harvard-acc / smaug

SMAUG: Simulating Machine Learning Applications Using Gem5-Aladdin
https://harvard-acc.github.io/smaug_docs
BSD 3-Clause "New" or "Revised" License
97 stars 27 forks source link

Support for Depthwise Conv layer #100

Open shivmgg opened 3 years ago

shivmgg commented 3 years ago

Hey,

Thanks for the amazing work! I would like to implement hardware-friendly DNNs like MobileNets, etc. I wanted to know if there is any upcoming support for depthwise conv layer? If no, may I know how can I add it?

xyzsam commented 3 years ago

I assume you want to model a hardware-accelerated block that runs depthwise convolutions. You will have to implement your own operator for this, which will require an accompanying Aladdin kernel. See https://harvard-acc.github.io/smaug_docs/add_python_operator.html for more details.