<width; i++){
-#ifdef lift5
b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
-#else
- int r= 3*(b0[i] + b2[i]);
- r+= r>>4;
- r+= r>>8;
- b1[i] += (r+W_CO)>>W_CS;
-#endif
}
}
#ifdef liftS
b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
#else
- b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + 8*5 + (5<<27)) / (5*16) - (1<<23);
+ b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23);
#endif
}
}
switch(type){
case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
- case DWT_X: spatial_decomposeX (buffer, width>>level, height>>level, stride<<level); break;
}
}
}
-static void horizontal_compose53i(DWTELEM *b, int width){
- DWTELEM temp[width];
+static void horizontal_compose53i(IDWTELEM *b, int width){
+ IDWTELEM temp[width];
const int width2= width>>1;
const int w2= (width+1)>>1;
int x;
b[width -1] = A3;
b[width2-1] = A2;
#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);
+ inv_lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1);
+ inv_lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
#endif
for(x=0; x<width2; x++){
b[2*x ]= temp[x ];
b[2*x ]= temp[x ];
}
-static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
int i;
for(i=0; i<width; i++){
}
}
-static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
int i;
for(i=0; i<width; i++){
cs->y = -1;
}
-static void spatial_compose53i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){
+static void spatial_compose53i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){
cs->b0 = buffer + mirror(-1-1, height-1)*stride;
cs->b1 = buffer + mirror(-1 , height-1)*stride;
cs->y = -1;
static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
int y= cs->y;
- DWTELEM *b0= cs->b0;
- DWTELEM *b1= cs->b1;
- 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);
+ IDWTELEM *b0= cs->b0;
+ IDWTELEM *b1= cs->b1;
+ IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
+ IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
{START_TIMER
if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
cs->y += 2;
}
-static void spatial_compose53i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){
+static void spatial_compose53i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){
int y= cs->y;
- DWTELEM *b0= cs->b0;
- DWTELEM *b1= cs->b1;
- DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
- DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
+ IDWTELEM *b0= cs->b0;
+ IDWTELEM *b1= cs->b1;
+ IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
+ IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
{START_TIMER
if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
cs->y += 2;
}
-static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){
+static void spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){
dwt_compose_t cs;
spatial_compose53i_init(&cs, buffer, height, stride);
while(cs.y <= height)
}
-void ff_snow_horizontal_compose97i(DWTELEM *b, int width){
- DWTELEM temp[width];
+void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){
+ IDWTELEM temp[width];
const int w2= (width+1)>>1;
- lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1);
- lift5(temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1);
- liftS(b , temp , temp+w2, 2, 1, 1, width, -W_BM, W_BO, W_BS, 0, 1);
- lift (b+1 , temp+w2, b , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1);
+ inv_lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1);
+ inv_lift (temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1);
+ inv_liftS(b , temp , temp+w2, 2, 1, 1, width, W_BM, W_BO, W_BS, 0, 1);
+ inv_lift (b+1 , temp+w2, b , 2, 1, 2, width, W_AM, W_AO, W_AS, 1, 0);
}
-static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
int i;
for(i=0; i<width; i++){
}
}
-static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *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;
-#else
- int r= 3*(b0[i] + b2[i]);
- r+= r>>4;
- r+= r>>8;
- b1[i] -= (r+W_CO)>>W_CS;
-#endif
}
}
-static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
int i;
for(i=0; i<width; i++){
}
}
-static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
int i;
for(i=0; i<width; i++){
}
}
-void ff_snow_vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
+void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *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;
-#else
- r= 3*(b2[i] + b4[i]);
- r+= r>>4;
- r+= r>>8;
- b3[i] -= (r+W_CO)>>W_CS;
-#endif
#ifdef liftS
b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS;
#else
cs->y = -3;
}
-static void spatial_compose97i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){
+static void spatial_compose97i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){
cs->b0 = buffer + mirror(-3-1, height-1)*stride;
cs->b1 = buffer + mirror(-3 , height-1)*stride;
cs->b2 = buffer + mirror(-3+1, height-1)*stride;
static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
int y = cs->y;
- DWTELEM *b0= cs->b0;
- DWTELEM *b1= cs->b1;
