SOPC Custom Peripheral

Hi there!

I am trying to accelerate the following block of code implemented in the NIOS II softcore processor:

//-----------------------------------------------------
void mlse(void)
{
    //Iterative MLSE detector
    for(int k=0 ; k<NUMBER_OF_ITERATIONS ; k++)
    {            
        B = pow(5,((double)(2*(k-NUMBER_OF_ITERATIONS+1)))/NUMBER_OF_ITERATIONS) ;
                    
        //perform asynchronous updates
        for(int p=0 ; p<SEQUENCE_LENGTH ; p++)
        {
            u = 0;
                        
            //Multiply
            for(int q=0 ; q<SEQUENCE_LENGTH ; q++)
                u += T*s ;
                        
            //Add
            u += I ;                        
                        
            sigmoid(B,p) ;  //s = g(u) 
        }
    }
}

All the variables are of type float. I was thinking of converting them to integer, accelerating with C2H and then convert back to float for the sigmoid() function, but then I will have a floating point operation inside the loop.

The other way I was considering of doing this was keeping the variables of type float and using the altfp_matrix_mult Megacore function. The altfp_matrix_mult function block looks like a good idea, but I have no idea of how to implement it using the NIOS II processor. I was reading up on custom instructions and as far as I can understand this block won't work as a custom instruction since it needs more than two floating point input values and outputs more than one value. I have read up a bit on custom peripherals but I have no idea how to implement this using a custom peripheral.

If someone could please help me out here as I am stuck, I don't know how to solve this!

Thank you ;)