[SYCL] Enable diagnosing of indirect implicit capture of "this"

[Fznamznon]

The compiler issues an error if a lambda passed directly to a SYCL kernel
function implicitly captures "this", but does not issue an error if the
lambda that captures "this" is itself captured by the kernel lambda.
The situation when SYCL kernel function captures lambda is a normal
workflow for SYCL reductions, so it is important to diagnose this case
as well.

[Fznamznon]

I've reported an issue with CI to tools team. I don't think that a small change in FE somehow could make llvm build hang.

[smanna12]

LGTM.

[Fznamznon]

BTW, the previous commit passed CI checks. It lacks only comment in the test.

[premanandrao]

LGTM, one question: does this work within nested lambdas?

[Fznamznon]

LGTM, one question: does this work within nested lambdas?

Yes, this should work with nested lambdas. I.e. when kernel lambda captures lambda that captures "this" - we issue an error. Or you mean something different? Or just one more level of nesting?

[premanandrao]

LGTM, one question: does this work within nested lambdas?

Yes, this should work with nested lambdas. I.e. when kernel lambda captures lambda that captures "this" - we issue an error. Or you mean something different? Or just one more level of nesting?

I did mean one more level of nesting. Could you add a test case with one more level of nesting (and the data member being captured only within the nested one)?

[Fznamznon]

LGTM, one question: does this work within nested lambdas?

Yes, this should work with nested lambdas. I.e. when kernel lambda captures lambda that captures "this" - we issue an error. Or you mean something different? Or just one more level of nesting?

I did mean one more level of nesting. Could you add a test case with one more level of nesting (and the data member being captured only within the nested one)?

Sure, thank you for the explanation. I've added it.