-
Notifications
You must be signed in to change notification settings - Fork 13
Open
Description
We are storing the size of synapse groups[1] in our connectivity matrix. We could also compute it from consecutive start indices of synapse groups in the connectivity matrix. Still, both would need two global memory reads, but one would save some memory. Though the saving would be in order of synapse groups and therefore likely negligible compared to the synapses themselves.
I feel like I should have thought of that in the past when trying to optimize for memory. Maybe we need the sizes somewhere else where we don't need the start indices, and hence accessing size with only one memory load is beneficial?
Check where we need synapse group sizes.
[1] synapse groups are all synapses for one (preID, postGroup) - pair