comparison cabac.h @ 4024:d550343b5dac libavcodec

shift CABACContext.range right, this reduces the number of shifts needed in get_cabac() and is slightly faster on P3 (and should be much faster on P4 as the P4 except the more recent variants lacks an integer shifter and so shifts have ~10 times longer latency then simple operations like adds)

4023:508f089d0b28

    const uint8_t *bytestream_end;
    PutBitContext pb;
}CABACContext;

extern uint8_t ff_h264_mlps_state[4*64];
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);

        c->low += c->low;
    }
}

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{

static void refill2(CABACContext *c){
    int i, x;

    x= c->low ^ (c->low-1);
    i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS+1)];

    x= -CABAC_MASK;

#if CABAC_BITS == 16
        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);

    c->low += x<<i;
    c->bytestream+= CABAC_BITS/8;
}

static inline void renorm_cabac_decoder(CABACContext *c){
    while(c->range < (0x200 << CABAC_BITS)){
        c->range+= c->range;
        c->low+= c->low;
        if(!(c->low & CABAC_MASK))
            refill(c);
    }

#ifdef ARCH_X86_DISABLED
    int temp;
#if 0
    //P3:683    athlon:475
    asm(
        "lea -0x2000000(%0), %2     \n\t"
        "shr $31, %2                \n\t"  //FIXME 31->63 for x86-64
        "shl %%cl, %0               \n\t"
        "shl %%cl, %1               \n\t"
        : "+r"(c->range), "+r"(c->low), "+c"(temp)
    );
#elif 0
    //P3:680    athlon:474
    asm(
        "cmp $0x2000000, %0         \n\t"
        "setb %%cl                  \n\t"  //FIXME 31->63 for x86-64
        "shl %%cl, %0               \n\t"
        "shl %%cl, %1               \n\t"
        : "+r"(c->range), "+r"(c->low), "+c"(temp)
    );
#elif 1
    int temp2;
    //P3:665    athlon:517
    asm(
        "lea -0x2000000(%0), %%eax  \n\t"
        "cdq                        \n\t"
        "mov %0, %%eax              \n\t"
        "and %%edx, %0              \n\t"
        "and %1, %%edx              \n\t"
        "add %%eax, %0              \n\t"

    );
#elif 0
    int temp2;
    //P3:673    athlon:509
    asm(
        "cmp $0x2000000, %0         \n\t"
        "sbb %%edx, %%edx           \n\t"
        "mov %0, %%eax              \n\t"
        "and %%edx, %0              \n\t"
        "and %1, %%edx              \n\t"
        "add %%eax, %0              \n\t"

    );
#else
    int temp2;
    //P3:677    athlon:511
    asm(
        "cmp $0x2000000, %0         \n\t"
        "lea (%0, %0), %%eax        \n\t"
        "lea (%1, %1), %%edx        \n\t"
        "cmovb %%eax, %0            \n\t"
        "cmovb %%edx, %1            \n\t"
        : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
    );
#endif
#else
    //P3:675    athlon:476
    int shift= (uint32_t)(c->range - (0x200 << CABAC_BITS))>>31;
    c->range<<= shift;
    c->low  <<= shift;
#endif
    if(!(c->low & CABAC_MASK))
        refill(c);

#ifndef BRANCHLESS_CABAC_DECODER
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"
        "shrl $23, %%ebx                        \n\t"
        "movzbl "MANGLE(ff_h264_lps_range)"(%%ebx, %%eax, 4), %%esi\n\t"
        "shll $17, %%esi                        \n\t"
        "movl "LOW      "(%2), %%ebx            \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%esi, %%edx                      \n\t"
        "cmpl %%edx, %%ebx                      \n\t"
        " ja 1f                                 \n\t"

#if 1
        //athlon:4067 P3:4110
        "lea -0x2000000(%%edx), %%ecx           \n\t"
        "shr $31, %%ecx                         \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
#else
        //athlon:4057 P3:4130
        "cmp $0x2000000, %%edx                  \n\t" //FIXME avoidable
        "setb %%cl                              \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
#endif
        "movzbl "MANGLE(ff_h264_mps_state)"(%%eax), %%ecx   \n\t"

        "addl %%ecx, %%ebx                      \n\t"
        "movl %%esi, "BYTE    "(%2)             \n\t"
        "jmp 2f                                 \n\t"
        "1:                                     \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%edx, %%ebx                      \n\t"
        "movl %%esi, %%edx                      \n\t"
        "shr $19, %%esi                         \n\t"
        "movzbl "MANGLE(ff_h264_lps_state)"(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx   \n\t"
        "shll %%cl, %%ebx                       \n\t"
        "shll %%cl, %%edx                       \n\t"
        "addl $1, %%eax                         \n\t"
        "test %%bx, %%bx                        \n\t"
        " jnz 2f                                \n\t"

