On Wed, 15 May 2019, Kewen.Lin wrote: > > on 2019/5/14 下午3:18, Richard Biener wrote: > > Hum. The function is somewhat of a hack, trying to produce > > "reasonable" DECL_RTL, exposing it in expr.[ch] with this > > name is eventually misleading. Also you fail to "outline" > > the most important part: > > > > FOR_EACH_VEC_ELT (decl_rtl_to_reset, i, obj) > > SET_DECL_RTL (obj, NULL_RTX); > > > > which IMHO would warrant making this machinery a class > > with the above done in its destructor? > > > > Good suggestion! In the IVOPTS implementation, it has one > interface free_loop_data to clean up some data structures > including this decl_rtl_to_reset. While for the use in > "PATCH v2 2/3", we have to clean it artificially once > expanding finishes. > > It's better to make it as a class and ensure the clean of > the vector in its destructor. > > > Maybe name the functions prepare_guessed_decl_rtl () > > and the new class guessed_decl_rtl? > > > OK. Or "tmp" instead of "guessed"?
Also works for me. Richard.