Add sycl = ::cl::sycl namespace alias

[psalz]

SYCL 2020 introduces a new canonical namespace, ::sycl, which is defined within the new <sycl/sycl.hpp> header. To ease the transition of the CTS to the new namespace, this creates an alias sycl = cl::sycl, while keeping the SYCL 1.2.1 header path <CL/sycl.hpp> for the time being.

For the record, I also tried going the reverse route, i.e., including <sycl/sycl.hpp>, and defining namespace cl { namespace sycl = ::sycl; }. However that didn't work, as DPC++ already exports that alias (while e.g. hipSYCL doesn't - the spec isn't super clear on this), and ComputeCpp 2.4 doesn't seem to support the new header at all.

[bader]

To clarify: I've added "Agenda" tag to discuss this description section:

For the record, I also tried going the reverse route, i.e., including <sycl/sycl.hpp>, and defining namespace cl { namespace sycl = ::sycl; }. However that didn't work, as DPC++ already exports that alias (while e.g. hipSYCL doesn't - the spec isn't super clear on this), and ComputeCpp 2.4 doesn't seem to support the new header at all.

[keryell]

This would work with triSYCL.

[bader]

This would work with triSYCL.

@keryell, can we interpret this comment as "Approved" from your side? :)

[keryell]

This would work with triSYCL.

@keryell, can we interpret this comment as "Approved" from your side? :)

Not really.
In theory yes for careful implementation with small footprint not trying to pollute all the namespaces around. :-)
Do you plan to update DPC++ accordingly to reduce its namespace pollution?
Perhaps we can discuss this inside the SYCL committee meeting.

[keryell]

I have the feeling that a CTS should test exactly what is in the spec and not define some aliases or new indirect namespaces because it might have some side effect, like noticed by @psalz.

[keryell]

Do we need to validate in the CTS that cl::sycl works as a deprecated compatibility mode?
I do not think we need.
Then, why not just changing in the CTS 2020 all the cl::sycl:: by sycl:: and do not set up an alias?

[psalz]

Then, why not just changing in the CTS 2020 all the cl::sycl:: by sycl:: and do not set up an alias?

That is of course also an option, I just thought that a more gradual change would reduce noise in the repo history. But it's all the same to me, really.

[keryell]

By looking at it again:

For the record, I also tried going the reverse route, i.e., including <sycl/sycl.hpp>, and defining namespace cl { namespace sycl = ::sycl; }. However that didn't work, as DPC++ already exports that alias (while e.g. hipSYCL doesn't - the spec isn't super clear on this), and ComputeCpp 2.4 doesn't seem to support the new header at all.

is the right hack to use for now, before changing globally cl::sycl:: to sycl::.
DPC++ should not pollute the global namespace and ComputeCpp should support <sycl/sycl.hpp>, right?