howardjchen/InnerProduct_CUDA

Use CUDA to accelerate the operations of a typical convolutional layer in often-used.

MASTER : NCTU IEE 2016 Fall
Computer Architecture Final Project

Part-I: Use CUDA to accelerate the operations of a typical convolutional layer in often-used.
(You can find the description slides here)

Three sub-directory

./data

This directory contains the input data for the base program

• /data/filt.txt - Store the values of filters
• /data/inNeu.txt - Store the values of input neurons

./innerProduct

This is the example to show you how to use CUDA to accelerate Inner Product

Usage

cd ./innerProduct
make
make run

./device

The program under this directory can show the device information

Usage

cd ./device
make
make run

Usage of the base program

Compile the code

make

Run the code

make run

Task

• Implement convLayerGPU() with CUDA
• Store your result in the outGPU
• Use NVIDIA Visual Profiler to analyze and improve your code

Evaluation

• convLayerCPU() will do the computation with C++ and store the output in the outCPU

• checker() will check whether the values stored in outCPU and outGPU are the same

• clock_gettime() is used to measure your preformance

• Lunch your CUDA kernels within two clock_gettime() functions (You are allowed to lunch multiple kernels in this project)

• Put cudaDeviceSynchronize() before the last clock_gettime()

• You must pass the checking to ensure your result is correct!

• We will compare the execution time to get the speedup

    Speedup = convLayerCPU_execTime / convLayerGPU_execTime
  

Grading Policy

• Completeness (30%)
  • Your result must be correct (Pass the check) (10%)
  • You get speedup compared to convLayerCPU() (10%)
  • You use NVIDIA Visual Profiler (NVVP) to help you (10%)
• Report (40%)
  • Describe your implementation algorithm and explain your results (10%)
  • Show how you use NVVP to help you find and solve perf. issues (10%)
  • Discussions on the optimizations you do (10%)
  • Feedback of this project (10%)
• Performance Rank (30%)
  • We will rank your CUDA kernels’ performance on GTX 680
  • The fastest one will get 30 points and the last one will get 1 points for this part
• Delay is not acceptable!

Other Rules

• It’s team work, 1 ~ 3 people in one team
  • Register here before deadline
• Compress your code and report into one zip file and upload to E3
  • Name your package as: LeaderID_FP1.zip
  • One team only need to upload one package to E3
  • Please name your report as: LeaderID_Report_FP1.pdf
  • Make sure TA can compile and run your code with "make" and "make run" on the provided server
• Any CUDA library is forbidden to use in this project

Useful Reference

• LeNet: Gradient Based Learning Applied to Document Recognition
• AlexNet: ImageNet Classification with Deep Convolutional Neural Networks
• CNN: Standford CS231n Convolutional Neural Networks for Visual Recognition
• CUDA Tutorial: CUDA C/C++ Basics
• CNN with CUDA: Optimizing Convolution Operations in CUDA with Adaptive Tiling convolution on gpu
• GPU Profiling: GPU Performance Analysis and Optimisation
• GPU Profiling: CUDA Profiling Documentation

TA: Chien-Yu Lin
Email: myislin@gmail.com