Base code and optimized code for the benchmarks used in the PolyMage paper published at ASPLOS 2015
Switch branches/tags
Nothing to show
Clone or download
bondhugula Merge pull request #2 from bollu/update-opencv-and-lint
Update opencv, fix PEP8 violations
Latest commit d20264e May 21, 2016
Failed to load latest commit information.
apps fixing makefile dependences Jan 2, 2016
images adding copyright information for the images Mar 24, 2015
.gitignore ignore the .so and .pyc build artifacts May 21, 2016
LICENSE Initial commit Mar 16, 2015 update README Mar 24, 2015

=================== PolyMage Benchmarks


  1. Python 2.x

  2. OpenCV 2.4.7 or higher (with QT/GTK support, video codec support for the video demo), Python bindings for OpenCV. On a Fedora Linux system, these can be installed with

$ sudo yum -y install opencv python-opencv

  1. g++ (GNU C++ compiler) version 4.8 or higher or Intel C/C++ compiler (icpc) 12.0 or higher (recommended: icpc 14.0 or higher)

How to run the benchmarks

All seven benchmarks are in the apps/ dir. For any of the benchmarks, say camera_pipe/, refers to the fully optimized Polymage code. refers to the base code (naive C code parallelized and vectorized) and corresponds to PolyMage (base+vec) from the paper.

Set the enviroment variable GOMP_CPU_AFFINITY=0-{NUM_THREADS} for pinning threads to reduce variance between runs

$ make camera_pipe

will build and run the optimized PolyMage code

$ make camera_pipe_naive

will build and run the base code.

$ make graph

will generate a PNG file with the pipeline diagram.

For the video demos, please see video_demo/README.txt