comparison cabac.h @ 4002:ec426fa57dfe libavcodec

adds some useful comments after some of the #else, #elseif, #endif preprocessor directives to make it clearer which code block depends on which #define xx

4001:34fdffe98bd0

        :"=&a"(bit) //FIXME this is fragile gcc either runs out of registers or misscompiles it (for example if "+a"(bit) or "+m"(*state) is used
        :"r"(state), "r"(c)
        : "%ecx", "%ebx", "%edx", "%esi"
    );
    bit&=1;
#else
    asm volatile(
        "movzbl (%1), %%eax                     \n\t"
        "movl "RANGE    "(%2), %%ebx            \n\t"
        "movl "RANGE    "(%2), %%edx            \n\t"
        "shrl $23, %%ebx                        \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
        "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
        "addl %%ecx, %%esi                      \n\t" //new range
        "andl %%edx, %%ecx                      \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"

        :"=&a"(bit)
        :"r"(state), "r"(c)
        : "%ecx", "%ebx", "%edx", "%esi"
    );
    bit&=1;
#endif
#else
    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;

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

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

    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
#endif
    return bit;
}

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

4002:ec426fa57dfe

        :"=&a"(bit) //FIXME this is fragile gcc either runs out of registers or misscompiles it (for example if "+a"(bit) or "+m"(*state) is used
        :"r"(state), "r"(c)
        : "%ecx", "%ebx", "%edx", "%esi"
    );
    bit&=1;
#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"

        "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
        "addl %%ecx, %%esi                      \n\t" //new range
        "andl %%edx, %%ecx                      \n\t"
        "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_mps_state)"(%%eax), %%ecx   \n\t"
        "movb %%cl, (%1)                        \n\t"

        :"=&a"(bit)
        :"r"(state), "r"(c)
        : "%ecx", "%ebx", "%edx", "%esi"
    );
    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;

        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;

    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 */
#endif /* ARCH_X86 */
    return bit;
}

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