h264.c

arm平台下的H264编码和解码源代码

        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
    }
}

#if 0
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
    int i;
    //FIXME try int temp instead of block
    
    for(i=0; i<4; i++){
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
        const int z0= d0 + d3;
        const int z3= d0 - d3;
        const int z1= d1 + d2;
        const int z2= d1 - d2;
        
        block[0 + 4*i]=   z0 +   z1;
        block[1 + 4*i]= 2*z3 +   z2;
        block[2 + 4*i]=   z0 -   z1;
        block[3 + 4*i]=   z3 - 2*z2;
    }    

    for(i=0; i<4; i++){
        const int z0= block[0*4 + i] + block[3*4 + i];
        const int z3= block[0*4 + i] - block[3*4 + i];
        const int z1= block[1*4 + i] + block[2*4 + i];
        const int z2= block[1*4 + i] - block[2*4 + i];
        
        block[0*4 + i]=   z0 +   z1;
        block[1*4 + i]= 2*z3 +   z2;
        block[2*4 + i]=   z0 -   z1;
        block[3*4 + i]=   z3 - 2*z2;
    }
}
#endif

//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
    int i;
    const int * const quant_table= quant_coeff[qscale];
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
    const unsigned int threshold2= (threshold1<<1);
    int last_non_zero;

    if(seperate_dc){
        if(qscale<=18){
            //avoid overflows
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
            const unsigned int dc_threshold2= (dc_threshold1<<1);

            int level= block[0]*quant_coeff[qscale+18][0];
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
                if(level>0){
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
                    block[0]= level;
                }else{
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
                    block[0]= -level;
                }
//                last_non_zero = i;
            }else{
                block[0]=0;
            }
        }else{
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
            const unsigned int dc_threshold2= (dc_threshold1<<1);

            int level= block[0]*quant_table[0];
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
                if(level>0){
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
                    block[0]= level;
                }else{
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
                    block[0]= -level;
                }
//                last_non_zero = i;
            }else{
                block[0]=0;
            }
        }
        last_non_zero= 0;
        i=1;
    }else{
        last_non_zero= -1;
        i=0;
    }

    for(; i<16; i++){
        const int j= scantable[i];
        int level= block[j]*quant_table[j];

//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
        if(((unsigned)(level+threshold1))>threshold2){
            if(level>0){
                level= (bias + level)>>QUANT_SHIFT;
                block[j]= level;
            }else{
                level= (bias - level)>>QUANT_SHIFT;
                block[j]= -level;
            }
            last_non_zero = i;
        }else{
            block[j]=0;
        }
    }

    return last_non_zero;
}

static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
    const uint32_t a= ((uint32_t*)(src-stride))[0];
    ((uint32_t*)(src+0*stride))[0]= a;
    ((uint32_t*)(src+1*stride))[0]= a;
    ((uint32_t*)(src+2*stride))[0]= a;
    ((uint32_t*)(src+3*stride))[0]= a;
}

static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
}

static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
    
    ((uint32_t*)(src+0*stride))[0]= 
    ((uint32_t*)(src+1*stride))[0]= 
    ((uint32_t*)(src+2*stride))[0]= 
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
}

static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
    
    ((uint32_t*)(src+0*stride))[0]= 
    ((uint32_t*)(src+1*stride))[0]= 
    ((uint32_t*)(src+2*stride))[0]= 
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
}

static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
    
    ((uint32_t*)(src+0*stride))[0]= 
    ((uint32_t*)(src+1*stride))[0]= 
    ((uint32_t*)(src+2*stride))[0]= 
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
}

static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
    ((uint32_t*)(src+0*stride))[0]= 
    ((uint32_t*)(src+1*stride))[0]= 
    ((uint32_t*)(src+2*stride))[0]= 
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
}


#define LOAD_TOP_RIGHT_EDGE\
    const int t4= topright[0];\
    const int t5= topright[1];\
    const int t6= topright[2];\
    const int t7= topright[3];\

#define LOAD_LEFT_EDGE\
    const int l0= src[-1+0*stride];\
    const int l1= src[-1+1*stride];\
    const int l2= src[-1+2*stride];\
    const int l3= src[-1+3*stride];\

#define LOAD_TOP_EDGE\
    const int t0= src[ 0-1*stride];\
    const int t1= src[ 1-1*stride];\
    const int t2= src[ 2-1*stride];\
    const int t3= src[ 3-1*stride];\

static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
    const int lt= src[-1-1*stride];
    LOAD_TOP_EDGE
    LOAD_LEFT_EDGE

