Porting Firefox to Apple Silicon

We talk often on this forum of how innaccessible giants like Google and Amazon are for the little guy. I thus found this point particularly interesting:<p>&gt;&quot;Attempts to contact the vendor through regular support channels were unsuccessful so we ended up searching LinkedIn and managed to find an engineer working on the core antivirus detection. They immediately understood the seriousness of the problem and took prompt action to get a fix shipped, thus preventing quite the disaster for the users of this product. It’s notable that without this last-ditch effort we would have been effectively blocked from releasing a native Apple Silicon version for an indefinite period.&quot;

&gt; The Apple Silicon chips are one of the first desktop chips that are a heterogeneous design with distinct performance and efficiency cores. We’re revising much of our core threading and thread pooling architecture to handle the distinction better, improve efficiency, and eventually be able to schedule less performance-critical tasks on the efficiency cores.<p>I found this bit interesting. Likely more prevalent in mobile apps, but perhaps shifting desktop code to Big.Little approach and using core affinity will result in a lot less wasted energy.

Got a kick out of one of the bugzilla links:<p><a href="https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=34572" rel="nofollow">https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=34572</a><p>&quot;Use native context menus on Mac OS&quot;<p>&quot;Opened 21 years ago&quot;

A bit tangential to the main topic of the post. They mention that they are working on another optimizing compiler Ion which will replace the cranelift compiler (which is still in nightly) as the new compiler for WebAssembly.<p>They link the issue [1] tracking the change which also speaks about disabling cranelift.<p>To my knowledge cranelift was made for the purpose of compiling WebAssembly in Firefox, so I am not sure if I am missing something here (it&#x27;s not yet production ready maybe). The Cranelift README[2] mentions that it will be a backend for IonMonkey.<p>I am a complete layman here so I am curious if someone here has a better understanding.<p>[1] <a href="https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1687626" rel="nofollow">https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1687626</a><p>[2] <a href="https:&#x2F;&#x2F;github.com&#x2F;bytecodealliance&#x2F;wasmtime&#x2F;tree&#x2F;main&#x2F;cranelift" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;bytecodealliance&#x2F;wasmtime&#x2F;tree&#x2F;main&#x2F;crane...</a>

I&#x27;m surprised that Rosetta 2 isn&#x27;t installed by default. It seems that for the next couple of years the vast majority of people will need at least one x86 app.<p>I guess split-architecture applications were also not foreseen as it is clear that the auto-install prompt doesn&#x27;t work very well in that case.

&gt; If the user visits such a site, Firefox will automatically download and install such a proprietary EME&#x2F;CDM module. This presented a problem to us as we would be dependent on those third-party vendors to publish ARM64 versions of those decoders.<p>Wait, modern browsers still download and run native binaries at the request of certain sites? How is this different from the days when native plugins like Flash were massive security liabilities? I thought we didn&#x27;t do that anymore?

&gt; <i>Rust in particular was a concern. Firefox depends on Rust code, and we require a working Rust compiler to build the browser. Although Apple Silicon support for Rust was underway, it took until mid-August for there to be functional compiler builds, which limited the amount of progress possible for Firefox.</i><p>Lack of rust support for 64-bit ARM was a bit surprising to me, especially given the velocity in which people have been rewriting certain components in Rust.<p>Take for example ffmpeg failing to compile because librsvg was rewritten in rust: <a href="https:&#x2F;&#x2F;trac.macports.org&#x2F;ticket&#x2F;61668" rel="nofollow">https:&#x2F;&#x2F;trac.macports.org&#x2F;ticket&#x2F;61668</a>

Love you firefox. Keep fighting the good fight. Keep that budget figured out.

As they linked to the LibreOffice bug that we hit,[1] it might be worthwhile explaining how the cross platform architecture works in LibreOffice.<p>The widgeting&#x2F;graphics library is actually run by something called VCL (the Visual Component Library). It&#x27;s a bit of a mess to be honest, but the simplified version is that there is a class called OutputDevice that the rest of the app uses, which basically acts as a fascade over a platform specific class called SalGraphics (there are a number of other platform specific classes, SalGraphics is what I focus on here).<p>Basically it is a class that implements a bunch of primitive drawing functions which call on abstract functions. We then implement these functions in a platform specific class.<p>To see the guts of the Mac class, see AquaSalGraphics [2] - and no, none of know why it was named &quot;Aqua&quot;... our codebase is <i>old</i>.<p>FWIW, OutputDevice has serious issues. I have detailed them in a mailing list post. [3]<p>1. <a href="https:&#x2F;&#x2F;bugs.documentfoundation.org&#x2F;show_bug.cgi?id=138122" rel="nofollow">https:&#x2F;&#x2F;bugs.documentfoundation.org&#x2F;show_bug.cgi?id=138122</a><p>2. <a href="https:&#x2F;&#x2F;opengrok.libreoffice.org&#x2F;xref&#x2F;core&#x2F;vcl&#x2F;inc&#x2F;quartz&#x2F;salgdi.h?r=e7ace1d0#133" rel="nofollow">https:&#x2F;&#x2F;opengrok.libreoffice.org&#x2F;xref&#x2F;core&#x2F;vcl&#x2F;inc&#x2F;quartz&#x2F;sa...</a><p>3. <a href="https:&#x2F;&#x2F;lists.freedesktop.org&#x2F;archives&#x2F;libreoffice&#x2F;2020-December&#x2F;086546.html" rel="nofollow">https:&#x2F;&#x2F;lists.freedesktop.org&#x2F;archives&#x2F;libreoffice&#x2F;2020-Dece...</a>

For who curious, the antivirus that caused problem looks like Norton 360.<p><a href="https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1682834#c39" rel="nofollow">https:&#x2F;&#x2F;bugzilla.mozilla.org&#x2F;show_bug.cgi?id=1682834#c39</a>

&gt; It’s notable that without this last-ditch effort we would have been effectively blocked from releasing a native Apple Silicon version for an indefinite period.<p>Effectively blocked from releasing it for the single-digit-percentage of people who run an antivirus on a Mac.<p>Does anyone have credible numbers on this?

