Hi All,

This patch fixes the placement of the single-use requirement for xorsign.

on `x = a * copysign (1, b)` the intention was that `copysign (1, b)` be
single use, and not `x`. Requiring `x` to be single use blocks transformation
where we do want it to occur.

Regtested on aarch64-none-linux-gnu and no regressions
(only target currently supporting this).

Ok for trunk?

gcc/
2017-08-22  Tamar Christina  <tamar.christ...@arm.com>

        PR middle-end/19706
        * tree-ssa-math-opts.c (convert_expand_mult_copysign):
        Fix single-use check.

-- 
diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
index 87940b6e00db0d34a472504cf70923d1d334eccb..f26d105c16145bc77dd5e3a05b13f22aff43d4d8 100644
--- a/gcc/tree-ssa-math-opts.c
+++ b/gcc/tree-ssa-math-opts.c
@@ -3200,21 +3200,20 @@ convert_expand_mult_copysign (gimple *stmt, gimple_stmt_iterator *gsi)
   type = TREE_TYPE (lhs);
   machine_mode mode = TYPE_MODE (type);
 
-  if (HONOR_SNANS (type) || !has_single_use (lhs))
+  if (HONOR_SNANS (type))
     return false;
 
   if (TREE_CODE (treeop0) == SSA_NAME && TREE_CODE (treeop1) == SSA_NAME)
     {
       gimple *call0 = SSA_NAME_DEF_STMT (treeop0);
-      if (!is_copysign_call_with_1 (call0))
+      if (!is_copysign_call_with_1 (call0) || !has_single_use (treeop0))
 	{
 	  call0 = SSA_NAME_DEF_STMT (treeop1);
-	  if (!is_copysign_call_with_1 (call0))
+	  if (!is_copysign_call_with_1 (call0) || !has_single_use (treeop1))
 	    return false;
 
 	  treeop1 = treeop0;
 	}
-
 	if (optab_handler (xorsign_optab, mode) == CODE_FOR_nothing)
 	  return false;
 

Reply via email to