http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55888
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> 2013-01-07 10:45:37 UTC --- (gdb) call debug_bitmap_set (set) debug[0] := { message_7(D) (0007), rv_26 (0024), _30 (0034), {plus_expr,i_62,1} (0021), vntemp_50 (0044), {convert_expr,vntemp_50} (0069) } but value_expressions[44] only contains len_31. So the question is where we added vntemp_50 (well, the bisect shows us where). This is a VN value-number created for an expression that might need to be inserted somewhere ((const size_t) _41). We fail to add that to its value. Alternatively we fail to do the valueization "properly" when translating an SSA_NAME. In fact as original the valueization after PHI-translating SSA names looks superfluous. The fix for PR55862 probably simply made us prune sth from the antic set which resolved the oscillation. Back to the drawing-board for PR55862 ...