        "movl "BYTE     "(%2), %%ecx            \n\t"

        "addl $2, %%ecx                         \n\t"
        "movl %%ecx, "BYTE    "(%2)             \n\t"

        "leal -1(%%ebx), %%ecx                  \n\t"
        "xorl %%ebx, %%ecx                      \n\t"
        "shrl $17, %%ecx                        \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx   \n\t"
        "neg %%ecx                              \n\t"
        "add $7, %%ecx                          \n\t"

        "shll %%cl , %%esi                      \n\t"

#else /* BRANCHLESS_CABAC_DECODER */
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"
        "shrl $23, %%ebx                        \n\t"
        "movzbl "MANGLE(ff_h264_lps_range)"(%%ebx, %%eax, 4), %%esi\n\t"
        "shll $17, %%esi                        \n\t"
        "movl "LOW      "(%2), %%ebx            \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%esi, %%edx                      \n\t"
#ifdef CMOV_IS_FAST //FIXME actually define this somewhere
        "cmpl %%ebx, %%edx                      \n\t"
        "cmova %%edx, %%esi                     \n\t"
        "sbbl %%ecx, %%ecx                      \n\t"
        "andl %%ecx, %%edx                      \n\t"
        "subl %%edx, %%ebx                      \n\t"
        "xorl %%ecx, %%eax                      \n\t"
#else /* CMOV_IS_FAST */
        "movl %%edx, %%ecx                      \n\t"
        "subl %%ebx, %%edx                      \n\t"
        "sarl $31, %%edx                        \n\t" //lps_mask
        "subl %%ecx, %%esi                      \n\t" //RangeLPS - range
        "andl %%edx, %%esi                      \n\t" //(RangeLPS - range)&lps_mask

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

//eax:state ebx:low edx:mask esi:range
        "movzbl "MANGLE(ff_h264_mlps_state)"+128(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"

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

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

        "addl $2, %%ecx                         \n\t"
        "movl %%ecx, "BYTE    "(%2)             \n\t"

        "leal -1(%%ebx), %%ecx                  \n\t"
        "xorl %%ebx, %%ecx                      \n\t"
        "shrl $17, %%ecx                        \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx   \n\t"
        "neg %%ecx                              \n\t"
        "add $7, %%ecx                          \n\t"

        "shll %%cl , %%esi                      \n\t"

    );
    bit&=1;
#endif /* BRANCHLESS_CABAC_DECODER */
#else /* ARCH_X86 */
    int s = *state;
    int RangeLPS= ff_h264_lps_range[s][c->range>>(CABAC_BITS+7)]<<(CABAC_BITS+1);
    int bit, lps_mask attribute_unused;

    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)){
            refill2(c);
        }
    }
#else /* BRANCHLESS_CABAC_DECODER */
    lps_mask= (c->range - c->low)>>31;

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

    s^=lps_mask;
    *state= (ff_h264_mlps_state+128)[s];
    bit= s&1;

    lps_mask= ff_h264_norm_shift[c->range>>(CABAC_BITS+3)];
    c->range<<= lps_mask;
    c->low  <<= lps_mask;
    if(!(c->low & CABAC_MASK))
        refill2(c);
#endif /* BRANCHLESS_CABAC_DECODER */

static int get_cabac(CABACContext *c, uint8_t * const state){
    return get_cabac_inline(c,state);
}

static int get_cabac_bypass(CABACContext *c){
    c->low += c->low;

    if(!(c->low & CABAC_MASK))
        refill(c);

    if(c->low < c->range){
        return 0;
    }else{
        c->low -= c->range;
        return 1;
    }
}

/**
 *
 * @return the number of bytes read or 0 if no end
 */
static int get_cabac_terminate(CABACContext *c){
    c->range -= 4<<CABAC_BITS;
    if(c->low < c->range){
        renorm_cabac_decoder_once(c);
        return 0;
    }else{
        return c->bytestream - c->bytestream_start;
    }

4024:d550343b5dac

    const uint8_t *bytestream_end;
    PutBitContext pb;
}CABACContext;

extern uint8_t ff_h264_mlps_state[4*64];
extern uint8_t ff_h264_lps_range[4][2*64];  ///< 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[512];


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);

        c->low += c->low;
    }
}

static void put_cabac(CABACContext *c, uint8_t * const state, int bit){
    int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + *state];

