common.h

使用stl技术,(还没看,是听说的)

#   define UPDATE_CACHE(name, gb)\
    if(name##_bit_count >= 0){\
        name##_cache+= ((name##_buffer_ptr[0]<<8) + name##_buffer_ptr[1]) << name##_bit_count;\
        name##_buffer_ptr+=2;\
        name##_bit_count-= 16;\
    }\

#endif

#   define SKIP_CACHE(name, gb, num)\
        name##_cache <<= (num);\

#   define SKIP_COUNTER(name, gb, num)\
        name##_bit_count += (num);\

#   define SKIP_BITS(name, gb, num)\
        {\
            SKIP_CACHE(name, gb, num)\
            SKIP_COUNTER(name, gb, num)\
        }\

#   define LAST_SKIP_BITS(name, gb, num) SKIP_BITS(name, gb, num)
#   define LAST_SKIP_CACHE(name, gb, num) SKIP_CACHE(name, gb, num)

#   define SHOW_UBITS(name, gb, num)\
        NEG_USR32(name##_cache, num)

#   define SHOW_SBITS(name, gb, num)\
        NEG_SSR32(name##_cache, num)

#   define GET_CACHE(name, gb)\
        ((uint32_t)name##_cache)

static inline int get_bits_count(GetBitContext *s){
    return (s->buffer_ptr - s->buffer)*8 - 16 + s->bit_count;
}

#elif defined A32_BITSTREAM_READER

#   define MIN_CACHE_BITS 32

#   define OPEN_READER(name, gb)\
        int name##_bit_count=(gb)->bit_count;\
        uint32_t name##_cache0= (gb)->cache0;\
        uint32_t name##_cache1= (gb)->cache1;\
        uint32_t * name##_buffer_ptr=(gb)->buffer_ptr;\

#   define CLOSE_READER(name, gb)\
        (gb)->bit_count= name##_bit_count;\
        (gb)->cache0= name##_cache0;\
        (gb)->cache1= name##_cache1;\
        (gb)->buffer_ptr= name##_buffer_ptr;\

#   define UPDATE_CACHE(name, gb)\
    if(name##_bit_count > 0){\
        const uint32_t next= be2me_32( *name##_buffer_ptr );\
        name##_cache0 |= NEG_USR32(next,name##_bit_count);\
        name##_cache1 |= next<<name##_bit_count;\
        name##_buffer_ptr++;\
        name##_bit_count-= 32;\
    }\

#ifdef ARCH_X86
#   define SKIP_CACHE(name, gb, num)\
        asm(\
            "shldl %2, %1, %0		\n\t"\
            "shll %2, %1		\n\t"\
            : "+r" (name##_cache0), "+r" (name##_cache1)\
            : "Ic" ((uint8_t)num)\
           );
#else
#   define SKIP_CACHE(name, gb, num)\
        name##_cache0 <<= (num);\
        name##_cache0 |= NEG_USR32(name##_cache1,num);\
        name##_cache1 <<= (num);
#endif

#   define SKIP_COUNTER(name, gb, num)\
        name##_bit_count += (num);\

#   define SKIP_BITS(name, gb, num)\
        {\
            SKIP_CACHE(name, gb, num)\
            SKIP_COUNTER(name, gb, num)\
        }\

#   define LAST_SKIP_BITS(name, gb, num) SKIP_BITS(name, gb, num)
#   define LAST_SKIP_CACHE(name, gb, num) SKIP_CACHE(name, gb, num)

#   define SHOW_UBITS(name, gb, num)\
        NEG_USR32(name##_cache0, num)

#   define SHOW_SBITS(name, gb, num)\
        NEG_SSR32(name##_cache0, num)

#   define GET_CACHE(name, gb)\
        (name##_cache0)

static inline int get_bits_count(GetBitContext *s){
    return ((uint8_t*)s->buffer_ptr - s->buffer)*8 - 32 + s->bit_count;
}

