[CM] heronstalker:2006-06-26 [Release, Mac OS 10.4, i386] (O

Posted July 27th, 2006, 3:32 pm

Hello all, this is only my second posting to the forums, but I've decided to put up the optimized builds I have been using for personal use that runs on Intel Macs. Please note that this build is NOT a universal binary. Though I find UBs an awesome concept, I'm only working with a Mac Mini here, and I like to conserve disk space and keep only the stuff I need, so this build will only run on the new Intel based Macs. Here's a link to my <a href="http://www2.hawaii.edu/~coreya/camino/camino.mozconfig">.mozconfig</a> file.

This is is build:
<a href="http://www2.hawaii.edu/~coreya/camino/Camino-1.0.2-Intel.dmg">Camino 1.0.2 <i>x86</i> </a>

Tell me what you think. Enjoy!

Posted July 27th, 2006, 6:24 pm

It works great for me. Thanks! One questions, why not use "march=prescott" instead of "march=i686"?

Posted July 27th, 2006, 6:52 pm

Great to see we've got optimized builds for Intel now :]

Posted July 27th, 2006, 7:03 pm

davedit wrote:Great to see we've got optimized builds for Intel now :]

I've been trying for ages to do Intel builds, but I still can't get my G5 to cross-compile. I dunno why. All I get is some unintelligable crash log when I try to launch it (see this post for an example crash log). So I've basically given up.

Posted July 27th, 2006, 7:38 pm

wrc_fan wrote:It works great for me. Thanks! One questions, why not use "march=prescott" instead of "march=i686"?

I'm seriously debating whether or not if I should build with the prescott option instead of the i686 flag. The other night when the release code of Firefox 1.5.0.5 came out, I did do an optimized build using the prescott flag, but honestly the build just "feels" slower, so now I'm thinking I should trash that build and redo it with the i686 option. If anyone else out there is using prescott and getting better performance please let me know and from that point foward I will post either i686 or prescott for future releases of Camino. I am sort of wanting to shy away from using the prescott flag, as the GCC online manual says prescott was only meant for Pentium 4 processors and includes SSE3 optimizations. Pentium 4 is the keyword here because all of us know how badly Intel screwed up the design on that chip, and since the pipeline in the Core chips are less than half the size of Prescott's 31 stage pipe, i686 seems to be the way to go for the time being. If anyone else is doing Intel builds out there for OS X please let me know what you think about this. Here is my .mozconfig I used to compile Firefox 1.5.0.5 the other night:

. $topsrcdir/browser/config/mozconfig

mk_add_options MOZ_MAKE_FLAGS=-j4
mk_add_options MOZ_CO_PROJECT=browser

ac_add_options --enable-application=browser
ac_add_options --enable-reorder
ac_add_options --enable-strip
ac_add_options --enable-static
ac_add_options --enable-static-libs
ac_add_options --enable-svg
ac_add_options --enable-canvas
ac_add_options --enable-pthreads
ac_add_options --enable-optimize="-O3 -ftree-vectorize -fast -march=prescott -mfpmath=sse,387 -fforce-addr -mieee-fp -msse3 -msse2 -msse -mmmx -fsched-spec-load-dangerous -ftree-loop-linear -fivopts"

ac_add_options --disable-debug
ac_add_options --disable-tests
ac_add_options --disable-shared

ac_add_options --with-macos-sdk=/Developer/SDKs/MacOSX10.4u.sdk

ac_add_options --enable-official-branding

ac_add_options --without-system-png
ac_add_options --without-system-mng
ac_add_options --without-system-nspr
ac_add_options --without-system-zlib
ac_add_options --without-system-jpeg

It could also be that the compile options -ftree-vectorize -fsched-spec-load-dangerous -ftree-loop-linear -fivopts are slowing things down, but open source is all about trying new things, right?

Posted July 27th, 2006, 8:02 pm

If you've got a fast enough Mac, you could try builds with each of those options (or combinations thereof) disabled, and see if things are better, but I think you're onto something with your identification of the deep pipelines in the P4 as the culprit.

cl

Posted July 27th, 2006, 10:35 pm

I am now writing this post in a build of Firefox I did using the i686 optimization flag. It feels twice as fast to startup and about 1.25 - 1.4X faster in terms of connecting to Google and rendering the homepage. My guess is the prescott optimization is not the thing to do when building for the newest generation of Intel processors. I'm going to keep researching the GCC online manual and other sources to see what people have to say about x86 optimizations, but for the mean time it loooks like I'll be building future Camino releases using i686.

Posted July 28th, 2006, 5:29 am

I'll add this build to PimpMyCamino if that's OK. Any particiular feelings on what you'd like to call it? 'Heronstalker's Intel x86 builds' snappy enough?!

Posted July 28th, 2006, 5:41 am

Thanks for the inclusion on the pimpin website. It's like 2 AM in the middle of the Pacific where I am writnig this, so it could sound dumb, but what about 'Heronstalker's Core Camino Customized Builds' instead, seeing as to how we're going to have the Intel 'Core' architecture for a while. You guys are the jury on all this, but thanks for even considering adding my small contribution to your site.

Posted July 28th, 2006, 6:04 am

Great, I'll add that later tonight! Size doesn't matter, but I wouldn't call the first optimised builds for Intel a small contribution. Thanks for making it available!

Posted July 28th, 2006, 6:42 am

heronstalker. Great work so far!
Intel Mac's will not benefit that much from optimized builds as PowerPC Mac's. Simply because the official builds are already compiled using GCC 4.0 and 10.4uSDK. But I am sure there are some small improvements to gain some places...

By the way, here are a couple of useful links about CPU type optimization:
http://gentoo-wiki.com/Safe_Cflags
http://developer.apple.com/documentation/..../i386-and-x86_002d64-Options

Posted July 28th, 2006, 4:14 pm

Thanks for the link to the Gentoo wiki, krmathis. It gave me some fresh ideas for the G4 builds I do for my iBook as well as more insight on what the flags mean for the x86 architecure since the GCC online manual can be somewhat ambiguious. As the wiki says to use the prescott optimization flag for Yonah processors, I guess I will go with that the next time a new Camino release comes out. I'm stilll up in the air on whether to use -O2 or -O3 for optimization level, however. My thinking is that the new Yonah chips have enough L2 cache (2 MB) to use the -O3 flag and get all the code in cache, but I'm by no means a Jon "Hannibal" Stokes, so I could be wrong. Since it only takes about 43 minutes to build on a 1.66 Core Duo Yonah chip, I'll spin out both -O3 and -O2 when the next release and post the one that feels faster. I'm glad all of you are enjoying the build that's out so far.

Posted July 28th, 2006, 7:02 pm

I've tried benchmarking O3 and O2 versions on a G4 7450, and couldn't tell a significant difference between them. YMMV and all that.

cl

Posted July 29th, 2006, 10:56 am

Is there a canonical set of benchmarking tools? I'm curious about the performance difference between the Intel-optimized build and the universal one (and also how the optimized builds of FF and Camino compare).

Posted July 29th, 2006, 11:07 am

Not really; when we looked into this seriously about 18 months ago, we ended up using three or four different Web sites that measured various aspects of page-loading time. It's all in this thread if you're curious:

http://forums.mozillazine.org/viewtopic.php?t=191100

cl

[CM] heronstalker:2006-06-26 [Release, Mac OS 10.4, i386] (O

Who is online