Visual Fortran 10.2 Cuda

Greetings.
I have a problem with Fortran 10.2 and CUDA. Explain how to start an example please http://www.pgroup.com/lit/samples/matmul.CUF
or another, I newbie =)
installed Win7 x64, Visual Studio 2008 prof, PGI Visual Fortran 2008 10.2, Cuda toolkit 2.3 and Cuda SDK 2.3

Hi DarkMatter,

The basic steps are as follows:

  1. Open PVF
  2. Create a new project, selecting “PGI Visual Fortran -> x64” and “empty project”. (or Win32 for 32-bits).
  3. In the solution explorer, add matmul.CUF to the source files.
  4. Right-mouse click on the project properties, select “Fortran->Language”, and then change the “CUDA Fortran” drop-down to “yes”.
  5. Select the menu option “Build->Build Solution” to build the executable.
  6. Finally, select the menu option “Debug->Start Without Debugging” to run the executable.

Let us know if you encounter any issues.

  • Mat

Hello.
Thanks for the answer.
Yes, I so did, result such:

arrays sized 512 by 1024 by 512
calling mmul
Kernel time excluding data xfer: 0.000000 microseconds
Megaflops excluding data xfer: Inf
Total time including data xfer: 75000.00 microseconds
Megaflops including data xfer: 3579.139
C(1,1) = 0.000000
C(2,2) = 0.000000
C( 1 , 1 ) = 0.000000 should be
3.5791870E+11 error= 1.000000
C( 2 , 1 ) = 0.000000 should be
3.5792414E+11 error= 1.000000
C( 3 , 1 ) = 0.000000 should be
3.5792929E+11 error= 1.000000
C( 4 , 1 ) = 0.000000 should be
3.5793420E+11 error= 1.000000
C( 5 , 1 ) = 0.000000 should be
3.5794000E+11 error= 1.000000
C( 6 , 1 ) = 0.000000 should be
3.5794498E+11 error= 1.000000
C( 7 , 1 ) = 0.000000 should be
3.5795023E+11 error= 1.000000
C( 8 , 1 ) = 0.000000 should be
3.5795563E+11 error= 1.000000
C( 9 , 1 ) = 0.000000 should be
3.5796101E+11 error= 1.000000
C( 10 , 1 ) = 0.000000 should be
3.5796596E+11 error= 1.000000
262144 ERRORS FOUND!!!

Hi DarkMatter,

What type of NVIDIA card do you have?

If it’s a card that only supports Compute Capability 1.1, then you must select this CC version in the Fortran->Language Properties tab. (Note that in the next release, 10.3, the compiler will auto-detect at runtime with CC version to use)

If you have CC1.3, then most likely you have a configuration issue with your device, such as an older driver. To use CUDA Fortran you need a driver that is CUDA 2.3 capable (such as >190).

Hope this helps,
Mat

Hi.
Video card 8800 GTS, I think it CC 1.1.
Some minutes later: yes, works without errors.
Big thanks!
P.S. when version 1.3 will be accessible?

Hi DarkMatter,

when version 1.3 will be accessible?

CC 1.3 is available now on several NVIDIA cards. PGI 10.3 will be available early next week.

  • Mat