#endif

/**
 * read mpeg1 dc style vlc (sign bit + mantisse with no MSB).
 * if MSB not set it is negative 
 * @param n length in bits
 * @author BERO  
 */
static inline int get_xbits(GetBitContext *s, int n){
    register int tmp;
    register int32_t cache;
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)
    cache = GET_CACHE(re,s);
    if ((int32_t)cache<0) { //MSB=1
        tmp = NEG_USR32(cache,n);
    } else {
    //   tmp = (-1<<n) | NEG_USR32(cache,n) + 1; mpeg12.c algo
    //   tmp = - (NEG_USR32(cache,n) ^ ((1 << n) - 1)); h263.c algo
        tmp = - NEG_USR32(~cache,n);
    }
    LAST_SKIP_BITS(re, s, n)
    CLOSE_READER(re, s)
    return tmp;
}

static inline int get_sbits(GetBitContext *s, int n){
    register int tmp;
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)
    tmp= SHOW_SBITS(re, s, n);
    LAST_SKIP_BITS(re, s, n)
    CLOSE_READER(re, s)
    return tmp;
}

/**
 * reads 0-17 bits.
 * Note, the alt bitstream reader can read upto 25 bits, but the libmpeg2 reader cant
 */
static inline unsigned int get_bits(GetBitContext *s, int n){
    register int tmp;
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)
    tmp= SHOW_UBITS(re, s, n);
    LAST_SKIP_BITS(re, s, n)
    CLOSE_READER(re, s)
    return tmp;
}

unsigned int get_bits_long(GetBitContext *s, int n);

/**
 * shows 0-17 bits.
 * Note, the alt bitstream reader can read upto 25 bits, but the libmpeg2 reader cant
 */
static inline unsigned int show_bits(GetBitContext *s, int n){
    register int tmp;
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)
    tmp= SHOW_UBITS(re, s, n);
//    CLOSE_READER(re, s)
    return tmp;
}

unsigned int show_bits_long(GetBitContext *s, int n);

static inline void skip_bits(GetBitContext *s, int n){
 //Note gcc seems to optimize this to s->index+=n for the ALT_READER :))
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)
    LAST_SKIP_BITS(re, s, n)
    CLOSE_READER(re, s)
}

static inline unsigned int get_bits1(GetBitContext *s){
#ifdef ALT_BITSTREAM_READER
    int index= s->index;
    uint8_t result= s->buffer[ index>>3 ];
    result<<= (index&0x07);
    result>>= 8 - 1;
    index++;
    s->index= index;

    return result;
#else
    return get_bits(s, 1);
#endif
}

static inline unsigned int show_bits1(GetBitContext *s){
    return show_bits(s, 1);
}

static inline void skip_bits1(GetBitContext *s){
    skip_bits(s, 1);
}

void init_get_bits(GetBitContext *s,
                   const uint8_t *buffer, int buffer_size);

int check_marker(GetBitContext *s, const char *msg);
void align_get_bits(GetBitContext *s);
int init_vlc(VLC *vlc, int nb_bits, int nb_codes,
             const void *bits, int bits_wrap, int bits_size,
             const void *codes, int codes_wrap, int codes_size);
void free_vlc(VLC *vlc);

/**
 *
 * if the vlc code is invalid and max_depth=1 than no bits will be removed
 * if the vlc code is invalid and max_depth>1 than the number of bits removed
 * is undefined
 */
#define GET_VLC(code, name, gb, table, bits, max_depth)\
{\
    int n, index, nb_bits;\
\
    index= SHOW_UBITS(name, gb, bits);\
    code = table[index][0];\
    n    = table[index][1];\
\
    if(max_depth > 1 && n < 0){\
        LAST_SKIP_BITS(name, gb, bits)\
        UPDATE_CACHE(name, gb)\
\
        nb_bits = -n;\
\
        index= SHOW_UBITS(name, gb, nb_bits) + code;\
        code = table[index][0];\
        n    = table[index][1];\
        if(max_depth > 2 && n < 0){\
            LAST_SKIP_BITS(name, gb, nb_bits)\
            UPDATE_CACHE(name, gb)\
\
            nb_bits = -n;\
\
            index= SHOW_UBITS(name, gb, nb_bits) + code;\
            code = table[index][0];\
            n    = table[index][1];\
        }\
    }\
    SKIP_BITS(name, gb, n)\
}

#define GET_RL_VLC(level, run, name, gb, table, bits, max_depth)\
{\
    int n, index, nb_bits;\
\
    index= SHOW_UBITS(name, gb, bits);\
    level = table[index].level;\
    n     = table[index].len;\
\
    if(max_depth > 1 && n < 0){\
        LAST_SKIP_BITS(name, gb, bits)\
        UPDATE_CACHE(name, gb)\
\
        nb_bits = -n;\
\
        index= SHOW_UBITS(name, gb, nb_bits) + level;\
        level = table[index].level;\
        n     = table[index].len;\
    }\
    run= table[index].run;\
    SKIP_BITS(name, gb, n)\
}

