see http://wdk.dyndns.org/flags.png.
Guess which one is -Os, 5th is -O3. Far right is: "-Wno-deprecated -O2 -pipe -Wall -funroll-loops -fprefetch-loop-arrays -finline-limit=1200 -falign-functions=32" 10 runs per flag combination, spikes are because its on a server in use. -Os is consistently as bad as this, but it is impossible to try every combination. Other flags can vary depending on the application used for testing, cpu etc. This is part of a project I am doing and more detail will be available when finished (some of the runs shown here are not representative of previous runs - consistency with the small differences some flags make is a problem on an in-use system over long time spans). Note that a graph like this takes ~10 hours to run, and many more to test and set up. It has been done using a perl driver script to manage the process. Note that I am on a dialup, so accessing the png may be a bit slow. BillK On Thu, 2003-10-23 at 05:05, Stroller wrote: > On Oct 22, 2003, at 5:43 pm, Hall Stevenson wrote: > > > At 11:59 AM 10/22/2003, you wrote: > >> Arg! I'm in the middle of recompiling my WHOLE kdelibs using -Os at > >> the advice of another poster here. > >> I really would like a definitive answer here, and don't have the time > >> for multiple recompilations myself. > > > > When you do figure out which is the "best" flag to use, let us know if > > your system is truly noticeably faster. My money says you won't notice > > the difference. The compiled files will be bigger, but probably not > > faster (to a human). > > It was the previous posters' contention that -O2 would not be > noticeably faster than -Os, but would produce larger binaries & take > far longer to compile. It was Mr Kenworthy's emphasis in the statement > "-Os was by *FAR* the slowest" that suggested the difference might be > noticeable. > > As I posted within the last week, the laptop with which I am concerned > is a PII 400 with only 64meg of RAM. More RAM for this proprietary > form-factor sub-notebook might be argued to be prohibitively expensive, > at $160 or so for 64meg. I think I'll probably have to bite the bullet, > but I think that on this machine performance may be more noticeable > than on most. Unfortunately compilations take far longer also. At > present I am more interested in getting compilation to support > prelinking, which I'm finding quite a confusing process, but since I > have made several big recompiles in the last few days it would be nice > to know that I'm using the optimal flags. > > Stroller. > > > -- > [EMAIL PROTECTED] mailing list -- William Kenworthy <[EMAIL PROTECTED]> -- [EMAIL PROTECTED] mailing list