Javascript Performance Thread
-
- Posts: 269
- Joined: February 7th, 2013, 3:19 pm
Re: Javascript Performance Thread
Time for another bump
Bug 1162199 enabled unboxed objects by default.
Meanwhile, V8 got a huge speed-up in octane-mandreel-latency, my guess is that this change did it: https://chromium.googlesource.com/v8/v8 ... 9ff52c1a3d
I wonder if something similar is applicable to SpiderMonkey.
[edit] Oh, it got backed out But at least it's close [/edit]
Bug 1162199 enabled unboxed objects by default.
Meanwhile, V8 got a huge speed-up in octane-mandreel-latency, my guess is that this change did it: https://chromium.googlesource.com/v8/v8 ... 9ff52c1a3d
I wonder if something similar is applicable to SpiderMonkey.
[edit] Oh, it got backed out But at least it's close [/edit]
-
- Posts: 7418
- Joined: July 28th, 2009, 4:52 pm
Re: Javascript Performance Thread
About mandreel-latency, it's a problem with how it works - https://bugzilla.mozilla.org/show_bug.cgi?id=1162272 / https://github.com/chromium/octane-benchmark/issues/29
-
- Posts: 1030
- Joined: June 28th, 2008, 4:57 am
Re: Javascript Performance Thread
This happened a while back but I don't think it was ever mentioned here: The Backtracking register allocator is now the default, and in fact the Linear Scan register allocator was removed, opening up a lot of possibilities for cleanup and further development (since the backtracking allocator is now the only supported allocator).
Beyond that, hmm, I know Terrence Cole has been cleaning up the garbage collection logic a lot (Chrome's garbage collector is in some ways less advanced than SM's, but it's also a lot simpler, more straightforward and faster. Terrence's work is in some ways long overdue, having been delayed by big projects like Generational GC). The rest of the stuff that comes to mind for me recently has mostly been about standards compliance, and cleanup like Boris Zbarsky's work to remove object parents from the internal representation.
One big project I know about, which is still in its early stages as far as I know, is djvj's work to incrementalize parsing and make it asynchronous. That could really help with pause times on first load, but it's a big undertaking.
Beyond that, hmm, I know Terrence Cole has been cleaning up the garbage collection logic a lot (Chrome's garbage collector is in some ways less advanced than SM's, but it's also a lot simpler, more straightforward and faster. Terrence's work is in some ways long overdue, having been delayed by big projects like Generational GC). The rest of the stuff that comes to mind for me recently has mostly been about standards compliance, and cleanup like Boris Zbarsky's work to remove object parents from the internal representation.
One big project I know about, which is still in its early stages as far as I know, is djvj's work to incrementalize parsing and make it asynchronous. That could really help with pause times on first load, but it's a big undertaking.
-
- Posts: 874
- Joined: July 11th, 2011, 2:57 am
Re: Javascript Performance Thread
Thank for the info, Greeneyes.
-
- Posts: 471
- Joined: February 25th, 2013, 9:52 pm
Re: Javascript Performance Thread
https://bugzilla.mozilla.org/show_bug.cgi?id=1161516 Re-use baseline caches in ionmonkey
https://bugzilla.mozilla.org/show_bug.cgi?id=1157624 Odin: Split parsing/validation from codegen
https://bugzilla.mozilla.org/show_bug.cgi?id=1154987 Move JS parsing off of main thread and enable incremental parsing
https://bugzilla.mozilla.org/show_bug.cgi?id=1157624 Odin: Split parsing/validation from codegen
https://bugzilla.mozilla.org/show_bug.cgi?id=1154987 Move JS parsing off of main thread and enable incremental parsing
-
- Posts: 269
- Joined: February 7th, 2013, 3:19 pm
Re: Javascript Performance Thread
AWFY got an update too, with reported regressions linked to the graph. Here's what it looks like: http://arewefastyet.com/regressions/#/r ... on/1592575
-
- Posts: 269
- Joined: February 7th, 2013, 3:19 pm
Re: Javascript Performance Thread
Woah, Microsoft got me impressed: http://blogs.windows.com/msedgedev/2015 ... soft-edge/
-
- Posts: 5472
- Joined: May 13th, 2012, 10:43 am
Re: Javascript Performance Thread
Timvde wrote:Woah, Microsoft got me impressed: http://blogs.windows.com/msedgedev/2015 ... soft-edge/
I'd be more impressed if they backported it to win7...
