Ted's PGO build

[RyanVM]

uh, the post right above yours?

[swsnyder]

swsnyder: would be interesting to rebuild that with debug symbols (set MOZ_DEBUG_SYMBOLS=1 in your mozconfig) and get a stack, to find out where it's crashing. Wouldn't surprise me if SeaMonkey has some modules of its own that are fragile to PGO. We can turn those off in their respective Makefiles till they get sorted out.

I pulled the trunk on 03 Mar., rebuilt with profiling, and now see no crashes.

Maybe I just missed the sqlite exclusion in that last checkout. In any case the profiled build of SeaMonkey 2.0a1pre is working well for me.

[Incognito129]

i have msvc++ 2005 redist. installed but the PGO build keeps crashing. Is there anything else I need? I'm using winxp sp2.

[Ted Mielczarek]

Incognito: no, in fact our recent trunk builds should not even need the VC redist since we're shipping our own CRT. Are you submitting crash reports? These builds appear to be working fine for everyone else, so there's something unique about your setup.

[trieste]

i have msvc++ 2005 redist. installed but the PGO build keeps crashing. Is there anything else I need? I'm using winxp sp2.

Are you using the PGO build that Ted posted in this thread? There are issues with it, since it was created to test for possible bugs. Just use the latest nightlies.

[Foxicated]

Are you using the PGO build that Ted posted in this thread? There are issues with it, since it was created to test for possible bugs. Just use the latest nightlies.

Where can I download the latest nightlies ( PGO enabled ) please? You don't mean the trunk builds I take it?

[trieste]

Yeah the trunk builds are all humming nicely on PGO.

[Foxicated]

Yeah the trunk builds are all humming nicely on PGO.

Alright then, thanks!
Downloading .....

[tqft]

Linux here - ubuntu gutsy - no luck building a PGO build, it seems to have regressed - updated version off gcc to latest svn [other programs - vlc git - would not build as well] and still no luck - now won't get to the part where it opens ff.

I haven't seen much traffic on the bug lately - does anyone know anything about when some more action may happen. I know some of the patches are waiting for movement in the ff3 betas - but I can't find much info.

[OldSkoOL]

Has anyone found any of the latest nightlies to be as fast or faster than teds build?

For me, teds build is the fastest - everything else recently is always slower.

[Ted Mielczarek]

tqft: can you be more specific on what you're seeing? We only tested with GCC 4.1, so with newer versions YMMV.

OldSkoOL: it's probably a factor of the profiling input. Right now we're just profiling on startup/shutdown. At least one of my builds was profiled on the SunSpider test, so it would obviously be faster there. (and likely in overall JS perf) We will add some more profiling input eventually.

[tqft]

Ted:
/media/sdb2/mozilla/mozilla/netwerk/cookie/src/nsCookieService.cpp:1: error: ‘nsCookieService.gcda’ is version ‘401p’, expected version ‘404e’
cc1plus: warnings being treated as errors
/media/sdb2/mozilla/mozilla/netwerk/cookie/src/nsCookie.cpp:1: error: ‘nsCookie.gcda’ is version ‘401p’, expected version ‘404e’
make[8]: *** [nsCookie.o] Error 1
and that is where it bails
With prev version gcc 4.1 didn't get that and current gcc does build non-profiled ff
Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9b5pre) Gecko/2008030908 Minefield/3.0b5pre ID:2008030908

[Ted Mielczarek]

tqft: interesting, but that sounds like a GCC bug. :-/

[tqft]

Ted, I noticed you removed yoruself from the bug and it has gone to P2 - that is the linux pgo enabled build bug. Is it still a possibility or is the reason the tree is still closed that bug? Or will those changes be backed out and linux pgo will come after ff3 ?

[Ted Mielczarek]

tqft: there just hasn't been a lot of momentum to turn it on. It's probably late in the game to get it done now, but maybe if someone champions it. It's not going to be me though.