Index

Show enters and exits. Hide enters and exits.

00:09:49boyscoutAdd ability to load options from a file - 878ca1b - Evan Phoenix
00:09:50boyscoutConfig files can load other config files - 97de677 - Evan Phoenix
00:12:08evanok, enough config file stuff :D
00:23:55rueTee hee, I have a new favourite term: Java is an enormity.
00:24:32evanrad
00:24:39evancscope -R vm
00:24:41evansegfaults.
00:25:25rueTrying to prune stuff?
00:25:39evana little
00:25:49evantrying to clean up CallFrame a little
00:26:20evanso the JIT doesn't have to so much work to build one that the VM understands
00:31:45rueThat should help a bit. Have you looked into how well LLVM groks ByteArray and Tuple?
00:32:59evanit groks them fine
00:33:08evanit's easy to explain to them how to fetch raw parts
00:33:13evanbe it pointers or bytes
00:34:02rueBeing able to "inline" that access should be helpful
00:34:12evanalready doing that for Tuple#at
00:34:13evanyeah
00:34:28evani'll hand inline the other 3 too
00:34:30evanthey're easy
01:14:39slavahi evan
01:14:50evanallo slava
01:16:19headiusafternoon
01:16:24evanheadius: hey there
01:16:27evanwe're off by one week
01:16:30evani'm in SF next week
01:16:47headiusahh I was wondering about that
01:16:47headiusoh well
01:24:33evanmy parents are in town this week
01:24:35evanthats why i'm not up there
02:03:03boyscoutRefactor CallFrame::name - 84d2003 - Evan Phoenix
02:03:03boyscoutMove CallFrame::name to CallFrame::msg - 2f7908a - Evan Phoenix
02:27:59headiusevan: ahhh I see
02:28:07evanglances about
02:28:07headiuswell, you can hang out with laurent next week :D
02:28:15evanah!
02:28:16evanhehe
02:28:25evanit was partly because of WWDC
02:28:30evanMarcus is in town again
02:28:41headiusahhh
02:30:13evanI forgot JavaOne and friends was this week
02:30:27evanfor I am a bad nerd.
02:31:29headiusyeah, crazytown
02:32:50evani'm just happy I can stay in the EY apt
02:32:57evani'm sure there are no hotels
02:44:58headiusevan: yeah, probably would be tougher to get a hotel at least
02:45:36evanheadius: yeah
02:45:48evanso, because our paths are intertwinned
02:45:54evani'm trying to slim down my call frame
02:45:55evan:D
02:47:48headiusyou probably have somewhat less cost than we do because of alloca, but it can't hurt
02:48:02evanyeah
02:48:10evanit's not so much the allocation i'm working on
02:48:15evanbut the ammount of data in it.
02:48:46headiusyeah, in 1.4 I'm hoping that --fast will actually only initialize what's needed
02:48:53headiusrather than an all-or-nothing thing
02:49:04evanyeah, for instance
02:49:16evani'm trying to get my StaticScope object in the call frame lazily used
02:49:22evanbecause it's only there so that instance_eval works right.
02:49:31headiusah, yes
02:52:44headiusevan: yeah, we keep trying to reduce frame size, and probably will also fix stack size at 1000 or something soon
02:53:08evanreally?
02:53:09headiusI keep feeling like there's some magic way we could pass this data across calls that doesn't involve heap frames
02:53:11evanpython style?
02:53:38headiuswell, it would be fixed but still possibly to go over it with a perf penalty
02:53:39evangoing to just preallocate all the frames?
02:53:43headiusyes
02:53:45headiuswe already do that
02:55:00evaninteresting
04:36:04slavaevan: how's the JIT going?
21:16:28evanslava: poke
22:55:18lypanovwibbles
22:55:41evanyo yo
22:55:51lypanovyoyo evan. how goes?
22:55:57evanpretty good
22:55:58evanyou?
22:56:38lypanovcompletely awesome
22:56:45evangood to hear!
22:56:52lypanovhow goes to llvm?
22:56:56lypanovteh*
22:57:05evangood good
22:57:13evanworking great now that i've got the architecture setup properly
22:57:20evanand LLVM complication is shunted to a background thread
22:57:27evanit's never decreasing performance
22:57:33evanonly increas.
22:58:06lypanovawesome!
22:58:38lypanovgetting any good speed ups?
22:59:11evanyeah
22:59:15evansmall wins right now
22:59:22evanabout 15% improvement on the spec run
22:59:27evanArray#each is 100% faster
22:59:41evanbecause i've taught the JIT a bit about how to make it fast
22:59:49lypanovmaybe time for me to fuck around with it for a bit
22:59:52evanie, it knows how to perform Tuple#at directly
23:00:03evanno method or function call
23:00:11lypanovis getting married in 2 weeks, will have time free to code
23:00:12evana guard and then a memory move
23:00:19evanhey! congrats!
23:00:26lypanov:D thx
23:00:54evanbig wedding? small?
23:00:58lypanovnew iphone, time to hack, and a wife. couldn't get better
23:01:19lypanov(i'm assuming the iphone will be released anyway :P )
23:03:14lypanovevan: not sure really. 70. no idea if thats big or small
23:03:23evansmall.
23:03:24lypanovguess small :)
23:03:36lypanovwould prefer 20 :P
23:03:49evanhehe
23:03:49boyscoutLazily use CallFrame::static_scope - 09beb70 - Evan Phoenix
23:03:49boyscoutMake usage of CallFrame::top_scope conditional - 931fb04 - Evan Phoenix
23:03:49boyscoutRemove CallFrame::stack_size - 584c8cf - Evan Phoenix
23:03:49boyscoutRemove CallFrame::args - 5f346f2 - Evan Phoenix
23:03:56evanwtf.
23:04:03evani thought that push failed.
23:04:06evanbizarre.
23:27:00slavahie van
23:27:02slavaevan*
23:29:06evanallo
23:29:16evani was curious how you profile things
23:29:28slavaI have a counting profiler
23:31:44lypanovnot the special profiling instructions?
23:31:53lypanovnot worth the effort?
23:46:28rueevan: Apropos "compilation" misspelling :P
23:49:53evaneh?
23:50:37rue"LLVM complication"
23:51:05evanhah
23:51:49jackdempseyhahaha
23:51:55jackdempseygreat typo
23:54:42rueKeep thinking there should be a Nil::to<>() or something