[Coder] Which types are supported for constant folding?

In MATLAB HDL coder I try to do constant folding on the output of a function.

The output type is a 1x262144 uint16.

The coder complains with:

The input to coder.const cannot be reduced to a constant: Unsupported type.

Which types are supported?

[Denis Gurchenkov]

Most likely, you are hitting a case where the functon internally creates a variable-sized array (that is the most commonly seen "unsupported type"). There is no convenient way of figuring out what exactly is the cause of the error, other than opening the compilation report and starting on the body of the failing function, and hovering the mouse over various expressions and seeing which one has a non-constant size.

[Jan Siegmund]

Thanks, I'll have a look.

[Jan Siegmund]

I found the matrix being responsible for the issue. It is created from a concatenation in a for loop.

However the size can be determined at compile time wihtout doubt.

I tried to help the coder with

coder.varsize('a', [1 2^b], [0 0])

where b is compile time constant, but it still complains, that it cannot create the vector from concatenation

if the size is locked. Is there any other way?

Answer 1

[Jan Siegmund]

I did understand constant folding wrong:

I thought one must help the coder as follows:

a = coder.const(my_prod(3,2));

, because the gui of the fixed point converter was still complaining about a unsupported function in

my_prod

But the statement is already recognized as constant.

coder.const

just needs to be used, if you call a function multiple times with different input parameters

and you want the coder to create the function multiple times, each time with all of the parameters constant.