Can ifx in OneAPI HPC toolkit compile openmp codes which offloads to NVIDIA GPU

[victor_888888]

Can ifx in OneAPI HPC toolkit compile openmp codes which offloads to NVIDIA GPU?

when I compile openmp codes with ifx with the command"ifx -qopenmp -fopenmp-targets=spir64 test.f90 -o test.exe" and offload it to NVIDIA GPU "./test.exe".

There is an error of "./test.exe: error while loading shared libraries: libomptarget.so: cannot open shared object file: No such file or directory".

Thank you very much!

[ShivaniK_Intel · Accepted solution]

Hi,

Thanks for reaching out to us.

>>>" Can ifx in OneAPI HPC toolkit compile OpenMP codes which offloads to NVIDIA GPU?"

ifx(Intel Fortran compiler(Beta)) is a part of the Intel oneAPI HPC Toolkit that supports only Intel discrete and integrated GPUs.

>>"ifx -qopenmp -fopenmp-targets=spir64 test.f90 -o test.exe"

When you specify spir64, the compiler generates an x86 + SPIR64 (64-bit Standard Portable Intermediate Representation) fat binary for Intel® GPU devices.

For more information refer to the below link:

https://www.intel.com/content/www/us/en/develop/documentation/oneapi-dpcpp-cpp-compiler-dev-guide-and-reference/top/compiler-reference/compiler-options/compiler-option-details/offload-openmp-and-parallel-processing-options/fopenmp-targets-qopenmp-targets.html

Could you please let us know whether you have any other issues?

Thanks & Regards

Shivani

[victor_888888]

Thanks a lot. You have solved my question. 

[ShivaniK_Intel]

Hi,

Thanks for accepting our solution. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Thanks & Regards

Shivani

[JohnNichols]

Intel said:

ifx(Intel Fortran compiler(Beta)) is a part of the Intel oneAPI HPC Toolkit that supports only Intel discrete and integrated GPUs.

--------------------------------------------------------------

Considering that the low end Dell machines for instance use Intel Graphics, but the high end machines typically use the Quadro Boards for the lap tops, and some one who is a serious FORTRAN user is likely to use the higher end computers, perhaps Intel should support at least NVIDIA.  

Just a thought --  but my reason for this is if you mix FFT, AI and producing drawing output, you are looking at a 2 hour run time on a good machine for 15 data sets of 3million records each.  Of course you can use the machine for other things, but .... 

[Steve_Lionel]

The stated intention is that oneAPI will support GPUs and accelerators from multiple vendors. That hasn't happened yet.

[JohnNichols]

Thanks,