comparison cabac.h @ 3993:8b7c59b7af01 libavcodec

make state transition tables global as they are constant and the code is slightly faster that way

3992:a09b5b667229

    int range;
    int outstanding_count;
#ifdef STRICT_LIMITS
    int symCount;
#endif
    uint8_t lps_state[2*64];      ///< transIdxLPS
    uint8_t mps_state[2*64];      ///< transIdxMPS
    const uint8_t *bytestream_start;
    const uint8_t *bytestream;
    const uint8_t *bytestream_end;
    PutBitContext pb;
}CABACContext;

extern uint8_t ff_h264_lps_range[2*65][4];  ///< rangeTabLPS
extern const uint8_t ff_h264_mps_state[64];
extern const uint8_t ff_h264_lps_state[64];
extern const uint8_t ff_h264_norm_shift[128];


void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
void ff_init_cabac_decoder(CABACContext *c, const uint8_t *buf, int buf_size);
void ff_init_cabac_states(CABACContext *c,
                          uint8_t const *mps_state, uint8_t const *lps_state, int state_count);


static inline void put_cabac_bit(CABACContext *c, int b){
    put_bits(&c->pb, 1, b);
    for(;c->outstanding_count; c->outstanding_count--){

static void put_cabac(CABACContext *c, uint8_t * const state, int bit){
    int RangeLPS= ff_h264_lps_range[*state][c->range>>6];

    if(bit == ((*state)&1)){
        c->range -= RangeLPS;
        *state= c->mps_state[*state];
    }else{
        c->low += c->range - RangeLPS;
        c->range = RangeLPS;
        *state= c->lps_state[*state];
    }

    renorm_cabac_encoder(c);

#ifdef STRICT_LIMITS

#ifdef ARCH_X86
    int bit;

#define LOW          "0"
#define RANGE        "4"
#define LPS_STATE   "12"
#define MPS_STATE   "12+2*64"
#define BYTESTART   "12+4*64"
#define BYTE        "16+4*64"
#define BYTEEND     "20+4*64"
#ifndef BRANCHLESS_CABAC_DECODER
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"

        " ja 1f                                 \n\t"
        "cmp $0x2000000, %%edx                  \n\t" //FIXME avoidable
        "setb %%cl                              \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
        "movzbl "MPS_STATE"(%2, %%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "test %%bx, %%bx                        \n\t"
        " jnz 2f                                \n\t"
        "movl "BYTE     "(%2), %%esi            \n\t"

        "movl %%esi, %%edx                      \n\t"
        "shr $19, %%esi                         \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx   \n\t"
        "shll %%cl, %%ebx                       \n\t"
        "shll %%cl, %%edx                       \n\t"
        "movzbl "LPS_STATE"(%2, %%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"
        "addl $1, %%eax                         \n\t"
        "test %%bx, %%bx                        \n\t"
        " jnz 2f                                \n\t"

        "subl %%ecx, %%ebx                      \n\t"
        "xorl %%edx, %%eax                      \n\t"
#endif

//eax:state ebx:low edx:mask esi:range
        "movzbl "MPS_STATE"(%2, %%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"

        "movl %%esi, %%edx                      \n\t"
//eax:bit ebx:low edx:range esi:range

    c->range -= RangeLPS;
#ifndef BRANCHLESS_CABAC_DECODER
    if(c->low < c->range){
        bit= s&1;
        *state= c->mps_state[s];
        renorm_cabac_decoder_once(c);
    }else{
        bit= ff_h264_norm_shift[RangeLPS>>19];
        c->low -= c->range;
        *state= c->lps_state[s];
        c->range = RangeLPS<<bit;
        c->low <<= bit;
        bit= (s&1)^1;

        if(!(c->low & 0xFFFF)){

    c->low -= c->range & lps_mask;
    c->range += (RangeLPS - c->range) & lps_mask;

    s^=lps_mask;
    *state= c->mps_state[s];
    bit= s&1;

    lps_mask= ff_h264_norm_shift[c->range>>(CABAC_BITS+3)];
    c->range<<= lps_mask;
    c->low  <<= lps_mask;

