Re: [PATCH 2/6] Indirect call graph edges

[Jan Hubicka <hubicka@ucw.cz>]

> On Mon, Feb 22, 2010 at 4:44 PM, Jan Hubicka <hubicka@ucw.cz> wrote:
> >> 2010-02-10  Martin Jambor  <mjambor@suse.cz>
> >>
> >>       * cgraph.h (struct cgraph_node): New field indirect_calls.
> >>       (struct cgraph_indirect_call_info): New type.
> >>       (struct cgraph_edge): Removed field indirect_call. New fields
> >>       indirect_info, indirect_inlining_edge and indirect_unknown_callee.
> >>       (cgraph_create_indirect_edge): Declare.
> >>       (cgraph_make_edge_direct): Likewise.
> >>       (enum LTO_cgraph_tags): New item LTO_cgraph_indirect_edge.
> >>       * ipa-prop.h (struct ipa_param_call_note): Removed.
> >>       (struct ipa_node_params): Removed field param_calls.
> >>       * cgraph.c (cgraph_add_edge_to_call_site_hash): New function..
> >>       (cgraph_edge): Search also among the indirect edges, use
> >>       cgraph_add_edge_to_call_site_hash to add edges to the call site hash.
> >>       (cgraph_set_call_stmt): Possibly turn an indirect edge into a direct
> >>       one, use cgraph_add_edge_to_call_site_hash to add edges to the call
> >>       site hash.
> >>       (initialize_inline_failed): Assign a reason to indirect edges.
> >>       (cgraph_create_edge_1): New function.
> >>       (cgraph_create_edge): Moved some functionality to
> >>       cgraph_create_edge_1.
> >>       (cgraph_create_indirect_edge): New function.
> >>       (cgraph_edge_remove_callee): Add an assert checking for
> >>       non-indirectness.
> >>       (cgraph_edge_remove_caller): Special-case indirect edges.
> >>       (cgraph_remove_edge): Likewise.
> >>       (cgraph_set_edge_callee): New function.
> >>       (cgraph_redirect_edge_callee): Use cgraph_set_edge_callee.
> >>       (cgraph_make_edge_direct): New function.
> >>       (cgraph_update_edges_for_call_stmt_node): Do nothing only when also
> >>       the declaration of the call statement matches.
> >>       (cgraph_node_remove_callees): Special-case indirect edges.
> >>       (cgraph_clone_edge): Likewise.
> >>       (cgraph_clone_node): Clone also the indirect edges.
> >>       (dump_cgraph_node): Dump indirect_inlining_edge flag instead of
> >>       indirect_call, dump count of indirect_calls edges.
> >>       * ipa-prop.c (ipa_note_param_call): Create indirect edges instead of
> >>       creating notes.  New parameter node.
> >>       (ipa_analyze_call_uses): New parameter node, pass it on to
> >>       ipa_note_param_call.
> >>       (ipa_analyze_stmt_uses): Likewise.
> >>       (ipa_analyze_params_uses): Pass node to ipa_analyze_stmt_uses.
> >>       (print_edge_addition_message): Work on edges rather than on notes.
> >>       (update_call_notes_after_inlining): Likewise.
> >>       (ipa_free_node_params_substructures): Do not free notes.
> >>       (ipa_node_duplication_hook): Do not duplicate notes.
> >>       (ipa_write_param_call_note): Removed.
> >>       (ipa_read_param_call_note): Removed.
> >>       (ipa_write_indirect_edge_info): New function.
> >>       (ipa_read_indirect_edge_info): Likewise.
> >>       (ipa_write_node_info): Do not stream notes, do stream information
> >>       in indirect edges.
> >>       (ipa_read_node_info): Likewise.
> >>       (lto_ipa_fixup_call_notes): Removed.
> >>       * ipa-cp.c (pass_ipa_cp): Set stmt_fixup to NULL.
> >>       * ipa-inline.c (pass_ipa_inline): Likewise.
> >>       * cgraphunit.c (verify_cgraph_node): Check also indirect edges.
> >>       * cif-code.def (INDIRECT_UNKNOWN_CALL): New reason.
> >>       * tree-inline.c (copy_bb): Removed an unnecessary double check for
> >>       is_gimple_call.
> >>       * tree-inline.c (get_indirect_callee_fndecl): Do not consider indirect
> >>       edges.
> >>       * lto-cgraph.c (output_outgoing_cgraph_edges): New function.
> >>       (output_cgraph): Stream also indirect edges.
> >>       (lto_output_edge): Added capability to stream indirect edges..
> >>       (input_edge): Likewise.
> >>       (input_cgraph_1): Likewise.
> >>
> >>       * testsuite/gcc.dg/lto/20091209-1_0.c: New testcase.
> >
> > OK (for pretty-ipa).
> > As Richard commented, we will need way of collecting list of indirect callers
> > for need of passes propagating data across callgraph.  That can be done
> > incrementally.  We need way to get list of possible destinations that should
> > be fed in by PTA.
> 
> Yeah.  We currently throw away the points-to solution for function
> pointer SSA names ...

Yep, also the patch does not provide quite a callgraph with indirect edges, it
just adds edges for callback calls that are interesting for ipa-cp/inlining.  I
guess we will need to merge stuff with PTA work first before getting to full
callgraph with indirect calls.

I just didn't wanted to have separate ipa-prop specific datastructure for indirect
calls when we should go for callgraph edges in not so long run.

Honza
> 
> Richard.
> 
> > Honza
> >