DPC++ compatibility tool error, oneAPI samples, Devcloud

[Eltablawy__Alaa]

Hello,

I'm trying to run one of the oneAPI samples for Cuda code migration on the devcloud.

The code I'm running is https://github.com/oneapi-src/oneAPI-samples/tree/master/Tools/Migration/folder-options-dpct

When I run the command in the README.md, I get the following:

Case1:

dpct --in-root=foo --out-root=result/foo foo/main.cu foo/bar/util.cu --extra-arg="-Ifoo/bar/"

dpct exited with code: -5 (Error: Path for CUDA header files is invalid or not available. Use --cuda-include-path to specify the correct path to the header files)

When I add the path to include/cude, I get the following:

Case2:

dpct --cuda-include-path=<path to cuda>/include --in-root=foo --out-root=result/foo foo/main.cu foo/bar/util.cu --extra-arg="-Ifoo/bar/"

Processing: /home/u52/workstation/CudaToDPC++/oneAPI-samples/Tools/Migration/folder-options-dpct/foo/main.cu
Skipping /home/u52/workstation/CudaToDPC++/oneAPI-samples/Tools/Migration/folder-options-dpct/foo/main.cu. Compile command not found.
Processing: /home/u52/workstation/CudaToDPC++/oneAPI-samples/Tools/Migration/folder-options-dpct/foo/bar/util.cu
Skipping /home/u52/workstation/CudaToDPC++/oneAPI-samples/Tools/Migration/folder-options-dpct/foo/bar/util.cu. Compile command not found.
dpct exited with code: 1 (Migration not necessary; no CUDA code detected)

is the DPC++ compatibility tool supported on Devcloud? and is there any documentation for using the tool on the Devcloud?

[RahulV_intel]

Hi,

I have tried the steps (given in the tutorial) in my environment. The migration and compilation/run were successful in my case.

Could you please specify your CUDA toolkit?

DPCT requirements:

https://software.intel.com/content/www/us/en/develop/articles/intel-dpc-compatibility-tool-system-requirements.html

Thanks,

Rahul

[Eltablawy__Alaa]

Thanks, Rahul for your answer, I used the toolkit include from 10.2. 

Which toolkit you used in your env. ?

Regards,

Alaa

[RahulV_intel]

Hi,

Apologies for the late response. I have the same CUDA toolkit version.

As a possible workaround, could you try changing your CUDA application to a different directory and give it a try? (CUDA toolkit installation directory and CUDA application directory should be different). Let me know if it helps.

Also, try to migrate a simple CUDA vector add code and let me know how it goes.

dpct --cuda-include-path=<path to cuda>/include vecadd.cu

Thanks,

Rahul

[RahulV_intel]

Hi,

Could you please confirm if your issue is solved? Let us know if you face any issues.

Thanks,

Rahul

[Eltablawy__Alaa]

Thanks Rahul, yes, I have the issue resolved. The solution to my issue us as follow:

1- as you suggested, changing the CUDA application directory helped in resolving the following issue:

Processing: /home/.../oneAPI-samples/Tools/Migration/vector-add-dpct/src/vector_add.cu
Skipping /home/.../oneAPI-samples/Tools/Migration/vector-add-dpct/src/vector_add.cu. Compile command not found.
dpct exited with code: 1 (Migration not necessary; no CUDA code detected)

2- Changing the CUDA application path not to be the same as CUDA lib, helped with that issue:

dpct exited with code: -5 (Error: Path for CUDA header files is invalid or not available. Use --cuda-include-path to specify the correct path to the header files)

So I used the following command to migrate: dpct --cuda-include-path=../include --in-root=. src/vector_add.cu --out-root=output/

Also, without specifying the --in-root option, it produced the same output.

Thanks for your help! 

[RahulV_intel]

Hi,

Since your issue is resolved, could you let me know if I can close this thread from my end?

Thanks,

Rahul

[Eltablawy__Alaa]

Sure, it's resolved, and you can close the issue.

Thanks!

Alaa

[RahulV_intel]

Hi,

Thanks for the confirmation. Intel will no longer monitor this thread. Further discussions on this thread will be considered community only.

Regards,

Rahul