https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114052
--- Comment #7 from Jan Hubicka <hubicka at ucw dot cz> --- > I see it doesn't do anything if mark_dfs_back_edges returns false, so it > will claim the function is finite even when it calls a non-finite function? > So I assume this is local analysis only and call edges will be handled > elsewhere? Yes, the side effects flags are transitively closed in callgraph at the IPA propagation stage. > > I found another similar compute, fill_always_executed_in in LIM > (that considers all non-PURE calls as eventually looping, relying > solely on gimple_has_side_effects here). There are stmts with side effects that are always finite. There is stmt_may_terminate_bb_p and stmt_may_terminate_function_p which does more careful checking (and most of it should be unified I guess). I also had patches (never pushed) to track whether given callgraph edge is always executed. This is useful, for example, to bypass inliner heuristics bounds on stack frame size. It is also useful for profile propagation. So having this analysis factored out would be useful here, too. > > In the end I want to have this on a stmt granularity (stmts before > calls are OK, stmts after not). Yep, if you have info whether BB is always executed, you still need to walk its statements. > > I guess greedily walking loop blocks along edges or walking in RPO oder > and tracking whether there's no possible exit on the way to X would be > the way to go. That is what I would do. Maybe it would be nice to see though loops known to be finite, so I guess if you check that it contains no stmts that can terminate execution, then one can consider them safe....
