Saturday, June 2, 2012

10.0.5 available (7450 test and general RC)

10.0.5 is available for testing prior to release. Please ensure it operates to your satisfaction. After the release of 13 when I was reviewing code, it dawned on me that we have a low-hanging inefficiency in our methodjit backend that we can simply remove in optimized builds. Because this removes an assertion, even a dead one, I eventually decided not to ship it in 10.0.5, but it will be shipped in 14 and then ultimately 10.0.6. The tweak reduces JavaScript compiler latency and improves performance on V8 by a consistent 1-2%, but it is likely to improve small scripts even more because compiler latency is a larger proportion of their runtime. I have issued a 7450-only test build for this tweak, since I do my 10.x testing on a 7450 system, but if you want to build your own with it and test it yourself the patch is in issue 160. Please test the official build (below) before testing this one. Don't compare performance to 13; 13 is already significantly faster and this would simply add icing to the cake.

A polite note for pedants and download aggregators, since I am starting to get spurious mail and bug reports about this: the availability of this build doesn't mean it's generally available; this is for last-minute RC testing. Our release dates correspond broadly with Mozilla's and version alerts will not be updated until Monday.

10.0.5 also marks the general availability of language pack installers, for stable branch users only. Many thanks to Chris and our localizers! If you have not tested them yet, please choose German, French or Polish from the Downloads tab, and report your results. The installers will work on any 10.x version of TenFourFox from 10.0.3 on up; they are distributed as actual installers, not as standard Firefox .xpis. Do not use them with the unstable versions.

I am considering further improvements to JavaScript compiler speed and a straightforward one is to simply mark all the trivial codegen functions in PPCAssembler.h as inline. The problem is I'm not sure this will make it any faster, since gcc is probably inlining some of them anyway; about the only thing I am sure about is that it will definitely make the binary larger. I'll be doing some testing of this when I begin the port for 14. This change will probably not be backported to stable unless it is a massive win.

The port for 14 will begin with beta 8 (14 betas will start with 5 to harmonize numbering with Fennec). For now, 10.0.5 release notes and official arch builds follow; if there are no issues these will finalize on Monday:

23 comments:

  1. The G3, 7450 and test build all seem to work without a problem.

    ReplyDelete
  2. Thanks for the post! I'm using my 21" G5 as a media center and the old Safari lacks some useful plugins.

    ReplyDelete
  3. Looks like almost one in ten who updated to 10.0.5 so far also downloaded a langpack installer. Very interesting to finally get some numbers. Also, the number of downloads of the previous version 10.0.4 suggests a pretty large user base for TenFourFox (>20.000??).

    ReplyDelete
  4. Somewhere in that ballpark based on server checkins, although we likely have a bit of overcount from people who have multiple machines. I expect the number to increase somewhat now that 3.6 is dead dead dead.

    ReplyDelete
  5. Heya guys, I just got me an old ibook G3 with 600 Mhz (don't know yet if I got lucky and it is the latest one with a 16 MB Radeon Mobility which would make it possible to use Quartz Extreme, but probably not) off of ebay and plan to dual boot Ubuntu 12.04 ppc and OSX 10.4.11.

    The latter will of couse use ur awesome TenFourFox, but now I'm wondering is there an equivalent for Ubuntu ppc? Also what do you guys think what kind of performance can I expect on Ubuntu and OSX.

    I have experience with an old G4 400 Yikes under OSX 10.4.11, which over all is slow but still good useable.

    Thx in advance!

    ReplyDelete
  6. It is just me or this files are corrupted?I download them,unzip them and them when I try to open it says I can't open it because they may be damaged or corrupted.
    Delu, ubuntu 12.04 is too heavy for that ibook.get 10.3.9 ou 10.4.11..maybe linux mint or debian

    ReplyDelete
  7. Shouldn't a light Ubuntu or linux mint work better on a G3 with 600Mhz compared to the heavy OSX 10.4.11. Also wouldn't it be safer? As far as I understand you basically can compile a lot of modern apps on ppc linux, apps you would never get on OSX 10.4.11 (maybe through X11 but even slower there) - !?

    ReplyDelete
    Replies
    1. "Shouldn't a light Ubuntu or linux mint work better on a G3 with 600Mhz compared to the heavy OSX 10.4.11."

      I don't think Ubuntu is any lighter on an old G3 and it's not very easy to install as it's not officially supported. Especially G4 Macs benefit a lot from OS X's altivec optimizations.

      "Also wouldn't it be safer?"

      I don't think old PPC OS X is much of a target for anyone anymore, but theoretically it would be safer.

      "As far as I understand you basically can compile a lot of modern apps on ppc linux, apps you would never get on OSX 10.4.11 (maybe through X11 but even slower there) - !?"

