--- Log opened Sun Jul 26 00:00:22 2009
08:07 -!- Netsplit clarke.freenode.net <-> irc.freenode.net quits: vegard
08:33 -!- Netsplit over, joins: vegard
09:20 <@ahuillet> hi
10:21 < penberg_home> salut
11:07  * ahuillet is nervous about Bull
11:08 <@ahuillet> I'm nervous about anything unusual in my life anyway. :)
11:15 < penberg_home> :)
11:16 <@ahuillet> it would be good if work was taking place over IRC, that medium I know very well
11:16 <@ahuillet> person to person contact is more difficult
11:17 < penberg_home> yeah?
11:18 < penberg_home> I don't think it's difficult, but it is different for sure
11:18 < penberg_home> I usually end up being more of an asshole on-line than in person ;)
11:18 < penberg_home> simply because when working online, I tend to optimize the amount of communication needed
11:19 < penberg_home> which can easily be interpreted as being a total ass
11:19 <@ahuillet> I kind of have the opposite problem, I think
11:19 <@ahuillet> in person I don't know how to properly phrase things so I go straight to the point
11:19 <@ahuillet> it improves once I know the person I'm talking to, but establishing contact is not too easy for me
11:19 <@ahuillet> then again I doubt it is for any person
11:20 < penberg_home> probably, yeah
11:20 < penberg_home> so are you going to be working in a team or...?
11:21 <@ahuillet> I don't really know yet
11:21 <@ahuillet> I'm probably going to work on my own and have one person help me get started
11:22 < penberg_home> so it's the unknown that makes you nervous? :)
11:22 <@ahuillet> not sure if I get an office or what yet either
11:22 <@ahuillet> yeah pretty much :)
11:22 < penberg_home> how long will you be working there?
11:22 <@ahuillet> eight weeks
11:22 < penberg_home> ok, that's nice.
11:22 < penberg_home> any overlap with school?
11:22 <@ahuillet> none at all, it's required by school
11:23 <@ahuillet> I end the internship on a friday and start school again the next monday
11:23 < penberg_home> interesting
11:23 < penberg_home> I don't think we have that here.
11:23 < penberg_home> at university anyway
11:23 <@ahuillet> internships required by school?
11:23 < penberg_home> yeah
11:23 <@ahuillet> in engineer schools in france you have 2 or 3 months on second year, and 5 to 8 months on the third
11:24 <@ahuillet> I don't know about university but they have some interships to do as well
11:24 <@ahuillet> university is less relevant in france than in other countries anyway
11:24 <@ahuillet> oh, and I don't even know what I'll have to do !
11:25 <@ahuillet> they have this piece of hardware intended to serve as a memory barrier (with an interesting trick)
11:25 <@ahuillet> and I have to get it working under their MPI implementation
11:25 <@ahuillet> they're supposed to have had an intern before who took care of the kernel driver part
11:26 <@ahuillet> but I was told it was likely I'd have to do everything (kernel driver, userspace lib, MPI integration, testing)
11:26 <@ahuillet> that said back in december when I talked with my mentor he did not know if the hardware would be available to me at all
11:26 <@ahuillet> so I may have no hardware at all, an FPGA, or the real ASIC
11:33 < penberg_home> :)
11:33 < penberg_home> nice
11:34 <@ahuillet> their trick for a barrier is funny, they do a LOAD assembly instruction on a "special" address
11:34 <@ahuillet> and make the memory controller wait
11:35 <@ahuillet> if I understand correctly they were designing an ASIC for some purpose I forgot, and had extra space on the silicon so they ask the high performance computing team @Bull if they had something they wanted implemented in hardware
11:36 <@ahuillet> they don't have the slightest idea whether their hardware barrier will speed up MPI barriers or not :)
11:36 <@ahuillet> *asked
11:41 < penberg_home> :)
12:23 -!- tgrabiec [n=tomekg@afhk112.neoplus.adsl.tpnet.pl] has joined #jato
12:41 < penberg_home> tgrabiec: http://www.cs.cmu.edu/afs/cs.cmu.edu/project/cmcl/link.iwarp/member/OldFiles/bwolen/papers-others/ken-pldi99.ps
12:41 < penberg_home> interesting read.
