Closed johnramsden closed 3 years ago
If you want to keep end to end runtime, You can edit the "export_emb()" function for end-to-end latency of populating mram and "apply_emb()" for end-to-end latency of the lookup for each inference cycle:
Please rebase to master so you can run in random mode finely.
Add initial runtime recording for:
This will just give us output if we compiled debug at which point it will be printed to standard out.
Eventually we may want to pass an additional variable from python which keeps track of total run times. The problem is because the function does not hold state between runs we cannot keep track of all the lookups that might happen, so they will all be printed to standard out. This might be acceptable but it will mean parsing video output after run time to determine total time spent when multiple lookups are done.