why is my image processing time additive when I have it spread to seperate timed structures on a quad-core realtime desktop computer?

[arnieb]

I am developing a vision application on a quad-core realtime desktop computer and 2 NI-1430 boards.  There are 3 cameras.  Each has its own timed structure loop with its own cpu core.  We generated the image processing vi's with Vision Assistant.  I expected the image processing for each camera to happen in parallel, but the total elapsed time is the same as if the image processing was serial.

Are the Vision Assistant generated vi's not designed for this?  Do I have to set some properties to facilitate multi-core processing? For example, do the shared lower-level vi's have to declared re-entrant?

[muks]

Can you post your vi or post a screen shot showing how you have designed the parallel loops?

[BCho]

Hey arnieb,

It would be interesting to see your code in 'parallel mode' as opposed to 'serial mode.' In 'serial mode' are you processing the image from one camera and then the other? So, do you actually wire error wires (or other form of execution order control) connecting the processing code from one camera to the processing code of another? How does the processing time of your code compare to processing images from a single camera?

[arnieb]

I have each of the 3 cameras in different timed structure loops completely seperated from each other and assigned its own cpu core.

I did find out from NI support engineers that the vision assistant vi's are only guaranteed to be single thread, i.e. not rentrant.  However, I should be able to define them as rentrant in the VI Properties-Execution tab.  This did saved a few milliseconds, but not enough, and still not close to acting parallel.

In regards to the first reply,  I am working on a small representation of what I am doing.

thank you.