On Fri, Oct 4, 2013 at 6:43 PM, Andrew MacLeod <amacl...@redhat.com> wrote: > This patch mostly re-factors tree-flow-inline.h out of existence. > > the gimple-df data structure has found anew home in gimple-ssa.h, and this > actually seems like a very appropriate place for it as it holds a lot fo the > ssa specific stuff in it. > > The remaining inline functions in tree-flow-inline.h are spread to the wind > a bit. > - 2 were no longer used, s they are deleted. > - The various stmt_uid functions went to gimple.h, along with some stats > macros and misc stuff. didn't seem to be a more appropriate place right now > based on their usage patterns. > - tree-hasher.h was including tree-flow.h simply to get the definition of > 'struct int_tree_map'. man. Its a hash table function only, so it belongs > here. > - contains_view_convert_expr_p was only called from tree-sra.c, so I moved > it there and made it static. > - ranges_overlap_p .. this was used almost exclusively by SSA aliasing code, > so this seemed appropriate for tree-ssa-alias.h. > - gimple_ssa_operands goes to tree-ssa-operands.c and becomes static as this > was the only consumer > - is_global_var and may_be_aliased are tree specific.. may_be_aliased is > used by dse.c, and by putting it into tree.c, that file is tantalizing close > to being able to NOT include tree-ssa.h or any gimple stuff.. > local_variable_can_escape() is the only routine in there that is even aware > of ssa or gimple and something seems wrong there. thats on my list to take > care of shortly. > > Bootstraps on x86_64-unknown-linux-gnu and no new regressions. OK?
Ok. THanks, Richard. > Andrew >