[llvm] [Hexagon] Have fast-math-libcalls.ll use the hexagonv65 cpu (PR #149647)

[Brian Cain via llvm-commits]

================

----------------
androm3da wrote:

Instead of replacing the target CPU, why not just add a new RUN line?

That way you get no reduction in coverage. 

Btw - did I understand that using v68 changes the behavior wrt when the compiler emits library calls?  Is that expected?  I always assumed the differences among these arch revisions were pretty mild. Though I guess if v68 added a relevant opcode we might not need to call for those?

https://github.com/llvm/llvm-project/pull/149647