common.h

wince 平台下的h264 压缩代码程序

    } else {
	bit_buf<<=bit_left;
        bit_buf |= value >> (n - bit_left);
#ifdef UNALIGNED_STORES_ARE_BAD
        if (3 & (int) s->buf_ptr) {
            s->buf_ptr[0] = bit_buf >> 24;
            s->buf_ptr[1] = bit_buf >> 16;
            s->buf_ptr[2] = bit_buf >>  8;
            s->buf_ptr[3] = bit_buf      ;
        } else
#endif
        *(uint32_t *)s->buf_ptr = be2me_32(bit_buf);
        //printf("bitbuf = %08x\n", bit_buf);
        s->buf_ptr+=4;
	bit_left+=32 - n;
        bit_buf = value;
    }

    s->bit_buf = bit_buf;
    s->bit_left = bit_left;
}
#endif


#ifdef ALT_BITSTREAM_WRITER
static inline void put_bits(PutBitContext *s, int n, unsigned int value)
{
#    ifdef ALIGNED_BITSTREAM_WRITER
#        ifdef ARCH_X86
    asm volatile(
	"movl %0, %%ecx			\n\t"
	"xorl %%eax, %%eax		\n\t"
	"shrdl %%cl, %1, %%eax		\n\t"
	"shrl %%cl, %1			\n\t"
	"movl %0, %%ecx			\n\t"
	"shrl $3, %%ecx			\n\t"
	"andl $0xFFFFFFFC, %%ecx	\n\t"
	"bswapl %1			\n\t"
	"orl %1, (%2, %%ecx)		\n\t"
	"bswapl %%eax			\n\t"
	"addl %3, %0			\n\t"
	"movl %%eax, 4(%2, %%ecx)	\n\t"
	: "=&r" (s->index), "=&r" (value)
	: "r" (s->buf), "r" (n), "0" (s->index), "1" (value<<(-n))
	: "%eax", "%ecx"
    );
#        else
    int index= s->index;
    uint32_t *ptr= ((uint32_t *)s->buf)+(index>>5);
    
    value<<= 32-n; 
    
    ptr[0] |= be2me_32(value>>(index&31));
    ptr[1]  = be2me_32(value<<(32-(index&31)));
//if(n>24) printf("%d %d\n", n, value);
    index+= n;
    s->index= index;
#        endif
#    else //ALIGNED_BITSTREAM_WRITER
#        ifdef ARCH_X86
    asm volatile(
	"movl $7, %%ecx			\n\t"
	"andl %0, %%ecx			\n\t"
	"addl %3, %%ecx			\n\t"
	"negl %%ecx			\n\t"
	"shll %%cl, %1			\n\t"
	"bswapl %1			\n\t"
	"movl %0, %%ecx			\n\t"
	"shrl $3, %%ecx			\n\t"
	"orl %1, (%%ecx, %2)		\n\t"
	"addl %3, %0			\n\t"
	"movl $0, 4(%%ecx, %2)		\n\t"
	: "=&r" (s->index), "=&r" (value)
	: "r" (s->buf), "r" (n), "0" (s->index), "1" (value)
	: "%ecx"
    );
#        else
    int index= s->index;
    uint32_t *ptr= (uint32_t*)(((uint8_t *)s->buf)+(index>>3));
    
    ptr[0] |= be2me_32(value<<(32-n-(index&7) ));
    ptr[1] = 0;
//if(n>24) printf("%d %d\n", n, value);
    index+= n;
    s->index= index;
#        endif
#    endif //!ALIGNED_BITSTREAM_WRITER
}
#endif


static inline uint8_t* pbBufPtr(PutBitContext *s)
{
#ifdef ALT_BITSTREAM_WRITER
	return s->buf + (s->index>>3);
#else
	return s->buf_ptr;
#endif
}

/**
 *
 * PutBitContext must be flushed & aligned to a byte boundary before calling this.
 */
static inline void skip_put_bytes(PutBitContext *s, int n){
#ifndef WINCE //assert    
	assert((put_bits_count(s)&7)==0);
#endif
#ifdef ALT_BITSTREAM_WRITER
//FIXME may need some cleaning of the buffer
	s->index += n<<3;
#else
#ifndef WINCE //assert
	assert(s->bit_left==32);
#endif
	s->buf_ptr += n;
#endif    
}

