Official firefox-pgo thread

Ranguvar · 2009-11-16 06:40:52

Glad I could help, thanks

And yes, that's the case. It's the "name" of a revision in Mercurial -- a SHA-1 hash chunk, like Git uses.
If you want to use the very latest code, you can just comment out the "hg up" line in the PKGBUILD.

Pank · 2009-11-16 07:11:05

Ranguvar,
There does not seem to be random crashes with your i686 build of 3.6b2, as oppose to my own builds.
At the moment I am rebuilding with pkgrel 2.
Note: I have not had problems with compiling, only random crashes.

Could this be caused by makepkg.conf?

Edit: My own build of pkgrel 2 still crashes. . . There must be some kind of mal-configuration in my system. . .

Last edited by Pank (2009-11-16 07:27:42)

Ranguvar · 2009-11-16 07:47:17

I don't think your makepkg.conf could do much harm. The PKGBUILD sets its own CFLAGS, CXXFLAGS, and LDFLAGS. Maybe, though. Doing a `diff -u` between your makepkg.conf and the default one couldn't hurt. Also, what CPU do you have? The PKGBUILD sets -march=native, perhaps you have some CPU which causes it problems? Try with -march=i686?

Pank · 2009-11-16 07:55:13

I have a Core2Duo running i686.
Here is a snip from my makepkg.conf. I believe it is copy-pasted from the wiki

#########################################################################
# ARCHITECTURE, COMPILE FLAGS
#########################################################################
#
CARCH="i686"
CHOST="i686-pc-linux-gnu"

#-- Exclusive: will only run on i686
# -march (or -mcpu) builds exclusively for an architecture
# -mtune optimizes for an architecture, but builds for whole processor family
CFLAGS="-march=core2 -O2 -pipe -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"
#LDFLAGS=""
#-- Make Flags: change this for DistCC/SMP systems
MAKEFLAGS="-j3"

Should I try to change the FLAGS in the PKGBUILD?

Thanks,
Rasmus

Pank · 2009-11-16 11:13:22

I have done some testing. Here are the results:

Compiling using CFLAGS="native -O2 -pipe" etc. Firefox crashes instantly.
Compiling using CFLAGS="-march=core2 -O2 -pipe -fomit-frame-pointer" etc Firefox crashes instantly.
Compiling using CFLAGS="-march=i686" etc Firefox seems to be stable.

I don't know why this suddenly makes the compiled Fx unstable. This is not a problem with Fx 3.5. . .

If somebody could provide a build optimized for core2duo I could test it to determine whether any optimized build is invalid or whether it is a local problem.

--Rasmus

Ranguvar · 2009-11-16 12:10:39

I also compile for Core 2 (-march=core2) in my personal builds (albeit 64-bit), so that is very likely not the problem. But if you want, I'll make a 32-bit Core 2 build.

harryNID · 2009-11-17 00:28:05

Well I guess I'm on the same bandwith as Pank. I thought my makepkg.conf (w custom CFLAGS/LDFLAGS) could have been the culprit. I have found it does make a difference what you have set in there as mplayer-minimal-svn will error out during a build if you have any volatile flags set (I know, I know, I'm being naughty). I have to set it to CFLAGS="-march=native -O2 -pipe -fomit-frame-pointer" and get rid of my custom LDFLAGS before it will compile successfully. The funny thing is its PKGBUILD unsets both flags too. It's weird I should have to do that!

Anyway, after setting makepkg to a sane default of CFLAGS="-march=native -O2 -pipe firefox-pgo-beta 3.6b2-2 is still doing the same after I recompiled (seg faulting). I removed that version and tried Ranguvar's "i686 generic" and voila it's working. So far no seg fault and Sunspider is working again albeit slightly slower than my PGO build of a2. I hope this beta-testing of sorts helps. I'm going to give the generic build a tour of the grounds per say and see if anything happens. I will let you know.

Thanks guys

Ranguvar · 2009-11-17 04:50:46

Huh. I'm also getting errors when I try to do a 32-bit compile with "-march=core2 -O2 -pipe".

After extensive testing, it appears that '-march=core2' breaks 32-bit compiles, but not 64-bit ones. '-march=core2 -mno-ssse3' also breaks. For optimization options very similar to '-march=core2' that should not break, I advise those that are having problems to try '-march=nocona -mssse3'.

