Since the number of SHAVEs is limited to sixteen in total, the order of allocating SHAVEs and the number of allocated SHAVEs for various purposes matters most. When applying the default settings of the afore mentioned parameters (shaves_scaling=4, shaves_conversion=4, shaves_nn=4), the SHAVEs are occupied as follows:
SHAVE number |
Purpose |
|---|---|
SHAVE 00-03 |
Scaling |
SHAVE 04-07 |
Format conversion |
SHAVE 08-11 |
Reserved |
SHAVE 12-15 |
Neural network execution; with CNN hardware acceleration |
Note:
Refer to vid_pipeline.init() and neural_network.init() for details on parameterization.