Clear CAM or Set Receiving Neurons to Listen to Nothing

[Andrew Hansen]

Hi all,

So I'm having trouble clearing CAM using the netparser module. This is the error I recieve when I select ClearNetwork in the JavaFX GUI:

I'm currently perusing the Neuron.h, Neuron.cpp, and netparser.cpp files. Is there any way to clear CAM without the above problem? At the very least, would it be possible to set the post-synaptic neurons to listen to "nothing," effectively clearing them (in terms of activity)? Could I increase the size of the ring-buffer in ringbuffer.c to better accomodate ClearNetwork? Additionally, would it be possible to emulate ClearNetwork by having neurons listen to a virtual source chip which outputs no current? If so, how should I set the virtual source chip ID as in below:

I'm trying to fix this for students in a class this coming Friday, so any and all suggestions are welcome. Thank you for your time.

Best,

Andrew

[Federico Corradi]

Hi Andrew,

The message that you are seeing "Dropped event packet" is there to warn you that the USB is dropping events while you are programming all the CAM. Therefore, while you are programming the CAM the data that you are receiving from the USB is corrupted, note that this is because you are sending a lot of programming events and you are filling up the bus with these.

After you have cleared or programmed all the CAMs, the error should disappear as the USB should be able to handle all the traffic again. In general, programming all the CAMs with for big loops and generating tons of input configurations will always lead to "Dropped event packet" error, but you can neglect it if you are not interested in the event-stream while you have not yet finished programming your network.

Maybe Luca / Carsten that are reading us in copy can comment on the ring buffer questions. In the meanwhile, I hope this helps you. 

[Andrew Hansen]

Hi Federico,

Thank you for the response. That helps clear things up. In class we will ideally have a student assigned to each chip. As clearing all CAM is rather slow, do you think it would be feasible for me to simply write a function that only allows the students to write/clear the first say 16 CAM slots of any given neuron (and not need to clear all 64 per neuron), and clear single cores at a time? We don't necessarily need to work with dense networks, but it would be helpful given the time constraints to have in class. By feasible I mean integratable with cAER without throwing several errors or disrupting other functions.

Best,

Andrew