- DWTELEM *b2= cs->b2;
- DWTELEM *b3= cs->b3;
- 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);
+ IDWTELEM *b0= cs->b0;
+ IDWTELEM *b1= cs->b1;
+ IDWTELEM *b2= cs->b2;
+ IDWTELEM *b3= cs->b3;
+ IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
+ IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
{START_TIMER
if(y>0 && y+4<height){
cs->y += 2;
}
-static void spatial_compose97i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){
+static void spatial_compose97i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){
int y = cs->y;
- DWTELEM *b0= cs->b0;
- DWTELEM *b1= cs->b1;
- DWTELEM *b2= cs->b2;
- DWTELEM *b3= cs->b3;
- DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
- DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
+ IDWTELEM *b0= cs->b0;
+ IDWTELEM *b1= cs->b1;
+ IDWTELEM *b2= cs->b2;
+ IDWTELEM *b3= cs->b3;
+ IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
+ IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
{START_TIMER
if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
cs->y += 2;
}
-static void spatial_compose97i(DWTELEM *buffer, int width, int height, int stride){
+static void spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){
dwt_compose_t cs;
spatial_compose97i_init(&cs, buffer, height, stride);
while(cs.y <= height)
switch(type){
case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
- /* not slicified yet */
- case DWT_X: /*spatial_composeX(buffer, width>>level, height>>level, stride<<level); break;*/
- av_log(NULL, AV_LOG_ERROR, "spatial_composeX neither buffered nor slicified yet.\n"); break;
}
}
}
-static void ff_spatial_idwt_init(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
+static void ff_spatial_idwt_init(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
int level;
for(level=decomposition_count-1; level>=0; level--){
switch(type){
case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
- /* not slicified yet */
- case DWT_X: spatial_composeX(buffer, width>>level, height>>level, stride<<level); break;
}
}
}
-static void ff_spatial_idwt_slice(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
+static void ff_spatial_idwt_slice(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
const int support = type==1 ? 3 : 5;
int level;
if(type==2) return;
break;
case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
break;
- case DWT_X: break;
}
}
}
break;
case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
break;
- case DWT_X: break;
}
}
}
}
-static void ff_spatial_idwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
- if(type==2){
- int level;
- for(level=decomposition_count-1; level>=0; level--)
- spatial_composeX (buffer, width>>level, height>>level, stride<<level);
- }else{
+static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
dwt_compose_t cs[MAX_DECOMPOSITIONS];
int y;
ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
for(y=0; y<height; y+=4)
ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
- }
}
-static int encode_subband_c0run(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){
+static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
const int w= b->width;
const int h= b->height;
int x, y;
p= parent[px + py*2*stride];
}
if(/*ll|*/l|lt|t|rt|p){
- int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p));
+ int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
put_rac(&s->c, &b->state[0][context], !!v);
}else{
}
}
if(v){
- int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p));
- int l2= 2*ABS(l) + (l<0);
- int t2= 2*ABS(t) + (t<0);
+ int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
+ int l2= 2*FFABS(l) + (l<0);
+ int t2= 2*FFABS(t) + (t<0);
- put_symbol2(&s->c, b->state[context + 2], ABS(v)-1, context-4);
+ put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0);
}
}
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, IDWTELEM *src, IDWTELEM *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);
}
}
if(/*ll|*/l|lt|t|rt|p){
- int context= av_log2(/*ABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
+ int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
v=get_rac(&s->c, &b->state[0][context]);
if(v){
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 y;
- const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
+ const int qlog= av_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(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){
qadd= 0;
qmul= 1<<QEXPSHIFT;
}
for(y=start_y; y<h; y++){
int x = 0;
int v;
- DWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
- memset(line, 0, b->width*sizeof(DWTELEM));
+ IDWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
+ memset(line, 0, b->width*sizeof(IDWTELEM));
v = b->x_coeff[new_index].coeff;
x = b->x_coeff[new_index++].x;
while(x < w)
return;
}
-static void reset_contexts(SnowContext *s){
+static void reset_contexts(SnowContext *s){ //FIXME better initial contexts
