Possible bug with ABS_EXPR and complex operands

[Peter Gerwinski]

Hello,

in the big `switch' statement in expand_expr(), case ABS_EXPR,
the absolute value of a complex operand gets the mode of the
complex type.  IMHO it should have the mode of the underlying
type - abs(complex)=real.

I am suggesting the change below.

    Peter

Thu 30 Sep 1999 27:25  Peter Gerwinski <peter@gerwinski.de>

        * expr.c (expand_expr): the absolute value of a complex
        expression is of the underlying type.

--- expr.c.orig	Thu Jul  1 00:59:55 1999
+++ expr.c	Fri Oct  1 03:24:24 1999
@@ -7273,10 +7274,17 @@
     case ABS_EXPR:
       op0 = expand_expr (TREE_OPERAND (exp, 0), subtarget, VOIDmode, 0);
 
-      /* Handle complex values specially.  */
-      if (GET_MODE_CLASS (mode) == MODE_COMPLEX_INT
-	  || GET_MODE_CLASS (mode) == MODE_COMPLEX_FLOAT)
-	return expand_complex_abs (mode, op0, target, unsignedp);
+      /* Handle complex values specially.
+         It is the mode of the operand, not the mode of the return
+	 value that is tested here. ABS(complex) does not return
+	 complex type.  */
+      {
+	enum machine_mode op0_mode =
+	  TYPE_MODE (TREE_TYPE (TREE_OPERAND (exp, 0)));
+	if (GET_MODE_CLASS (op0_mode) == MODE_COMPLEX_INT
+	    || GET_MODE_CLASS (op0_mode) == MODE_COMPLEX_FLOAT)
+	  return expand_complex_abs (op0_mode, op0, target, unsignedp);
+      }
 
       /* Unsigned abs is simply the operand.  Testing here means we don't
 	 risk generating incorrect code below.  */