[SYCL] Start fully-qualifying our inline kernel names in forward decl…

[erichkeane]

…arations.

This has the ability to make a number easy-to-make mistakes no longer
issues when using named kernels. This causes us to differentiate the
inline-named kernels when they are in different namespaces, which will
hopefully make running into duplicate-kernel-name issues more difficult.

[erichkeane]

@AaronBallman

[elizabethandrews]

Change looks ok to me. Just had a couple of questions to clarify kernel naming requirements.

[erichkeane]

a couple of questions

Which were?

[elizabethandrews]

a couple of questions

Which were?

That's strange. My inline comments disappeared.....

[elizabethandrews]

Can you point me to where this 'technically violates SYCL rules'? Also, what happens when the declarations are in the same namespace but the context is technically different? For e.g.

    // namespace N1 { void foo() { kernel<class K>(...); }}
    // and
    // namespace N1 { void bar() { kernel<class K>(...); }}

[erichkeane]

I think this is supposed to be part of this line:

The kernel name must be forward declarable at namespace scope (including global namespace scope) and may not be forward declared other than at namespace scope. If it isn’t forward declared but is specified as a template argument in a kernel invoking interface, as described in Section 4.9.4.2, then it may not conflict with a name in any enclosing namespace scope.

here: https://www.khronos.org/registry/SYCL/specs/sycl-2020/html/sycl-2020.html#sec:naming.kernels

That at least prohibits your example (since they are the same name in the same enclosing namespace), but I don't know what prevents something like:

namespace N1 { void foo() { kernel<class K>(...); }}
namespace N2 { void foo() { kernel<class K>(...); }}

[erichkeane]

@AlexeySachkov : Can you help with the above standards interpretation there?

[elizabethandrews]

Ok. Thank you for clarification

[AaronBallman]

Some small nits from me, but otherwise LG.

[erichkeane]

@bader : Looks like the Lit_With_Cuda bot is out of disk space?

[premanandrao]

LGTM

[bader]

@bader : Looks like the Lit_With_Cuda bot is out of disk space?

Yes. I've seen this issue in my PR as well and I reported it already.
@DoyleLi, @tfzhu, FYI.

[bader]

Yes. I've seen this issue in my PR as well and I reported it already.
@DoyleLi, @tfzhu, FYI.

FYI: I've disabled the machine with that problem to avoid future problems.
Now we have only one worker to process all the requests, but I hope there will be no "false negatives" anymore.

[bader]

Also, I've restarted the Lit_With_Cuda job and results are expected in 1.5-2 hours.