This is a port of a Hogbom image cleaning algorithm benchmark originally written by Ben Humphreys of CSIRO ATNF. The port uses Hemi to provide both a fast parallel CUDA implementation for GPUs and an OpenMP parallel implementation for CPUs.
Hemi enables most of the code to be used for both implementations.
The Hemi port was done by Mark Harris.
tHogbomCleanHemi depends on Hemi and cub. Hemi is a simple library designed to make writing portable CUDA C/C++ code easy. In many cases, code written using Hemi can compile and run on either the CPU or GPU.
cub is a CUDA block-level library of parallel algorithms. tHogbomCleanHemi uses cub::BlockReduce to implement a fast image peak pixel finding step on the GPU. The 4-5 lines of code that call CUB are the only CUDA-specific code in tHogbomCleanHemi.
You must clone the cub and Hemi libraries from github. Hemi is available at http://github.com/harrism/hemi. cub is available at https://github.com/NVlabs/cub.
Edit the Makefile and change the value of HEMIDIR and CUBDIR to the relative path of your clones of hemi and cub, respectively.
To run on the GPU, you need a CUDA-capable GPU. More info about CUDA.
You should update the makefile to change the -arch flag to nvcc to specify the GPU architecture you are interested in. In the distribution it is set to -arch=sm_20
By default, running make will build the code to run on a CUDA GPU. To build for CPUs, comment out the first line below and uncomment the second line.
HEMICC = nvcc -x cu
#HEMICC = $(CXX)
Then run
> make
You can also change the value of CXX to icc to use the Intel compiler, if you have it installed.
Make sure the dirty.img and psf.img files are in the same location as the executable, and then run it.
> ./tHogbomCleanHemi
You can skip the (slow) "golden" step which generates the baseline (correct) images for comparison, by passing the -skipgolden command line option.
> ./tHogbomCleanHemi -skipgolden
Make sure you run the app at least once before you do this, to generate and save the image files on disk. -skipgolden loads the files from disk.
The following results use CUDA 5.0 and the Intel compiler (ICC) 13.0.1
Running CUDA version on a Tesla K20X GPU:
[harrism@machine tHogbomCleanHemi]$ ./tHogbomCleanHemi
Reading dirty image and psf image
Iterations = 1000
Image dimensions = 4096x4096
+++++ Forward processing (CPU Golden) +++++
Found peak of PSF: Maximum = 1 at location 2048,2048
Time 34.71 (s)
Time per cycle 34.71 (ms)
Cleaning rate 28.8101 (iterations per second)
Done
+++++ Forward processing (CUDA) +++++
Using CUDA Device 0: Tesla K20Xm( 14 SMs)
Found peak of PSF: Maximum = 1 at location 2048,2048
Time 1.19 (s)
Time per cycle 1.19 (ms)
Cleaning rate 840.336 (iterations per second)
Done
Verifying model...Pass
Verifying residual...Pass
Running OpenMP version on dual Xeon (Sandy Bridge) CPUs:
[harrism@machine tHogbomCleanHemi]$ ./tHogbomCleanHemi
Reading dirty image and psf image
Iterations = 1000
Image dimensions = 4096x4096
+++++ Forward processing (CPU Golden) +++++
Found peak of PSF: Maximum = 1 at location 2048,2048
Time 36.17 (s)
Time per cycle 36.17 (ms)
Cleaning rate 27.6472 (iterations per second)
Done
+++++ Forward processing (OpenMP) +++++
Using 12 OpenMP threads
Found peak of PSF: Maximum = 1 at location 2048,2048
Time 6.18 (s)
Time per cycle 6.18 (ms)
Cleaning rate 161.812 (iterations per second)
Done
Verifying model...Pass
Verifying residual...Pass
@copyright (c) 2011 CSIRO Australia Telescope National Facility (ATNF) Commonwealth Scientific and Industrial Research Organisation (CSIRO) PO Box 76, Epping NSW 1710, Australia atnf-enquiries@csiro.au
This file is part of the ASKAP software distribution.
The ASKAP software distribution is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
@author Ben Humphreys ben.humphreys@csiro.au
@author Ported to use Hemi & CUDA by Mark Harris mharris@nvidia.com
Hemi is licensed under the Apache License, version 2.0 Hemi is available at http://github.com/harrism/hemi