ipa: fix handling of multiple speculations (PR93318)

This patch started as work to resole Richard's comment on quadratic lookups
in resolve_speculation. While doing it I however noticed multiple problems
in the new speuclative call code which made the patch quite big. In
particular:
 1) Before applying speculation we consider only targets with at lest
    probability 1/2.
    If profile is sane at most two targets can have probability greater or
    equal to 1/2. So the new multi-target speculation code got enabled only
    in very special scenario when there ae precisely two target with precise
    probability 1/2 (which is tested by the single testcase).

    As a conseuqence the multiple target logic got minimal test coverage and
    this made us to miss several ICEs.
 2) Profile updating in profile merging, tree-inline and indirect call
    expansion was wrong which led to inconsistent profiles (as already seen
    on the testcase).
 3) Code responsible to turn speculative call to direct call was broken for
    anything with more than one target.
 4) There were multiple cases where call_site_hash went out of sync which
    eventually leads to an ICE..
 5) Some code expects that all speculative call targets forms a sequence in
    the callee linked list but there is no code to maintain that invariant
    nor a verifier.
Fixing this it became obvious that the current API of speculative_call_info is
not useful because it really builds on fact tht there are precisely three
components (direct call, ref and indirect call) in every speculative call
sequence.  I ended up replacing it with iterator API for direct call
(first_speculative_call_target, next_speculative_call_target) and accessors for
the other coponents updating comment in cgraph.h.

Finally I made the work with call site hash more effetive by updating edge
manipulation to keep them in sequence. So first one can be looked up from the
hash and then they can be iterated by callee.

There are other things that can be improved (for example the speculation should
start with most common target first), but I will try to keep that for next
stage1. This patch is mostly about getting rid of ICE and profile corruption
which is a regression from GCC 9.

gcc/ChangeLog:

	PR lto/93318
	* cgraph.c (cgraph_add_edge_to_call_site_hash): Update call site
	hash only when edge is first within the sequence.
	(cgraph_edge::set_call_stmt): Update handling of speculative calls.
	(symbol_table::create_edge): Do not set target_prob.
	(cgraph_edge::remove_caller): Watch for speculative calls when updating
	the call site hash.
	(cgraph_edge::make_speculative): Drop target_prob parameter.
	(cgraph_edge::speculative_call_info): Remove.
	(cgraph_edge::first_speculative_call_target): New member function.
	(update_call_stmt_hash_for_removing_direct_edge): New function.
	(cgraph_edge::resolve_speculation): Rewrite to new API.
	(cgraph_edge::speculative_call_for_target): New member function.
	(cgraph_edge::make_direct): Rewrite to new API; fix handling of
	multiple speculation targets.
	(cgraph_edge::redirect_call_stmt_to_callee): Likewise; fix updating
	of profile.
	(verify_speculative_call): Verify that targets form an interval.
	* cgraph.h (cgraph_edge::speculative_call_info): Remove.
	(cgraph_edge::first_speculative_call_target): New member function.
	(cgraph_edge::next_speculative_call_target): New member function.
	(cgraph_edge::speculative_call_target_ref): New member function.
	(cgraph_edge;:speculative_call_indirect_edge): New member funtion.
	(cgraph_edge): Remove target_prob.
	* cgraphclones.c (cgraph_node::set_call_stmt_including_clones):
	Fix handling of speculative calls.
	* ipa-devirt.c (ipa_devirt): Fix handling of speculative cals.
	* ipa-fnsummary.c (analyze_function_body): Likewise.
	* ipa-inline.c (speculation_useful_p): Use new speculative call API.
	* ipa-profile.c (dump_histogram): Fix formating.
	(ipa_profile_generate_summary): Watch for overflows.
	(ipa_profile): Do not require probablity to be 1/2; update to new API.
	* ipa-prop.c (ipa_make_edge_direct_to_target): Update to new API.
	(update_indirect_edges_after_inlining): Update to new API.
	* ipa-utils.c (ipa_merge_profiles): Rewrite merging of speculative call
	profiles.
	* profile-count.h: (profile_probability::adjusted): New.
	* tree-inline.c (copy_bb): Update to new speculative call API; fix
	updating of profile.
	* value-prof.c (gimple_ic_transform): Rename to ...
	(dump_ic_profile): ... this one; update dumping.
	(stream_in_histogram_value): Fix formating.
	(gimple_value_profile_transformations): Update.

