snow.c

从FFMPEG转换而来的H264解码程序,VC下编译..

#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)))

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 offset;
    DWTELEM * buffer;

    assert(buf->data_stack_top >= 0);

    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;

    return buffer;
}

static void slice_buffer_release(slice_buffer * buf, int line)
{
    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;
}

static void slice_buffer_flush(slice_buffer * buf)
{
    int i;
    for (i = 0; i < buf->line_count; i++)
    {
        if (buf->line[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_freep(&buf->data_stack[i]);
    }
    assert(buf->data_stack);
    av_freep(&buf->data_stack);
    assert(buf->line);
    av_freep(&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){
    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){
    int i;

    if(v){
        const int a= FFABS(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 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);
    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);
    }
}

#ifndef lift5
static av_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);
    }
}
#endif

#ifndef liftS
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))
    if(mirror_left){
        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
        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);
    }
}
#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


#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

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];