Hi, On Thu, Jun 07 2018, Jan Hubicka wrote: >> >> gcc/ChangeLog: >> >> 2018-04-24 Martin Liska <mli...@suse.cz> >> >> * ipa-cp.c (ipcp_store_bits_results): Use >> ipcp_transformation_sum. >> (ipcp_store_vr_results): Likewise. >> * ipa-prop.c (ipcp_grow_transformations_if_necessary): Renamed >> to ... >> (ipcp_transformation_initialize): ... this. >> (ipa_set_node_agg_value_chain): >> (ipa_node_params_t::duplicate): Use ipcp_transformation_sum. >> (write_ipcp_transformation_info): Likewise. >> (read_ipcp_transformation_info): Likewise. >> (ipcp_update_bits): Likewise. >> (ipcp_update_vr): Likewise. >> (ipcp_transform_function): Likewise. >> * ipa-prop.h: Rename ipcp_transformation_summary to >> ipcp_transformation. >> (class ipcp_transformation_t): New function summary. >> (ipcp_get_transformation_summary): Use ipcp_transformation_sum. >> (ipa_get_agg_replacements_for_node): Likewise. > > I will delegate this one to Martin Jambor. it looks fine to me in general > except again > I would like to use get() on places we do not create new info. >
I have read the whole series couple of weeks ago and think it is a step in the right direction, certainly when it comes to IPA-CP bits, so unless anything changed substantially, I am OK with it. I think most of the get_creates are actually necessary in this patch, we could probably often move ipcp_transformation_initialize to a caller, but let's do it as a follow-up (it is not performance critical in any way anyhow). Martin
