Intel® oneAPI Base Toolkit
Support for the core tools and libraries within the base toolkit that are used to build and deploy high-performance data-centric applications.
Announcements
The Intel sign-in experience has changed to support enhanced security controls. If you sign in, click here for more information.
358 Discussions

dpct not able to find CUDA header files

FARSHAD_A_Intel
Employee
1,909 Views

I have upgraded to oneAPI Beta 8. When I run dpct, I see a different behavior:

$dpct -p compile_commands.json

The directory "dpct_output" is used as "out-root"
dpct exited with code: -5 (Error: Path for CUDA header files is invalid or not available. Specify with --cuda-include-path)

 

CUDA header files are available through $PATH, $CUDADIR, $CUDA_INCLUDE_PATH and $CUDA_PATH.  

 

When explicitly specifying the path to include files, I see the same behavior:

 

$ dpct --cuda-include-path=/usr/local/cuda/include -p compile_commands.json
The directory "dpct_output" is used as "out-root"
dpct exited with code: -5 (Error: Path for CUDA header files is invalid or not available. Specify with --cuda-include-path)

 

I used to use CUDA 11.0.228 with Beta07. I understand Beta07 supported CUDA 10.1 but since Beta07 worked with CUDA 11, I did not bother to downgrade. I wonder if above error message is due to CUDA version dpct sees or something different?

Labels (2)
0 Kudos
1 Solution
FARSHAD_A_Intel
Employee
1,856 Views

Hello,

 

I downloaded CUDA 10.2 and installed it side by side with existing 11.0 without downgrading drivers, installed binaries, etc. DPCT seem to be happy now and not generating the cryptic error message. thanks.

View solution in original post

7 Replies
RahulV_intel
Moderator
1,892 Views

Hi,

 

As per the latest oneAPI beta08 DPCT requirements, the only supported CUDA versions are: 8.0, 9.x, 10.1, 10.2.

 

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

 

Since the requirements are not met, DPCT will throw an error if you try to migrate.

 

DPCT only needs CUDA headers from the supported versions. So, if you could manage to get headers from these supported versions (given above) and provide its path to the --cuda-include-path flag, there are pretty high chances that it could work, without having to downgrade your current CUDA version.

 

Let me know if it helps.

 

Regards,

Rahul

 

FARSHAD_A_Intel
Employee
1,885 Views
I am suspicious but willing to give it a try. Is there any way I can grab the header files quickly without downloading the entire package? I hope this would be a quick exercise.
RahulV_intel
Moderator
1,877 Views

Hi,


You can give it a try on their Github pages. The other option would be to copy the headers from a machine that has one of the supported CUDA versions installed, on to your machine.


--Rahul


FARSHAD_A_Intel
Employee
1,857 Views

Hello,

 

I downloaded CUDA 10.2 and installed it side by side with existing 11.0 without downgrading drivers, installed binaries, etc. DPCT seem to be happy now and not generating the cryptic error message. thanks.

RahulV_intel
Moderator
1,851 Views

Hi,


Good to know that it works.


Let me know if I can close the thread from my end?



Thanks,

Rahul


FARSHAD_A_Intel
Employee
1,839 Views
RahulV_intel
Moderator
1,824 Views

Thanks for the confirmation.


Intel will no longer monitor this thread. However, it will remain open for community participation.


Reply