Open andresovela opened 4 months ago
The profiling info with times for each op is cycle accurate. Did you mean something else?
Usually you estimate an algorithm's processing load using an MCPS number. That allows people to be able to tell how much processing the model needs in terms of "processor usage".
There's a difference between a model that runs in 10 ms using all available cores 100% of the time, and a model that runs in 10 ms but allows the processor to enter sleep mode for 90% of the time.
I'm interested in getting MCPS figures for different models, but I'm not sure how can I get cycle information on the xcore.ai/using the tools, especially considering multithreading.
Is there a proven way to calculate this?