Hi!
On Fri, Feb 10, 2023 at 10:06:03AM +0100, Gerald Pfeifer wrote:
> Yes, thank you! Two minor suggestions/questions below:
>
> > --- a/htdocs/gcc-13/changes.html
> > +++ b/htdocs/gcc-13/changes.html
> > + <code>-fexcess-precision=fast</code>. The option affects mainly
>
> Here I'd say "mainly affects".
>
> > + IA-32/x86-64 where when defaulting to x87 math and in some cases on
> > + Motorola 68000 <code>float</code> and <code>double</code> expressions
> > + are evaluated in <code>long double</code> precision and S/390,
> > System z,
> > + IBM z Systems where <code>float</code> expressions are evaluated in
> > + <code>double</code> precision.
>
> The "where when" part proved a bit tricky for my brain. :-)
>
> I think it is precise, but am wondering whether
>
> ...IA-32/x64 using x87 math and in some cases on Motorola 68000, where
> <code>float</code> and <code>double</code> expressions are evaluated...
>
> might work? What do you think?
Thanks, committed with those tweaks.
Martin would like to see some note in porting_to.html for it too,
here is my attempt to do so:
diff --git a/htdocs/gcc-13/porting_to.html b/htdocs/gcc-13/porting_to.html
index 9a9a3147..5cbeefb6 100644
--- a/htdocs/gcc-13/porting_to.html
+++ b/htdocs/gcc-13/porting_to.html
@@ -120,5 +120,29 @@ the operand as an lvalue.<br>
}
</code></pre>
+<h3 id="excess-precision">Excess precision changes</h3>
+<p>GCC 13 implements in C++ <a href="changes.html#cxx">excess precision
support</a>
+which has been implemented just in the C front-end before. The new behavior is
+enabled by default in <code>-std=c++</code><em>NN</em> modes and when
+<code>FLT_EVAL_METHOD</code> is 1 or 2 affects behavior of floating point
+constants and expressions. E.g. for <code>FLT_EVAL_METHOD</code> equal
+to 2 on ia32:
+
+<pre><code>
+#include <stdlib.h>
+void foo (void) { if (1.1f + 3.3f != 1.1L + 3.3L) abort (); }
+void bar (void) { double d = 4.2; if (d == 4.2) abort (); }
+</code></pre>
+
+will not abort with standard excess precision, because constants and
expressions
+in <code>float</code> or <code>double</code> are evaluated in precision of
+<code>long double</code> and demoted only on casts or assignments, but will
+abort with fast excess precision, where whether something is evaluated in
+precision of <code>long double</code> or not depends on what evaluations are
+done in the i387 floating point stack or are spilled from it.
+
+The <code>-fexcess-precision=fast</code> option can be used to request the
+previous behavior.
+
</body>
</html>
Jakub
