re PR middle-end/79805 (ICE (verify_flow_info failed) with -fnon-call-exceptions -O)

PR middle-end/79805
	* internal-fn.def (ATOMIC_BIT_TEST_AND_SET, ATOMIC_BIT_TEST_AND_RESET,
	ATOMIC_BIT_TEST_AND_COMPLEMENT, ATOMIC_COMPARE_EXCHANGE): Remove
	ECF_NOTHROW.
	* gimple-fold.c (fold_builtin_atomic_compare_exchange): Set
	gimple_call_nothrow_p flag based on whether original builtin can throw.
	If it can, emit following stmts on the fallthrough edge.
	* tree-ssa-ccp.c (optimize_atomic_bit_test_and): Similarly, except
	don't create new bb if inserting just debug stmts on the edge, try to
	insert them on the fallthru bb or just reset debug stmts.

	* g++.dg/opt/pr79805.C: New test.

From-SVN: r245882

gcc/ChangeLog

@@ -1,3 +1,16 @@
+2017-03-03  Jakub Jelinek  <jakub@redhat.com>
+
+	PR middle-end/79805
+	* internal-fn.def (ATOMIC_BIT_TEST_AND_SET, ATOMIC_BIT_TEST_AND_RESET,
+	ATOMIC_BIT_TEST_AND_COMPLEMENT, ATOMIC_COMPARE_EXCHANGE): Remove
+	ECF_NOTHROW.
+	* gimple-fold.c (fold_builtin_atomic_compare_exchange): Set
+	gimple_call_nothrow_p flag based on whether original builtin can throw.
+	If it can, emit following stmts on the fallthrough edge.
+	* tree-ssa-ccp.c (optimize_atomic_bit_test_and): Similarly, except
+	don't create new bb if inserting just debug stmts on the edge, try to
+	insert them on the fallthru bb or just reset debug stmts.
+
 2017-03-03  Segher Boesssenkool  <segher@kernel.crashing.org>
 
 	PR target/43763

gcc/gimple-fold.c

@@ -3533,6 +3533,8 @@ fold_builtin_atomic_compare_exchange (gimple_stmt_iterator *gsi)
   tree itype = TREE_VALUE (TREE_CHAIN (TREE_CHAIN (parmt)));
   tree ctype = build_complex_type (itype);
   tree expected = TREE_OPERAND (gimple_call_arg (stmt, 1), 0);
+  bool throws = false;
+  edge e = NULL;
   gimple *g = gimple_build_assign (make_ssa_name (TREE_TYPE (expected)),
 				   expected);
   gsi_insert_before (gsi, g, GSI_SAME_STMT);
@@ -3558,19 +3560,39 @@ fold_builtin_atomic_compare_exchange (gimple_stmt_iterator *gsi)
   gimple_set_vdef (g, gimple_vdef (stmt));
   gimple_set_vuse (g, gimple_vuse (stmt));
   SSA_NAME_DEF_STMT (gimple_vdef (g)) = g;
-  if (gimple_call_lhs (stmt))
+  tree oldlhs = gimple_call_lhs (stmt);
+  if (stmt_can_throw_internal (stmt))
+    {
+      throws = true;
+      e = find_fallthru_edge (gsi_bb (*gsi)->succs);
+    }
+  gimple_call_set_nothrow (as_a <gcall *> (g),
+			   gimple_call_nothrow_p (as_a <gcall *> (stmt)));
+  gimple_call_set_lhs (stmt, NULL_TREE);
+  gsi_replace (gsi, g, true);
+  if (oldlhs)
     {
-      gsi_insert_before (gsi, g, GSI_SAME_STMT);
       g = gimple_build_assign (make_ssa_name (itype), IMAGPART_EXPR,
 			       build1 (IMAGPART_EXPR, itype, lhs));
-      gsi_insert_before (gsi, g, GSI_SAME_STMT);
-      g = gimple_build_assign (gimple_call_lhs (stmt), NOP_EXPR,
-			       gimple_assign_lhs (g));
+      if (throws)
+	{
+	  gsi_insert_on_edge_immediate (e, g);
+	  *gsi = gsi_for_stmt (g);
+	}
+      else
+	gsi_insert_after (gsi, g, GSI_NEW_STMT);
+      g = gimple_build_assign (oldlhs, NOP_EXPR, gimple_assign_lhs (g));
+      gsi_insert_after (gsi, g, GSI_NEW_STMT);
     }
-  gsi_replace (gsi, g, true);
   g = gimple_build_assign (make_ssa_name (itype), REALPART_EXPR,
 			   build1 (REALPART_EXPR, itype, lhs));
-  gsi_insert_after (gsi, g, GSI_NEW_STMT);
+  if (throws && oldlhs == NULL_TREE)
+    {
+      gsi_insert_on_edge_immediate (e, g);
+      *gsi = gsi_for_stmt (g);
+    }
+  else
+    gsi_insert_after (gsi, g, GSI_NEW_STMT);
   if (!useless_type_conversion_p (TREE_TYPE (expected), itype))
     {
       g = gimple_build_assign (make_ssa_name (TREE_TYPE (expected)),

gcc/internal-fn.def

@@ -205,11 +205,13 @@ DEF_INTERNAL_FN (GOACC_TILE, ECF_NOTHROW | ECF_LEAF, NULL)
    current target.  */
 DEF_INTERNAL_FN (SET_EDOM, ECF_LEAF | ECF_NOTHROW, NULL)
 
