> From: Dhruv Chawla <dhr...@nvidia.com>
>
> This patch modifies afdo_set_bb_count to propagate profile information
> to outline copies of functions if they are not inlined. This information
> gets lost otherwise.
>
> Signed-off-by: Dhruv Chawla <dhr...@nvidia.com>
>
> gcc/ChangeLog:
>
> * gcc/auto-profile.cc (count_info): Adjust comments.
> (function_instance::in_afdo_source_profile): New function.
> (function_instance::set_in_afdo_source_profile): Likewise.
> (function_instance::get_callsites): Likewise.
> (autofdo_source_profile::add_function_instance): Likewise.
> (function_instance::in_afdo_source_profile_): New member.
> (autofdo_source_profile::~autofdo_source_profile): Use
> in_afdo_source_profile to prevent a double-free bug.
> (afdo_set_bb_count): Propagate inline information to non-inlined
> outline copy.
> @@ -230,6 +224,10 @@ public:
> return head_count_;
> }
>
> + bool in_afdo_source_profile () const { return in_afdo_source_profile_; }
> +
> + void set_in_afdo_source_profile () { in_afdo_source_profile_ = true; }
> /* Map from source location to count_info. */
> position_count_map pos_counts;
> +
> + /* If this is an inline instance tracked in afdo_source_profile. */
> + bool in_afdo_source_profile_;
What about calling it is_inline_instance? I think it is more
self-explanatory of what it is useful for.
> @@ -814,6 +825,24 @@ autofdo_source_profile::update_inlined_ind_target (gcall
> *stmt,
> return true;
> }
>
> +/* Add a new function_instance entry if an inlined function is found in the
> + profile that doesn't have a corresponding entry in the map. Return false
> if
> + the function_instance already exists, true if it doesn't. */
> +
> +bool
> +autofdo_source_profile::add_function_instance (function_instance *fun)
> +{
> + int name = fun->name ();
> + if (auto fun_it = map_.find (name); fun_it != map_.end ())
> + {
> + fun_it->second->merge (fun);
Assume that we have call chain
foo->bar->baz
and assume that this was inlined in the train run, while after early
inlining we inline bar->baz but not foo->bar.
Will we then also recursively merge foo->bar->baz profile into bar->baz?
> @@ -1144,6 +1173,31 @@ afdo_set_bb_count (basic_block bb, const stmt_set
> &promoted)
> gimple *stmt = gsi_stmt (gsi);
> if (gimple_clobber_p (stmt) || is_gimple_debug (stmt))
> continue;
> + if (gimple_code (stmt) == GIMPLE_CALL)
> + {
> + tree fn = gimple_call_fndecl (stmt);
> + function_instance *cur
> + = afdo_source_profile->get_function_instance_by_decl (
> + current_function_decl);
> + if (!cur || !fn)
> + continue;
I think we want to do the merging in a global pass before annotation
starts (to solve the dependency).
I think you can relatively cheaply walk all functions in callgraph, all
edges and for each edge see if it the statement is inlined in profile
and do the merging...
Honza