int plane_index, level, orientation;
for(plane_index=0; plane_index<3; plane_index++){
static int pix_norm1(uint8_t * pix, int line_size, int w)
{
int s, i, j;
- uint32_t *sq = squareTbl + 256;
+ uint32_t *sq = ff_squareTbl + 256;
s = 0;
for (i = 0; i < w; i++) {
}
static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
- BlockNode *left, BlockNode *top, BlockNode *tr){
+ const BlockNode *left, const BlockNode *top, const BlockNode *tr){
if(s->ref_frames == 1){
*mx = mid_pred(left->mx, top->mx, tr->mx);
*my = mid_pred(left->my, top->my, tr->my);
}else{
const int *scale = scale_mv_ref[ref];
- *mx = mid_pred(left->mx * scale[left->ref] + 128 >>8,
- top ->mx * scale[top ->ref] + 128 >>8,
- tr ->mx * scale[tr ->ref] + 128 >>8);
- *my = mid_pred(left->my * scale[left->ref] + 128 >>8,
- top ->my * scale[top ->ref] + 128 >>8,
- tr ->my * scale[tr ->ref] + 128 >>8);
+ *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
+ (top ->mx * scale[top ->ref] + 128) >>8,
+ (tr ->mx * scale[tr ->ref] + 128) >>8);
+ *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
+ (top ->my * scale[top ->ref] + 128) >>8,
+ (tr ->my * scale[tr ->ref] + 128) >>8);
}
}
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;
+ int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
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 block_w= 1<<(LOG2_MB_SIZE - level);
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
+ const BlockNode *left = x ? &s->block[index-1] : &null_block;
+ const BlockNode *top = y ? &s->block[index-w] : &null_block;
+ const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
+ const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
+ const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
+ const 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];
const int shift= 1+qpel;
MotionEstContext *c= &s->m.me;
int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
- int mx_context= av_log2(2*ABS(left->mx - top->mx));
- int my_context= av_log2(2*ABS(left->my - top->my));
+ int mx_context= av_log2(2*FFABS(left->mx - top->mx));
+ int my_context= av_log2(2*FFABS(left->my - top->my));
int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
int ref, best_ref, ref_score, ref_mx, ref_my;
s->m.mb_stride=2;
s->m.mb_x=
s->m.mb_y= 0;
- s->m.me.skip= 0;
+ c->skip= 0;
- assert(s->m.me. stride == stride);
- assert(s->m.me.uvstride == uvstride);
+ assert(c-> stride == stride);
+ assert(c->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);
assert(ref_my >= c->ymin);
assert(ref_my <= c->ymax);
- ref_score= s->m.me.sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
+ ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
if(s->ref_mvs[ref]){
my= ref_my;
}
}
- //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2
+ //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
// subpel search
+ base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
pc= s->c;
pc.bytestream_start=
pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], 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;
+ score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
block_s= block_w*block_w;
sum = pix_sum(current_data[0], stride, block_w);
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;
+ iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
// assert(score==256*256*256*64-1);
assert(iscore < 255*255*256 + s->lambda2*10);
}
}
-static always_inline int same_block(BlockNode *a, BlockNode *b){
+static av_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{
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
+ const BlockNode *left = x ? &s->block[index-1] : &null_block;
+ const BlockNode *top = y ? &s->block[index-w] : &null_block;
+ const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
+ const 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, pmy;
int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
- int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 16*!!b->ref;
- int my_context= av_log2(2*ABS(left->my - top->my)) + 16*!!b->ref;
+ int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
+ int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
if(s->keyframe){
const int rem_depth= s->block_max_depth - level;
const int index= (x + y*w) << rem_depth;
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
+ const BlockNode *left = x ? &s->block[index-1] : &null_block;
+ const BlockNode *top = y ? &s->block[index-w] : &null_block;
+ const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
+ const 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){
}
if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
- int type;
+ int type, mx, my;
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 ref = 0;
int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
- 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));
+ int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
+ int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));
