https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110334
--- Comment #12 from rguenther at suse dot de <rguenther at suse dot de> --- On Mon, 26 Jun 2023, hubicka at ucw dot cz wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110334 > > --- Comment #11 from Jan Hubicka <hubicka at ucw dot cz> --- > Hi, > what about this. It should make at least quite basic inlining to happen > to always_inline. I do not think many critical always_inlines have > indirect calls in them. The test for lto is quite bad and I can > work on solving this incrementally (it would be nice to have this > tested and possibly backport it). > > diff --git a/gcc/ipa-inline.cc b/gcc/ipa-inline.cc > index efc8df7d4e0..dcec07e49e1 100644 > --- a/gcc/ipa-inline.cc > +++ b/gcc/ipa-inline.cc > @@ -702,6 +702,38 @@ can_early_inline_edge_p (struct cgraph_edge *e) > if (!can_inline_edge_p (e, true, true) > || !can_inline_edge_by_limits_p (e, true, false, true)) > return false; > + /* When inlining regular functions into always-inline functions > + during early inlining watch for possible inline cycles. */ > + if (DECL_DISREGARD_INLINE_LIMITS (caller->decl) > + && lookup_attribute ("always_inline", DECL_ATTRIBUTES (caller->decl)) > + && (!DECL_DISREGARD_INLINE_LIMITS (callee->decl) > + || !lookup_attribute ("always_inline", DECL_ATTRIBUTES > (callee->decl)))) > + { > + /* If there are indirect calls, inlining may produce direct call. > + TODO: We may lift this restriction if we avoid errors on formely > + indirect calls to always_inline functions. Taking address > + of always_inline function is generally bad idea and should > + have been declared as undefined, but sadly we allow this. */ > + if (caller->indirect_calls || e->callee->indirect_calls) why disallow caller->indirect_calls? > + return false; > + for (cgraph_edge *e2 = callee->callees; e2; e2 = e2->next_callee) I don't think this flys - it looks quadratic. Can we compute this in the inline summary once instead? As for indirect calls, can we maybe mark initial direct GIMPLE call stmts as "always-inline" and only look at that marking, thus an indirect call will never become "always-inline"? Iff cgraph edges prevail during all early inlining we could mark call edges for this purpose? > + { > + struct cgraph_node *callee2 = e2->callee->ultimate_alias_target (); > + /* As early inliner runs in RPO order, we will see uninlined > + always_inline calls only in the case of cyclic graphs. */ > + if (DECL_DISREGARD_INLINE_LIMITS (callee2->decl) > + || lookup_attribute ("always_inline", callee2->decl)) > + return false; > + /* With LTO watch for case where function is later replaced > + by always_inline definition. > + TODO: We may either stop treating noninlined cross-module always > + inlines as errors, or we can extend decl merging to produce > + syntacic alias and honor always inline only in units it has > + been declared as such. */ > + if (flag_lto && callee2->externally_visible) > + return false; > + } > + } > return true; > } > > @@ -3034,18 +3066,7 @@ early_inliner (function *fun) > > if (!optimize > || flag_no_inline > - || !flag_early_inlining > - /* Never inline regular functions into always-inline functions > - during incremental inlining. This sucks as functions calling > - always inline functions will get less optimized, but at the > - same time inlining of functions calling always inline > - function into an always inline function might introduce > - cycles of edges to be always inlined in the callgraph. > - > - We might want to be smarter and just avoid this type of inlining. */ > - || (DECL_DISREGARD_INLINE_LIMITS (node->decl) > - && lookup_attribute ("always_inline", > - DECL_ATTRIBUTES (node->decl)))) > + || !flag_early_inlining) > ; > else if (lookup_attribute ("flatten", > DECL_ATTRIBUTES (node->decl)) != NULL) > >