gcc/testsuite/ChangeLog:

	* g++.dg/tree-prof/indir-call-prof.C: Update template.
	* gcc.dg/tree-prof/crossmodule-indircall-1.c: Add more targets.
	* gcc.dg/tree-prof/crossmodule-indircall-1a.c: Add more targets.
	* gcc.dg/tree-prof/indir-call-prof.c: Update template.

1 files changed, 58 insertions, 183 deletions

diff --git a/gcc/ipa-utils.c b/gcc/ipa-utils.c
index 587ad5c..23e7f71 100644
--- a/gcc/ipa-utils.c
+++ b/gcc/ipa-utils.c
@@ -674,201 +674,75 @@ ipa_merge_profiles (struct cgraph_node *dst,
       for (e = dst->indirect_calls, e2 = src->indirect_calls; e;
 	   e2 = (e2 ? e2->next_callee : NULL), e = e->next_callee)
 	{
-	  profile_count count = gimple_bb (e->call_stmt)->count;
-	  /* Below code are introduced by r279373 of "Fix merging of common
-	     traget info.".
-
-	     ipa-icf runs after ipa-profile, common_target_id with
-	     common_target_probablity are useless in ipa-icf since they are
-	     moved from cgraph.h to ipa-profile.c and processed already.
-	     Need double circulation to find out each mapped direct speculative
-	     edge and do prob merge.  Not easy to construct a case to cover all
-	     circumstances here.  For src and dst both have multiple speculative
-	     targets, only N:N maps are implemented, 2:0, 2:1, 0:2, 1:2 are not
-	     implemented yet as too complicated and no test cases to cover.  */
-	  if (copy_counts)
+	  if (!e->speculative && !e2->speculative)
 	    {
-	      /* copy if both e and e2 have same num_speculative_call_targets.
-	       */
-	      if (e->num_speculative_call_targets_p ()
-		  == e2->num_speculative_call_targets_p ())
-		{
-		  int num_specs = e->num_speculative_call_targets_p ();
-		  cgraph_edge *direct, *indirect, *next_direct;
-		  cgraph_edge *direct2, *indirect2, *next_direct2;
-		  ipa_ref *ref;
-		  for (next_direct = e; next_direct && num_specs--;
-		       next_direct = direct->next_callee)
-		    {
-		      next_direct->speculative_call_info (direct, indirect,
-							  ref);
-
-		      int num_specs2 = e2->num_speculative_call_targets_p ();
-		      for (next_direct2 = e2; next_direct2 && num_specs2--;
-			   next_direct2 = direct2->next_callee)
-			{
-			  if (e2 && e2->speculative)
-			    next_direct2->speculative_call_info (direct2,
-								 indirect2,
-								 ref);
-			  if (direct->speculative_id == direct2->speculative_id
-			      && direct->lto_stmt_uid == direct2->lto_stmt_uid)
-			    {
-			      direct->target_prob = direct2->target_prob;
-			      break;
-			    }
-			}
-		    }
-		}
-	      else
-		gcc_assert (e->num_speculative_call_targets_p ()
-			    && e->num_speculative_call_targets_p ());
+	      /* FIXME: we need to also merge ipa-profile histograms
+		 because with LTO merging happens from lto-symtab before
+		 these are converted to indirect edges.  */
+	      e->count = gimple_bb (e->call_stmt)->count;
+	      continue;
 	    }
-	  else if (e->num_speculative_call_targets_p ()
-		   || e2->num_speculative_call_targets_p ())
+
+	  /* When copying just remove all speuclations on dst and then copy
+	     one from src.  */
+	  if (copy_counts)
 	    {
-	      if (e->num_speculative_call_targets_p ()