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

static void refill2(CABACContext *c){
    int i, x;

    x= c->low ^ (c->low-1);
    i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS-1)];

    x= -CABAC_MASK;

#if CABAC_BITS == 16
        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);

    c->low += x<<i;
    c->bytestream+= CABAC_BITS/8;
}

static inline void renorm_cabac_decoder(CABACContext *c){
    while(c->range < 0x100){
        c->range+= c->range;
        c->low+= c->low;
        if(!(c->low & CABAC_MASK))
            refill(c);
    }

#ifdef ARCH_X86_DISABLED
    int temp;
#if 0
    //P3:683    athlon:475
    asm(
        "lea -0x100(%0), %2         \n\t"
        "shr $31, %2                \n\t"  //FIXME 31->63 for x86-64
        "shl %%cl, %0               \n\t"
        "shl %%cl, %1               \n\t"
        : "+r"(c->range), "+r"(c->low), "+c"(temp)
    );
#elif 0
    //P3:680    athlon:474
    asm(
        "cmp $0x100, %0             \n\t"
        "setb %%cl                  \n\t"  //FIXME 31->63 for x86-64
        "shl %%cl, %0               \n\t"
        "shl %%cl, %1               \n\t"
        : "+r"(c->range), "+r"(c->low), "+c"(temp)
    );
#elif 1
    int temp2;
    //P3:665    athlon:517
    asm(
        "lea -0x100(%0), %%eax      \n\t"
        "cdq                        \n\t"
        "mov %0, %%eax              \n\t"
        "and %%edx, %0              \n\t"
        "and %1, %%edx              \n\t"
        "add %%eax, %0              \n\t"

    );
#elif 0
    int temp2;
    //P3:673    athlon:509
    asm(
        "cmp $0x100, %0             \n\t"
        "sbb %%edx, %%edx           \n\t"
        "mov %0, %%eax              \n\t"
        "and %%edx, %0              \n\t"
        "and %1, %%edx              \n\t"
        "add %%eax, %0              \n\t"

    );
#else
    int temp2;
    //P3:677    athlon:511
    asm(
        "cmp $0x100, %0             \n\t"
        "lea (%0, %0), %%eax        \n\t"
        "lea (%1, %1), %%edx        \n\t"
        "cmovb %%eax, %0            \n\t"
        "cmovb %%edx, %1            \n\t"
        : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
    );
#endif
#else
    //P3:675    athlon:476
    int shift= (uint32_t)(c->range - 0x100)>>31;
    c->range<<= shift;
    c->low  <<= shift;
#endif
    if(!(c->low & CABAC_MASK))
        refill(c);

#ifndef BRANCHLESS_CABAC_DECODER
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"
        "andl $0xC0, %%ebx                      \n\t"
        "movzbl "MANGLE(ff_h264_lps_range)"(%%eax, %%ebx, 2), %%esi\n\t"
        "movl "LOW      "(%2), %%ebx            \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%esi, %%edx                      \n\t"
        "movl %%edx, %%ecx                      \n\t"
        "shll $17, %%ecx                        \n\t"
        "cmpl %%ecx, %%ebx                      \n\t"
        " ja 1f                                 \n\t"

#if 1
        //athlon:4067 P3:4110
        "lea -0x100(%%edx), %%ecx               \n\t"
        "shr $31, %%ecx                         \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
#else
        //athlon:4057 P3:4130
        "cmp $0x100, %%edx                      \n\t" //FIXME avoidable
        "setb %%cl                              \n\t"
        "shl %%cl, %%edx                        \n\t"
        "shl %%cl, %%ebx                        \n\t"
#endif
        "movzbl "MANGLE(ff_h264_mps_state)"(%%eax), %%ecx   \n\t"

        "addl %%ecx, %%ebx                      \n\t"
        "movl %%esi, "BYTE    "(%2)             \n\t"
        "jmp 2f                                 \n\t"
        "1:                                     \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%ecx, %%ebx                      \n\t"
        "movl %%esi, %%edx                      \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"

        "movl "BYTE     "(%2), %%ecx            \n\t"

        "addl $2, %%ecx                         \n\t"
        "movl %%ecx, "BYTE    "(%2)             \n\t"

        "leal -1(%%ebx), %%ecx                  \n\t"
        "xorl %%ebx, %%ecx                      \n\t"
        "shrl $15, %%ecx                        \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx   \n\t"
        "neg %%ecx                              \n\t"
        "add $7, %%ecx                          \n\t"

        "shll %%cl , %%esi                      \n\t"

