https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64164
--- Comment #17 from Richard Biener <rguenth at gcc dot gnu.org> --- To add to all this - IMHO copyrename should go - it's purpose was to have more SSA names with user-decls and thus debug info for them. This should now be dealt with debug insns (in a way better and correct way). Yes, out-of-SSA coalescing could be changed to allow coalescing of SSA names with a user-DECL and anonymous SSA names (or SSA names with a DECL_IGNORED_P decl). But that will make the conflict graph much larger(?). The decision past then was to retain previous behavior which wouldn't have coalesced this case (because the DECL of the now anonymous SSA name was different). In theory we could just go for it, hoping the conflict graph will not get much larger. But it might happen that we end up coalescing things in a way that there ends up being no DECL for a register and thus we lose in debug quality (not sure if we could compensate by inserting debug stmts on the edges to compensate for that - we couldn't do that if it requires splitting the edge). Not sure if always coalescing to the variable with the user-decl won't make debug info wrong, like in bb1: _1 = 2 * _3; goto <bb 3>; bb2: i_4 = 5 / _3; bb3: _3 = PHI <i_4 (2), _1 (1)> if we coalesce to _1 to i_4 then the expression 2 * _3 will be computed to a register with decl 'i'. So to avoid wrong debug we'd have to always coalesce to an anonymous entity - but that would lose the association of 'i' to 5 / _3 in bb2 unless we manage to compensate for that somehow with debug-insns we magically emit during RTL expansion.
