Varec-CC

Deep Learning Accelerator Stack

Introduction

1. varecc/ includes varec compiler, runtime driver source files.
2. examples/ includes some end-to-end examples
3. models/ includes pre-built YOLO models
4. hardware/ includes pre-built hardware bitstreams
5. patch/ includes some patches of tvm.

Prepare & Build

1. Check TVM for dependencies.
2. Run make to clone tvm and build.
3. Run source setup.sh to set environment variables.
4. Run pip3 install -r requirements.txt for packages needed by examples.

Examples

Simulation

1. Set “TARGET” to “sim” in varecc/config/varec_config.json
2. Run cd tvm/build && make varec to rebuild libvarec.so
3. In directory examples/, run python3 yolo.py

On-board (PYNQ images)

1. Set “TARGET” to “pynq” in varecc/config/varec_config.json
2. Copy varecc/ and tvm/ (without build/) from host to PYNQ (/home/xilinx/)
3. (on PYNQ, once) In directory tvm/, run mkdir build && cp cmake/config.cmake build/ && cd build && cmake .. && make runtime varec
4. (on PYNQ) In directory varecc/, run sudo ./scripts/start_rpc.sh to start RPC
5. (on the host) In directory examples/, run python3 yolo.py