Please let me know how that works for you all
If you have problems and don't use a Core 2 CPU, try messing around with -march as I did. Find the best setting that works.
See here for info on CPU settings for GCC: http://gcc.gnu.org/onlinedocs/gcc-4.4.2 … tions.html

toxygen · 2009-11-17 06:47:25

dropping by to say hi, so far ff3.7a1 from Ranguvar's firefox-pgo-minefield pkg is working great. no crashes, my most used pages (youtube, picture pages, video pages, heavy multimedia pages) have loaded with no issues. havent tested html5 too much as i'm not expecting too much improvement there until support matures.
say what you will about firefox, bloated or such, i love using it. with the right combination of extensions, config options, and compile options, it is one sleek piece of software.

Last edited by toxygen (2009-11-17 06:49:10)

Pank · 2009-11-17 15:50:58

I tried to compile with -march=nocona -mssse3 (whatever it means). It still breaks on my system. I only changed it in the PKGBUILD though.

Find the best setting that works.

It is very time consuming, you know

Also: Since 3.6 every build has been successful. This was definitely not the case for 3.5. As a consequence of this and the many times I have build Firefox 3.6 I have dropped the pre-PGO build. Does this add anything beside increased chance of creating a good build?

I will try the 3.7a package and report back.
edit: I also get segmentation fault with 3.7a1-PGO...

--Rasmus

Last edited by Pank (2009-11-17 17:16:35)

Ranguvar · 2009-11-17 20:18:45

Pank wrote:

...I have dropped the pre-PGO build. Does this add anything beside increased chance of creating a good build?

No, but "good build" matters even after the actual compile It may prevent some segfaults.

Pank wrote:

I also get segmentation fault with 3.7a1-PGO...

I wonder if something is odd with your system. Can you give me a backtrace? Of 3.6b2-2. I doubt I'll be able to do much with it, but maybe.

Pank · 2009-11-18 08:54:30

Firefox starts up a lot of processes. I attached gdb to the /usr/bin/firefox. I also tried to attach it to run-mozilla.sh but the result seems to be similar. I have not done a lot of debugging, so if something is missing please let me know.

GNU gdb (GDB) 7.0
Copyright (C) 2009 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "i686-pc-linux-gnu".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
(gdb) handle SIG33 pass nostop noprint
Signal        Stop    Print    Pass to program    Description
SIG33         No    No    Yes        Real-time event 33
(gdb) set pagination 0
(gdb) f[Kattach 5628
Attaching to process 5628
Reading symbols from /bin/bash...(no debugging symbols found)...done.
Reading symbols from /lib/libreadline.so.6...(no debugging symbols found)...done.
Loaded symbols for /lib/libreadline.so.6
Reading symbols from /lib/libdl.so.2...(no debugging symbols found)...done.
Loaded symbols for /lib/libdl.so.2
Reading symbols from /lib/libc.so.6...(no debugging symbols found)...done.
Loaded symbols for /lib/libc.so.6
Reading symbols from /lib/libncursesw.so.5...(no debugging symbols found)...done.
Loaded symbols for /lib/libncursesw.so.5
Reading symbols from /lib/ld-linux.so.2...(no debugging symbols found)...done.
Loaded symbols for /lib/ld-linux.so.2
0xb7749424 in __kernel_vsyscall ()
(gdb) continue 
Continuing.

Program exited with code 0213.
(gdb) backtrace full
No stack.
(gdb) info registers 
The program has no registers now.
(gdb) z[Kx/16i $pc
No registers.
(gdb) thread apply all backtrace
(gdb) quit

Ranguvar · 2009-11-18 11:11:38

That's really odd... I still suck at GDB, but that looks like it's pointing to either the kernel itself, or glibc!
I'm not sure what to do... my only offer is if you have time, can you rebuild glibc and also perhaps Bash/readline and ncurses with the strip makepkg option off, so there will be debugging symbols, and then re-run the backtrace?
No promises though.

harryNID · 2009-11-21 00:59:24

Well I can also confirm the CFLAGS="native -O2 -pipe" issue. I finally got a working build when I switched to CFLAGS="-march=i686". I'm on a P4 and I also tried "-march=pentium4 -O2 -pipe" but that caused a segmentation fault as well. It appears pentium4 is also broke. I am going to do some more builds and see what happens. I will post my results.

toods · 2009-11-23 10:30:14

'Clobber_all' Build Problem'

The fix I use until Mozilla sort it properly is to edit line 233 of 'client.mk' as follows:

- $(MAKE) -f $(TOPSRCDIR)/client.mk maybe_clobber_profiledbuild
+ $(MAKE) -f $(TOPSRCDIR)/client.mk clobber_all

After applying this patch the 2 PGO builds progress as expected. I don't understand why the 'libffi' directory doesn't like 'maybe_clobber_profiledbuild'. I guess it's an issue with 'makefil.in' but I don't know enough about 'make' to figure it out. I guess it's for Mozilla to do that .