-		  == e2->num_speculative_call_targets_p ())
+	      while (e->speculative)
+		cgraph_edge::resolve_speculation (e, NULL);
+	      e->count = gimple_bb (e->call_stmt)->count;
+	      if (e2->speculative)
 		{
-		  int num_specs = e->num_speculative_call_targets_p ();
-		  cgraph_edge *direct, *indirect, *next_direct;
-		  cgraph_edge *direct2, *indirect2, *next_direct2;
-		  ipa_ref *ref;
-		  for (next_direct = e; next_direct && num_specs--;
-		       next_direct = direct->next_callee)
+		  for (cgraph_edge *e3 = e2->first_speculative_call_target ();
+		       e3;
+		       e3 = e3->next_speculative_call_target ())
 		    {
-		      next_direct->speculative_call_info (direct, indirect,
-							  ref);
-
-		      int num_specs2 = e2->num_speculative_call_targets_p ();
-		      for (next_direct2 = e2; next_direct2 && num_specs2--;
-			   next_direct2 = direct2->next_callee)
-			{
-			  if (e2 && e2->speculative)
-			    next_direct2->speculative_call_info (direct2,
-								 indirect2,
-								 ref);
-			  if (direct->speculative_id == direct2->speculative_id
-			      && direct->lto_stmt_uid == direct2->lto_stmt_uid)
-			    {
-			      sreal scale1
-				= e->count.ipa ().to_sreal_scale (count);
-			      sreal scale2
-				= e2->count.ipa ().to_sreal_scale (count);
-
-			      if (scale1 == 0 && scale2 == 0)
-				scale1 = scale2 = 1;
-			      sreal sum = scale1 + scale2;
-			      int scaled_prob1
-				= (((sreal)direct->target_prob)
-				   * scale1 / sum).to_int ();
-			      int scaled_prob2
-				= (((sreal)direct2->target_prob)
-				   * scale2 / sum).to_int ();
-			      if (symtab->dump_file)
-				{
-				  fprintf (
-				    symtab->dump_file,
-				    "Merging speculative id %i prob %i"
-				    " and %i prob %i with scales %f %f\n",
-				    direct->speculative_id, direct->target_prob,
-				    direct2->speculative_id,
-				    direct2->target_prob, scale1.to_double (),
-				    scale2.to_double ());
-				  fprintf (symtab->dump_file,
-					   "Combined BB count ");
-				  count.dump (symtab->dump_file);
-				  fprintf (symtab->dump_file,
-					   " dst edge count ");
-				  e->count.dump (symtab->dump_file);
-				  fprintf (symtab->dump_file,
-					   " src edge count ");
-				  e2->count.dump (symtab->dump_file);
-				  fprintf (symtab->dump_file, "\n");
-				}
-			      direct->target_prob = scaled_prob1 + scaled_prob2;
-			      break;
-			    }
-			}
+		      cgraph_edge *ns;
+		      ns = e->make_speculative
+			 (dyn_cast <cgraph_node *>
+			    (e3->speculative_call_target_ref ()->referred),
+			     e3->count, e3->speculative_id);
+		      /* Target may differ from ref (for example it may be
+			 redirected to local alias.  */
+		      ns->redirect_callee (e3->callee);
 		    }
 		}
-	      else if (e->num_speculative_call_targets_p ())
-		{
-		  /* Process if only dst is speculative.  */
-		  gcc_assert (!e->num_speculative_call_targets_p ());
-		}
-	      else if (e2->num_speculative_call_targets_p ())
-		{
-		  /* Process if only src is speculative.  */
-		  gcc_assert (!e2->num_speculative_call_targets_p ());
-		}
+	      continue;
 	    }
 