// deprecated, dont use get_vlc for new code, use get_vlc2 instead or use GET_VLC directly
static inline int get_vlc(GetBitContext *s, VLC *vlc)
{
    int code;
    VLC_TYPE (*table)[2]= vlc->table;
    
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)

    GET_VLC(code, re, s, table, vlc->bits, 3)    

    CLOSE_READER(re, s)
    return code;
}

/**
 * parses a vlc code, faster then get_vlc()
 * @param bits is the number of bits which will be read at once, must be 
 *             identical to nb_bits in init_vlc()
 * @param max_depth is the number of times bits bits must be readed to completly
 *                  read the longest vlc code 
 *                  = (max_vlc_length + bits - 1) / bits
 */
static always_inline int get_vlc2(GetBitContext *s, VLC_TYPE (*table)[2],
                                  int bits, int max_depth)
{
    int code;
    
    OPEN_READER(re, s)
    UPDATE_CACHE(re, s)

    GET_VLC(code, re, s, table, bits, max_depth)

    CLOSE_READER(re, s)
    return code;
}

//#define TRACE

#ifdef TRACE

static inline void print_bin(int bits, int n){
    int i;
    
    for(i=n-1; i>=0; i--){
        printf("%d", (bits>>i)&1);
    }
    for(i=n; i<24; i++)
        printf(" ");
}

static inline int get_bits_trace(GetBitContext *s, int n, char *file, char *func, int line){
    int r= get_bits(s, n);
    
    print_bin(r, n);
    printf("%5d %2d %3d bit @%5d in %s %s:%d\n", r, n, r, get_bits_count(s)-n, file, func, line);
    return r;
}
static inline int get_vlc_trace(GetBitContext *s, VLC_TYPE (*table)[2], int bits, int max_depth, char *file, char *func, int line){
    int show= show_bits(s, 24);
    int pos= get_bits_count(s);
    int r= get_vlc2(s, table, bits, max_depth);
    int len= get_bits_count(s) - pos;
    int bits2= show>>(24-len);
    
    print_bin(bits2, len);
    
    printf("%5d %2d %3d vlc @%5d in %s %s:%d\n", bits2, len, r, pos, file, func, line);
    return r;
}
static inline int get_xbits_trace(GetBitContext *s, int n, char *file, char *func, int line){
    int show= show_bits(s, n);
    int r= get_xbits(s, n);
    
    print_bin(show, n);
    printf("%5d %2d %3d xbt @%5d in %s %s:%d\n", show, n, r, get_bits_count(s)-n, file, func, line);
    return r;
}

#define get_bits(s, n)  get_bits_trace(s, n, __FILE__, __PRETTY_FUNCTION__, __LINE__)
#define get_bits1(s)    get_bits_trace(s, 1, __FILE__, __PRETTY_FUNCTION__, __LINE__)
#define get_xbits(s, n) get_xbits_trace(s, n, __FILE__, __PRETTY_FUNCTION__, __LINE__)
#define get_vlc(s, vlc)            get_vlc_trace(s, (vlc)->table, (vlc)->bits, 3, __FILE__, __PRETTY_FUNCTION__, __LINE__)
#define get_vlc2(s, tab, bits, max) get_vlc_trace(s, tab, bits, max, __FILE__, __PRETTY_FUNCTION__, __LINE__)

#define tprintf printf

#else //TRACE
#define tprintf(_arg...) {}
#endif

/* define it to include statistics code (useful only for optimizing
   codec efficiency */
//#define STATS

#ifdef STATS

enum {
    ST_UNKNOWN,
    ST_DC,
    ST_INTRA_AC,
    ST_INTER_AC,
    ST_INTRA_MB,
    ST_INTER_MB,
    ST_MV,
    ST_NB,
};

extern int st_current_index;
extern unsigned int st_bit_counts[ST_NB];
extern unsigned int st_out_bit_counts[ST_NB];

void print_stats(void);
#endif

/* misc math functions */
extern const uint8_t ff_log2_tab[256];

static inline int av_log2(unsigned int v)
{
    int n;

    n = 0;
    if (v & 0xffff0000) {
        v >>= 16;
        n += 16;
    }
    if (v & 0xff00) {
        v >>= 8;
        n += 8;
    }
    n += ff_log2_tab[v];

