On Fri, Oct 18, 2013 at 2:48 AM, Andrew MacLeod <amacl...@redhat.com> wrote: > This patch finally renames tree-flow.h to tree-cfg.h. It now contains just > the prototypes for tree-cfg.h.
Yay! > I've also removed all the #include's from tree-cfg.h, and relocated them > (temporarily) to tree-ssa.h which is acting as the tree-ssa module header. > > tree-flow.h is removed from all includes, and a few places were tweaked to > include just the bit(s) they may have needed from the old tree-flow.h list. > (for instance, gimple.c only needed bitmap.h, nothing else :-P) > > virtual_operand_p is removed from tree-ssa-operands.h and relocated to > gimple.c. It actually has nothing to do with operand processing, merely > queries whether a given VAR_DECL is the global variable for VOPs. > (VAR_DECL_CHECK (NODE)->base.u.bits.saturating_flag). So I dont think it > even counts as a border routine for gimple-ssa.h. Moving this prevented > tree-ssa-operands.h from being required in a couple of files that shouldn't > need it.. You could argue it belongs to tree-ssa-alias.h as it is a GIMPLE alias machinery specific var ... > A set of 8 more patches will follow this (very simple ones :-) which reduce > the include list out of tree-ssa.h to just files which are commonly used. > > rs6000 and alpha were including tree-flow.h in their config/target.c file. I > built stage 1 for each target as a cross compiler to confirm builds dont > break due to something unresolved. rs6000.c no longer needed it, alpha > needs num_ssa_names, so it requires gimple-ssa.h (instead of tree-flow.h). > There is a compilation error in alpha (like on many other targets) for > target_flags_explicit.... when it fails, that is the only unresolved > external left.... the situation which existed before this patch. > > gimple-ssa.h cannot be compiled without tree-ssa-operands.h (it requires > some structs and fields), so I #include tree-ssa-operands.h *back* into > gimple-ssa.h where it was a week ago... not sure why I took it out. oops. > > Bootstraps on x86_64-unknown-linux-gnu with no new regressions. > > OK? Ok. Thanks, Richard. > Andrew >
