On Thu, Sep 12, 2013 at 3:53 PM, Andrew MacLeod <amacl...@redhat.com> wrote: > It turns out that ssa_undefined_value_p() is also used in tree-complex.c and > tree-ssa-pre.c. doh. > > That routine is *almost* pass independent... only the last line uses > 'possibly_undefined_names'. Since the late warning pass initializes and > then clears that pointer_set when done, those other 2 passes obviously don't > need that part. > > So I moved the majority of ssa_undefined_value_p() to tree-ssa.c and added > a new local routine 'has_undefined_value_p()' to tree-ssa-uninit.c. It > first calls ssa_undefined_value_p() and then does the pass dependent check. > > I also moved the early warning pass stuff from tree-ssa.c into > tree-ssa-uninit.c where it belongs. > > Now there is nothing exported from tree-ssa-uninit.c, and the one pass > independent routine is now in tree-ssa.[ch]. > > Bootstrapped on x86_64-unknown-linux-gnu. regression testing underway. > Assuming no regressions, OK?
Ok. Thanks! Richard. > Andrew
