Resample Incorrect Assumption During Codegen.

The resample function is using the incorrect variant when generating code.

My entrypoint function looks something like this:

function outputs = my_foo(input_struct)
    % resample
    [y,ty]=resample(input_struct.x, input_struct.tx, 1/5);
    
    %rest of my function
end

And my coder config looks something like

cfg = coder.config('mex');

input_struct.x = coder.typeof(double(0), [1 inf]);
input_struct.tx = coder.typeof(double(0), [1 inf]);

struct_type = coder.typeof(input_struct)

spec = { ...
    'my_foo' '-args' {struct_type} ...
}

codegen('-config', cfg, spec{:})

When run, this produces an error on the resample line saying "Expected Q to be integer-valued." This indicates that the wrong version of the the resample function is being used for code generation.

Resample should be using the "[y,ty] = resample(x,tx,___)" version, but it appears to be trying to use the [y,b] = resample(x,p,q,___) version.

However, when I replace my struct definition with

input_struct.x = zeros(1, 10);
input_struct.tx = zeros(1, 10);

struct_type = coder.typeof(input_struct)

The build is fine, and the mex function does produce the same output as my matlab function.

So my question is this, is there any way to force the codegen to use the correct resample version if I know my input type is going to use vectors of length at least 10?

Answer 1

[Darshan Ramakant Bhat]

Thanks for the detailed description and reproduction steps.

This looks like a bug in the codegen support for this function. I have made an internal note to investigate the issue. Please continue with the current workaround that you have figured out.

I will post here if we find a better workaround for this issue after our investigation.