http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60381
Alexandre Oliva <aoliva at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|ASSIGNED |UNCONFIRMED
Ever confirmed|1 |0
--- Comment #10 from Alexandre Oliva <aoliva at gcc dot gnu.org> ---
Fixed by reversal of the faulty patch. We should eventually tweak
vt_add_function_parameter so as to preserve VALUEs recursively, adding
dataflow_set binds for all of them, like we do for top-level MEMs (and their
addresses) and REGs. We'll probably have to work harder to preserve these
intermediate equivalences in spite of the removal of the base VALUEs when their
REGs are reset and their loc lists become empty. Just marking the VALUEs as
preserved won't stop that, and when we get to preserving them in the separate
preserved table, they'll be gone already. Regarding preserved values as
non-useless will likely cause lots of values to be preserved that currently
aren't. This may very well get us better debug info, but I'm not sure at what
cost.
