Variable fold/reduction in Simulink

I am looking for a method to perform a fold/reduction across a variable number of inputs in the form of a vector signal. I specifically would like it to work without manually rewiring to multiple ports whenever I change the number of inputs. There currently exists one for add/sum, but there is no equivalent for general binary functions. The closest solution seems to be with delay + feedback in a systolic fashion like with CORDIC iterations however I need a method without delay and removing the delay in this solution gives an algebraic loop error. I also need the method to be HDL Coder compatible.

Answer 1

[Darshak]

Hello @Eric,

I ran into a similar challenge where I needed to apply a reduction operation across a vector signal in Simulink and make sure it is HDL coder compatible. I found that this can be done using the following approach:

• Use a MATLAB Function block inside a Subsystem (this is required for HDL Coder).
• Make sure your input vector is fixed size and set the data type to fixed-point using “fixdt” because double isn’t supported for synthesis.
• Implement the reduction as a simple loop applying the binary function over the vector elements. Keep it purely combinational to avoid algebraic loops.
• Define your binary operation inside a helper function so it’s easy to swap out.

This approach works well with HDL Coder too You may use the following function for reference:

function result = reduce_vector(u)
%#codegen
assert(~isempty(u));
N = length(u);
result = u(1);

for i = 2:N
    result = my_binary_op(result, u(i));
end
end

function out = my_binary_op(a, b)
% Change to your own binary operation

out = min(a, b);   
end

You can refer to the following documentation for more information on HDL Coder - https://www.mathworks.com/help/hdlcoder/ug/supported-simulink-blocks.html

I hope this helps.