OpenMP Devices runtime library wish list

[jimdempseyatthecove]

While this can be implemented using DPC++ call, it would be more effective if the OpenMP runtime library supported procedures to

enumerate devices (assuming you have more than one)

and

obtain features of the device

name

vendor

is_CPU

is_GPU

max_compute_units

max_clock_frequency

global_mem_size

fp64

...

Jim Dempsey

[JohnNichols]

name = James Dempsey

vendor = his mother, 

is_CPU = has core 9 of course for the weekend

is_GPU = NVIDIA

max_compute_units = one arithmetic sum at a time

max_clock_frequency = he fully uses the 24 hours in a day 

global_mem_size = one human brain, less the loss due to all the human things

fp64= I doubt it

After I post one reply I get

Correct the highlighted errors and try again.

• Maximum flood limit reached.

[jimdempseyatthecove]

fp64= yes for newer Intel GPUs

Platform: Intel(R) oneAPI Unified Runtime over Level-Zero
  Device: Intel(R) Arc(TM) Pro B50 Graphics
    Vendor: Intel(R) Corporation
    Type: GPU
    Max Compute Units: 128
    Max Clock Frequency (MHz): 2600
    Global Memory Size (MB): 15876
    Double Precision Support: Yes
Platform: Intel(R) OpenCL
  Device: Intel(R) Core(TM) i9-14900K
    Vendor: Intel(R) Corporation
    Type: CPU
    Max Compute Units: 32
    Max Clock Frequency (MHz): 0
    Global Memory Size (MB): 65291
    Double Precision Support: Yes
Platform: Intel(R) OpenCL Graphics
  Device: Intel(R) Arc(TM) Pro B50 Graphics
    Vendor: Intel(R) Corporation
    Type: GPU
    Max Compute Units: 128
    Max Clock Frequency (MHz): 2600
    Global Memory Size (MB): 15876
    Double Precision Support: Yes

I can now (selectively) use an Intel GPU for my simulation code.

Jim