# IEEE2017

**Repository Path**: sheenisme/IEEE2017

## Basic Information

- **Project Name**: IEEE2017
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-11-21
- **Last Updated**: 2021-11-21

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

DESCRIPTION

These are the benchmarks from the IEEE paper "Domain-Specific Optimization and Generation of High-Performance GPU Code for Stencil Computations"
The paper is in the main directory (ieee2017gpustencil.pdf).

The package contains:
a. the GPU benchmarks generated by StencilGen
b. scripts for benchmarking on a Pascal device



DEPENDENCIES

We tested the framework on ubuntu 16.04 and Red Hat Enterprise Linux Server release 6.7 using a 
Pascal P100 and Volta V100 card, with GCC 5.3.0, and NVCC 9.0. The following are hardware requirements
for the framework:
1. cmake >= 3.8
2. GCC >=4.9
3. NVCC 9.0




STEPS TO INSTALL

1. We present single- and double-precision benchmarks for Pascal P100. The single-precision benchmarks are in folder 'dsl-float', and the double-precision benchmarks are in folder 'dsl-double'.
2. Go to either 'dsl-float', or 'dsl-fouble', and
	2a. set the paths to NVCC and NVPROF in setup.sh
	2b. simply run './nvcc-run.sh' in the directory. This will invoke 'run.sh' script within each benchmark
3. In order to run the code for Volta, change the compute capability to 'sm_70' in each run.sh

We will simplify the scripts in some time.




MORE INFORMATION

For more information or questions, contact the authors at <rawat.15@osu.edu> or <vaidya.56@osu.edu>