-/* Atomic functions.  */
-DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_SET, ECF_LEAF | ECF_NOTHROW, NULL)
-DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_COMPLEMENT, ECF_LEAF | ECF_NOTHROW, NULL)
-DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_RESET, ECF_LEAF | ECF_NOTHROW, NULL)
-DEF_INTERNAL_FN (ATOMIC_COMPARE_EXCHANGE, ECF_LEAF | ECF_NOTHROW, NULL)
+/* Atomic functions.  These don't have ECF_NOTHROW because for
+   -fnon-call-exceptions they can throw, otherwise we set
+   gimple_call_nothrow_p on it.  */
+DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_SET, ECF_LEAF, NULL)
+DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_COMPLEMENT, ECF_LEAF, NULL)
+DEF_INTERNAL_FN (ATOMIC_BIT_TEST_AND_RESET, ECF_LEAF, NULL)
+DEF_INTERNAL_FN (ATOMIC_COMPARE_EXCHANGE, ECF_LEAF, NULL)
 
 /* To implement [[fallthrough]].  */
 DEF_INTERNAL_FN (FALLTHROUGH, ECF_LEAF | ECF_NOTHROW, NULL)

gcc/testsuite/ChangeLog

@@ -1,3 +1,8 @@
+2017-03-03  Jakub Jelinek  <jakub@redhat.com>
+
+	PR middle-end/79805
+	* g++.dg/opt/pr79805.C: New test.
+
 2017-03-03  Andrew Senkevich  <andrew.senkevich@intel.com>
 
 	* gcc.target/i386/avx512vpopcntdq-check.h: New.

gcc/testsuite/g++.dg/opt/pr79805.C

@@ -0,0 +1,219 @@
+// PR middle-end/79805
+// { dg-do compile }
+// { dg-options "-O2 -fnon-call-exceptions" }
+
+struct A { A (); ~A (); };
+
+void bar (void);
+
+int
+f0 (int *d, int f)
+{
+  A z;
+  int e = __atomic_compare_exchange_n (d, &f, 1, 1, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
+  return e;
+}
+
+int
+f1 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__sync_fetch_and_or (a, mask) & mask) != 0;
+}
+
+int
+f2 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  unsigned int t1 = __atomic_fetch_or (a, mask, __ATOMIC_RELAXED);
+  unsigned int t2 = t1 & mask;
+  return t2 != 0;
+}
+
+long int
+f3 (long int *a, int bit)
+{
+  A z;
+  unsigned long int mask = (1ul << bit);
+  return (__atomic_fetch_or (a, mask, __ATOMIC_SEQ_CST) & mask) == 0;
+}
+
+int
+f4 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 7);
+  return (__sync_fetch_and_or (a, mask) & mask) != 0;
+}
+
+int
+f5 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 13);
+  return (__atomic_fetch_or (a, mask, __ATOMIC_RELAXED) & mask) != 0;
+}
+
+int
+f6 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 0);
+  return (__atomic_fetch_or (a, mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+void
+f7 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  if ((__sync_fetch_and_xor (a, mask) & mask) != 0)
+    bar ();
+}
+
+void
+f8 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  if ((__atomic_fetch_xor (a, mask, __ATOMIC_RELAXED) & mask) == 0)
+    bar ();
+}
+
+int
+f9 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__atomic_fetch_xor (a, mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+int
+f10 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 7);
+  return (__sync_fetch_and_xor (a, mask) & mask) != 0;
+}
+
+int
+f11 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 13);
+  return (__atomic_fetch_xor (a, mask, __ATOMIC_RELAXED) & mask) != 0;
+}
+
+int
+f12 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 0);
+  return (__atomic_fetch_xor (a, mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+int
+f13 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__sync_fetch_and_and (a, ~mask) & mask) != 0;
+}
+
+int
+f14 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__atomic_fetch_and (a, ~mask, __ATOMIC_RELAXED) & mask) != 0;
+}
+
+int
+f15 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__atomic_fetch_and (a, ~mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+int
+f16 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 7);
+  return (__sync_fetch_and_and (a, ~mask) & mask) != 0;
+}
+
+int
+f17 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 13);
+  return (__atomic_fetch_and (a, ~mask, __ATOMIC_RELAXED) & mask) != 0;
+}
+
+int
+f18 (int *a)
+{
+  A z;
+  unsigned int mask = (1u << 0);
+  return (__atomic_fetch_and (a, ~mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+unsigned long int
+f19 (unsigned long int *a, int bit)
+{
+  A z;
+  unsigned long int mask = (1ul << bit);
+  return (__atomic_xor_fetch (a, mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}
+
+unsigned long int
+f20 (unsigned long int *a)
+{
+  A z;
+  unsigned long int mask = (1ul << 7);
+  return (__atomic_xor_fetch (a, mask, __ATOMIC_SEQ_CST) & mask) == 0;
+}
+
+int
+f21 (int *a, int bit)
+{
+  A z;
+  unsigned int mask = (1u << bit);
+  return (__sync_fetch_and_or (a, mask) & mask);
+}
+
+unsigned long int
+f22 (unsigned long int *a)
+{
+  A z;
+  unsigned long int mask = (1ul << 7);
+  return (__atomic_xor_fetch (a, mask, __ATOMIC_SEQ_CST) & mask);
+}
+
+unsigned long int
+f23 (unsigned long int *a)
+{
+  A z;
+  unsigned long int mask = (1ul << 7);
+  return (__atomic_fetch_xor (a, mask, __ATOMIC_SEQ_CST) & mask);
+}
+
+unsigned short int
+f24 (unsigned short int *a)
+{
+  A z;
+  unsigned short int mask = (1u << 7);
+  return (__sync_fetch_and_or (a, mask) & mask) != 0;
+}
+
+unsigned short int
+f25 (unsigned short int *a)
+{
+  A z;
+  unsigned short int mask = (1u << 7);
+  return (__atomic_fetch_or (a, mask, __ATOMIC_SEQ_CST) & mask) != 0;
+}

