How to install MAGAMA

Open discussion for MAGMA

How to install MAGAMA

Postby David » Tue May 24, 2011 9:45 am

Hi, I'm a MAGAMA newer and have a strong interest in it. I used CUDA C before and know about MAGAMA recently. I know MAGAMA is used under Linux platform and tried to install it but failed. I want to familiarize how MAGAMA does matrix algebra on GPU by running some simple codes. I'll really appreciate if someone can guide me the detailed steps of installation.

Regards,

David
David
 
Posts: 5
Joined: Mon May 09, 2011 4:00 pm

Re: How to install MAGAMA

Postby mateo70 » Thu May 26, 2011 12:09 pm

Hi David,

To try MAGMA, take one of the make.inc close to your configuration and adapt it to your specific case by giving your compiler, the path to your libraries, the GPU family you are targeting. And then just type make.
If you have any problems, with the compilation, just reply to this post with the detail of the make.inc you used and the error obtained, if so, as detailed as possible. We will try to help you.

Mathieu
mateo70
 
Posts: 41
Joined: Tue Mar 08, 2011 12:38 pm

Re: How to install MAGAMA

Postby David » Sun May 29, 2011 3:52 pm

Hi, Mathieu,

Thanks for your reply. My operating system is 32-bit Ubuntu 10.04. Let me show what I did and errors I got.

Firstly, I cannot compile and execute CUDA sample in Linux.

I downloaded and installed CUDA driver, toolkits and SDK (3.2 version) at http://developer.nvidia.com/cuda-toolki ... oads#Linux. Then I set environment variables as adding following two lines

export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib:$LD_LIBRARY_PATH

into ~/.bashrc. and adding " <CUDA_INSTALL_PATH>/cuda/lib" into /etc/ld.so.conf.

When I build the SDK as typing commands as below:

@ubuntu:~$ cd [to my NVIDIA_GPU_Computing_SDK/C]
@ubuntu:~$ make

errors are showed as

make[1]: *** [obj/i386/release/cutil.cpp.o] Error 1
make[1]: Leaving directory `/home/darkwafer/NVIDIA_GPU_Computing_SDK/C/common'
make: *** [lib/libcutil.so] Error 2


I think I should pass CUDA installation and then install BLAS and LAPACK. Finally, I'm allowed install MAGMA, right?

Regards,

David
David
 
Posts: 5
Joined: Mon May 09, 2011 4:00 pm

Re: How to install MAGAMA

Postby David » Tue May 31, 2011 1:07 am

Hi, Mathieu,

After configuring my Linux recently, I've successfully installed CUDA. Could you guide me how to install BLAS and LAPACK, so that I can install MAGMA. Thanks a lot.

Regards,

David
David
 
Posts: 5
Joined: Mon May 09, 2011 4:00 pm

Re: How to install MAGAMA

Postby mateo70 » Tue May 31, 2011 9:33 pm

The easiest way to install blas and lapack if you are not interrested by the performances is to use the plasma installer with the configuration by default. (http://icl.cs.utk.edu/plasma/software/index.html)
Otherwise you can install MKl or ACML which will provide you BLAS and LAPACK.

Mathieu
mateo70
 
Posts: 41
Joined: Tue Mar 08, 2011 12:38 pm

Re: How to install MAGAMA

Postby David » Wed Jun 01, 2011 11:59 am

Thanks, Mathieu. I'm not sure how to install PLASMA in Linux. Do you have detailed steps with commands to install it? I googled the detailed steps when I installed CUDA in linux. As PLASMA and MAGMA in Linux, I don't find too much information online. I'll appreciate if you can share your experiences or related materials with me.

Regards,

David
David
 
Posts: 5
Joined: Mon May 09, 2011 4:00 pm

Re: How to install MAGAMA

Postby mateo70 » Wed Jun 08, 2011 3:29 pm

Hi,

concerning PLASMA use the PLASMA installer you'll find on the link I gave you last time. And once you have untar the archive on your machine, just run :
Code: Select all
./setup.py --prefix=directory_where_you_want_to_install


This will give you an unoptimized version of PLASMA using all the reference libraries from NetLib. If you want to use another blas library, you can for example do for MKL:
Code: Select all
./setup.py --prefix=directory_where_you_want_to_install  --blaslib="-L${MKL_ROOT}/lib/intel64 -Wl,--start-group -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"


After that for MAGMA, copy an example of make.inc and update this file with your own information. You can copy/past most of the information from the make.inc generated by the plasma installer.

Regards,
Mathieu
mateo70
 
Posts: 41
Joined: Tue Mar 08, 2011 12:38 pm

Re: How to install MAGAMA

Postby NLewkow » Fri Jun 24, 2011 2:35 pm

I am also trying to get MAGMA installed and have been unsuccessful thus far. I have PLASMA installer to make sure everything that I need for MAGMA is installed.

Atlas was installed from the repository as
Code: Select all
yum install atlas.x86_64

returns that it was already installed.

When I run the PLASMA installer scripts I am passing it
Code: Select all
-bash-4.2$ ./setup.py --prefix=/home/lewkow/Bin --blaslib="-lf77blas -lcblas -latlas"

as directed by the installer, but the output from the script says

Code: Select all
BLAS library is...  -lf77blas -lcblas -latlas
CBLAS library is... Check if included in Blas Library
LAPACK library is... Check if included in Blas library
LAPACKE Interface library is... Check if included in Lapack library

========================================
  BLAS installation/verification
========================================
Checking if provided BLAS works...

BLAS: provided BLAS cannot be used! aborting...
error is:
****************************************
gfortran -o tmpf tmpf.f -lf77blas -lcblas -latlas   -lm
/usr/bin/ld: cannot find -lf77blas
/usr/bin/ld: cannot find -lcblas
/usr/bin/ld: cannot find -latlas
collect2: ld returned 1 exit status

****************************************

Any ideas/suggestions would be great.

Nick
NLewkow
 
Posts: 7
Joined: Fri Jun 24, 2011 11:13 am


Return to User discussion

Who is online

Users browsing this forum: No registered users and 3 guests