12:44 -!- t_grabiec [n=tomekg@cdv86.neoplus.adsl.tpnet.pl] has joined #jato
12:46 -!- tgrabiec [n=tomekg@afhk112.neoplus.adsl.tpnet.pl] has quit [Read error: 110 (Connection timed out)]
13:28 -!- penberg [n=penberg@cs146249.pp.htv.fi] has joined #jato
13:35 < t_grabiec> this disconnections are making me irritated. I loose conversation and my sentences are lost so I have to check the logs to find out
13:35 <@ahuillet> t_grabiec : check out the irc logs of jato-irc-logger :)
13:38 < penberg> ahuillet, t_grabiec: so I've been reading on garbage collection + jit a bit
13:38 < t_grabiec> penberg: I've read that paper you sent the link to
13:39 < penberg> from jit point of view, we need to provide a per-method "reference map" that can be used to look up where _references_ are at a given EIP
13:39 < penberg> things that are either on the stack or in registers.
13:39 < penberg> so there's some overlap with register allocation (and spilling)
13:39 < penberg> and exception handling stack walking
13:40 < penberg> t_grabiec: yeah?
13:40 < penberg> t_grabiec: http://www.research.ibm.com/people/d/dfb/talks/Bacon01Java.ppt
13:40 < penberg> seems pretty decent too
13:40 < t_grabiec> I was wandering what are we going to do with those write-reorder issues
13:41 < penberg> which are those?
13:44 < t_grabiec> I'm not sure about that, but I think we need memory barriers to be executed by all CPUs when GC is started no?
13:46 < t_grabiec> penberg: so are you going to implement GC or trick one of us to do it ? ;)
14:23 < t_grabiec> penberg: I fixed the Caused by: problem, patches sent
14:24 -!- t_grabiec [n=tomekg@cdv86.neoplus.adsl.tpnet.pl] has quit ["Leaving"]
14:25 -!- tgrabiec [n=tomekg@cdv86.neoplus.adsl.tpnet.pl] has joined #jato
14:25 < tgrabiec> ahuillet: answering your question, on october the 6th we'll have the 2nd wedding anniversary
14:51 -!- Eduard_Munteanu [n=edi@79.115.194.81] has joined #jato
14:51 < Eduard_Munteanu> Hi.
15:08 < tgrabiec> penberg, ahuillet: there is one thing to note. When we will want to implement jsr/ret support without inlining, all our algorithms must work with any number of basic block successors & predecesors
15:09 < tgrabiec> because subroutine basic block has as many successors/predecesors as the number of its call sites
15:13 <@ahuillet> yup
15:14 <@ahuillet> it should be ok
16:01 < penberg_home> tgrabiec: yup, I am trying to trick you into looking at the _root set_ discovery part
16:01 < penberg_home> which is the jit -> gc interface
16:02 < tgrabiec> aha
16:02 < tgrabiec> I'll think about it
16:04 < penberg_home> especially as I think there's quite a bit of overlap with the exception handling part
16:05 < tgrabiec> yup, exception handlers will cause some problems I guess
16:08 < penberg_home> yeah
16:15 < penberg_home> Eduard_Munteanu: salut!
16:15 < penberg_home> Eduard_Munteanu: nice set of patches you had there
16:19 < penberg> tgrabiec: hmm
16:19 < penberg> the first patch
16:20 < tgrabiec> what about it?
16:20 < penberg> is this behavior explained in the JVM spec?
16:20 < penberg> looks ad hoc to me
16:20 < tgrabiec> penberg: it's not about JVM it's about classpath
16:21 < tgrabiec> there's a comment in Throwable.java
16:21 < penberg> yeah?
