Iris Xe MAX node is missing double precision support

[tom91136]

Hi

When running SYCL programs (I assume this applies to any OpenCL program as well) on Iris Xe MAX nodes, the runtime doesn't seem to support doubles.

To reproduce:

> qsub -l nodes=1:iris_xe_max:ppn=2 -I
# once job started:
> git clone git@github.com:UoB-HPC/BabelStream.git
> cd BabelStream
> make -f SYCL.make  COMPILER=DPCPP
> ./sycl-stream --list # compiled program is called sycl-stream
> ./sycl-stream --device 2 # pick Iris Xe GPU

The output is as follows:

BabelStream
Version: 3.4
Implementation: SYCL
Running kernels 100 times
Precision: double
Array size: 268.4 MB (=0.3 GB)
Total size: 805.3 MB (=0.8 GB)
Using SYCL device Intel(R) Graphics [0x4905]
Driver: 21.01.18793
Reduction kernel config: 384 groups of size 512
terminate called after throwing an instance of 'cl::sycl::compile_program_error'
  what():  The program was built for 1 devices
Build program log for 'Intel(R) Graphics [0x4905]':

error: double type is not supported on this platform
error: backend compiler failed build.
 -11 (CL_BUILD_PROGRAM_FAILURE)
Aborted

To confirm device capability, running `clinfo` yields the following:

Device Name                                     Intel(R) Graphics [0x4905]
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 3.0 NEO 
  Driver Version                                  21.01.18793
  # -- snip --
  Preferred / native vector sizes                 
    char                                                16 / 16      
    short                                                8 / 8       
    int                                                  4 / 4       
    long                                                 1 / 1       
    half                                                 8 / 8        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (n/a)
  Half-precision Floating-point support           (cl_khr_fp16)
  # -- snip --
  Double-precision Floating-point support         (n/a)
  # -- snip --

 This doesn't seem right,  double precision is supported on gen9 nodes and I was able to run BabelStream correctly.

Perhaps the driver is outdated?

[AthiraM_Intel]

Hi,

Thanks for reaching out to us.

Please export the below environment variables before running the commands.

export OverrideDefaultFP64Settings=1 

export IGC_EnableDPEmulation=1 

If you face any further issue please let us know.

Thanks.

[AthiraM_Intel]

Hi,

Could you please give us an update? Is your issue resolved?

Thanks.

[AthiraM_Intel]

Hi,

 We are assuming that the solution provided helped and would no longer be monitoring this issue. Please raise a new thread if you have further issues.

Thanks.