&gt; The Apple Silicon chips are one of the first desktop chips that are a heterogeneous design with distinct performance and efficiency cores. We’re revising much of our core threading and thread pooling architecture to handle the distinction better, improve efficiency, and eventually be able to schedule less performance-critical tasks on the efficiency cores.<p>Isn&#x27;t this at the wrong abstraction level? I would expect this to be a job for the OS scheduler.

&gt; Apple introduced a translation cache that likely removes this overhead completely for most applications but it does not work for code that is output by a JIT. With the native build, this second translation is avoided completely and we’re back to having a snappy browser.<p>Indeed while Rosetta does have support for JITs (which is really impressive in and of itself), every piece of machine code generated by the JIT has to be translated on the fly.<p>While the hiccup at the initial run is not too costly &#x2F; annoying for a regular application being AOT-compiled in its entirety and Apple can then shove the result somewhere nearby, for a JIT it&#x27;s basically constant, continuous overhead which can&#x27;t be cached because it won&#x27;t be around next run. I&#x27;m not surprised that the gains are significant there.

Wait, what? People run anti-virus on Macs? What proportion of the userbase is this?<p>It&#x27;s good to hear from Mozilla doing some browser developmenmt, and not making bizarre political announcements that an authoritarian shutdown of a social network by a cartel of tech giants is &quot;not enough&quot;.

Oh hey, the author is also the guy who developed the amazing open source AlphaGo reimplementation Leela Zero: <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Leela_Zero" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Leela_Zero</a>

does firefox use llvm to compile ?

What an interesting and informative article! Nicely done!

Is “apple silicon” the port job here? Isn’t this more correctly described as a port of their existing ARM64 target to macOS?

What porting? Isn&#x27;t C++ the most portable language in existence? Oh, there is also Rust. But isn&#x27;t it just uses an LLVM codegen, same as Swift or clang? So, there is Rust&#x27;s stdlib.<p>Seems like these abstractions are not exactly zero-cost?

I wonder if Mozilla regrets laying off 25% of their engineering team [0] given their rapidly declining browser market share [1], and their rapidly declining performance. Not to mention their forced adoption of the new Firefox on Android which disables all add-ons except Mozilla approved ones, despite their promises to correct this [2].<p>Mozilla _used_ to be about open internet and security, but that&#x27;s just a false pretense at this point [3][4].<p>I believe it&#x27;s time to embrace Chromium &#x2F; Blink, throw away the idea of internet freedom and just use the best performing browser of the week.<p>[0] <a href="https:&#x2F;&#x2F;thenextweb.com&#x2F;insights&#x2F;2020&#x2F;08&#x2F;11&#x2F;mozilla-firefox-layoffs-250-employees&#x2F;" rel="nofollow">https:&#x2F;&#x2F;thenextweb.com&#x2F;insights&#x2F;2020&#x2F;08&#x2F;11&#x2F;mozilla-firefox-l...</a><p>[1] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Usage_share_of_web_browsers" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Usage_share_of_web_browsers</a><p>[2] <a href="https:&#x2F;&#x2F;www.androidpolice.com&#x2F;2020&#x2F;09&#x2F;03&#x2F;firefox-update-faces-backlash-due-to-missing-features-and-few-supported-extensions&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.androidpolice.com&#x2F;2020&#x2F;09&#x2F;03&#x2F;firefox-update-face...</a><p>[3] <a href="https:&#x2F;&#x2F;blog.mozilla.org&#x2F;blog&#x2F;2021&#x2F;01&#x2F;08&#x2F;we-need-more-than-deplatforming&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.mozilla.org&#x2F;blog&#x2F;2021&#x2F;01&#x2F;08&#x2F;we-need-more-than-d...</a><p>[4] <a href="https:&#x2F;&#x2F;www.theverge.com&#x2F;2017&#x2F;12&#x2F;16&#x2F;16784628&#x2F;mozilla-mr-robot-arg-plugin-firefox-looking-glass" rel="nofollow">https:&#x2F;&#x2F;www.theverge.com&#x2F;2017&#x2F;12&#x2F;16&#x2F;16784628&#x2F;mozilla-mr-robo...</a>

FYI you still cannot get a full native experience with Chrome and Firefox, due to plugins that have not been ported.<p>For example Chrome ships with an x64 version of Widevine, a plugin that is required to watch live streams on YouTube TV (and perhaps other services with live TV). Currently, YouTube TV does not work natively in Chrome or Firefox.<p>All that said, it will work fine if you run Rosetta -- the x64 decoder will run in Rosetta.