Hi SYCL developers,
In the SYCL 1.2.1 manual, some atomic math member functions require the data type T != float.
We are using SYCL to write Op Kernel for deep learning framework, and floats are the main data type we are dealing with, I am wondering if there is a plan to add supports for floats?
Thanks!
Screenshot from 2020-04-27 10-10-51

Hi,

The Khronos SYCL working group is definitely aware of this, and I think that you should expect to see support at some point in the future.

There is an extension that adds additional floating point atomic support (amongst other things) at SYCL_INTEL_extended_atomics. Does that extension enable what you’re looking for?

Thanks,
Mike

Hi Mike,

Thanks for pointing me to the extension, it has what I am looking for, but not sure its included by DPCPP or not, let me check it out.

Thanks again.

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.