-	  /* When call is speculative, we need to re-distribute probabilities
-	     the same way as they was.  This is not really correct because
-	     in the other copy the speculation may differ; but probably it
-	     is not really worth the effort.  */
-	  if (e->speculative)
+	  /* Iterate all speculations in SRC, see if corresponding ones exist
+	     int DST and if so, sum the counts.  Otherwise create new
+	     speculation.  */
+	  int max_spec = 0;
+	  for (cgraph_edge *e3 = e->first_speculative_call_target ();
+	       e3;
+	       e3 = e3->next_speculative_call_target ())
+	    if (e3->speculative_id > max_spec)
+	      max_spec = e3->speculative_id;
+	  for (cgraph_edge *e3 = e2->first_speculative_call_target ();
+	       e3;
+	       e3 = e3->next_speculative_call_target ())
 	    {
-	      cgraph_edge *direct, *indirect;
-	      cgraph_edge *direct2 = NULL, *indirect2 = NULL;
-	      ipa_ref *ref;
-
-	      e->speculative_call_info (direct, indirect, ref);
-	      gcc_assert (e == indirect);
-	      if (e2 && e2->speculative)
-	        e2->speculative_call_info (direct2, indirect2, ref);
-	      if (indirect->count > profile_count::zero ()
-		  || direct->count > profile_count::zero ())
+	      cgraph_edge *te
+		 = e->speculative_call_for_target
+			 (dyn_cast <cgraph_node *>
+			    (e3->speculative_call_target_ref ()->referred));
+	      if (te)
+		te->count = te->count + e3->count;
+	      else
 		{
-		  /* We should mismatch earlier if there is no matching
-		     indirect edge.  */
-		  if (!e2)
-		    {
-		      if (symtab->dump_file)
-		        fprintf (symtab->dump_file,
-				 "Mismatch in merging indirect edges\n");
-		    }
-		  else if (!e2->speculative)
-		    indirect->count += e2->count;
-		  else if (e2->speculative)
-		    {
-		      if (DECL_ASSEMBLER_NAME (direct2->callee->decl)
-			  != DECL_ASSEMBLER_NAME (direct->callee->decl))
-			{
-			  if (direct2->count >= direct->count)
-			    {
-			      direct->redirect_callee (direct2->callee);
-			      indirect->count += indirect2->count
-						 + direct->count;
-			      direct->count = direct2->count;
-			    }
-			  else
-			    indirect->count += indirect2->count + direct2->count;
-			}
-		      else
-			{
-			   direct->count += direct2->count;
-			   indirect->count += indirect2->count;
-			}
-		    }
+		  e->count = e->count + e3->count;
+		  cgraph_edge *ns;
+		  ns = e->make_speculative
+			 (dyn_cast <cgraph_node *>
+			    (e3->speculative_call_target_ref ()
+			     ->referred),
+			  e3->count,
+			  e3->speculative_id + max_spec + 1);
+		  /* Target may differ from ref (for example it may be
+		     redirected to local alias.  */
+		  ns->redirect_callee (e3->callee);
 		}
-	      else
-		/* At the moment we should have only profile feedback based
-		   speculations when merging.  */
-		gcc_unreachable ();
 	    }
-	  else if (e2 && e2->speculative)
-	    {
-	      cgraph_edge *direct, *indirect;
-	      ipa_ref *ref;
-
-	      e2->speculative_call_info (direct, indirect, ref);
-	      e->count = count;
-	      e->make_speculative (direct->callee, direct->count);
-	    }
-	  else
-	    e->count = count;
 	}
       if (!preserve_body)
         src->release_body ();
@@ -882,7 +756,8 @@ ipa_merge_profiles (struct cgraph_node *dst,
   src->decl = oldsrcdecl;
 }
 
-/* Return true if call to DEST is known to be self-recusive call withing FUNC.   */
+/* Return true if call to DEST is known to be self-recusive
+   call withing FUNC.  */
 
 bool
 recursive_call_p (tree func, tree dest)