Device is supported : NO (GTX1080/Xeon 5122)

1

Should ComputeCPP community edition work on my Fedora 33 Xeon 5122 GTX1080 System?

I thought it would since I have CUDA HPC SDK working(in path by default), and Clang 11 OpenMP offload working(not in path by default).

When I try computecpp_info I see:

 ********************************************************************************

ComputeCpp Info (CE 2.3.0 2020/11/19)

SYCL 1.2.1 revision 3

********************************************************************************


Device Info:

Discovered 2 devices matching:
  platform    : <any>
  device type : <any>

--------------------------------------------------------------------------------
Device 0:

  Device is supported                     : NO - Please refer to the website for more info about the PTX backend on the Platform Support page
  Bitcode targets                         : ptx64 
  CL_DEVICE_NAME                          : GeForce GTX 1080
  CL_DEVICE_VENDOR                        : NVIDIA Corporation
  CL_DRIVER_VERSION                       : 460.27.04
  CL_DEVICE_TYPE                          : CL_DEVICE_TYPE_GPU 
--------------------------------------------------------------------------------
Device 1:

  Device is supported                     : NO - No bitcode target found
  Bitcode targets                         : 
  CL_DEVICE_NAME                          : pthread-Intel(R) Xeon(R) Gold 5122 CPU @ 3.60GHz
  CL_DEVICE_VENDOR                        : GenuineIntel
  CL_DRIVER_VERSION                       : 1.5
  CL_DEVICE_TYPE                          : CL_DEVICE_TYPE_CPU 

If you encounter problems when using any of these OpenCL devices, please consult
this website for known issues:
https://developer.codeplay.com/products/computecpp/ce/guides/platform-support?version=2.3.0

********************************************************************************
2

Apologies for the delay in responding to this, it slipped through without me seeing it.

ComputeCpp has discontinued support for Nvidia GPUs, so whilst the compiler option is still enabled we make no guarantees.

Our recommendation is to use the Nvidia implementation we have been working on for the DPC++ project. There are instructions for building DPC++ with this enabled here.

I’d also recommend watching this video if you have the time to get an introduction to the work. We will be publishing roadmap information soon on the continued development of this.

3

DPC++ cannot be installed without an Intel graphics driver - which cannot be installed without an Intel CPU that has onboard Graphics.

Any other options for Nvidia device support? Or AMD support for that matter? ( hipSYCL is Linux only - otherwise I’d use that.)

Thanks!

4

Hi Andrew,
You would need Intel drivers if you want to use DPC++ with Intel hardware. If you want to use DPC++ with the CUDA/Nvidia device support we have added you don’t. The minimum requirement is CUDA 10.2.
However we have not done any support for Windows yet and it sounds like that is something you are looking for? You could use the WSL as an alternative way to use Linux on your Windows machine but at the moment you need an “insider” build as described on the Nvidia website here.
Cheers,
Rod.