Thread Subject: C S Function Rwork vs Global Variables

I am currently working on a C s function that uses considerable memory (10000+ doubles) for storing intermediate calculations. I am interested in optimizing the memory usage & runtime performance for both simulations and real time workshop code generation.

In my experience, I have three resonable options available:

1) Create one long Rwork vector, and index the addresses to each of the local variables I use at every function call.

2) Declare a global pointer, and use malloc to fill this with the required memory during initialization. Then index my variables as above.

3) Declare multiple global pointers, with multiple calls to malloc for each variable during initialization. This means I do not have to index my variables, but multiple malloc calls is also not suggeted.

If anyone has any suggestions for the best strategy, this would be appreciated!

On Nov 8, 2:56 pm, "Jonathan Currie" <jonathan.cur...@hotmail.com>
wrote:
> I am currently working on a C s function that uses considerable memory (10000+ doubles) for storing intermediate calculations. I am interested in optimizing the memory usage & runtime performance for both simulations and real time workshop code generation.
>
> In my experience, I have three resonable options available:
>
> 1) Create one long Rwork vector, and index the addresses to each of the local variables I use at every function call.
>
> 2) Declare a global pointer, and use malloc to fill this with the required memory during initialization. Then index my variables as above.
>
> 3) Declare multiple global pointers, with multiple calls to malloc for each variable during initialization. This means I do not have to index my variables, but multiple malloc calls is also not suggeted.
>
> If anyone has any suggestions for the best strategy, this would be appreciated!

I'd recommend the first strategy. If you plan on having more than one
instance of this S-Function in your model, and if these intermediate
values need to be different for each instance, the global variable
strategy will fail since global variables are common to all S-Function
instances.

HTH,
Ashish.

Praetorian <ashish.sadanandan@gmail.com> writes:

> I'd recommend the first strategy. If you plan on having more than one
> instance of this S-Function in your model, and if these intermediate
> values need to be different for each instance, the global variable
> strategy will fail since global variables are common to all S-Function
> instances.

Simplicity in the implementation rules, therefore use strategy 3).
Multiple pointers don't use that much more memory than one single
pointer. And 10000 doubles are 80 kiB, that means approx. 13500
instances of your S-function fit into 1 GiB. Think about that.

--
Ralph Schleicher <http://ralph-schleicher.de>

Development * Consulting * Training
Mathematical Modeling and Simulation
Software Tools