- Grantius
- Posts: 1545
- Joined: June 28th, 2011, 4:14 pm
- Contact:
Re: Javascript Performance Thread
Timvde wrote:Woah, Microsoft got me impressed: http://blogs.windows.com/msedgedev/2015 ... soft-edge/
I'm glad they blog about these kind of things, very interesting
Micro gaming box: AMD A10-7800 APU, 8gb RAM M350 ITX case (size of a book), Windows 10/Ubuntu
Tablet/Laptop: Asus Transformer T100, Intel Atom 2GB RAM, Windows 10 x86
Mobile:Xiaomi Redmi Note 3 Pro
Tablet/Laptop: Asus Transformer T100, Intel Atom 2GB RAM, Windows 10 x86
Mobile:Xiaomi Redmi Note 3 Pro
-
- Posts: 471
- Joined: February 25th, 2013, 9:52 pm
Re: Javascript Performance Thread
Timvde wrote:Woah, Microsoft got me impressed: http://blogs.windows.com/msedgedev/2015 ... soft-edge/
Nice!
But i would hate if Firefox is slower than IE in Octane . Where would our bragging rights go?
- joeg
- Posts: 2616
- Joined: October 10th, 2003, 12:37 pm
- Location: How can you be in two places at once, when you're not anywhere at all?
Re: Javascript Performance Thread
Using the betas (I just moved up to v39), I have script problems with the German site "der Spiegel".
http://www.spiegel.de/
Fx frequently "stops responding" or I get the message "a javascript is busy". It doesn't happen with any of the few sites I regularly visit (BBC, CNN, EBay)
What's wrong with the Spiegel site, please - just out of interest?
Thanks.
http://www.spiegel.de/
Fx frequently "stops responding" or I get the message "a javascript is busy". It doesn't happen with any of the few sites I regularly visit (BBC, CNN, EBay)
What's wrong with the Spiegel site, please - just out of interest?
Thanks.
Although every day is Judgment Day, I nonetheless feel like a room without a roof.
- KilliK
- Posts: 612
- Joined: June 18th, 2004, 7:11 am
Re: Javascript Performance Thread
Timvde wrote:Woah, Microsoft got me impressed: http://blogs.windows.com/msedgedev/2015 ... soft-edge/
wow, now Firefox is even slower than the new IE/Edge? :/
- Omega X
- Posts: 8225
- Joined: October 18th, 2007, 2:38 pm
- Location: A Parallel Dimension...
Re: Javascript Performance Thread
For now anyway. Google is barely managing to stay an inch ahead in benchmarks.
-
- Posts: 1030
- Joined: June 28th, 2008, 4:57 am
Re: Javascript Performance Thread
Whatever the MS engineers have been up to, the results are impressive. And that's great! More competition. But let's be reasonable - the results on these benchmarks are very good in all browsers at this point, because they've gotten so much attention. I think work on avoiding bad performance cliffs and improving general responsiveness is more important at this point - and we don't know how Chakra does in that area (though I think they've traditionally been good at responsiveness).
One point I thought was interesting was about the cross-script inlining. Gecko has very good security for these because of compartment-per-global and the centralized infrastructure for cross-compartment security wrappers. However I don't think that stuff has a lot of JIT support. In theory once you have a cross-compartment wrapper, accessing the cross-compartment object should be fast, but there are probably opportunities for gains there. Of course, JITing such security sensitive stuff is also scary. I'd be curious to know what Chakra's security story is for these. IIRC Chrome's is a mess, though they're working on changing that (and it doesn't mean they have exploitable holes in practice).
One point I thought was interesting was about the cross-script inlining. Gecko has very good security for these because of compartment-per-global and the centralized infrastructure for cross-compartment security wrappers. However I don't think that stuff has a lot of JIT support. In theory once you have a cross-compartment wrapper, accessing the cross-compartment object should be fast, but there are probably opportunities for gains there. Of course, JITing such security sensitive stuff is also scary. I'd be curious to know what Chakra's security story is for these. IIRC Chrome's is a mess, though they're working on changing that (and it doesn't mean they have exploitable holes in practice).
-
- Posts: 471
- Joined: February 25th, 2013, 9:52 pm
Re: Javascript Performance Thread
maybe the FF team should send the Chakra team a cake, to celebrate their perf improvements.