    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
    src[0+2*stride]=
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
    src[0+1*stride]=
    src[1+2*stride]=
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
    src[0+0*stride]=
    src[1+1*stride]=
    src[2+2*stride]=
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
    src[1+0*stride]=
    src[2+1*stride]=
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
    src[2+0*stride]=
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
}

static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
    LOAD_TOP_EDGE    
    LOAD_TOP_RIGHT_EDGE    
//    LOAD_LEFT_EDGE    

    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
    src[1+0*stride]=
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
    src[2+0*stride]=
    src[1+1*stride]=
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
    src[3+0*stride]=
    src[2+1*stride]=
    src[1+2*stride]=
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
    src[3+1*stride]=
    src[2+2*stride]=
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
    src[3+2*stride]=
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
}

static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
    const int lt= src[-1-1*stride];
    LOAD_TOP_EDGE    
    LOAD_LEFT_EDGE    
    //const __attribute__((unused)) int unu= l3;

    src[0+0*stride]=
    src[1+2*stride]=(lt + t0 + 1)>>1;
    src[1+0*stride]=
    src[2+2*stride]=(t0 + t1 + 1)>>1;
    src[2+0*stride]=
    src[3+2*stride]=(t1 + t2 + 1)>>1;
    src[3+0*stride]=(t2 + t3 + 1)>>1;
    src[0+1*stride]=
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
    src[1+1*stride]=
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
    src[2+1*stride]=
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
}

static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
    LOAD_TOP_EDGE    
    LOAD_TOP_RIGHT_EDGE    
    //const __attribute__((unused)) int unu= t7;

    src[0+0*stride]=(t0 + t1 + 1)>>1;
    src[1+0*stride]=
    src[0+2*stride]=(t1 + t2 + 1)>>1;
    src[2+0*stride]=
    src[1+2*stride]=(t2 + t3 + 1)>>1;
    src[3+0*stride]=
    src[2+2*stride]=(t3 + t4+ 1)>>1;
    src[3+2*stride]=(t4 + t5+ 1)>>1;
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
    src[1+1*stride]=
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
    src[2+1*stride]=
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
    src[3+1*stride]=
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
}

static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
    LOAD_LEFT_EDGE    

    src[0+0*stride]=(l0 + l1 + 1)>>1;
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
    src[2+0*stride]=
    src[0+1*stride]=(l1 + l2 + 1)>>1;
    src[3+0*stride]=
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
    src[2+1*stride]=
    src[0+2*stride]=(l2 + l3 + 1)>>1;
    src[3+1*stride]=
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
    src[3+2*stride]=
    src[1+3*stride]=
    src[0+3*stride]=
    src[2+2*stride]=
    src[2+3*stride]=
    src[3+3*stride]=l3;
}
    
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
    const int lt= src[-1-1*stride];
    LOAD_TOP_EDGE    
    LOAD_LEFT_EDGE    
    //const __attribute__((unused)) int unu= t3;

    src[0+0*stride]=
    src[2+1*stride]=(lt + l0 + 1)>>1;
    src[1+0*stride]=
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
    src[0+1*stride]=
    src[2+2*stride]=(l0 + l1 + 1)>>1;
    src[1+1*stride]=
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
    src[0+2*stride]=
    src[2+3*stride]=(l1 + l2+ 1)>>1;
    src[1+2*stride]=
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
    src[0+3*stride]=(l2 + l3 + 1)>>1;
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
}

static void pred16x16_vertical_c(uint8_t *src, int stride){
    int i;
    const uint32_t a= ((uint32_t*)(src-stride))[0];
    const uint32_t b= ((uint32_t*)(src-stride))[1];
    const uint32_t c= ((uint32_t*)(src-stride))[2];
    const uint32_t d= ((uint32_t*)(src-stride))[3];
    
    for(i=0; i<16; i++){
        ((uint32_t*)(src+i*stride))[0]= a;
        ((uint32_t*)(src+i*stride))[1]= b;
        ((uint32_t*)(src+i*stride))[2]= c;
        ((uint32_t*)(src+i*stride))[3]= d;
    }
}

static void pred16x16_horizontal_c(uint8_t *src, int stride){
    int i;

    for(i=0; i<16; i++){
        ((uint32_t*)(src+i*stride))[0]=
        ((uint32_t*)(src+i*stride))[1]=
        ((uint32_t*)(src+i*stride))[2]=
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
    }
}

static void pred16x16_dc_c(uint8_t *src, int stride){
    int i, dc=0;

    for(i=0;i<16; i++){
        dc+= src[-1+i*stride];
    }
    
    for(i=0;i<16; i++){
        dc+= src[i-stride];
    }

    dc= 0x01010101*((dc + 16)>>5);

    for(i=0; i<16; i++){
        ((uint32_t*)(src+i*stride))[0]=
        ((uint32_t*)(src+i*stride))[1]=
        ((uint32_t*)(src+i*stride))[2]=
        ((uint32_t*)(src+i*s