[PATCH] Redirect call within specific target attribute among MV clones (PR ipa/82625).

[Martin Liška <mliska at suse dot cz>]

Hi.

When having a pair of target clones where foo calls bar, if the target
attribute are equal we can redirect the call and not use ifunc dispatcher.

Patch survives regression tests on x86_64-linux-gnu.
Ready for trunk?

Martin

gcc/ChangeLog:

2018-10-04  Martin Liska  <mliska@suse.cz>

	PR ipa/82625
	* multiple_target.c (redirect_to_specific_clone): New function.
	(ipa_target_clone): Use it.
	* tree-inline.c: Fix comment.

gcc/testsuite/ChangeLog:

2018-10-04  Martin Liska  <mliska@suse.cz>

	PR ipa/82625
	* g++.dg/ext/pr82625.C: New test.
---
 gcc/multiple_target.c              | 51 ++++++++++++++++++++++++++++++
 gcc/testsuite/g++.dg/ext/pr82625.C | 36 +++++++++++++++++++++
 gcc/tree-inline.c                  |  2 +-
 3 files changed, 88 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/ext/pr82625.C


diff --git a/gcc/multiple_target.c b/gcc/multiple_target.c
index a610d9a3345..2d892f201c5 100644
--- a/gcc/multiple_target.c
+++ b/gcc/multiple_target.c
@@ -451,6 +451,54 @@ expand_target_clones (struct cgraph_node *node, bool definition)
   return ret;
 }
 
+/* When NODE is a target clone, consider all callees and redirect
+   to a clone with equal target attributes.  That prevents multiple
+   multi-versioning dispatches and a call-chain can be optimized.  */
+
+static void
+redirect_to_specific_clone (cgraph_node *node)
+{
+  cgraph_function_version_info *fv = node->function_version ();
+  if (fv == NULL)
+    return;
+
+  tree attr_target = lookup_attribute ("target", DECL_ATTRIBUTES (node->decl));
+  if (attr_target == NULL_TREE)
+    return;
+
+  /* We need to remember NEXT_CALLER as it could be modified in the loop.  */
+  for (cgraph_edge *e = node->callees; e ; e = e->next_callee)
+    {
+      cgraph_function_version_info *fv2 = e->callee->function_version ();
+      if (!fv2)
+	continue;
+
+      tree attr_target2 = lookup_attribute ("target",
+					    DECL_ATTRIBUTES (e->callee->decl));
+
+      /* Function is not calling proper target clone.  */
+      if (!attribute_list_equal (attr_target, attr_target2))
+	{
+	  while (fv2->prev != NULL)
+	    fv2 = fv2->prev;
+
+	  /* Try to find a clone with equal target attribute.  */
+	  for (; fv2 != NULL; fv2 = fv2->next)
+	    {
+	      cgraph_node *callee = fv2->this_node;
+	      attr_target2 = lookup_attribute ("target",
+					       DECL_ATTRIBUTES (callee->decl));
+	      if (attribute_list_equal (attr_target, attr_target2))
+		{
+		  e->redirect_callee (callee);
+		  e->redirect_call_stmt_to_callee ();
+		  break;
+		}
+	    }
+	}
+    }
+}
+
 static unsigned int
 ipa_target_clone (void)
 {
@@ -464,6 +512,9 @@ ipa_target_clone (void)
   for (unsigned i = 0; i < to_dispatch.length (); i++)
     create_dispatcher_calls (to_dispatch[i]);
 
+  FOR_EACH_FUNCTION (node)
+    redirect_to_specific_clone (node);
+
   return 0;
 }
 
diff --git a/gcc/testsuite/g++.dg/ext/pr82625.C b/gcc/testsuite/g++.dg/ext/pr82625.C
new file mode 100644
index 00000000000..47bd2df1104
--- /dev/null
+++ b/gcc/testsuite/g++.dg/ext/pr82625.C
@@ -0,0 +1,36 @@
+/* { dg-do compile } */
+/* { dg-require-ifunc "" } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+
+__attribute__ ((target ("default")))
+static unsigned foo(const char *buf, unsigned size) {
+  return 1;
+}
+
+__attribute__ ((target ("avx")))
+static unsigned foo(const char *buf, unsigned size) {
+  return 2;
+}
+
+__attribute__ ((target ("default")))
+unsigned bar() {
+  char buf[4096];
+  unsigned acc = 0;
+  for (int i = 0; i < sizeof(buf); i++) {
+    acc += foo(&buf[i], 1);
+  }
+  return acc;
+}
+
+__attribute__ ((target ("avx")))
+unsigned bar() {
+  char buf[4096];
+  unsigned acc = 0;
+  for (int i = 0; i < sizeof(buf); i++) {
+    acc += foo(&buf[i], 1);
+  }
+  return acc;
+}
+
+/* { dg-final { scan-tree-dump-times "return 4096;" 1 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "return 8192;" 1 "optimized" } } */
diff --git a/gcc/tree-inline.c b/gcc/tree-inline.c
index ff8ee8ce78f..913425394e0 100644
--- a/gcc/tree-inline.c
+++ b/gcc/tree-inline.c
@@ -2631,7 +2631,7 @@ copy_loops (copy_body_data *id,
     }
 }
 
-/* Call cgraph_redirect_edge_call_stmt_to_callee on all calls in BB */
+/* Call redirect_call_stmt_to_callee on all calls in BB.  */
 
 void
 redirect_all_calls (copy_body_data * id, basic_block bb)