Prevent Clang/LLVM from inlining or removing MPIR_Breakpoint as dead function

[kiranchandramohan]

Many parallel debuggers depend on the MPIR_Breakpoint function. Clang/LLVM can inline or remove MPIR_Breakpoint as dead function. Can the optnone attribute be added to the MPIR_Breakpoint function to prevent it from being optimized away ?

file : orte/orted/orted_submit.c

#ifdef __clang__
__attribute__((optnone))
#endif
ORTE_DECLSPEC void* MPIR_Breakpoint(void);

[kiranchandramohan]

Thanks Ralph.

[ghost]

Thanks @rhc54 . If I understand the merge #4626 correctly this is only in master so far, but not in 3.1.x or 3.0.x. Any chance to backport this into 3.0.x and 3.1.x, as otherwise hitting MPIR_Breakpoint won't work in those versions when compiled with Clang/LLVM?

[rhc54]

Done: #4715 and #4716

[ghost]

Thanks Ralph.