type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
}
}
-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){
+static void mc_block(uint8_t *dst, const 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++){
}
#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){\
+static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const 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);\
}
}
-void ff_snow_inner_add_yblock(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
+void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
int y, x;
- DWTELEM * dst;
+ IDWTELEM * dst;
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);
+ //FIXME ugly misuse of obmc_stride
+ const uint8_t *obmc1= obmc + y*obmc_stride;
+ const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
+ const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
+ const 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]
v <<= 8 - LOG2_OBMC_MAX;
if(FRAC_BITS != 8){
- v += 1<<(7 - FRAC_BITS);
v >>= 8 - FRAC_BITS;
}
if(add){
}
//FIXME name clenup (b_w, block_w, b_width stuff)
-static always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, DWTELEM *dst, uint8_t *dst8, const 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 offset_dst, int plane_index){
+static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const 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 offset_dst, 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;
STOP_TIMER("inner_add_yblock")
}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);
+ //FIXME ugly misuse of obmc_stride
+ const uint8_t *obmc1= obmc + y*obmc_stride;
+ const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
+ const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
+ const 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]
v <<= 8 - LOG2_OBMC_MAX;
if(FRAC_BITS != 8){
- v += 1<<(7 - FRAC_BITS);
v >>= 8 - FRAC_BITS;
}
if(add){
#endif
}
-static always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, DWTELEM * old_buffer, int plane_index, int add, int mb_y){
+static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * 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;
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];
+ IDWTELEM * line = sb->line[y];
for(x=0; x<w; x++)
{
// int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
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];
+ IDWTELEM * line = sb->line[y];
for(x=0; x<w; x++)
{
line[x] -= 128 << FRAC_BITS;
STOP_TIMER("predict_slice")
}
-static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){
+static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *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;
STOP_TIMER("predict_slice")
}
-static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){
+static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *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++)
const int obmc_stride= plane_index ? block_size : 2*block_size;
const int ref_stride= s->current_picture.linesize[plane_index];
uint8_t *src= s-> input_picture.data[plane_index];
- DWTELEM *dst= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
+ IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
const int b_stride = s->b_width << s->block_max_depth;
const int w= p->width;
const int h= p->height;
b->type|= BLOCK_INTRA;
b->color[plane_index]= 0;
- memset(dst, 0, obmc_stride*obmc_stride*sizeof(DWTELEM));
+ memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
for(i=0; i<4; i++){
int mb_x2= mb_x + (i &1) - 1;
}
*b= backup;
- return clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we shouldnt need cliping
+ return av_clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we should not need clipping
}
static inline int get_block_bits(SnowContext *s, int x, int y, int w){
const int b_stride = s->b_width << s->block_max_depth;
const int b_height = s->b_height<< s->block_max_depth;
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+w<b_stride ? &s->block[index-b_stride+w] : tl;
+ const BlockNode *b = &s->block[index];
+ const BlockNode *left = x ? &s->block[index-1] : &null_block;
+ const BlockNode *top = y ? &s->block[index-b_stride] : &null_block;
+ const BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left;
+ const BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
int dmx, dmy;
-// int mx_context= av_log2(2*ABS(left->mx - top->mx));
-// int my_context= av_log2(2*ABS(left->my - top->my));
+// int mx_context= av_log2(2*FFABS(left->mx - top->mx));
+// int my_context= av_log2(2*FFABS(left->my - top->my));
if(x<0 || x>=b_stride || y>=b_height)
return 0;
//FIXME try accurate rate
//FIXME intra and inter predictors if surrounding blocks arent the same type
if(b->type & BLOCK_INTRA){
- return 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])));
+ return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
