This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tuples] Tuplify invariant motion
>>>>> "Zdenek" == Zdenek Dvorak <rakdver@kam.mff.cuni.cz> writes:
> Hi,
> this patch enables lcSSA form creation and tuplifies lim, as well as
> fixing a few things to make it work:
> 1) gimple_could_trap_p is factored out of tree_could_trap_p.
> 2) the remove_eh_info argument of gsi_remove is renamed to make it
> clear that it really means "do not destroy anything permanent,
> I am going to reinsert this statement elsewhere"; and we now
> do not free the operands of the statement if this argument is false.
> 3) gimple_find_edge_insert_loc is made to use gsi_last_bb, so that
> commit_edge_insertions updates stmt->basic block mapping correctly
> The patch causes one misscompilation (gcc.c-torture/execute/pr34176.c)
> in the testsuite (virtual operands are wrong, making lim perform an
> illegal transformation). I will try to fix this, but I would like to
> commit the bulk of the patch now (possibly with lim disabled, if
> preferable) in order to avoid having to fix conflicts with other
> changes. OK?
Ok, provided you look into the misscompilation :).
Thanks.
Aldy