cxmatmul.cpp

opencv库在TI DM6437上的移植,目前包括两个库cv.lib和cxcore.lib的工程

                else if( a != d )
                {
                    int c_step0 = 1;
                    if( c == zero )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        double t0 = a[0]*b[0] + a[1]*b[b_step];
                        double t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step];
                        d[0] = t0*alpha + c[0]*beta;
                        d[d_step] = t1*alpha + c[c_step]*beta;
                    }
                }
                else
                    break;
                EXIT;
            case 3:
                if( len == d_size.width && b != d )
                {
                    for( i = 0; i < d_size.height; i++, d += d_step, a += a_step, c += c_step )
                    {
                        double t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2];
                        double t1 = a[0]*b[1] + a[1]*b[b_step+1] + a[2]*b[b_step*2+1];
                        double t2 = a[0]*b[2] + a[1]*b[b_step+2] + a[2]*b[b_step*2+2];
                        d[0] = t0*alpha + c[0]*beta;
                        d[1] = t1*alpha + c[1]*beta;
                        d[2] = t2*alpha + c[2]*beta;
                    }
                }
                else if( a != d )
                {
                    int c_step0 = 1;
                    if( c == zero )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        double t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2];
                        double t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step] + a[a_step+2]*b[b_step*2];
                        double t2 = a[a_step*2]*b[0] + a[a_step*2+1]*b[b_step] + a[a_step*2+2]*b[b_step*2];

                        d[0] = t0*alpha + c[0]*beta;
                        d[d_step] = t1*alpha + c[c_step]*beta;
                        d[d_step*2] = t2*alpha + c[c_step*2]*beta;
                    }
                }
                else
                    break;
                EXIT;
            case 4:
                if( len == d_size.width && b != d )
                {
                    for( i = 0; i < d_size.height; i++, d += d_step, a += a_step, c += c_step )
                    {
                        double t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2] + a[3]*b[b_step*3];
                        double t1 = a[0]*b[1] + a[1]*b[b_step+1] + a[2]*b[b_step*2+1] + a[3]*b[b_step*3+1];
                        double t2 = a[0]*b[2] + a[1]*b[b_step+2] + a[2]*b[b_step*2+2] + a[3]*b[b_step*3+2];
                        double t3 = a[0]*b[3] + a[1]*b[b_step+3] + a[2]*b[b_step*2+3] + a[3]*b[b_step*3+3];
                        d[0] = t0*alpha + c[0]*beta;
                        d[1] = t1*alpha + c[1]*beta;
                        d[2] = t2*alpha + c[2]*beta;
                        d[3] = t3*alpha + c[3]*beta;
                    }
                }
                else if( d_size.width <= 16 && a != d )
                {
                    int c_step0 = 1;
                    if( c == zero )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        double t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2] + a[3]*b[b_step*3];
                        double t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step] +
                                    a[a_step+2]*b[b_step*2] + a[a_step+3]*b[b_step*3];
                        double t2 = a[a_step*2]*b[0] + a[a_step*2+1]*b[b_step] +
                                    a[a_step*2+2]*b[b_step*2] + a[a_step*2+3]*b[b_step*3];
                        double t3 = a[a_step*3]*b[0] + a[a_step*3+1]*b[b_step] +
                                    a[a_step*3+2]*b[b_step*2] + a[a_step*3+3]*b[b_step*3];
                        d[0] = t0*alpha + c[0]*beta;
                        d[d_step] = t1*alpha + c[c_step]*beta;
                        d[d_step*2] = t2*alpha + c[c_step*2]*beta;
                        d[d_step*3] = t3*alpha + c[c_step*3]*beta;
                    }
                }
                else
                    break;
                EXIT;
            }
        }

        if( type == CV_32F )
        {
            float* d = D->data.fl;
            const float *a = A->data.fl, *b = B->data.fl, *c = C->data.fl;
            size_t d_step = D->step/sizeof(d[0]),
                   a_step = A->step/sizeof(a[0]),
                   b_step = B->step/sizeof(b[0]),
                   c_step = C->step/sizeof(c[0]);

            if( !c )
                c = zerof;

