[BUG][GCC] *Serious* optimization bug: Any gcc specialistsout there?

Andres Valloud avalloud at exobox.com
Fri Aug 4 20:29:57 UTC 2000


Hi.

Here's some mail from the GCC list.

~~~~~~~~~~~~~
In a message dated 11/19/99 12:17:09 AM Pacific Standard Time,
rth at cygnus.com 
writes:

> It would be
>  > nice if gcc could expand the safe versions of these functions in
>  > -fno-fast-math mode.
>  
>  The code would be too big; you'd gain nothing.

Code size is a problem only for a minority of the functions in
mathinline.h 
e.g. asin(), pow(), which aren't speeded up by in-lining and aren't so
well 
implemented anyway.  The greater problem is the lack of protection on
extreme 
values of data.

Tim
tprince at computer.org
~~~~~~~~~~~~~

~~~~~~~~~~~~~
On Fri, Nov 19, 1999 at 08:49:25AM -0500, N8TM at aol.com wrote:
> The greater problem is the lack of protection on extreme 
> values of data.

Exactly.  It's the argument reduction loop that kills sin/cos
for expansion inline, when looking for Proper versions.


r~
~~~~~~~~~~~~~


I don't understand the Proper versions part, but...

Andres.





More information about the Squeak-dev mailing list