snow.c

tcpmp播放器的flv插件

//#define TYPE_NOCOLOR  4
    uint8_t level; //FIXME merge into type?
}BlockNode;

#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)

typedef struct x_and_coeff{
    int16_t x;
    uint16_t coeff;
} x_and_coeff;

typedef struct SubBand{
    int level;
    int stride;
    int width;
    int height;
    int qlog;                                   ///< log(qscale)/log[2^(1/6)]
    DWTELEM *buf;
    int buf_x_offset;
    int buf_y_offset;
    int stride_line; ///< Stride measured in lines, not pixels.
    x_and_coeff * x_coeff;
    struct SubBand *parent;
    uint8_t state[/*7*2*/ 7 + 512][32];
}SubBand;

typedef struct Plane{
    int width;
    int height;
    SubBand band[MAX_DECOMPOSITIONS][4];
}Plane;

/** Used to minimize the amount of memory used in order to optimize cache performance. **/
typedef struct {
    DWTELEM * * line; ///< For use by idwt and predict_slices.
    DWTELEM * * data_stack; ///< Used for internal purposes.
    int data_stack_top;
    int line_count;
    int line_width;
    int data_count;
    DWTELEM * base_buffer; ///< Buffer that this structure is caching.
} slice_buffer;

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)

    AVCodecContext *avctx;
    RangeCoder c;
    DSPContext dsp;
    AVFrame input_picture;
    AVFrame current_picture;
    AVFrame last_picture;
    AVFrame mconly_picture;
//     uint8_t q_context[16];
    uint8_t header_state[32];
    uint8_t block_state[128 + 32*128];
    int keyframe;
    int always_reset;
    int version;
    int spatial_decomposition_type;
    int temporal_decomposition_type;
    int spatial_decomposition_count;
    int temporal_decomposition_count;
    DWTELEM *spatial_dwt_buffer;
    int colorspace_type;
    int chroma_h_shift;
    int chroma_v_shift;
    int spatial_scalability;
    int qlog;
    int lambda;
    int lambda2;
    int mv_scale;
    int qbias;
#define QBIAS_SHIFT 3
    int b_width;
    int b_height;
    int block_max_depth;
    Plane plane[MAX_PLANES];
    BlockNode *block;
    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)
}SnowContext;

typedef struct {
    DWTELEM *b0;
    DWTELEM *b1;
    DWTELEM *b2;
    DWTELEM *b3;
    int y;
} dwt_compose_t;

#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 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);  
  
    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;

    assert(line >= 0 && line < buf->line_count);
    assert(buf->line[line]);

    offset = buf->line_width * line;
    buffer = buf->line[line];
    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);
}

static void slice_buffer_flush(slice_buffer * buf)
{
    int i;
    for (i = 0; i < buf->line_count; i++)
    {
        if (buf->line[i])
        {
//      av_log(NULL, AV_LOG_DEBUG, "slice_buffer_flush: line: %d \n", i);
            slice_buffer_release(buf, i);
        }
    }
}

static void slice_buffer_destroy(slice_buffer * buf)
{
    int i;
    slice_buffer_flush(buf);
  
    for (i = buf->data_count - 1; i >= 0; i--)
    {
        assert(buf->data_stack[i]);
        av_free(buf->data_stack[i]);
    }
    assert(buf->data_stack);
    av_free(buf->data_stack);
    assert(buf->line);
    av_free(buf->line);
}

#ifdef	__sgi
// Avoid a name clash on SGI IRIX
#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;
}

static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
    int i;

    if(v){
        const int a= ABS(v);
        const int e= av_log2(a);
#if 1
        const int el= FFMIN(e, 10);   
        put_rac(c, state+0, 0);

        for(i=0; i<el; i++){
            put_rac(c, state+1+i, 1);  //1..10
        }
        for(; i<e; i++){
            put_rac(c, state+1+9, 1);  //1..10
        }
        put_rac(c, state+1+FFMIN(i,9), 0);

        for(i=e-1; i>=el; i--){
            put_rac(c, state+22+9, (a>>i)&1); //22..31
        }
        for(; i>=0; i--){
            put_rac(c, state+22+i, (a>>i)&1); //22..31
        }

        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++){
                put_rac(c, state+1+i, 1);  //1..10
            }
            put_rac(c, state+1+i, 0);

            for(i=e-1; i>=0; i--){
                put_rac(c, state+22+i, (a>>i)&1); //22..31
            }

            if(is_signed)
                put_rac(c, state+11 + e, v < 0); //11..21
        }else{
            for(i=0; i<e; i++){
                put_rac(c, state+1+FFMIN(i,9), 1);  //1..10
            }
            put_rac(c, state+1+FFMIN(i,9), 0);

            for(i=e-1; i>=0; i--){
                put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31
            }

            if(is_signed)
                put_rac(c, state+11 + FFMIN(e,10), v < 0); //11..21
        }
#endif
    }else{
        put_rac(c, state+0, 1);
    }
}

static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
    if(get_rac(c, state+0))
        return 0;
    else{
        int i, e, a;
        e= 0;
        while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
            e++;
        }

        a= 1;
        for(i=e-1; i>=0; i--){
            a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
        }

        if(is_signed && get_rac(c, state+11 + FFMIN(e,10))) //11..21
            return -a;
        else
            return a;
    }
}

static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
    int i;
    int r= log2>=0 ? 1<<log2 : 1;

    assert(v>=0);
    assert(log2>=-4);

    while(v >= r){
        put_rac(c, state+4+log2, 1);
        v -= r;
        log2++;
        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);
    }
}

static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
    int i;
    int r= log2>=0 ? 1<<log2 : 1;
    int v=0;

    assert(log2>=-4);

    while(get_rac(c, state+4+log2)){
        v+= r;
        log2++;
        if(log2>0) r+=r;
    }
    
    for(i=log2-1; i>=0; i--){
        v+= get_rac(c, state+31-i)<<i;
    }

    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){
    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){
        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
        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);
    }
}

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){
    const int mirror_left= !highpass;
    const int mirror_right= (width&1) ^ highpass;
    const int w= (width>>1) - 1 + (highpass & width);
    int i;

    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 += 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;
        r += r>>8;
        dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse);
    }
}

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){
    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))
    if(mirror_left){
        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
        dst += dst_step;
        src += src_step;