16:22 < tgrabiec> http://pastebin.com/d7c945b6f
16:22 < tgrabiec> the point is that initCause() does not work for all classes
16:23 < tgrabiec> because for example ExceptionInInitializerError(cause) stores the 'cause' in it's own 'exception' field, and overrides getCause() method to return the 'exception' field value
16:24 < penberg> hmmh
16:24 < penberg> ok
16:45 < penberg> tgrabiec: http://research.sun.com/jtech/pubs/04-g1-paper-ismm.pdf
16:45 < penberg> that describes the latest hotspot GC
16:45 < penberg> I wonder how much effort it takes for us to support copying GCs
16:45 -!- ahuillet [n=ahuillet@29.152.72-86.rev.gaoland.net] has left #jato ["Cette connasse d'Ève a bouffé la pomme !"]
16:45 < penberg> that is, moving objects around.
16:46 < Eduard_Munteanu> penberg_home, thanks.
16:48 < tgrabiec> penberg: I'll take a look at that paper later, I'm working on some patches now :)
16:49 < penberg> tgrabiec: oh, sure sure
16:49 < penberg> just throwing ideas into the air
16:50 < Eduard_Munteanu> Does make test fail on your machines as well?
16:50 < Eduard_Munteanu> failure == compile error
16:51 < penberg> Eduard_Munteanu: I didn
16:51 < penberg> I didn't test it on x86-64
16:51 < Eduard_Munteanu> penberg, no, on x86-32. If it's only on x86-64 I'll fix it.
16:52 < Eduard_Munteanu> that is 'make test', not 'make regression'
16:56 < tgrabiec> penberg: are you going to merge my "[RFC][PATCH 22/22] x86: print mixed native and java backtrace." ?
17:26 < penberg_home> tgrabiec: I already did
17:26 < penberg_home> Eduard_Munteanu: on 32-bit it works
17:27 < tgrabiec> penberg_home: oh, cool. I just found out that it's broken when print_trace() is called from unit tests because vm/stack-trace.c is stubbed then
17:27 < tgrabiec> I'm working on it
17:39 < penberg_home> k
18:05 -!- tgrabiec [n=tomekg@cdv86.neoplus.adsl.tpnet.pl] has quit [Read error: 60 (Operation timed out)]
18:32 -!- tgrabiec [n=tomekg@buf226.neoplus.adsl.tpnet.pl] has joined #jato
18:37 < tgrabiec> penberg_home: I can't see the backtrace patch in master. I'll resend it with new series
18:38 < penberg_home> I think I reverted it
18:39 < penberg_home> no
18:39 < penberg_home> hmm
18:39 < penberg_home> which patch are you talking about?
18:39 < tgrabiec> [RFC][PATCH 22/22] x86: print mixed native and java backtrace.
18:39 < penberg_home> ok, maybe I just forgot to apply it?
18:39 < penberg_home> please do resend
18:39 < tgrabiec> k
18:48 < tgrabiec> I'm leaving, see you tomorrow!
18:48 -!- tgrabiec [n=tomekg@buf226.neoplus.adsl.tpnet.pl] has quit ["Leaving"]
18:52 -!- Eduard_Munteanu [n=edi@79.115.194.81] has quit ["Lost terminal"]
20:24 < penberg> vegard: ping
20:31 < penberg> vegard: never mind
20:31 < penberg> I figured it out alreayd
21:39 -!- ahuillet [n=ahuillet@29.152.72-86.rev.gaoland.net] has joined #jato
21:39 -!- mode/#jato [+o ahuillet] by ChanServ
21:51 < penberg_home> ahuillet: any progress with the bug btw?
21:52 <@ahuillet> nothing yet
21:52 <@ahuillet> I'm a real electric battery right now, can't concentrate on anything :)
22:00 < penberg_home> :)
23:44 -!- penberg [n=penberg@cs146249.pp.htv.fi] has quit [Read error: 60 (Operation timed out)]
--- Log closed Mon Jul 27 00:00:22 2009