Patch for Quad.cc compile error on Sun/GCC

[Carsten Clark]

2009/5/4 John W. Eaton <jwe at octave.org>:
>
> I checked in the change, but what about
>
>  quad (@(x) x, single (0), single (2))
>
> ?  Are you using the Sun Fortran compiler?  My guess is that it will
> work, and that GCC and the Sun Fortran compiler will handle floats in
> a compatible way, unlike for doubles.

I was using g77 from GCC 4.  This works:

$ octave -qf
octave:1> quad ( @(x) x, 0, 2 )
ans =  2
octave:2> quad ( @(x) x, single(0), single(2) )
ans =  2.0000

I rebuilt using g++4 and this Sun Studio Fortran compiler:

$ /opt/SunStudioExpress/bin/f77 -V
NOTICE: Invoking /opt/SunStudioExpress/bin/f90 -f77 -ftrap=%none -V
f90: Sun Ceres Fortran 95 8.3 SunOS_i386 2008/10/22

and that works, too (your guess is right!):

octave:1> quad ( @(x) x, 0, 2 )
ans =  2
octave:2> quad ( @(x) x, single(0), single(2) )
ans =  2

Actually, the GCC manual
(http://gcc.gnu.org/onlinedocs/gcc-4.3.3/gcc/Interoperation.html)
states that the compatibility problem that access_double() solves only
occurs on SPARC systems.  I'm on an Intel PC.  Changing the #ifdefs in
Quad.cc and sun-utils.h to __sparc, as in the attached patch, I can
still run the above test code successfully, with both g77 and Sun
Studio f77.  (I verified that __sparc is not defined for me and the
access_double() lines are not compiled.)

I've checked a SPARC system at work, and it does define __sparc and
sparc.  The Sun Studio 12 User's Guide
(http://docs.sun.com/app/docs/doc/819-5267/6n7c46edt) also lists these
symbols as defined on SPARC only.  However, I don't have GCC 4 on that
system, so probably can't try building Octave.

At any rate, access_double() and assign_double() aren't necessary on
my system, and it looks like this should be true of Solaris/Intel
systems in general, unless I'm a fluke for some reason.

Regards,
Carsten Clark
-------------- next part --------------
diff -r 7ea76c8a59f7 liboctave/ChangeLog
--- a/liboctave/ChangeLog	Mon May 04 21:22:36 2009 -0400
+++ b/liboctave/ChangeLog	Tue May 05 02:17:13 2009 -0400
@@ -1,3 +1,8 @@
+2009-05-05  Carsten Clark  <tantumquantum+gnuoctave at gmail.com>
+
+	* Quad.cc (user_function), sun-utils.h: use access_double() and
+	assign_double() only on SPARC, not on all Sun.
+
 2009-05-04  Carsten Clark  <tantumquantum+gnuoctave at gmail.com>
 
 	* Quad.cc (float_user_function): Remove Sun/GCC special case.
diff -r 7ea76c8a59f7 liboctave/Quad.cc
--- a/liboctave/Quad.cc	Mon May 04 21:22:36 2009 -0400
+++ b/liboctave/Quad.cc	Tue May 05 02:17:13 2009 -0400
@@ -77,7 +77,7 @@
 {
   BEGIN_INTERRUPT_WITH_EXCEPTIONS;
 
-#if defined (sun) && defined (__GNUC__)
+#if defined (__sparc) && defined (__GNUC__)
   double xx = access_double (x);
 #else
   double xx = *x;
@@ -87,7 +87,7 @@
 
   double xresult = (*user_fcn) (xx);
 
-#if defined (sun) && defined (__GNUC__)
+#if defined (__sparc) && defined (__GNUC__)
   assign_double (result, xresult);
 #else
   *result = xresult;
diff -r 7ea76c8a59f7 liboctave/sun-utils.h
--- a/liboctave/sun-utils.h	Mon May 04 21:22:36 2009 -0400
+++ b/liboctave/sun-utils.h	Tue May 05 02:17:13 2009 -0400
@@ -24,12 +24,12 @@
 #define octave_sun_utils_h 1
 
 // This is only needed to dereference pointers to doubles if mixing
-// GCC and Sun f77/cc compiled code.  See the GCC manual (where the
+// GCC and Sun SPARC f77/cc compiled code.  See the GCC manual (where the
 // function access_double() is described) and the Sun f77 manual,
 // which explains that doubles are not always aligned on 8 byte
 // boundaries.
 
-#if defined (sun) && defined (__GNUC__)
+#if defined (__sparc) && defined (__GNUC__)
 
 inline double
 access_double (double *unaligned_ptr)