omarawad2/SCNN_GPU2

Implementation of a sparse CNN inference accelerator with compressed activation and weight memory using CUDA. Code optimized for GTX 980 card.

SCNN_GPU

CPU code compilation:

Command line compilation. First we need to configure the project:

cmake -H. -Bcmake-build-release -DCMAKE_BUILD_TYPE=Release

Then, we can proceed to build the project

cmake --build cmake-build-release/ --target all

Execute

./cmake-build-release/bin/SCNN_GPU

GPU code compilation:

Run script:

./compile.sh

Execute

./SCNN-GPU

Reference papers:

https://arxiv.org/pdf/1801.02108.pdf