This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: optimisation and AMD K7
- To: help-gcc at gnu dot org
- Subject: Re: optimisation and AMD K7
- From: n8tm at aol dot comnojunk (Tim Prince)
- Date: 27 Dec 1999 16:32:03 GMT
- Newsgroups: gnu.gcc.help
- Organization: AOL http://www.aol.com
- References: <Pine.LNX.3.96.991227133931.17421A-100000@prope.insa-lyon.fr>
- Xref: wodc7nx0 gnu.gcc.help:2303
>Which version of gcc or egcs and which
>options should I use to get the best perf or the smaller exe.
Your mileage surely will vary. gcc-2.95.2 does well on many examples; gcc-2.96
improves significantly in certain situations, and loses in others.
On x86 architectures, the option -Os, which is intended to produce smallest
code, often produces faster code as well, possibly in combination with
-funroll-loops or -funroll-all-loops.
I believe 2.96 has seen some work for k7. With 2.96, the combination
-march=pentiumpro -ffast-math activates conditional moves and the like, as well
as a number of bugs. The latest version of 2.96 has corrected some of the code
alignment problems of 2.95.2.
Tim Prince
tprince@computer.org