Intel® DevCloud
Help for those needing help starting or connecting to the Intel® DevCloud
1625 Discussions

Intel Iris Xe MAX discrete GPU on DevCloud?

Colin_Reinhardt
Beginner
962 Views

The Xe MAX info online says that the discrete GPU is available on DevCloud for testing/evaluation.  I would like to run some native OpenCL kernels on it.
Can I compile and run OpenCL kernels on DevCloud and specify/select it as the target OpenCL device?

Can anyone clarify specifically how I should do this via OpenCL API, e.g. what platform name/ID and device name/ID should I use?

Do any special compile flags need to be specified?

Thanks, Colin 

0 Kudos
3 Replies
ChithraJ_Intel
Moderator
949 Views


Hi Colin,


Thanks for posting in Intel Community Forums.

As going through your query, we could understand that you are trying to execute OpenCL Kernels from SYCL/DPC++ programs or simply, OpenCL-C kernel ingestion through a SYCL* program. In Devcloud, it is possible to perform OpenCL kernel ingestion through a SYCL/DPC++ programs while keeping your kernel code in OpenCL itself. So that, you can compile and execute the OpenCL kernel in SYCL environment and offload it on any device with specific device selector. Here is what you can try to solve your queries:

1. Initially, request Iris Xe MAX GPU's node available in devcloud. You can connect the node with any of the below commands.

qsub -I -l nodes=1:dual_gpu:ppn=2 (or) qsub -I -l nodes=1:quad_gpu:ppn=2

2. To compile and execute the OpenCL kernel in a SYCL environment, Please refer to below documentations:

A)"OpenCL-C kernel ingestion program" part in this article: https://software.intel.com/content/www/us/en/develop/articles/interoperability-dpcpp-sycl-opencl.html

B)"Executing OpenCL kernels from DPC++ programs" in 

https://software.intel.com/content/www/us/en/develop/articles/migrating-opencl-designs-to-dpcpp.html#inpage-nav-4-undefined

This documentation includes sample programs with build and run steps.

Note: There might be some issues while executing the above program. Because, to execute this program there should be OpenCL backend but by default, accelerators use Level0 backend. So you need to change the underlying backend to execute this program. As a work around, try the below command before running the executable:

export SYCL_BE=PI_OPENCL


Hope this helps for you. Please let us know in case of issues/queries.


Regards,

Chithra


0 Kudos
ChithraJ_Intel
Moderator
930 Views

Hi Colin,


Have you tried out the given solution? Is your issue resolved? Let us know the updates on this.


Regards,

Chithra


0 Kudos
ChithraJ_Intel
Moderator
908 Views

Hi Colin,


We haven't heard back anything from you. We would discontinue monitoring this thread by assuming that your issue got resolved. Please feel free to raise a new thread in case of further issues/queries.


Regards,

Chithra


0 Kudos
Reply