Look at inline vs force inline performance

[MrBurmark]

Jason Turner pointed out that force inline may cause the compiler to use more instructions and hence stress the instruction cache.
This could hurt performance, so we may want to evaluate where force inline helps and hurts with our current compilers, and then enable or disable it accordingly.

[rhornung67]

We have a macro constant RAJA_FORCEINLINE_RECURSIVE defined based on compiler in config.hpp.in; for example, https://github.com/LLNL/RAJA/blob/develop/include/RAJA/config.hpp.in#L392. Whether the macro constant is defined depends on the CMake configuration variable RAJA_ENABLE_FORCEINLINE_RECURSIVE.

The only RAJA files where the macro is used are include/RAJA/pattern/forall.hpp and include/RAJA/policy/MultiPolicy.hpp