Make tuple base public.

[rchen20]

While the test-kernel-single-loop-ForICount-8-OpenMPTarget test in RAJA (llnl/RAJA#1207), there was a compilation error in CAMP tuple. I'm not sure if this is the right solution, but making the base tuple_helper class publicly accessible solved the build error (although the test still failed with XLC).

Note: This solution does not solve the clang internal error llnl/RAJA#1216.

Error:
In file included from /usr/WS1/chen59/allraja/rajaatomicexhaustive/raja_git_atomicexhaustive/build_lc_blueos-xl_omptarget-2021.12.22/test/functional/kernel/single-loop-tile-icount-tcount/test-kernel-single-loop-ForICount-8-OpenMPTarget.cpp:11:
In file included from /usr/WS1/chen59/allraja/rajaatomicexhaustive/raja_git_atomicexhaustive/test/include/RAJA_test-base.hpp:15:
In file included from /usr/WS1/chen59/allraja/rajaatomicexhaustive/raja_git_atomicexhaustive/include/RAJA/RAJA.hpp:33:
In file included from /usr/WS1/chen59/allraja/rajaatomicexhaustive/raja_git_atomicexhaustive/include/RAJA/util/camp_aliases.hpp:31:
/usr/WS1/chen59/allraja/rajaatomicexhaustive/raja_git_atomicexhaustive/tpl/camp/include/camp/tuple.hpp:154:60: error: 'base' is a private member of 'camp::tuple<RAJA::TypedRangeSegment<unsigned long long, long long> >'
return static_cast<tpl_get_store<Tuple, index> const&>(t.base).get_inner();
^

[rchen20]

Tagging @trws @rhornung67 because I cannot add reviewers.

[trws]

Ok... I can see why that would fix that error, but that error is very, very wrong. The get function is a friend function of tuple, so that should all be kosher. Would you be willing to try making the friend functions inline into the class and see if that also works around it rather than exposing the base?

I know we haven't really done a great deal to preserve API or ABI, but that one might be nasty to fix later if it goes public now.

[rchen20]

Ok... I can see why that would fix that error, but that error is very, very wrong. The get function is a friend function of tuple, so that should all be kosher. Would you be willing to try making the friend functions inline into the class and see if that also works around it rather than exposing the base?

I know we haven't really done a great deal to preserve API or ABI, but that one might be nasty to fix later if it goes public now.

I'm not sure what you meant, but I put the get functions directly inside tuple, and that did not work. Also tried adding template<typename... Elements> friend struct tuple to each of tuple_helper, and tuple_storage, but that also did not work.

[rchen20]

@trws Looks like Kristi is running into this same issue on the RAJA CI with XL. https://lc.llnl.gov/gitlab/radiuss/RAJA/-/jobs/561894

[trws]

Ok, we need this for the release, but I really do not want to leave it this way. Would you add a TODO and an issue to come back and fix this?

[rchen20]

The clang docker instance is not starting up properly. Apart from that, this should be ready to go. I added a comment to the code to remind us to change this, and the issue is up as well.

[trws]

The clang thing is an upstream issue, I have a fix for it in another live PR, but this is cleared by the other clang. Thanks @rchen20.