Joao Seabra wrote:

-O3 (finline-functions to be exact) might create useless code that will
polute cache and throws away the advantages of using trace caches.

I use -O2 and this works fine for me.

Kind regards,

Joao Seabra


On Tue, 28 Oct 2003, Chris Graves wrote:



OK, so some folks swear by "-O3" optimizations, but alot of what I've
read suggests "-O2" produces "faster" binaries.

I have a Pentium III (coppermine) @ 900MHz with 512M RAM and 1G swap.
What CFLAGS do I want? (questionably unstable optimizations that do
produce better results are OK)

-chris



--
[EMAIL PROTECTED] mailing list




--
[EMAIL PROTECTED] mailing list



I have made a test with all the CFLAGS explained in the GCC info pages, using povray as the benchmark app (because it does lot of memory/integer/floating point operations) and these are just a few of my conclusions (when I come back home I'll post the full research, I wanted to do this before but never have the time, so now I see such a thread on the mailing list, I think this is the best opportunity):

(These are from my memory now, if I put (a little bit) wrong cflags, please apologize me.

-O3 = -O2 + -finline-functions -frename-regs

povray benchmarked faster with -O2, slower -O3, so I tested -O2 separatedly with these two additional -O3 CFLAGS:

-O2 -finline-functions = slower than -O2

-O2 -frename-regs = FASTER than -O2

I'll post my complete benchmarked results ASAP. It includes many more CFLAGS (but for my architecture only, an AMD Athlon-XP).

Salu2.

Javier Villavicencio.


-- [EMAIL PROTECTED] mailing list



Reply via email to