Live data from Hacker News

Firefox's JIT is getting significantly faster

groups.google.com

21–30 of 286 posts

Re: Firefox's JIT is getting significantly faster

#21
For everyone unable or unwilling to login:-

Hi all,

The SpiderMonkey (JS) team has been working on a significant update to our JITs called WarpBuilder (or just Warp) [0,1]. Before we enable Warp by default in Nightly (hopefully next cycle in 83) we need your help dogfooding it.

Warp improves performance by reducing the amount of internal type information that is tracked, optimizing for a broader spectrum of cases, and by leveraging the same CacheIR optimizations used by last year’s BaselineInterpreter work [2]. As a result, Warp has a much simpler design and improves responsiveness and page load performance significantly (we're seeing 5-15% improvements on many visual metrics tests). Speedometer is about 10% faster with Warp. The JS engine also uses less memory when Warp is enabled.

To enable Warp in Nightly:

1. Update to a recent Nightly 2. Go to about:config and set the "javascript.options.warp" pref to true 3. Restart the browser

We're especially interested in stability issues and real-world performance problems. Warp is currently slower on various synthetic JS benchmarks such as Octane (which we will continue investigating in the coming months) but should perform well on web content.

If you find any issues, please file bugs blocking:

https://bugzilla.mozilla.org/show_bug.cgi?id=1613592

If you notice any improvements, we'd love to hear about those too.

Finally, we want to thank our amazing contributors André Bargull and Tom Schuster for their help implementing and porting many optimizations.

Turning Warp on is only our first step, and we expect to see a lot of new optimization work over the next year as we build on this. We are excited for what the future holds here.

Thanks! The Warp team

[0] WarpBuilder still utilizes the backend of IonMonkey so we don't feel it has earned the WarpMonkey name just yet. [1] https://bugzilla.mozilla.org/show_bug.cgi?id=1613592 [2] https://hacks.mozilla.org/2019/08/the-baseline-interpreter-a...

Re: Firefox's JIT is getting significantly faster

#22

[flagged]

The last thing I want in my browser is a JVM.

Maybe I'm just ignorant, but haven't browsers nearly become just a VM for JavaScript? Rendering documents is sometimes incidental. I'm not sure Swing would actually be worse than HTML5.

Re: Firefox's JIT is getting significantly faster

#23
post #7

Here is the link that doesn't require a Google login: https://archive.is/SFFph

thank you! interestingly a very similar link appears to not require a google login, but requires javascript: https://groups.google.com/d/topic/mozilla.dev.platform/1PHhx... i wonder what's going on.

This did ask me to log in. It doesn't in private mode though.

I think if Google detects that you've _ever_ logged into Google, it's trying to authenticate you when you visit a Google Groups page. Private mode is my default workaround for Google Groups links these days.

Re: Firefox's JIT is getting significantly faster

#25
I believe in Mozilla.and their products. as much as they need to get their admin/ceo pay down.as much as they need to adjust the new Android Firefox UI

Unless there is another foundation with a browser promoting a free and open internet, my vote will always remain with them.

Re: Firefox's JIT is getting significantly faster

#27
post #13
post #6

The current title ("significantly faster") on HN is suboptimal. What is there written relating the speed is that the new implementation: "improves responsiveness and page load performance significantly (we're seeing 5-15% improvements on many visual metrics tests). Speedometer is about 10% faster with Warp." "Warp is currently slower on various synthetic JS benchmarks such as Octane (which we will continue investigat…

5-15% on web content is significantly faster to me. Also, like many people, I mainly use my browser for visiting websites instead of running things like Octane, but obviously it would be nicer if Octane were faster too.

>5-15% on web content is significantly faster to me.

Is it though? Would you even notice it?

If there was an already short processing action, e.g. 1 second, now it would be 850ms -- hardly noticable.

If there was some long processing action, say 5 minutes, now it would take 4 minutes and 25 seconds.

Hardly something to write home about in either case...

Re: Firefox's JIT is getting significantly faster

#28
post #11

It's nice to see some actual good news involving Firefox. With all of the leadership and financial struggles lately, it's made me worry that the browser I've been using for the last 17 years doesn't have long to live. Hopefully that is not the case.

I think it will take more time to say that for certain. Warp is an exception, but most of the recent improvements to Firefox were initially prototyped and developed in the experimental Servo engine which was to my knowledge intended to eventually replace all of gecko. Most of the team for that has been let go.

I think the thing to be worried about is: what will happen when they run out of stuff that's already in the pipeline? Will they be able to execute the sweeping changes they'll need to do to stay within reach of blink long term? Or will they have to concede like opera and edge did.

Re: Firefox's JIT is getting significantly faster

#29

[flagged]

The last thing I want in my browser is a JVM.

The brower already has a VM, and from the same people and the same principles who made the Java JVM JIT and the fast Smalltalk one back in the 90s...

Lars Bak, the v8 head:

"In 1994, he joined LongView Technologies LLC, where he designed and implemented high performance virtual machines for both Smalltalk and Java. After Sun Microsystems acquired LongView in 1997, Bak became engineering manager and technical lead in the HotSpot team at Sun's Java Software Division where he developed a high-performance Java virtual machine.".

Re: Firefox's JIT is getting significantly faster

#30
post #27
post #13

Earlier quoted context omitted.

5-15% on web content is significantly faster to me. Also, like many people, I mainly use my browser for visiting websites instead of running things like Octane, but obviously it would be nicer if Octane were faster too.

> 5-15% on web content is significantly faster to me. Is it though? Would you even notice it? If there was an already short processing action, e.g. 1 second, now it would be 850ms -- hardly noticable. If there was some long processing action, say 5 minutes, now it would take 4 minutes and 25 seconds. Hardly something to write home about in either case...

If I were to squeeze 10% out of a project I had spent significant time and resources optimizing I would be thrilled.
Post reply on HN