Do Interfaces use shared memory?

[atena]

Hi all,

I have heard sometimes from colleagues that when using interfaces between two tasks allocated in the same tile, they do not pass the whole data over the interface, but they send a reference to the memory address where the data is. This way, creating interfaces between tasks in a same tile becomes a way to share memory and a really efficient resource.

Nevertheless, I could not read any document where this is described. Could anyone confirm me this fact about interfaces or explain a bit how they work in the behind?

Thanks a LOT!!

Best!

[infiniteimprobability]

It depends.. Interfaces provide a unified way for calling methods (passing data and synchronising across tasks) but how it works under the bonnet (hood) depends on the specifics.

In general, normal interface calls across tasks do use channels underneath. This allows the compiler to use the channel hardware mechanism to get the core to wait for a comms event.

Distributable tasks are different - the allow the task to be a simple function call, except wrapped in locks to preserve the "CSP"ness of it and keep synchronisation. So no channel ends used if the distributable requirements are met..

You can explicitly send a pointer across an interface call (in which case you'll be limited to tasks on the same tile, which is checked by the compiler) and you can also use a distributable task between two other tasks as a very neat and well behaved shared memory manager..

Take a look at this example.. http://www.xcore.com/forum/viewtopic.php?p=18531#p18531

[atena]

Thank you VERY MUCH!
:)