+ + av_log2(2*FFABS(left->color[1] - b->color[1]))
+ + av_log2(2*FFABS(left->color[2] - b->color[2])));
}else{
pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
dmx-= b->mx;
dmy-= b->my;
- return 2*(1 + av_log2(2*ABS(dmx)) //FIXME kill the 2* can be merged in lambda
- + av_log2(2*ABS(dmy))
+ return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
+ + av_log2(2*FFABS(dmy))
+ av_log2(2*b->ref));
}
}
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 *dst= s->current_picture.data[plane_index];
uint8_t *src= s-> input_picture.data[plane_index];
- DWTELEM *pred= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
+ IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment
uint8_t tmp[ref_stride*(2*MB_SIZE+5)];
const int b_stride = s->b_width << s->block_max_depth;
for(y=y0; y<y1; y++){
const uint8_t *obmc1= obmc_edged + y*obmc_stride;
- const DWTELEM *pred1 = pred + y*obmc_stride;
+ const IDWTELEM *pred1 = pred + y*obmc_stride;
uint8_t *cur1 = cur + y*ref_stride;
uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
for(x=x0; x<x1; x++){
+#if FRAC_BITS >= LOG2_OBMC_MAX
int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
+#else
+ int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
+#endif
v = (v + pred1[x]) >> FRAC_BITS;
if(v&(~255)) v= ~(v>>31);
dst1[x] = v;
const int ref_stride= s->current_picture.linesize[plane_index];
uint8_t *dst= s->current_picture.data[plane_index];
uint8_t *src= s-> input_picture.data[plane_index];
- static const DWTELEM zero_dst[4096]; //FIXME
+ static const IDWTELEM 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;
return distortion + rate*penalty_factor;
}
-static always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){
+static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, 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;
}
/* 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, const uint8_t *obmc_edged, int *best_rd){
+static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){
int p[2] = {p0, p1};
return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
}
-static always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
+static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, 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[4]= {block[0], block[1], block[b_stride], block[b_stride+1]};
for(i=0; i<3; i++)
color[i]= get_dc(s, mb_x, mb_y, i);
- // get previous score (cant be cached due to OBMC)
+ // get previous score (cannot be cached due to OBMC)
if(pass > 0 && (block->type&BLOCK_INTRA)){
int color0[3]= {block->color[0], block->color[1], block->color[2]};
check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd);
}
}
-static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){
+static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, 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);
+ const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
+ const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
int x,y, thres1, thres2;
// START_TIMER
- if(s->qlog == LOSSLESS_QLOG) return;
+ if(s->qlog == LOSSLESS_QLOG){
+ for(y=0; y<h; y++)
+ for(x=0; x<w; x++)
+ dst[x + y*stride]= src[x + y*stride];
+ return;
+ }
bias= bias ? 0 : (3*qmul)>>3;
thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
if(i>=0){
i<<= QEXPSHIFT;
i/= qmul; //FIXME optimize
- src[x + y*stride]= i;
+ dst[x + y*stride]= i;
}else{
i= -i;
i<<= QEXPSHIFT;
i/= qmul; //FIXME optimize
- src[x + y*stride]= -i;
+ dst[x + y*stride]= -i;
}
}else
- src[x + y*stride]= 0;
+ dst[x + y*stride]= 0;
}
}
}else{
if(i>=0){
i<<= QEXPSHIFT;
i= (i + bias) / qmul; //FIXME optimize
- src[x + y*stride]= i;
+ dst[x + y*stride]= i;
}else{
i= -i;
i<<= QEXPSHIFT;
i= (i + bias) / qmul; //FIXME optimize
- src[x + y*stride]= -i;
+ dst[x + y*stride]= -i;
}
}else
- src[x + y*stride]= 0;
+ dst[x + y*stride]= 0;
}
}
}
}
}
-static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int start_y, int end_y){
+static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
const int w= b->width;
- const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
+ const int qlog= av_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;
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;
+ IDWTELEM * 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){
}
}
-static void dequantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride){
+static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *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 qlog= av_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;
}
}
-static void decorrelate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
+static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
const int w= b->width;
const int h= b->height;
int x,y;
}
}
-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){
+static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
const int w= b->width;
int x,y;
// START_TIMER
- DWTELEM * line;
- DWTELEM * prev;
+ IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
+ IDWTELEM * 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;
// STOP_TIMER("correlate")
}
-static void correlate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
+static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
const int w= b->width;
const int h= b->height;
int x,y;
memset(kstate, MID_STATE, sizeof(kstate));
put_rac(&s->c, kstate, s->keyframe);
- if(s->keyframe || s->always_reset)
+ if(s->keyframe || s->always_reset){
reset_contexts(s);
+ s->last_spatial_decomposition_type=
+ s->last_qlog=
+ s->last_qbias=
+ s->last_mv_scale=
+ s->last_block_max_depth= 0;
+ }
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->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);
+ put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
+ put_symbol(&s->c, s->header_state, s->qlog - s->last_qlog , 1);
+ put_symbol(&s->c, s->header_state, s->mv_scale - s->last_mv_scale, 1);
+ put_symbol(&s->c, s->header_state, s->qbias - s->last_qbias , 1);
+ put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
+
+ s->last_spatial_decomposition_type= s->spatial_decomposition_type;
+ s->last_qlog = s->qlog;
+ s->last_qbias = s->qbias;
+ s->last_mv_scale = s->mv_scale;
+ s->last_block_max_depth = s->block_max_depth;
}
static int decode_header(SnowContext *s){
memset(kstate, MID_STATE, sizeof(kstate));
s->keyframe= get_rac(&s->c, kstate);
- if(s->keyframe || s->always_reset)
+ if(s->keyframe || s->always_reset){
reset_contexts(s);
+ s->spatial_decomposition_type=
+ s->qlog=
+ s->qbias=
+ s->mv_scale=
+ s->block_max_depth= 0;
+ }
if(s->keyframe){
s->version= get_symbol(&s->c, s->header_state, 0);
if(s->version>0){
}
}
- s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0);
- if(s->spatial_decomposition_type > 2){
+ s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
+ if(s->spatial_decomposition_type > 1){
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);
+ s->qlog += get_symbol(&s->c, s->header_state, 1);
+ s->mv_scale += get_symbol(&s->c, s->header_state, 1);
+ s->qbias += get_symbol(&s->c, s->header_state, 1);
+ s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
if(s->block_max_depth > 1 || s->block_max_depth < 0){
av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
s->block_max_depth= 0;
width= s->avctx->width;
height= s->avctx->height;
- s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM));
+ s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM));
+ s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this doesnt belong here
s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0;
b->buf += b->stride>>1;
b->buf_y_offset = b->stride_line >> 1;
}
+ b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);
if(level)
b->parent= &s->plane[plane_index].band[level-1][orientation];
for(level=0; level<s->spatial_decomposition_count; level++){
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &s->plane[0].band[level][orientation];
- DWTELEM *buf= b->buf;
+ IDWTELEM *buf= b->ibuf;
const int w= b->width;
const int h= b->height;
const int stride= b->stride;
- const int qlog= clip(2*QROOT + b->qlog, 0, QROOT*16);
+ const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
const int qdiv= (1<<16)/qmul;
int x, y;
+ //FIXME this is ugly
+ for(y=0; y<h; y++)
+ for(x=0; x<w; x++)
+ buf[x+y*stride]= b->buf[x+y*stride];
if(orientation==0)
decorrelate(s, b, buf, stride, 1, 0);
for(y=0; y<h; y++)
for(x=0; x<w; x++)
coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
- if(orientation==0)
- correlate(s, b, buf, stride, 1, 0);
}
}
}
pict->quality= ff_rate_estimate_qscale(&s->m, 1);
+ if (pict->quality < 0)
+ return INT_MIN;
s->lambda= pict->quality * 3/2;
delta_qlog= qscale2qlog(pict->quality) - s->qlog;
s->qlog+= delta_qlog;
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;
+ IDWTELEM *ibuf= b->ibuf;
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);
+ memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
+ ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
+ ff_spatial_idwt(s->spatial_idwt_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];
+ int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
error += d*d;
}
}
// case PIX_FMT_YUV410P:
s->colorspace_type= 0;
break;
-/* case PIX_FMT_RGBA32:
+/* case PIX_FMT_RGB32:
s->colorspace= 1;
break;*/
default:
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;
- if(!(avctx->flags&CODEC_FLAG_QSCALE))
+ if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
pict->quality= ff_rate_estimate_qscale(&s->m, 0);
+ if (pict->quality < 0)
+ return -1;
+ }
}else{
s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
s->m.pict_type=
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;
+ s->spatial_idwt_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);
+ predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
if( plane_index==0
&& pict->pict_type == P_TYPE
pict->pict_type= FF_I_TYPE;