gcc/tree-ssa-ccp.c

@@ -2890,9 +2890,19 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
   gimple_set_location (g, gimple_location (call));
   gimple_set_vuse (g, gimple_vuse (call));
   gimple_set_vdef (g, gimple_vdef (call));
+  bool throws = stmt_can_throw_internal (call);
+  gimple_call_set_nothrow (as_a <gcall *> (g),
+			   gimple_call_nothrow_p (as_a <gcall *> (call)));
   SSA_NAME_DEF_STMT (gimple_vdef (call)) = g;
   gimple_stmt_iterator gsi = *gsip;
   gsi_insert_after (&gsi, g, GSI_NEW_STMT);
+  edge e = NULL;
+  if (throws)
+    {
+      maybe_clean_or_replace_eh_stmt (call, g);
+      if (after || (use_bool && has_debug_uses))
+	e = find_fallthru_edge (gsi_bb (gsi)->succs);
+    }
   if (after)
     {
       /* The internal function returns the value of the specified bit
@@ -2905,23 +2915,42 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
 			       use_bool ? build_int_cst (TREE_TYPE (lhs), 1)
 					: mask);
       new_lhs = gimple_assign_lhs (g);
-      gsi_insert_after (&gsi, g, GSI_NEW_STMT);
+      if (throws)
+	{
+	  gsi_insert_on_edge_immediate (e, g);
+	  gsi = gsi_for_stmt (g);
+	}
+      else
+	gsi_insert_after (&gsi, g, GSI_NEW_STMT);
     }
   if (use_bool && has_debug_uses)
     {
-      tree temp = make_node (DEBUG_EXPR_DECL);
-      DECL_ARTIFICIAL (temp) = 1;
-      TREE_TYPE (temp) = TREE_TYPE (lhs);
-      SET_DECL_MODE (temp, TYPE_MODE (TREE_TYPE (lhs)));
-      tree t = build2 (LSHIFT_EXPR, TREE_TYPE (lhs), new_lhs, bit);
-      g = gimple_build_debug_bind (temp, t, g);
-      gsi_insert_after (&gsi, g, GSI_NEW_STMT);
+      tree temp = NULL_TREE;
+      if (!throws || after || single_pred_p (e->dest))
+	{
+	  temp = make_node (DEBUG_EXPR_DECL);
+	  DECL_ARTIFICIAL (temp) = 1;
+	  TREE_TYPE (temp) = TREE_TYPE (lhs);
+	  SET_DECL_MODE (temp, TYPE_MODE (TREE_TYPE (lhs)));
+	  tree t = build2 (LSHIFT_EXPR, TREE_TYPE (lhs), new_lhs, bit);
+	  g = gimple_build_debug_bind (temp, t, g);
+	  if (throws && !after)
+	    {
+	      gsi = gsi_after_labels (e->dest);
+	      gsi_insert_before (&gsi, g, GSI_SAME_STMT);
+	    }
+	  else
+	    gsi_insert_after (&gsi, g, GSI_NEW_STMT);
+	}
       FOR_EACH_IMM_USE_STMT (g, iter, use_lhs)
 	if (is_gimple_debug (g))
 	  {
 	    use_operand_p use_p;
-	    FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
-	      SET_USE (use_p, temp);
+	    if (temp == NULL_TREE)
+	      gimple_debug_bind_reset_value (g);
+	    else
+	      FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
+		SET_USE (use_p, temp);
 	    update_stmt (g);
 	  }
     }