3993:8b7c59b7af01

    int range;
    int outstanding_count;
#ifdef STRICT_LIMITS
    int symCount;
#endif
    const uint8_t *bytestream_start;
    const uint8_t *bytestream;
    const uint8_t *bytestream_end;
    PutBitContext pb;
}CABACContext;

extern uint8_t ff_h264_lps_range[2*65][4];  ///< rangeTabLPS
extern uint8_t ff_h264_mps_state[2*64];     ///< transIdxMPS
extern uint8_t ff_h264_lps_state[2*64];     ///< transIdxLPS
extern const uint8_t ff_h264_norm_shift[128];


void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
void ff_init_cabac_decoder(CABACContext *c, const uint8_t *buf, int buf_size);
void ff_init_cabac_states(CABACContext *c);


static inline void put_cabac_bit(CABACContext *c, int b){
    put_bits(&c->pb, 1, b);
    for(;c->outstanding_count; c->outstanding_count--){

static void put_cabac(CABACContext *c, uint8_t * const state, int bit){
    int RangeLPS= ff_h264_lps_range[*state][c->range>>6];

    if(bit == ((*state)&1)){
        c->range -= RangeLPS;
        *state= ff_h264_mps_state[*state];
    }else{
        c->low += c->range - RangeLPS;
        c->range = RangeLPS;
        *state= ff_h264_lps_state[*state];
    }

    renorm_cabac_encoder(c);

#ifdef STRICT_LIMITS

#ifdef ARCH_X86
    int bit;

#define LOW          "0"
#define RANGE        "4"
#define BYTESTART   "12"
#define BYTE        "16"
#define BYTEEND     "20"
#ifndef BRANCHLESS_CABAC_DECODER
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"

        " ja 1f                                 \n\t"
        "cmp $0x2000000, %%edx                  \n\t" //FIXME avoidable
        "setb %%cl                              \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
        "movzbl "MANGLE(ff_h264_mps_state)"(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "test %%bx, %%bx                        \n\t"
        " jnz 2f                                \n\t"
        "movl "BYTE     "(%2), %%esi            \n\t"

        "movl %%esi, %%edx                      \n\t"
        "shr $19, %%esi                         \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx   \n\t"
        "shll %%cl, %%ebx                       \n\t"
        "shll %%cl, %%edx                       \n\t"
        "movzbl "MANGLE(ff_h264_lps_state)"(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"
        "addl $1, %%eax                         \n\t"
        "test %%bx, %%bx                        \n\t"
        " jnz 2f                                \n\t"

        "subl %%ecx, %%ebx                      \n\t"
        "xorl %%edx, %%eax                      \n\t"
#endif

//eax:state ebx:low edx:mask esi:range
        "movzbl "MANGLE(ff_h264_mps_state)"(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"

        "movl %%esi, %%edx                      \n\t"
//eax:bit ebx:low edx:range esi:range

    c->range -= RangeLPS;
#ifndef BRANCHLESS_CABAC_DECODER
    if(c->low < c->range){
        bit= s&1;
        *state= ff_h264_mps_state[s];
        renorm_cabac_decoder_once(c);
    }else{
        bit= ff_h264_norm_shift[RangeLPS>>19];
        c->low -= c->range;
        *state= ff_h264_lps_state[s];
        c->range = RangeLPS<<bit;
        c->low <<= bit;
        bit= (s&1)^1;

        if(!(c->low & 0xFFFF)){

    c->low -= c->range & lps_mask;
    c->range += (RangeLPS - c->range) & lps_mask;

    s^=lps_mask;
    *state= ff_h264_mps_state[s];
    bit= s&1;

    lps_mask= ff_h264_norm_shift[c->range>>(CABAC_BITS+3)];
    c->range<<= lps_mask;
    c->low  <<= lps_mask;