/**
 * Changes the end of the buffer.
 */
static inline void set_put_bits_buffer_size(PutBitContext *s, int size){
    s->buf_end= s->buf + size;
}

/* Bitstream reader API docs:
name
    abritary name which is used as prefix for the internal variables

gb
    getbitcontext

OPEN_READER(name, gb)
    loads gb into local variables

CLOSE_READER(name, gb)
    stores local vars in gb

UPDATE_CACHE(name, gb)
    refills the internal cache from the bitstream
    after this call at least MIN_CACHE_BITS will be available,

GET_CACHE(name, gb)
    will output the contents of the internal cache, next bit is MSB of 32 or 64 bit (FIXME 64bit)

SHOW_UBITS(name, gb, num)
    will return the nest num bits

SHOW_SBITS(name, gb, num)
    will return the nest num bits and do sign extension

SKIP_BITS(name, gb, num)
    will skip over the next num bits
    note, this is equinvalent to SKIP_CACHE; SKIP_COUNTER

SKIP_CACHE(name, gb, num)
    will remove the next num bits from the cache (note SKIP_COUNTER MUST be called before UPDATE_CACHE / CLOSE_READER)

SKIP_COUNTER(name, gb, num)
    will increment the internal bit counter (see SKIP_CACHE & SKIP_BITS)

LAST_SKIP_CACHE(name, gb, num)
    will remove the next num bits from the cache if it is needed for UPDATE_CACHE otherwise it will do nothing

LAST_SKIP_BITS(name, gb, num)
    is equinvalent to SKIP_LAST_CACHE; SKIP_COUNTER

for examples see get_bits, show_bits, skip_bits, get_vlc
*/

static inline int unaligned32_be(const void *v)
{
#ifdef CONFIG_ALIGN
	const uint8_t *p=(uint8_t *)v;
	return (((p[0]<<8) | p[1])<<16) | (p[2]<<8) | (p[3]);
#else
	return be2me_32( unaligned32(v)); //original
#endif
}

#ifdef ALT_BITSTREAM_READER
#   define MIN_CACHE_BITS 25

#   define OPEN_READER(name, gb)\
        int name##_index= (gb)->index;\
        int name##_cache= 0;\

#   define CLOSE_READER(name, gb)\
        (gb)->index= name##_index;\

#   define UPDATE_CACHE(name, gb)\
        name##_cache= unaligned32_be( ((uint8_t *)(gb)->buffer)+(name##_index>>3) ) << (name##_index&0x07);\

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

// FIXME name?
#   define SKIP_COUNTER(name, gb, num)\
        name##_index += (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_COUNTER(name, gb, num)
#   define LAST_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->index;
}
#elif defined LIBMPEG2_BITSTREAM_READER
//libmpeg2 like reader

#   define MIN_CACHE_BITS 17

#   define OPEN_READER(name, gb)\
        int name##_bit_count=(gb)->bit_count;\
        int name##_cache= (gb)->cache;\
        uint8_t * name##_buffer_ptr=(gb)->buffer_ptr;\

#   define CLOSE_READER(name, gb)\
        (gb)->bit_count= name##_bit_count;\
        (gb)->cache= name##_cache;\
        (gb)->buffer_ptr= name##_buffer_ptr;\

#ifdef LIBMPEG2_BITSTREAM_READER_HACK

#   define UPDATE_CACHE(name, gb)\
    if(name##_bit_count >= 0){\
        name##_cache+= (int)be2me_16(*(uint16_t*)name##_buffer_ptr) << name##_bit_count;\
        ((uint16_t*)name##_buffer_ptr)++;\
        name##_bit_count-= 16;\
    }\

#else

#   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);
}

/**
 * init GetBitContext.
 * @param buffer bitstream buffer, must be FF_INPUT_BUFFER_PADDING_SIZE bytes larger then the actual read bits
 * because some optimized bitstream readers read 32 or 64 bit at once and could read over the end