yiweifengyan / XOR-Net

The repository for ICPADS paper XOR-Net

The Experiment Source Code of XOR-Net Paper

Target Platform:

• Hardware: GreenWaves GAP8.
• Compiling Environment: GAP_SDK with PULP OS suport.

Contents:

• main.c: The experiment source code.
• Makefile: The Makefile that configs GAP8.
• Example Result.txt: The command line output of one example execution.

Usage:

• Install GreenWaves GAP_SDK.
• Config the Makefile to include the correct "pulp_rules.mk".
• Run this command under the source code folder: "make clean all run".

How to execute it on Windows/Linux/Mac

• Extract the convolution functions.
• Replace the B_int_r (insert a single bit) and B_popc (popcnt) with platform specific APIs.
• Write your own benchmark code and execute.

References

• https://github.com/GreenWaves-Technologies/gap_sdk
• https://github.com/GreenWaves-Technologies/benchmarks

If you find this repository useful, please cite:

@inproceedings{zhu2020xor,  
 title={XOR-Net: An Efficient Computation Pipeline for Binary Neural Network Inference on Edge Devices},  
 author={Zhu, Shien and Duong, Luan HK and Liu, Weichen},
 booktitle={2020 IEEE 26th International Conference on Parallel and Distributed Systems (ICPADS)},
 pages={124--131},
 year={2020},
 organization={IEEE}
}