    return n;
}

static inline int av_log2_16bit(unsigned int v)
{
    int n;

    n = 0;
    if (v & 0xff00) {
        v >>= 8;
        n += 8;
    }
    n += ff_log2_tab[v];

    return n;
}


/* median of 3 */
static inline int mid_pred(int a, int b, int c)
{
    int vmin, vmax;
    vmax = vmin = a;
    if (b < vmin)
        vmin = b;
    else
	vmax = b;

    if (c < vmin)
        vmin = c;
    else if (c > vmax)
        vmax = c;

    return a + b + c - vmin - vmax;
}

static inline int clip(int a, int amin, int amax)
{
    if (a < amin)
        return amin;
    else if (a > amax)
        return amax;
    else
        return a;
}

/* math */
extern const uint8_t ff_sqrt_tab[128];

int64_t ff_gcd(int64_t a, int64_t b);

static inline int ff_sqrt(int a)
{
    int ret=0;
    int s;
    int ret_sq=0;
    
    if(a<128) return ff_sqrt_tab[a];
    
    for(s=15; s>=0; s--){
        int b= ret_sq + (1<<(s*2)) + (ret<<s)*2;
        if(b<=a){
            ret_sq=b;
            ret+= 1<<s;
        }
    }
    return ret;
}

/**
 * converts fourcc string to int
 */
static inline int ff_get_fourcc(const char *s){
    assert( strlen(s)==4 );

    return (s[0]) + (s[1]<<8) + (s[2]<<16) + (s[3]<<24);
}

#define MKTAG(a,b,c,d) (a | (b << 8) | (c << 16) | (d << 24))
#define MKBETAG(a,b,c,d) (d | (c << 8) | (b << 16) | (a << 24))


void ff_float2fraction(int *nom_arg, int *denom_arg, double f, int max);


#ifdef ARCH_X86
#define MASK_ABS(mask, level)\
            asm volatile(\
		"cdq			\n\t"\
		"xorl %1, %0		\n\t"\
		"subl %1, %0		\n\t"\
		: "+a" (level), "=&d" (mask)\
	    );
#else
#define MASK_ABS(mask, level)\
            mask= level>>31;\
            level= (level^mask)-mask;
#endif


#if __CPU__ >= 686 && !defined(RUNTIME_CPUDETECT)
#define COPY3_IF_LT(x,y,a,b,c,d)\
asm volatile (\
    "cmpl %0, %3	\n\t"\
    "cmovl %3, %0	\n\t"\
    "cmovl %4, %1	\n\t"\
    "cmovl %5, %2	\n\t"\
    : "+r" (x), "+r" (a), "+r" (c)\
    : "r" (y), "r" (b), "r" (d)\
);
#else
#define COPY3_IF_LT(x,y,a,b,c,d)\
if((y)<(x)){\
     (x)=(y);\
     (a)=(b);\
     (c)=(d);\
}
#endif

#ifdef ARCH_X86
static inline long long rdtsc()
{
	long long l;
	asm volatile(	"rdtsc\n\t"
		: "=A" (l)
	);
	return l;
}

#define START_TIMER \
static uint64_t tsum=0;\
static int tcount=0;\
static int tskip_count=0;\
uint64_t tend;\
uint64_t tstart= rdtsc();\

#define STOP_TIMER(id) \
tend= rdtsc();\
if(tcount<2 || tend - tstart < 4*tsum/tcount){\
    tsum+= tend - tstart;\
    tcount++;\
}else\
    tskip_count++;\
if(256*256*256*64%(tcount+tskip_count)==0){\
    fprintf(stderr, "%Ld dezicycles in %s, %d runs, %d skips\n", tsum*10/tcount, id, tcount, tskip_count);\
}
#endif

#define CLAMP_TO_8BIT(d) ((d > 0xff) ? 0xff : (d < 0) ? 0 : d)

/* avoid usage of various functions */
#define malloc please_use_av_malloc
#define free please_use_av_free
#define realloc please_use_av_realloc

#define CHECKED_ALLOCZ(p, size)\
{\
    p= av_mallocz(size);\
    if(p==NULL && (size)!=0){\
        perror("malloc");\
        goto fail;\
    }\
}

#endif /* HAVE_AV_CONFIG_H */

#endif /* COMMON_H */