[rocprofiler-systems] Add Fortran main detection to rocprof-sys-instrument to avoid instrumenting around C "main" wrapper

[kcossett-amd]

Motivation

Currently, rocprof-sys-instrument will force the main_func variable to be either main or _main. This variable essentially dictates the primary function to instrument around.

In Fortran programs, there is usually a C-style main which will call the Fortran main. This is usually not a problem as it will consider the Fortran main as a subroutine and instrument it. However, in the case where the Fortran main is non-returning (i.e. Dyninst determine that its FuncReturnStatus == NORETURN), it will not be instrumented. (See SWDEV-558265 for more details).

The solution is to check if any of the Fortran main symbols exist and use them if possible.

Technical Details

Added check for Fortran main function. Uses regex.

• I will note that Cray compilers are very annoying here. The symbol name given to the Fortran main is the same as what the user uses in their PROGRAM statement. HOWEVER: The cray compiler will generate a "main" symbol with the same start address as the fortran main, fixing any issues. Essentially, in cray fortran, it instruments the fortran main by default.

Remove unused variable _init_arg0.

Test Plan

Tested using:

• openmp-vv. With amdflang, it will now center around _QQmain (Fortran main symbol for LLVM compilers).
• intervals_mpi.f90 (this is part of the MPI CTest PR and is the original program where this issue was discovered). With gfortran, binary-rewrite will now center around MAIN__ (Fortran main symbol for gfortran).
• Custom hello.f90 program that forces the Fortran main to have FuncReturnStatus == NORETURN. With gfortran, it now centers itself around MAIN__.

Test Result

• For openmp-vv, the omp traces are still present and the CTests pass.
• For intervals_mpi.f90, the CTest passes and works on every workflow.
  • NOTE: To see this, please take a look at PR $ , it basically contains the same code here.
• For the custom program, I now see the MPI traces.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[dgaliffiAMD]

Thanks, @kcossett-amd. Your change looks good.

A couple additions that I would like to suggest:

• Add to the CHANGELOG
• Your note about the Cray compiler would be good to add to the document. There is a "Troubleshooting" section at the end of https://github.com/ROCm/rocm-systems/blob/develop/projects/rocprofiler-systems/docs/how-to/instrumenting-rewriting-binary-application.rst.
• If you haven't already added it to your other PR, I would recommend adding an example & test that would have fallen into this failing case.

[kcossett-amd]

@dgaliffiAMD The cray compiler issue was resolved (it generates "main" symbol that has same address as the Fortran main. So for PROGRAM hello, "hello_" and "main" have same address). Still, I added a general note to troubleshooting reminding the user that --main-function option exists.

[dgaliffiAMD]

Looks good to me.
@ROCm/rocm-documentation, please review documentation changes.

[prbasyal-amd]

LGTM