Bill.

Last edited by toods (2009-11-23 10:31:21)

Mogger · 2009-11-24 22:40:52

Any heads up regarding this?

c++ -o nsThreadUtils.o -c -I../../dist/include/system_wrappers -include /home/hans/builds/firefox-pgo/src/mozilla-1.9.1/config/gcc_hidden.h -DOSTYPE=\"Linux2.6\" -DOSARCH=Linux -DTARGET_XPCOM_ABI=\"x86-gcc3\" -I/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/../build  -I/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue -I. -I../../dist/include/string -I../../dist/include   -I../../dist/include/xpcom -I/usr/include/nspr    -I/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo/dist/sdk/include    -fPIC   -fno-rtti -fno-exceptions -Wall -Wpointer-arith -Woverloaded-virtual -Wsynth -Wno-ctor-dtor-privacy -Wno-non-virtual-dtor -Wcast-align -Wno-invalid-offsetof -Wno-long-long -march=native -O2 -pipe -fno-strict-aliasing -fshort-wchar -pthread -pipe  -DNDEBUG -DTRIMMED -fprofile-use -Os -freorder-blocks -fno-reorder-functions    -DMOZILLA_CLIENT -include ../../mozilla-config.h -Wp,-MD,.deps/nsThreadUtils.pp /home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In destructor 'virtual nsRunnable::~nsRunnable()':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_ZN10nsRunnableD0Ev' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is 286e163d instead of 66f43ab
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In destructor 'virtual nsRunnable::~nsRunnable()':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_ZN10nsRunnableD1Ev' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is f4038c8a instead of da02d91c
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'PRBool hasPendingEvents(nsIThread*)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z16hasPendingEventsP9nsIThread' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is fbe13b8 instead of d8933d06
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z16hasPendingEventsP9nsIThread' while reading counter 'indirect_call'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is fbe13b8 instead of d8933d06
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'nsresult NS_DispatchToMainThread(nsIRunnable*, PRUint32)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z23NS_DispatchToMainThreadP11nsIRunnablej' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is e632f58b instead of e7c6253c
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z23NS_DispatchToMainThreadP11nsIRunnablej' while reading counter 'indirect_call'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is e632f58b instead of e7c6253c
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'PRBool NS_ProcessNextEvent(nsIThread*, PRBool)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: warning: no coverage for function '_Z19NS_ProcessNextEventP9nsIThreadi' found
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: warning: no coverage for function '_Z19NS_ProcessNextEventP9nsIThreadi' found
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'PRBool NS_HasPendingEvents(nsIThread*)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z19NS_HasPendingEventsP9nsIThread' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is 8970e9ad instead of 4123204e
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'nsresult NS_ProcessPendingEvents(nsIThread*, PRIntervalTime)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z23NS_ProcessPendingEventsP9nsIThreadj' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is f224dd79 instead of a3ad36b
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z23NS_ProcessPendingEventsP9nsIThreadj' while reading counter 'indirect_call'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is f224dd79 instead of a3ad36b
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp: In function 'nsresult NS_DispatchToCurrentThread(nsIRunnable*)':
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z26NS_DispatchToCurrentThreadP11nsIRunnable' while reading counter 'arcs'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is f084b40 instead of 2f5603ca
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: error: coverage mismatch for function '_Z26NS_DispatchToCurrentThreadP11nsIRunnable' while reading counter 'indirect_call'
/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/xpcom/glue/nsThreadUtils.cpp:240: note: checksum is f084b40 instead of 2f5603ca
make[6]: *** [nsThreadUtils.o] Error 1
make[6]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo/xpcom/glue'
make[5]: *** [libs] Error 2
make[5]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo/xpcom'
make[4]: *** [libs_tier_xpcom] Error 2
make[4]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo'
make[3]: *** [tier_xpcom] Error 2
make[3]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo'
make[2]: *** [default] Error 2
make[2]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1/ff-pgo'
make[1]: *** [build] Error 2
make[1]: Leaving directory `/home/hans/builds/firefox-pgo/src/mozilla-1.9.1'
make: *** [profiledbuild] Error 2
==> ERROR: Build Failed.
    Aborting...

