el1goluj / in-place-ds-algorithms

in-place-ds-algorithms

Table of Contents Introduction Algorithms Software Requirements Hardware Requirements Build Instructions

Introduction

Data Sliding (DS) algorithms are as a class of data manipulation primitives which perform unidirectional data shifting within a memory area.

Algorithms

Regular DS algorithms: Padding, Unpadding

Irregular DS algorithms: Select, Stream Compaction, Partition, Unique

This repository contains the implementations of the algorithms presented in this paper.

If you find this library useful, please cite:

@INPROCEEDINGS{dsalgorithms2015, author={Gomez-Luna, Juan and Chang, Li-Wen and Sung, I-Jui and Hwu, Wen-Mei and Guil, Nicolas}, booktitle={Parallel Processing (ICPP), 2015 44th International Conference on}, title={In-Place Data Sliding Algorithms for Many-Core Architectures}, year={2015}, pages={210-219}, keywords={Arrays;Compaction;Graphics processing units;Memory management;Parallel processing;Synchronization;in-place;relational algebra;stream compaction}, doi={10.1109/ICPP.2015.30}, ISSN={0190-3918}, month={Sept},}

Software Requirements

• Linux

Hardware Requirements

• NVIDIA Fermi generation of GPU or newer. Tested on:
  • GeForce GTX 580 (Fermi) with CUDA 6.5
  • Tesla K20 (Kepler) with CUDA 6.5
  • GeForce GTX 980 (Maxwell) with CUDA 6.5

OR

• AMD Hawaii generation of GPU or newer. Tested on:
  • Radeon R9 290 (Hawaii) with SDK 2.9.1
  • A10-7850K (Kaveri) with SDK 2.9.1

OR

• Intel CPU. Tested on:
  • Intel Core i7-3820 CPU with Intel OpenCL stack (driver version 1.2.0.8, Intel C Compiler version 14.0.1), and MxPA compiler

Build Instructions

Each benchmark folder contains a Makefile

Note: Compilation flags can be used to set thread block/work-group size, coarsening factor, SP/DP... See README.txt in each benchmark folder