            switch( len )
            {
            case 2:
                if( len == d_size.width && b != d )
                {
                    for( i = 0; i < d_size.height; i++, d += d_step, a += a_step, c += c_step )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step];
                        float t1 = a[0]*b[1] + a[1]*b[b_step+1];
                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[1] = (float)(t1*alpha + c[1]*beta);
                    }
                }
                else if( a != d )
                {
                    int c_step0 = 1;
                    if( c == zerof )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step];
                        float t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step];
                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[d_step] = (float)(t1*alpha + c[c_step]*beta);
                    }
                }
                else
                    break;
                EXIT;
            case 3:
                if( len == d_size.width && b != d )
                {
                    for( i = 0; i < d_size.height; i++, d += d_step, a += a_step, c += c_step )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2];
                        float t1 = a[0]*b[1] + a[1]*b[b_step+1] + a[2]*b[b_step*2+1];
                        float t2 = a[0]*b[2] + a[1]*b[b_step+2] + a[2]*b[b_step*2+2];
                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[1] = (float)(t1*alpha + c[1]*beta);
                        d[2] = (float)(t2*alpha + c[2]*beta);
                    }
                }
                else if( a != d )
                {
                    int c_step0 = 1;
                    if( c == zerof )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2];
                        float t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step] + a[a_step+2]*b[b_step*2];
                        float t2 = a[a_step*2]*b[0] + a[a_step*2+1]*b[b_step] + a[a_step*2+2]*b[b_step*2];

                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[d_step] = (float)(t1*alpha + c[c_step]*beta);
                        d[d_step*2] = (float)(t2*alpha + c[c_step*2]*beta);
                    }
                }
                else
                    break;
                EXIT;
            case 4:
                if( len == d_size.width && b != d )
                {
                    for( i = 0; i < d_size.height; i++, d += d_step, a += a_step, c += c_step )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2] + a[3]*b[b_step*3];
                        float t1 = a[0]*b[1] + a[1]*b[b_step+1] + a[2]*b[b_step*2+1] + a[3]*b[b_step*3+1];
                        float t2 = a[0]*b[2] + a[1]*b[b_step+2] + a[2]*b[b_step*2+2] + a[3]*b[b_step*3+2];
                        float t3 = a[0]*b[3] + a[1]*b[b_step+3] + a[2]*b[b_step*2+3] + a[3]*b[b_step*3+3];
                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[1] = (float)(t1*alpha + c[1]*beta);
                        d[2] = (float)(t2*alpha + c[2]*beta);
                        d[3] = (float)(t3*alpha + c[3]*beta);
                    }
                }
                else if( len <= 16 && a != d )
                {
                    int c_step0 = 1;
                    if( c == zerof )
                    {
                        c_step0 = 0;
                        c_step = 1;
                    }

                    for( i = 0; i < d_size.width; i++, d++, b++, c += c_step0 )
                    {
                        float t0 = a[0]*b[0] + a[1]*b[b_step] + a[2]*b[b_step*2] + a[3]*b[b_step*3];
                        float t1 = a[a_step]*b[0] + a[a_step+1]*b[b_step] +
                                   a[a_step+2]*b[b_step*2] + a[a_step+3]*b[b_step*3];
                        float t2 = a[a_step*2]*b[0] + a[a_step*2+1]*b[b_step] +
                                   a[a_step*2+2]*b[b_step*2] + a[a_step*2+3]*b[b_step*3];
                        float t3 = a[a_step*3]*b[0] + a[a_step*3+1]*b[b_step] +
                                   a[a_step*3+2]*b[b_step*2] + a[a_step*3+3]*b[b_step*3];
                        d[0] = (float)(t0*alpha + c[0]*beta);
                        d[d_step] = (float)(t1*alpha + c[c_step]*beta);
                        d[d_step*2] = (float)(t2*alpha + c[c_step*2]*beta);
                        d[d_step*3] = (float)(t3*alpha + c[c_step*3]*beta);
                    }
                }
                else
                    break;
                EXIT;
            }
        }
    }
 
    {
        int b_step = B->step;
        CvGEMMSingleMulFunc single_mul_func;
        CvMat tmat, *D0 = D;
        icvBLAS_GEMM_32f_t blas_func = 0;

        if( !inittab )
        {
            icvInitGEMMTable( &single_mul_tab, &block_mul_tab, &store_tab );
            inittab = 1;
        }

        single_mul_func = (CvGEMMSingleMulFunc)single_mul_tab.fn_2d[type];
        if( !single_mul_func )
            CV_ERROR( CV_StsUnsupportedFormat, "" );

        if( D->data.ptr == A->data.ptr || D->data.ptr == B->data.ptr )
        {
            int buf_size = d_size.width*d_size.height*CV_ELEM_SIZE(type);
            if( d_size.width <= CV_MAX_LOCAL_MAT_SIZE )
            {
                buffer = (uchar*)cvStackAlloc( buf_size );
                local_alloc = 1;
            }
            else
                CV_CALL( buffer = (uchar*)cvAlloc( buf_size ));