Trying to compile firefox-pgo 3.5.5-6 on i686. Default PKGBUILD and mozconfig, Firefox runs some tests before showing this error.

Last edited by Mogger (2009-11-24 23:04:54)

Ranguvar · 2009-11-24 22:47:02

It looks like your log got truncated at the end of each line?

Mogger · 2009-11-24 23:07:02

Ops sorry about that, bad copy-paste. Post above edited.

blasse · 2009-11-25 06:00:21

uncommenting potencial-pgo-fix.patch may help.

Mogger · 2009-11-25 22:45:10

Nope, same error. Without jemalloc PGO generates another error, earlier than the one I posted above I think.

Ranguvar · 2009-11-25 23:21:03

You shouldn't need to disable the jemalloc PGO patch to enable the potential PGO fix patch anymore.

toxygen · 2009-11-26 09:56:46

Ranguvar, have you been able to build the trunk-pgo (or just plain trunk) without getting segmentation fault errors when running the sunspider tests? I have tried the last couple of days, by commenting out the hg up line and by changing the revision manually on your PKGBUILD, and after installing, i get the crash, in normal and safe mode. only working version is leaving the PKGBUILD at default (20091115 version).

Ranguvar · 2009-11-26 17:04:02

I haven't tried, actually.
I'm waiting on at least two of these bugs to get fixed in mozilla-central, which should be soon

toxygen · 2009-11-26 18:00:00

at any rate, i just finished building 3.6b4 (updated version/sha256sum on the pkgbuild you provided in aur), and the jscript scores are decent (was getting 1800-1900 range, now at 960ms). the 3.7a1 score was around 870ms. 3.6b4 seems stable to me, where 3.6b3/2 crashed on launch a lot, and with sunspider tests. 3.6b4 has not crashed yet, even with extensions enabled.

blasse · 2009-11-26 18:04:28

3.6b4 will be updated today, I'm testing it with patches from bugzilla to fix these bugs

Arch Linux

#426 2009-11-16 06:40:52

Re: Official firefox-pgo thread

#427 2009-11-16 07:11:05

Re: Official firefox-pgo thread

#428 2009-11-16 07:47:17

Re: Official firefox-pgo thread

#429 2009-11-16 07:55:13

Re: Official firefox-pgo thread

#430 2009-11-16 11:13:22

Re: Official firefox-pgo thread

#431 2009-11-16 12:10:39

Re: Official firefox-pgo thread

#432 2009-11-17 00:28:05

Re: Official firefox-pgo thread

#433 2009-11-17 04:50:46

Re: Official firefox-pgo thread

#434 2009-11-17 06:47:25

Re: Official firefox-pgo thread

#435 2009-11-17 15:50:58

Re: Official firefox-pgo thread

#436 2009-11-17 20:18:45

Re: Official firefox-pgo thread

#437 2009-11-18 08:54:30

Re: Official firefox-pgo thread

#438 2009-11-18 11:11:38

Re: Official firefox-pgo thread

#439 2009-11-21 00:59:24

Re: Official firefox-pgo thread

#440 2009-11-23 10:30:14

Re: Official firefox-pgo thread

#441 2009-11-24 22:40:52

Re: Official firefox-pgo thread

#442 2009-11-24 22:47:02

Re: Official firefox-pgo thread

#443 2009-11-24 23:07:02

Re: Official firefox-pgo thread

#444 2009-11-25 06:00:21

Re: Official firefox-pgo thread

#445 2009-11-25 22:45:10

Re: Official firefox-pgo thread

#446 2009-11-25 23:21:03

Re: Official firefox-pgo thread

#447 2009-11-26 09:56:46

Re: Official firefox-pgo thread

#448 2009-11-26 17:04:02

Re: Official firefox-pgo thread

#449 2009-11-26 18:00:00

Re: Official firefox-pgo thread

#450 2009-11-26 18:04:28

Re: Official firefox-pgo thread

Board footer