Python notebook contains code for converting onnx model into a python dictionary
pending - Generating a set of instructions
- Systolic array
- weight and data fifo
- 2*2 max pooling
- BlockRam IP integration and weight and data filling
- RELU, Sigmoid, Softmax
Pending
-
Control logic
- Collecting output from sys arr and passing to activation and pooling layers
- redirecting outputs
- instruction decode
- pipelining
- If possible fused layer convolution
-
Top module
Each instruction lenght - 64 bit
opcode:- 8 bit
@dimension1 @dimension2
Tbd
instructions as of now
- read_inputs
- read_weights
- fillfifo
- convolution
- relu
- pooling
- store_outputs
- write_outputs
- reset_total
- reset_fifo
- reset_systolic