Problem using ndrange with Intel Graphics

[QM]

Hi,

I am using DPC++ to enable parallelization for my code. Using sycl::ndrange<1>{N, worker_number} works totally fine when run on CPU (using sycl::host_selector{}). However, when I switch to Intel graphics by using sycl::default_selector{}, the errors are caused by the use of ndrange ( switch to range approach works). Does anybody know what's wrong with it? 

Here is details of my dpcpp version: 

-Intel(R) oneAPI DPC++/C++ Compiler 2022.0.0 (2022.0.0.20211123)

-Target: x86_64-pc-windows-msvc

-Thread model: posix

Thanks,

Q

[QM]

I have posted this question to Intel OneAPI DPC++ forum as I believe it's more related there. But if anybody know the issue, please let me know.  Thank you.

[Ben_A_Intel]

Hi, what are your nd_range dimensions?  Specifically, what is the value of worker_number in the code snip above?

GPUs (and other types of accelerators) have limits on the maximum size of the local range (AKA "work-group size"), and if you are exceeding the device maximum you will get an error.

If you're sure that the value of worker_number is supported by your GPU can you please send additional information about the error you are seeing?  Thanks!

[QM]

Hi Ben,

Thanks for your reply,  local range is previously set at 64, then I changed it to 1, but still does not work. The error message is blank, yeah funny, but honestly, blank. Switching to sycl::host_selector{} and it works perfectly.

Regards,

Q 

[Ben_A_Intel]

Can you run your test with oneTrace enabled, especially with the CallLogging option ("-c")?  This might provide more hints why the error is occurring.

https://github.com/intel/pti-gpu/tree/master/tools/onetrace

If you can send the exact error you are seeing this will help also.  Thanks!

[SantoshY_Intel]

Hi,

We have not heard back from you. As suggested by @Ben_A_Intel , could you please try running your test with oneTrace and send the exact error you are seeing?

Thanks & Regards,

Santosh

[SantoshY_Intel]

Hi,

We have not heard back from you. This thread will no longer be monitored by Intel. If you need further assistance, please post a new question.

Thanks & Regards,

Santosh