Complete implementations of GPU basic reducer constructor test

[rchen20]

Summary

• This PR is a bugfix
• It does the following:
  • The basic reducer constructor test was not being compiled for the GPU due to lack of an instantiation of the test. This particular test should not be runnable on the GPU, but Gtest requires an implementation anyway. Hence, the implementation is empty.
  • Should help resolve CI pipeline failures in Renew CI specs per LLNL/radiuss-shared-ci#15 #1434 and Fix and activate vectorization tests with forall. #1289.

[rhornung67]

@rchen20 please don't merge this. Something is more broken here than it is on develop.

[rchen20]

@rchen20 please don't merge this. Something is more broken here than it is on develop.

I won't merge yet, but I doubt this PR is causing the linker warnings on Tioga because nothing here touches that. I'm running the CI on develop in another PR to check the Tioga output that we should expect #1476.

[rchen20]

@rchen20 please don't merge this. Something is more broken here than it is on develop.

I won't merge yet, but I doubt this PR is causing the linker warnings on Tioga because nothing here touches that. I'm running the CI on develop in another PR to check the Tioga output that we should expect #1476.

@rhornung67 Comparing the Tioga output from #1476 and this PR, they are roughly the same, except the failing test now passes. What occurs is that the ROCM/5.1.1 tests build, run, and pass with the linker warning spew. In the subsequent ROCM/5.2.3 build, the linker spew exceeds the log limit. This PR does not seem to affect the Tioga pipeline's excessive readelf warnings.

[adrienbernede]

Merged in #1434 and working. Thank you!

[adrienbernede]

@rhornung67 I think this could be merged into develop regardless of the CI branch. This PR addresses a bug affecting others (and develop).

[rhornung67]

Right. Robert will merge it. Later today, I'm going to look into the Spack config file that is generated for the build space issues. When I manually run CMake with the ROCm versions used on corona and tioga and compile, I don't see any of those warnings. I have a hunch that we may need to fiddle with the Spack spec for those builds.