s->keyframe=1;
s->current_picture.key_frame=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;
+ s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
+ }
+ }
+ }else{
+ for(y=0; y<h; y++){
+ for(x=0; x<w; x++){
+ s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
}
}
}
if(s->pass1_rc && plane_index==0){
int delta_qlog = ratecontrol_1pass(s, pict);
+ if (delta_qlog <= INT_MIN)
+ return -1;
if(delta_qlog){
//reordering qlog in the bitstream would eliminate this reset
ff_init_range_encoder(c, buf, buf_size);
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
- quantize(s, b, b->buf, b->stride, s->qbias);
+ quantize(s, b, b->ibuf, 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);
+ decorrelate(s, b, b->ibuf, b->stride, pict->pict_type == P_TYPE, 0);
+ encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : 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);
+ correlate(s, b, b->ibuf, b->stride, 1, 0);
}
}
// av_log(NULL, AV_LOG_DEBUG, "plane:%d bits:%d\n", plane_index, put_bits_count(&s->c.pb) - bits);
for(orientation=level ? 1 : 0; orientation<4; orientation++){
SubBand *b= &p->band[level][orientation];
- dequantize(s, b, b->buf, b->stride);
+ dequantize(s, b, b->ibuf, b->stride);
}
}
- ff_spatial_idwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
+ ff_spatial_idwt(s->spatial_idwt_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;
+ s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
}
}
}
{START_TIMER
- predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
+ predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
STOP_TIMER("pred-conv")}
}else{
//ME/MC only
}
}
}else{
- memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
- predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
+ memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
+ predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
}
}
if(s->avctx->flags&CODEC_FLAG_PSNR){
s->m.current_picture.quality = pict->quality;
s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
if(s->pass1_rc)
- ff_rate_estimate_qscale(&s->m, 0);
+ if (ff_rate_estimate_qscale(&s->m, 0) < 0)
+ return -1;
if(avctx->flags&CODEC_FLAG_PASS1)
ff_write_pass1_stats(&s->m);
s->m.last_pict_type = s->m.pict_type;
+ avctx->frame_bits = s->m.frame_bits;
+ avctx->mv_bits = s->m.mv_bits;
+ avctx->misc_bits = s->m.misc_bits;
+ avctx->p_tex_bits = s->m.p_tex_bits;
emms_c();
int plane_index, level, orientation, i;
av_freep(&s->spatial_dwt_buffer);
+ av_freep(&s->spatial_idwt_buffer);
av_freep(&s->m.me.scratchpad);
av_freep(&s->m.me.map);
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 + 3)) + 1, 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_idwt_buffer);
return 0;
}
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);
+ predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
for(y=0; y<h; y++){
for(x=0; x<w; x++){
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);
+ correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
+ dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
}
else
decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
if(s->qlog == LOSSLESS_QLOG){
for(; yq<slice_h && yq<h; yq++){
- DWTELEM * line = slice_buffer_get_line(&s->sb, yq);
+ IDWTELEM * 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);
+ predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);
y = FFMIN(p->height, slice_starty);
end_y = FFMIN(p->height, slice_h);
NULL
};
-#ifdef CONFIG_ENCODERS
+#ifdef CONFIG_SNOW_ENCODER
AVCodec snow_encoder = {
"snow",
CODEC_TYPE_VIDEO,
#undef malloc
#undef free
#undef printf
+#undef random
int main(){
int width=256;
ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
for(i=0; i<width*height; i++)
- if(ABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);
+ if(FFABS(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");
for(i=-256; i<256; i++){
START_TIMER
- put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1);
+ put_symbol(&s.c, s.header_state, i*i*i/3*FFABS(i), 1);
STOP_TIMER("put_symbol")
}
ff_rac_terminate(&s.c);
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 != %d\n", i, j);
+ if(j!=i*i*i/3*FFABS(i)) printf("fsck: %d != %d\n", i, j);
}
#endif
{
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(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
}
- if(ABS(height/2-y)<9 && level==2) printf("\n");
+ if(FFABS(height/2-y)<9 && level==2) printf("\n");
}
error= (int)(sqrt(error)+0.5);
errors[level][orientation]= error;
for(level=0; level<s.spatial_decomposition_count; level++){
printf(" {");
for(orientation=0; orientation<4; orientation++){
- printf("%8lld,", errors[level][orientation]/g);
+ printf("%8"PRId64",", errors[level][orientation]/g);
}
printf("},\n");
}
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(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
}
- if(ABS(height/2-y)<9) printf("\n");
+ if(FFABS(height/2-y)<9) printf("\n");
}
}