GCC option: -msse and -msse2

[Daihard]

Hi.

I would like to perform code optimization on the program I build using GCC
2.96. I currently only use "-O2 -march=i686," which will probably only
optimize the code for the general Pentium Pro architecture. I'd like to
optimize it specifically for Pentium III. Now, I've read somewhere that
using the "-msse" or "-msse2" switch will achive that goal. However, GCC
2.96 doesn't recognize "-msse2," so I guess I can only use "-msse." What
I'd like to know is, what are those switches for, and what are the
differences between the two? Or is there any other way to optimize for
Pentium III?

Thanks in advance!
Dai

[Tobias Oed]

Daihard wrote:

> Hi.
>
> I would like to perform code optimization on the program I build using GCC
> 2.96. I currently only use "-O2 -march=i686," which will probably only
> optimize the code for the general Pentium Pro architecture. I'd like to
> optimize it specifically for Pentium III. Now, I've read somewhere that
> using the "-msse" or "-msse2" switch will achive that goal. However, GCC
> 2.96 doesn't recognize "-msse2," so I guess I can only use "-msse." What
> I'd like to know is, what are those switches for, and what are the
> differences between the two? Or is there any other way to optimize for
> Pentium III?

-msse let's the compiler use the sse instructions of the cpu. These can use
the mmx registers to do floating point arithmetic (single precision). So
you have 8 more floating point registers and each can hold 2 floats.
Operations on these two floats can be done in parallel. -msse2 uses
registers in newer pentiums (or opteron) to do the same with doubles. I
couldn't see a measurable speedup in my benchmarks (but an illegal
instruction on athlonXP).
Tobias.

--
unix [url]http://www.faqs.org/faqs/by-newsgroup/comp/comp.unix.programmer.html[/url]
clc [url]http://www.eskimo.com/~scs/C-faq/top.html[/url]
fclc (french): [url]http://www.isty-info.uvsq.fr/~rumeau/fclc/[/url]