Adding CUDA multi-GPU support to OmicABEL for computing GWAS quickly.
C C++ Objective-C
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
doc Updated to revision 1167 Mar 21, 2013
inc Initial unmodified import of OmicABEL revision 1126. Mar 11, 2013
src
.gitignore
COPYING
DISCLAIMER.Linux
LICENSE Updated to revision 1167 Mar 21, 2013
Makefile
README
README.md
make.inc

README.md

cuOmicABEL

Adding CUDA multi-GPU support to OmicABEL for computing GWAS quickly.

Installing

You need the following dependencies:

GotoBLAS2

OmicABEL heavily relies on GotoBLAS2. Use the following to install it:

$ cd ~
$ wget http://www.tacc.utexas.edu/documents/13601/b58aeb8c-9d8d-4ec2-b5f1-5a5843b4d47b -O GotoBLAS2-1.13.tar.gz
$ tar -xzvf GotoBLAS2-1.13.tar.gz
$ cd GotoBLAS2
$ make TARGET=NEHALEM

You can drop the TARGET=NEHALEM option, but you will then get compile errors on modern Intel-processors because it doesn't recognize them. make clean in case that happens.

LAPACK

Same here; install it using the following commands:

$ cd ~
$ wget http://www.netlib.org/lapack/lapack-3.4.2.tgz
$ tar -xzvf lapack-3.4.2.tgz
$ mv lapack-3.4.2 lapack
$ cd lapack
$ cp make.inc.example make.inc

Edit that file so that at the bottom, it shows the following:

BLASLIB      = ~/GotoBLAS2/libgoto2.a

Then, compile it by running make. At some point, the tests will likely fail due to a linker error. We could fix this by linking the tests with pthreads, but we do not really care about them anyways; as long as there is a liblapack.a file, you're fine. Still need to create a differently named link to it for OmicABEL: ln -s liblapack.a lapack_LINUX.a.

cuOmicABEL

In case you installed GotoBLAS2 and LAPACK in other directories, you'll need to edit the make.inc file accordingly.

Besides that, cuOmicABEL should be compiled simply by running make.