      There's already vast amount of commercial software for OS X 10.4 that Linux doesn't run. Most of the old commercial software is old and not updated anymore, but still very usable if you don't need the very latest versions. With the latest Ubuntu, you would get up-to-date versions of lot of open source programs, but some of the most popular programs (Like Firefox, Thunderbird or OpenOffice) are still being compiled for OS X 10.4.

      PPC Linux has some other limitations that Intel Linux doesn't have. For example with OS X 10.4 you can still run older version of Flash player on sites where it's needed and it usually works quite fine (if a bit slow) although it hasn't been updated in ages. Flash player was never released for PPC Linux and even though there are open source alternatives, they don't work at all as well as the old OS X Flash player.

      I'm not saying Ubuntu is not a good choise for old Intel machines, but in my opinion OS X is gives a better experience on PowerMacs.

      Delete
  8. True I guess OSX might put Altivec to better use, tho I'm not sure if it can do much on with a G3. I'm wondering can Linux ppc gain any from Altivec or is it basically dormant/ unused, are there like any special kernels? How about compiling applications for linux ppc with altivec optimizations?

    Just stumbled across this distro which sounds interesting:
    http://www.archlinuxppc.org/

    ReplyDelete
  9. It's not just kernels, it's also libraries. libpixman comes to mind; I don't think HAVE_VMX is turned on, and even if it were, the pixman VMX support only covers composite operators and not blits or fills. That's something people could consider adding -- I looked at doing it myself but Mozilla only leverages the compositor ops, so there's no benefit for us (it uses CoreGraphics to do the blitting otherwise).

    Otherwise, I think Ville summarized the issues well. I like what people are doing with Linux on Power Macs and I've played with it myself, but I get more work done with 10.4 than Ubuntu.

    ReplyDelete
  10. Thx for your input guys. Oh now that you mention libraries, I remember a few OS9/ classis libraries for/ with Altivec optimizations which often ended up not being installed on most OS9 systems for some reason, even tho most applications probably couldn't make use of it anyhow I wonder if you guys could provide a link, I looked everywhere and could not find any working downloads. It is described you could extract them from your OS9 CD but I had no luck with that either.

    On a related note, ArchPPC looks promising but seems their build hardware recently failed and they are looking for donations or/ and are willing to pay for a (cheap) G4, so in spirit of ppc community support, if you can help out in one way or another please have a look :)

    https://bbs.archlinux.org/viewtopic.php?id=142220
    https://bbs.archlinux.org/viewtopic.php?pid=1108310#p1108310

    https://bbs.archlinux.org/viewforum.php?id=35

    ReplyDelete
  11. Why are TFF 9 onwards so sluggish compared to TFF 7 and 8? This shows a lot on my G4s. Any way to get this same speed back on the next update?

    ReplyDelete
  12. That doesn't really tell me a whole lot, but it's not the JIT, because 9 was still tracejit. You could see if 13 is any better, but if it's not, I'd start looking at addons you can jettison or clean up your profile. Short of that, on all the benchmarks that matter, 13 and even 10 are both faster than any previous version.

    ReplyDelete
  13. I don't have any add-ons apart from flash, but this shouldn't show when I don't go to flash websites. I've tested v. 7 and 8 and up on several computers (a powermac G4 mdd 1.42, a quicksilver g4 933, a powermac g4 533, a powerbook g4 1.67) and on all these computers, the results are the same. v.7 and 8 are a lot faster than 9 and up. I'm not talking about benchmarks, but the time to load pages is noticeably slower on v. 6, 9 and 10. It is such a difference that I can assure you if I'd use a timer and make statistics I could back up what I'm saying. I'm 100% this doesn't show on a G5 though, the less powerful the computer, the more it shows.

    ReplyDelete
  14. The G5 undoubtedly is fine, but I'm typing this message on a 1GHz iMac G4, which is similar to the machines you are reporting issues on. What doesn't get measured doesn't get fixed, but "overall slowness" isn't an actionable work item: I can't pick and choose pieces of Gecko to use (the whole point is to stay as close to Firefox as possible), I can't really test them individually, and I can't reproduce it in any case. I appreciate your frustration, but pervasive issues that are unreproducible by me are essentially unfixable by me, because there are an infinite number of places to look for needles in those kinds of haystacks that I can't even see in the first place. Certainly there are other browsers to leverage while support lasts, and there are Tobias' other projects you could look into too, which is about all I can suggest with the information at hand.

    ReplyDelete
  15. As I said, I could mesure what I'm seeing (using a timer) because it is only related to page loading time. But I'm lost; why is it slower than before? What have you changed? Because I changed nothing...

