Move cl::sycl::memory_order, atomic class and atomic operations to sycl namespace

[VerenaBeckham]

Developers can use sycl::memory_order.

[bader]

Should we add a note about this change to https://github.com/KhronosGroup/SYCL-Docs/blob/SYCL-2020/master/adoc/chapters/what_changed.adoc?

[gmlueck]

I don't think this PR changes anything relative to SYCL 1.2.1. We just removed a redundant definition of the memory_order enumeration from the spec. The enumeration is still available in SYCL 2020, and it is still defined elsewhere in the spec.

[AerialMantis]

I think this change makes sense, thanks! Though on it's own I this will be a breaking change for the deprecated atomic class as that relies on the definition of the memory_order in the cl::sycl namespace.

Technically in SYCL 2020 the sycl and cl::sycl namespaces are synonymous with each other if the CL/sycl.hpp header is included, so the intention is that this would be equivalent to having this in the sycl namespace, but having an interface still explicitly defined in the cl::sycl namespace suggests this is an exception to the rule.

From what I can tell this synopsis, the following synopsis for deprecated atomic free functions and a couple of member functions in the accessor synopsis are the only places where the cl::sycl namespace is still explicitly used, I think we should extend this PR to also switch from cl::sycl to sycl here to avoid any confusion.

[gmlueck]

From what I can tell this synopsis, the following synopsis for deprecated atomic free functions and a couple of member functions in the accessor synopsis are the only places where the cl::sycl namespace is still explicitly used, I think we should extend this PR to also switch from cl::sycl to sycl here to avoid any confusion.

I think this makes sense. At one point, we discussed the possibility of structuring the header file so that deprecated APIs are only available when <CL/sycl.hpp> is included, but that never made it into the spec. I suspect the fact that the atomic interface is documented in the cl::sycl namespace is a holdover from that thinking. At this point, there is no correlation between deprecated APIs and the cl::sycl namespace, so I think we should just document atomic in the sycl namespace.

[VerenaBeckham]

I also moved the atomic class and atomic operations into the sycl namespace.

[gmlueck]

I also moved the atomic class and atomic operations into the sycl namespace.

Thanks, this looks good!

[keryell]

Thanks!

[AerialMantis]

Looks good, thanks!

[tomdeakin]

One more review.

[tomdeakin]

WG approved to merge