portability of compiled binary code with openmp or openmpi

General OpenMP discussion

portability of compiled binary code with openmp or openmpi

Postby baddylover » Sun May 20, 2012 4:31 am

I haven't done anything yet but just ask beforehand:
if I compile a Fortran program by using gfortran+openmp in a Windows7 intel i5 4GB memory machine,
would the compiled binary code be usable in another machine running Windows XP AMD X4 945 8GB memory machine?
What about with OpenMPI?
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Thu Jul 05, 2012 9:21 pm

it works.
I compiled my fortran program by gfortran under windowsXP in an AMD 4-core machine.
the binary code program (.exe) runs parallel threads in two other intel- machines under Windows7.
The CPU usage is about 90% in each machine, and the results are as expected.
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Wed Jul 18, 2012 4:01 pm

well, it seems the compiled .exe file needs some .dll files of the gfortran compiler.
Does anyone know how to make an .exe file using multi-threads/cores executable under any Windows system without any special software like gfortran installed?
I ask this because I think it's important if I want to give my client an .exe file so my client can run it straight away.
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Thu Jul 26, 2012 7:56 pm

Any info would be appreciated. If it's not possible with gfortran, what about other compilers?
I wish to find a fortran compiler with openmp that can generate an executable binary code .exe program that runs and uses multi-threads/cores
under any Windows system without special softwares installed.
Also I wish the compiler works just fine for AMD processors such as the AMD 16-core bulldozer processors.
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby ftinetti » Fri Jul 27, 2012 6:42 am

Hi,

I'm not a Windows developer. I would try with
-static (a gcc option, which can be used with gfortran, I think), based on
http://gcc.gnu.org/onlinedocs/gfortran/ ... NU-Fortran
and
http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc_3.html

And, just in case:
-static-libgfortran, based on
http://gcc.gnu.org/onlinedocs/gfortran/ ... nk-Options

Also, enable full warning generation. Unfortunately, "this option has no effect" statement/s in the man pages does not help a lot... so I think you will have to try and see by yourself, since it applies to every library used by your source code.

BTW,

AMD processors such as the AMD 16-core bulldozer processors

I've seen that bulldozer is just about 8-core + SMT + replicated integer units, do you have non floating point compute intensive code?

HTH,

Fernando.
ftinetti
 
Posts: 580
Joined: Wed Feb 10, 2010 2:44 pm

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Mon Jul 30, 2012 9:26 pm

sorry, the AMD 16 core processor I mentioned is this one (I possibly used wrong name before):

http://www.cplonline.com.au/online-shop ... ory_id=134

AMD Opteron (Sixteen-Core) Model 6282 SE, 2.60 Ghz, Turbo Core, 16MB Cache, Socket G34, TDP 140W, TRAY/OEM version, Price: $1,279.00

Many thanks for the information on gfortran options. I'll certainly try various ways and report back here.
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Sun Oct 07, 2012 2:29 am

ftinetti wrote:Hi,

I'm not a Windows developer. I would try with
-static (a gcc option, which can be used with gfortran, I think), based on
http://gcc.gnu.org/onlinedocs/gfortran/ ... NU-Fortran
and
http://gcc.gnu.org/onlinedocs/gcc-3.0/gcc_3.html

And, just in case:
-static-libgfortran, based on
http://gcc.gnu.org/onlinedocs/gfortran/ ... nk-Options

Also, enable full warning generation. Unfortunately, "this option has no effect" statement/s in the man pages does not help a lot... so I think you will have to try and see by yourself, since it applies to every library used by your source code.

BTW,

AMD processors such as the AMD 16-core bulldozer processors

I've seen that bulldozer is just about 8-core + SMT + replicated integer units, do you have non floating point compute intensive code?

HTH,

Fernando.


It seems to me you are saying FX-8150 is only fast for non floating point compute intensive code, but not double precision computation?
How does it compare with AMD Phenom(tm) II X4 965 for double precision computation?
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am

Re: portability of compiled binary code with openmp or openm

Postby ftinetti » Mon Oct 08, 2012 7:19 am

It seems to me you are saying FX-8150 is only fast for non floating point compute intensive code, but not double precision computation?
How does it compare with AMD Phenom(tm) II X4 965 for double precision computation?

Well, since some time ago, AMD has decided to "share" a single (yet "wide") FPU among two (integer) cores, take a look at page 6 of
http://developer.amd.com/Assets/51803A_ ... SCREEN.pdf
or the statement
The FPU is a coprocessor model that is shared between the two cores of one AMD Family 15h compute unit.

in page 37 of
http://support.amd.com/us/Processor_Tec ... _guide.pdf
I'm not completely sure this really generates a huge slowdown, but I think that, at least, this is not a "full" or "traditional" dual core design. Unfortunately, I don't have any (literally...) FX-8150 nor Phenom 965 so I can't say it for sure... maybe you may share your info and comparisons?

Fernando.
ftinetti
 
Posts: 580
Joined: Wed Feb 10, 2010 2:44 pm

Re: portability of compiled binary code with openmp or openm

Postby baddylover » Mon Oct 08, 2012 5:03 pm

I'm sure AMD phenom ii X4 965 takes only a quarter of time by using parallely all four cores for double precision float number computation, see this thread:
viewtopic.php?f=3&t=1134

I'm not sure if AMD phenom ii X4 965 has four "full" cores.
I'm sure FX-8150 does not have 8 "full" cores as described in the previous post.
That's why i'm thinking maybe FX-8150 has the same performance as AMD phenom ii X4 965 for double precision float number computation.
baddylover
 
Posts: 25
Joined: Thu Apr 07, 2011 6:17 am


Return to Using OpenMP

Who is online

Users browsing this forum: Yahoo [Bot] and 7 guests