#else /* BRANCHLESS_CABAC_DECODER */
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"
        "andl $0xC0, %%ebx                      \n\t"
        "movzbl "MANGLE(ff_h264_lps_range)"(%%eax, %%ebx, 2), %%esi\n\t"
        "movl "LOW      "(%2), %%ebx            \n\t"
//eax:state ebx:low, edx:range, esi:RangeLPS
        "subl %%esi, %%edx                      \n\t"
#ifdef CMOV_IS_FAST
        "movl %%edx, %%ecx                      \n\t"
        "shl $17, %%edx                         \n\t"
        "cmpl %%ebx, %%edx                      \n\t"
        "cmova %%ecx, %%esi                     \n\t"
        "sbbl %%ecx, %%ecx                      \n\t"
        "andl %%ecx, %%edx                      \n\t"
        "subl %%edx, %%ebx                      \n\t"
        "xorl %%ecx, %%eax                      \n\t"
#else /* CMOV_IS_FAST */
FIXTHIS
        "movl %%edx, %%ecx                      \n\t"
        "subl %%ebx, %%edx                      \n\t"
        "sarl $31, %%edx                        \n\t" //lps_mask
        "subl %%ecx, %%esi                      \n\t" //RangeLPS - range
        "andl %%edx, %%esi                      \n\t" //(RangeLPS - range)&lps_mask

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

//eax:state ebx:low edx:mask esi:range

//eax:bit ebx:low esi:range

        "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx   \n\t"
        "shll %%cl, %%esi                       \n\t"
        "movzbl "MANGLE(ff_h264_mlps_state)"+128(%%eax), %%edx   \n\t"
        "movb %%dl, (%1)                        \n\t"
        "movl %%esi, "RANGE    "(%2)            \n\t"
        "shll %%cl, %%ebx                       \n\t"
        "movl %%ebx, "LOW      "(%2)            \n\t"
        "test %%bx, %%bx                        \n\t"
        " jnz 1f                                \n\t"

        "addl $2, %%ecx                         \n\t"
        "movl %%ecx, "BYTE    "(%2)             \n\t"

        "leal -1(%%ebx), %%ecx                  \n\t"
        "xorl %%ebx, %%ecx                      \n\t"
        "shrl $15, %%ecx                        \n\t"
        "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx   \n\t"
        "neg %%ecx                              \n\t"
        "add $7, %%ecx                          \n\t"

        "shll %%cl , %%esi                      \n\t"

    );
    bit&=1;
#endif /* BRANCHLESS_CABAC_DECODER */
#else /* ARCH_X86 */
    int s = *state;
    int RangeLPS= ff_h264_lps_range[0][2*(c->range&0xC0) + s];
    int bit, lps_mask attribute_unused;

    c->range -= RangeLPS;
#ifndef BRANCHLESS_CABAC_DECODER
    if(c->low < (c->range<<17)){
        bit= s&1;
        *state= ff_h264_mps_state[s];
        renorm_cabac_decoder_once(c);
    }else{
        bit= ff_h264_norm_shift[RangeLPS];
        c->low -= (c->range<<17);
        *state= ff_h264_lps_state[s];
        c->range = RangeLPS<<bit;
        c->low <<= bit;
        bit= (s&1)^1;

        if(!(c->low & 0xFFFF)){
            refill2(c);
        }
    }
#else /* BRANCHLESS_CABAC_DECODER */
    lps_mask= ((c->range<<17) - c->low)>>31;

    c->low -= (c->range<<17) & lps_mask;
    c->range += (RangeLPS - c->range) & lps_mask;

    s^=lps_mask;
    *state= (ff_h264_mlps_state+128)[s];
    bit= s&1;

    lps_mask= ff_h264_norm_shift[c->range];
    c->range<<= lps_mask;
    c->low  <<= lps_mask;
    if(!(c->low & CABAC_MASK))
        refill2(c);
#endif /* BRANCHLESS_CABAC_DECODER */

static int get_cabac(CABACContext *c, uint8_t * const state){
    return get_cabac_inline(c,state);
}

static int get_cabac_bypass(CABACContext *c){
    int range;
    c->low += c->low;

    if(!(c->low & CABAC_MASK))
        refill(c);

    range= c->range<<17;
    if(c->low < range){
        return 0;
    }else{
        c->low -= range;
        return 1;
    }
}

/**
 *
 * @return the number of bytes read or 0 if no end
 */
static int get_cabac_terminate(CABACContext *c){
    c->range -= 2;
    if(c->low < c->range<<17){
        renorm_cabac_decoder_once(c);
        return 0;
    }else{
        return c->bytestream - c->bytestream_start;
    }