Embedded coder without dynamic memory allocation: using a large input parameter.

I use the embedded coder to generate C code from my MatLab function with setting EnableDynamicMemoryAllocation = false for speed of the generated code. One of the input parameters to my function is a large multi-channel image (e.g. RGB). My code does not copy that large input parameter. My code only reads parts of that input parameter. Still, code generation fails on that input parameter with:

Computed maximum size exceeds maximum allowed number of elements (134217728).
The computed size of structure field 'data' is [:4 x :4096 x :100000].
Please consider enabling dynamic memory allocation to allow unbounded sizes.

But my code only needs to read from that data. The generated code does not need storage of that size since my code never copies that data. I tested that by modifying my function to not use that data at all. The same error was still reported.

How can I fix this ?

Update: A workaround seems to be to specify small sizes for the input parameter for code generation, and using large sizes for the assert()'s in the code.

Answer 1

[Ruchika Parag]

Hi @Hank, if you're running into the “Computed maximum size exceeds maximum allowed number of elements (134217728)” error during code generation with Embedded Coder and have EnableDynamicMemoryAllocation = false, it's because the code generator is attempting to allocate memory for the full possible size of the input. Even if your function doesn’t copy or operate on the entire input image (e.g., a large multi-channel RGB array), the coder still infers and reserves space for the maximum possible size unless you explicitly constrain it.

This is a known behavior when working with statically allocated memory. In your case, the inferred size of the input is something like 4 x 4096 x 100000, which far exceeds the static memory limit.

A few ways to resolve this:

1. Specify smaller upper bounds for code generation:Define tighter bounds on your input size when preparing your function for codegen. For instance, if you’re using codegen, you can specify realistic maximum sizes that won’t exceed the static limit. At the same time, you can keep broader limits using assert in the MATLAB code to validate inputs at runtime.
2. Enable dynamic memory allocation (if possible):If you can relax the constraint, set cfg.EnableDynamicMemoryAllocation = true in your code generation configuration object. This allows the generated code to handle variable-sized inputs without requiring a static memory allocation upfront. However, this might introduce a runtime performance cost or dynamic memory usage that you were trying to avoid.
3. Use chunking or streaming:Instead of passing the entire image at once, consider restructuring your code to process the input in smaller chunks. This keeps memory usage within bounds and avoids the need for dynamic memory.

In summary, if dynamic memory is disabled, the only way to avoid this error is to explicitly limit the input size used during code generation. Otherwise, you'll need to restructure the data handling logic or allow dynamic allocation. Hope this helps!