00:08:13
| * joshthecoder | quit (Quit: Leaving...) |
00:15:47
| * bradleymeck | quit (Quit: bradleymeck) |
00:32:30
| * pyrotechnick | joined |
01:03:45
| * pyrotechnick | part |
01:57:40
| * TheJH | quit (Ping timeout: 276 seconds) |
03:17:43
| * joshthecoder | joined |
04:21:02
| * [[zzz]] | joined |
04:24:20
| * [[zz]] | quit (Ping timeout: 248 seconds) |
05:27:11
| * [[zzz]] | changed nick to [[zz]] |
06:33:39
| * joshthecoder | quit (Quit: Leaving...) |
07:15:21
| * mraleph | joined |
08:17:13
| * rendar | joined |
08:19:11
| * RT|Chatzilla | quit (Quit: ChatZilla 0.9.86.1 [Firefox 2.0.0.22pre/2009081014]) |
08:24:56
| * RT|Chatzilla | joined |
10:21:50
| * C-Man | joined |
10:43:48
| * jamespage_ | joined |
10:46:20
| * Net147 | joined |
10:57:20
| * jamespage_ | quit (Ping timeout: 245 seconds) |
11:06:08
| <sanjoyd> | mraleph: out of curiosity, did you guys ever consider making crankshaft a tracing JIT? |
11:11:52
| * TheJH | joined |
11:17:49
| * Net147 | quit (Quit: HydraIRC -> http://www.hydrairc.com <- *I* use it, so it must be good!) |
11:18:07
| * Net147 | joined |
11:35:33
| <mraleph> | sanjoyd: no. |
11:35:58
| <mraleph> | sanjoyd: tracing works fine only for simple loopy code. |
11:36:37
| <mraleph> | sanjoyd: as soon as you get a real OOP program that is polymorphic and branchy all over the place tracing does not work as well as it is advertized. |
11:37:00
| <mraleph> | sanjoyd: good example will be DeltaBlue that is like 5 times faster on V8 then on LJ2. |
11:37:53
| <mraleph> | sanjoyd: why do you ask? |
11:38:40
| <sanjoyd> | mraleph: just curious, have been looking into LJ2 lately. |
11:39:46
| <mraleph> | it's true that tracing has its benefits over method jit on a loopy code |
11:41:09
| <sanjoyd> | Yeah, it seemed to me that you won't be able to do things like representation changes and other OOP things nicely in a tracing JIT. |
11:41:54
| <sanjoyd> | Like a value might be 31 bit integral in a specific trace, but all the trace exits will have to box it back, which may not have been necessary. |
11:45:49
| <mraleph> | sanjoyd: I guess if you have a hot side exit you will trace from it and then you can eliminate a boxing on the border between traces when you connect them. Though it is not clear how easy it is and whether LJ2 does that. |
11:45:58
| <mraleph> | should be pretty easy. |
11:48:49
| <sanjoyd> | mraleph: yeah, LJ2 does something like that in eliminating unneeded lua-stack operations. But we still have to wait for the side trace to get hot even when we have the info required to propagate the integer representation further. |
11:48:59
| * jamespage_ | joined |
11:49:45
| <sanjoyd> | I guess the central point is that a tracing JIT only sees a single trace, while a method JIT has the whole CFG to infer from. |
12:08:12
| <mraleph> | sanjoyd: well, yes. but it does not waste time propagating information from places that do not matter and neither does information from the cold places obscure the inference. |
12:08:46
| <mraleph> | sanjoyd: in method jit you would have to assign weights to loops somehow to drive inference right way in complex cases. |
12:08:56
| <mraleph> | in tracing JIT it happens naturally. |
12:11:35
| * Net147 | quit (Quit: HydraIRC -> http://www.hydrairc.com <- Would you like to know more?) |
12:15:26
| * jamespage_ | quit (Quit: Ex-Chat) |
14:41:27
| * bradleymeck | joined |
15:58:08
| * bradleymeck_ | joined |
16:00:38
| * bradleymeck | quit (Ping timeout: 246 seconds) |
16:00:39
| * bradleymeck_ | changed nick to bradleymeck |
16:06:46
| * RT|Chatzilla | quit (Quit: ChatZilla 0.9.86.1 [Firefox 2.0.0.22pre/2009081014]) |
16:15:32
| * decoder | quit (Ping timeout: 246 seconds) |
16:16:58
| * decoder | joined |
16:30:20
| * AndreasMadsen | joined |
16:53:09
| * AndreasMadsen | quit (Remote host closed the connection) |
17:08:12
| * AndreasMadsen | joined |
17:31:58
| * AndreasMadsen | quit (Remote host closed the connection) |
19:02:23
| * bradleymeck | quit (Quit: bradleymeck) |
19:03:16
| * AndreasMadsen | joined |
19:04:01
| * joshthecoder | joined |
19:33:35
| * AndreasMadsen | quit (Remote host closed the connection) |
19:45:14
| * AndreasMadsen | joined |
19:53:07
| <Ke> | is there somewhere a guide on how to sandbox v8 that is not chromium source code |
20:38:00
| * bnoordhuis | joined |
20:54:32
| * AndreasMadsen | quit (Remote host closed the connection) |
20:57:52
| * joshthecoder | quit (Quit: Leaving...) |
22:26:02
| * rendar | quit |
22:26:38
| * RT|Chatzilla | joined |
22:48:01
| * joshthecoder | joined |
22:52:23
| * joshthecoder | quit (Client Quit) |
22:58:30
| * TheJH | quit (Ping timeout: 260 seconds) |
23:25:58
| * bnoordhuis | quit (Ping timeout: 240 seconds) |