h264.c

FFmpeg is an audio/video conversion tool. It includes libavcodec, the leading open source codec libr

 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
 */
static void encode_rbsp_trailing(PutBitContext *pb){
    int length;
    put_bits(pb, 1, 1);
    length= (-put_bits_count(pb))&7;
    if(length) put_bits(pb, length, 0);
}
#endif

/**
 * identifies the exact end of the bitstream
 * @return the length of the trailing, or 0 if damaged
 */
static int decode_rbsp_trailing(uint8_t *src){
    int v= *src;
    int r;

    tprintf("rbsp trailing %X\n", v);

    for(r=1; r<9; r++){
        if(v&1) return r;
        v>>=1;
    }
    return 0;
}

/**
 * idct tranforms the 16 dc values and dequantize them.
 * @param qp quantization parameter
 */
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
    const int qmul= dequant_coeff[qp][0];
#define stride 16
    int i;
    int temp[16]; //FIXME check if this is a good idea
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

//memset(block, 64, 2*256);
//return;
    for(i=0; i<4; i++){
        const int offset= y_offset[i];
        const int z0= block[offset+stride*0] + block[offset+stride*4];
        const int z1= block[offset+stride*0] - block[offset+stride*4];
        const int z2= block[offset+stride*1] - block[offset+stride*5];
        const int z3= block[offset+stride*1] + block[offset+stride*5];

        temp[4*i+0]= z0+z3;
        temp[4*i+1]= z1+z2;
        temp[4*i+2]= z1-z2;
        temp[4*i+3]= z0-z3;
    }

    for(i=0; i<4; i++){
        const int offset= x_offset[i];
        const int z0= temp[4*0+i] + temp[4*2+i];
        const int z1= temp[4*0+i] - temp[4*2+i];
        const int z2= temp[4*1+i] - temp[4*3+i];
        const int z3= temp[4*1+i] + temp[4*3+i];

        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
    }
}

#if 0
/**
 * dct tranforms the 16 dc values.
 * @param qp quantization parameter ??? FIXME
 */
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
//    const int qmul= dequant_coeff[qp][0];
    int i;
    int temp[16]; //FIXME check if this is a good idea
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

    for(i=0; i<4; i++){
        const int offset= y_offset[i];
        const int z0= block[offset+stride*0] + block[offset+stride*4];
        const int z1= block[offset+stride*0] - block[offset+stride*4];
        const int z2= block[offset+stride*1] - block[offset+stride*5];
        const int z3= block[offset+stride*1] + block[offset+stride*5];

        temp[4*i+0]= z0+z3;
        temp[4*i+1]= z1+z2;
        temp[4*i+2]= z1-z2;
        temp[4*i+3]= z0-z3;
    }

    for(i=0; i<4; i++){
        const int offset= x_offset[i];
        const int z0= temp[4*0+i] + temp[4*2+i];
        const int z1= temp[4*0+i] - temp[4*2+i];
        const int z2= temp[4*1+i] - temp[4*3+i];
        const int z3= temp[4*1+i] + temp[4*3+i];

        block[stride*0 +offset]= (z0 + z3)>>1;
        block[stride*2 +offset]= (z1 + z2)>>1;
        block[stride*8 +offset]= (z1 - z2)>>1;
        block[stride*10+offset]= (z0 - z3)>>1;
    }
}
#endif

#undef xStride
#undef stride

static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
    const int qmul= dequant_coeff[qp][0];
    const int stride= 16*2;
    const int xStride= 16;
    int a,b,c,d,e;

    a= block[stride*0 + xStride*0];
    b= block[stride*0 + xStride*1];
    c= block[stride*1 + xStride*0];
    d= block[stride*1 + xStride*1];

    e= a-b;
    a= a+b;
    b= c-d;
    c= c+d;

    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
}

#if 0
static void chroma_dc_dct_c(DCTELEM *block){
    const int stride= 16*2;
    const int xStride= 16;
    int a,b,c,d,e;

    a= block[stride*0 + xStride*0];
    b= block[stride*0 + xStride*1];
    c= block[stride*1 + xStride*0];
    d= block[stride*1 + xStride*1];

    e= a-b;
    a= a+b;
    b= c-d;
    c= c+d;

    block[stride*0 + xStride*0]= (a+c);
    block[stride*0 + xStride*1]= (e+b);
    block[stride*1 + xStride*0]= (a-c);
    block[stride*1 + xStride*1]= (e-b);
}
#endif

/**
 * gets the chroma qp.
 */
static inline int get_chroma_qp(H264Context *h, int qscale){
    
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
}


/**
 *
 */
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
    int i;
    uint8_t *cm = cropTbl + MAX_NEG_CROP;

    block[0] += 32;

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

        block[0 + 4*i]= z0 + z3;
        block[1 + 4*i]= z1 + z2;
        block[2 + 4*i]= z1 - z2;
        block[3 + 4*i]= z0 - z3;
    }

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

        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
        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];\