    ReplyDelete
  16. That's exactly my point -- I don't have any way of proving "this change is why you find the browser slower." But for argument's sake let's say (even though I can't) that I can replicate the browser is slower; let's say I can also make that claim quantitatively, and let's even say (not even the best benchmark can say this) that I can make a stopwatch assessment completely free of local system bias. With a 40MB central library that tells me little or nothing about the problem; it certainly could be one problem, but more likely it's one of several things playing badly that in the aggregate result in degraded performance. I don't get to pick and choose components in Gecko generally; while we turn off things that are easy or necessary to disable, I don't have that kind of option for more subtle issues without seriously hurting the future maintainability of the browser. Stuff that I observe and that I believe I can repair, I do. I can't observe this, but more importantly (and again, I'm typing this from a G4 just like yours), I can't rule out that there is something different about your system than mine. Unless you determined the cause, proved it was part of Gecko and told me what you think is the best way to fix it, it's not getting fixed because you're depending on someone who doesn't see what you're seeing to fix the problem you're alleging.

    I'm not trying to be harsh, but I don't think I can help you here.

    ReplyDelete
  17. It should also be pointed out that many sites (e.g. Slashdot, Ars Technica, Gizmodo, CNN, NYTimes all come to mind) have put up new, heavier versions of themselves since TFF 7 came out -- it may simply be that the sites you like are using HTML or JS capabilities with the newer browsers that they don't try to enable with the older ones...

    ReplyDelete
  18. Just to get some numbers, I tested the load times of some popular websites (Facebook logged in with large friends list) with a fresh profile. Each website was loaded three times in each browser version. First time uncached, second and third time cached (as would be the case normally). After the first complete run, the whole test was repeated. The number for each browser version indicates the sum of all load times.

    PowerBook G4, 1.33 GHz, 2 GB RAM, 10.5.8, no big apps in the background, 12 Mbps line.

    Results seem to indicate that 10 is indeed not exactly the fastest version when it comes to load times. There's almost no difference between 7, 8 and 9, however, and 13 is greatly improved, especially on Facebook.

    TFF 7.0: 289 sec.
    Golem.de 15, 11, 11 | 21, 13, 13
    Amazon.com 9, 8, 8 | 9, 9, 7
    Facebook.com 17, 11, 19 | 22, 18, 12
    Apple.com 5, 2, 3 | 5, 3, 3
    Ebay.de 6, 6, 6 | 6, 5, 6

    TFF 8.0: 291 sec.
    Golem.de 16, 12, 11 | 19, 11, 11
    Amazon.com 11, 8, 10 | 10, 7, 7
    Facebook.com 19, 14, 16 | 16, 14, 14
    Apple.com 7, 3, 4 | 7, 4, 4
    Ebay.de 8, 6, 5 | 6, 4, 7

    TFF 9.0: 289 sec.
    Golem.de 14, 12, 14 | 14, 11, 12
    Amazon.com 11, 8, 8 | 11, 12, 8
    Facebook.com 18, 14, 17 | 16, 13, 15
    Apple.com 5, 7, 2 | 5, 7, 5
    Ebay.de 6, 4, 4 | 6, 5, 5

    TFF 10.0.5: 323 sec.
    Golem.de 16, 12, 13 | 18, 16, 13
    Amazon.com 12, 8, 8 | 12, 9, 8
    Facebook.com 20, 16, 22 | 21, 15, 15
    Apple.com 6, 10, 4 | 5, 7, 3
    Ebay.de 6, 5, 6 | 6, 5, 6

    TFF 13.0: 262 sec.
    Golem.de 14, 13, 12 | 17, 12, 12
    Amazon.com 11, 8, 6 | 11, 6, 7
    Facebook.com 13, 12, 15 | 15, 12, 13
    Apple.com 5, 3, 3 | 5, 3, 3
    Ebay.de 6, 5, 5 | 6, 4, 5

    Aurorafox 14.0a2: 282 sec.
    Golem.de 15, 13, 13 | 17, 16, 14
    Amazon.com 10, 5, 6 | 9, 8, 7
    Facebook.com 13, 14, 14 | 15, 14, 15
    Apple.com 5, 3, 6 | 7, 4, 7
    Ebay.de 6, 4, 4 | 7, 5, 6

    ReplyDelete
  19. This build seems to be more responsive on my Dual 1ghz 7450 quicksilver. I'm not sure if the load times are actually faster or if it's the library loading differences discussed in the previous post. Regardless, it is a lot more pleasant to browse with than the last several releases... it -feels- better.

    ReplyDelete
  20. The best way is to download "vanilla" 10.0.5 and see if you notice a change. It does seem smoother to me as well on my 10.0.5 system machines, but it is hard to quantify that because the scripts that benefit the most generally are the smallest and therefore are below the noise floor on most benchmarks. However, there's apparently no downside to it, so it will be formally part of 10.0.6 and 14 as mentioned previously.

    ReplyDelete

Due to an increased frequency of spam, comments are now subject to moderation.