Layup-HP3

This is the implementation of the Layup paper as part of our Semester Term Project for "High Performance Parallel Programming" for the class of Spring 2021.

File Structure

Ensure the following directory structure:

├── data/
├── layup/
    ├── bin/
    ├── obj/
    ├── src/
        ├── MNISTParser.h
        ├── helper_cuda.h
        ├── layers.cpp
        ├── layers.hpp
        ├── main.cpp 
        ├── model.cpp
        ├── model.hpp
        ├── tags
        ├── utils.cu
        ├── utils.cuh
    ├── tags       
├── stats/
├── README.md

Compilation

cd layup
make

Running Instructions

To run the VGG Model :

bin/vggnet --dir ./../data -b 20

To run a simple Conv-Net :

bin/conv-neuralnet --dir ./../data -b 20

To run any of the baseline add the following flag :

--pag   # To preallocate GPU Memory.
--tel   # To transfer every layer to CPU.

Note : To remove the logging/printing of average iteration time, please set the flag to 0 in the following line https://github.com/ShahRutav/Layup-HP3/blob/db3290722479105069d2ba652194e8b4ecd71057/layup/src/model.cpp#L19