Home

Welcome to your wiki!

This is the default page, edit it as you see fit. To add a new page simply reference it within brackets, e.g.: [SamplePage].

The wiki uses Markdown syntax.

Project Members:

• Emanuele Manca (admin)

[Emanuele Manca]

            ** CUDA-Quicksort **
          *  version 1.6.1  *

---

SYSTEM REQUIREMENTS

1. OS: linux, Windows

2. CPU: all

3. GPU: NVIDIA Fermi GPU architecture or higher (2.x capability or higher)

4. Builder: CUDA 5.5 or higher. For Older CUDA versions we recommend using Nsight Eclipse Edition (for linux) or Nsight Visual Studio Edition (for Windows)

NOTE: It is recommended to use NVIDIA Kepler GPU architecture with CUDA 5.5; using CUDA 4.0 or 4.2 would result in a slower runtime

---

HOW TO BUILD : LINUX

1. You can build the project through Nsight Eclipse Edition

2. The makefile can be found in the Release_linux folder.
   NOTE: This makefile works only with CUDA 6.0 or higher. You can use CUDA 5.5 but you must delete "-gencode arch=compute_32,code=sm_32" and "-gencode arch=compute_50,code=sm_50" in the file "Release_linux/src/subdir.mk" . Older CUDA versions do not support this makefile

3. Type 'make'. This builds the CUDA-Quicksort:

   make all - build all projects (the executable file will be created in the Relese_linux folder)
   make clean - clean project

---

HOW TO BUILD : Windows

1. You can build the project through Microsoft Visual Studio and Nsight Visual Studio Edition

2. The executable file will be created in the Relese_win\Win32 folder

---

HOW TO RUN : LINUX or Windows

To run the tests on CUDA-Quicksort, type the following: ./cuda-quickSort [test Flags] [distribution flags] [size flags] [--size SIZE] [--device identifier]

test Flags

default (no flag) or -t or --testGPU

    (run testGPU) The CUDA-Quicksort performance is compared over several configurations, e.g.:
        1. number of threads 
        2. size of the array to be sorted

-c or --compare:

    (run compare test) The CUDA-Quicksort performance is compared to the Radix Sort, Merge Sort, Bitonic Sort and Cederman and Tsigas' GPU Quick Sort performance.
    Results are stored in the following file: [executable folder]/compareTest.txt

-v or --validate

     (run validate test) A validity check is performed on CUDA-Quicksort

distribution flags

default or -a or --all

    The CUDA-Quicksort performance is compared over the following distributions: [uniform, gaussian, zero, staggered, bucket, sorted]

-u or --uniform

    The input will be a uniformly distributed random vector

-g or --gaussian

    The input will be a Gaussian distributed random vector

-s or --sorted

    The input will be a sorted vector

-b or --bucket

    The input will be a vector sorted into p buckets

-z or --zero

    The input will be a zero entropy vector

-r or --staggered

    The input will be a Staggered distributed random vector

size flags

 (size flags work only with testGPU and validate test, in compare test the size flags are "skipped")

default (no flags)

    Tests are performed on vectors of the following sizes: [1M, 2M, 4M, 8M, 16M, 32M]

-1 or --1M

    the input vector size will be  1M = 2^19

-2 or --2M

    the input vector size will be  2M = 2^20

-3 or --4M

    the input vector size will be  4M = 2^21

-4 or --8M

    the input vector size will be  8M = 2^22

-5 or --16M

    the input vector size will be  16M = 2^23

-6 or --32M

    the input vector size will be  32M = 2^24

--size SIZE

CUDA-Quicksort is performed on vectors of the size indicated by 'SIZE'

--device identifier

Use this option to choose which GPU should be used. default: --device 0