-O4 needed

N8TM@aol.com N8TM@aol.com
Fri Sep 11 20:33:00 GMT 1998


In a message dated 9/11/98 11:04:29 AM Pacific Daylight Time, patha@ida.liu.se
writes:

> -fstrict-aliasing

A useful option which hasn't had enough testing to be included in a standard
option package.  Let's give it a work-out and see if we can document whether
there are recognizable coding practices which break it, and whether there are
any needed compiler fixes.  And let's get g77 up to this new-found speed of
gcc!

I've built egcs-1.1 successfully with -fstrict-aliasing, for the following
configs:
i686-pc-cygwin32(-Os -fstrict-aliasing...)
i686-pc-linux-gnulibc1 (" ")
hppa1.1-hpux10.20 (-O2 -fstrict-aliasing -pipe)
sgi-mips-irix6.4 (-O2 -fstrict-aliasing)

and verified that the gcc/g++/g77 testsuites were not affected for the latter
two, nor the libio/libstdc++ for the hp.

If I understand the comment on the web page, strict-aliasing is under
consideration for inclusion in the -Os and -O2 options.

The combination -O2 -fstrict-aliasing -malign-double -march=pentiumpro
-funroll-loops puts egcs-1.1/gcc at the top of the Wintel compiler benchmark
list on Andy Smith's web page http://www.proweb.co.uk/~ajsmith



More information about the Gcc mailing list