mozilla :: #ionmonkey

19 Apr 2017
07:01jandemnbp: do you have an ETA for the bytecode cache?
09:34nbpjandem: I had some, but they are all in the past. So now I am taking more margins, and I hope to get it landed by the end of May (I will check the review feedback in a few minutes)
09:35nbpjandem: then I will have to tune it, and that I hope to get it tuned for 57.
09:35nbp^ in the worst case scenario
09:50jandemnbp: ok :)
09:58nbpjandem: apparently the end of May is sooner than I had expected (/me reading Bug 900784 comments)
10:29evilpiejandem: jonco: so where is the store?
10:30joncoevilpie: it's the next storePtr under the comment 'Mark iterator as active'
10:31evilpied'uh, I thought it would be before the post write barrier ;)
10:31evilpieThanks for fixing this
10:31jonconp, hopefully this improves things :)
10:34evilpieThis code is actually broken for unboxed objects, going to attach a patch for this
10:42bbouvieri'm dubious with the assert range code: it xors an input register, then tests to confirm it's zero -_-
10:43evilpie\o/
10:48jandemjonco: ping
10:51joncojandem: hey
10:58jandemjonco: hey i was wondering about nursery-allocated strings. Now with CGC it doesn't seem too difficult once we add/fix all the post barriers... Can you think of any showstoppers? There's the octane-splay pretenuring issue but i hope we can add an heuristic for that..
10:59joncojandem: I expect there are a bunch of places where we assume strings don't need barriers
10:59joncojandem: but other than that it should be ok
10:59joncojandem: I'm sure there will be something that comes up though :)
10:59jandemjonco: yeah :)
11:00joncojandem: didn't you try this before?
11:03jandemjonco: yeah i did a hackish prototype and the biggest issue was the barriers on octane-splay. But things improved a lot since then and websites allocate so many strings that i think it's worth doing soon
11:08jandemjonco: is it possible to see which GC things we allocate in the browser? If strings show up a lot on facebook/gmail I should probably spend a few weeks on this
11:09jandem(i can also hack in some logging in Allocate<> of course)
11:21jandemugh i found a pretty bad performance bug in a gmail profile
11:21* jandem patches
11:23jandemevilpie: it&#39;s possible gmail has GetName&#39;s we&#39;re not optimizing - i noticed some time under DoGetNameFallback
11:24evilpieinteresting
11:24jandemadding spew for GetName / BindName is on my list for this week
11:24evilpieyeah that would be useful
11:24jandemmaybe it&#39;s bogus but it was more time than for GetProp so i think there&#39;s something going on
11:47joncojandem: you can get some idea from about:memory I guess
12:02jandemjonco: yeah
15:30tcampbellevilpie: ping
15:33evilpietcampbell: pong
15:34tcampbellevilpie: Was wondering if you wanted to steal the JSOP_IN IC bug
15:34tcampbellevilpie: I have my patches with the mechanical work done, but you seem to be exploring the proxy, etc cases
15:35tcampbellevilpie: Otherwise, I&#39;ll go back to it in a few days and follow up to what you did
15:35tcampbell(trying to upload the patch to the bug regardless)
15:35evilpieNo rush, I am looking into some other stuff right now
15:36tcampbellevilpie: Sounds good. Thanks for figuring out the type issues
15:36evilpieI think we still have two months?
15:36tcampbellsomething like that
15:37tcampbellwill finally fix the spreadcall stuff now!
15:38evilpie:)
15:50tcampbellevilpie: I should have really read those comments in detail before I did my debugging yesterday. I hit all the same problems with temp registers and BoxPolicy
20 Apr 2017
No messages
   
Last message: 37 days and 12 hours ago