Parrot 2.0.0 "Inevitable" released! | parrot.org | Priorities: deprecated core PMCs, ops -> dynops, GC tuning and implementation | Roadmap: icanhaz.com/parrotroadmap | Latest modified TT's: icanhaz.com/parrotbugs
Set by moderator on 2 February 2010.
00:04 tetragon joined
Coke I have no idea. =-) 00:08
00:15 cotto_w0rk joined
Whiteknight cotto_w0rk: not too hard. I've set it up a few times 00:20
00:21 payload joined
bacek_at_work Aloha! 00:56
cotto_w0rk hi bacek_at_work
bacek_at_work hi cotto_w0rk 00:57
dukeleto o hai
00:58 patspam joined, hiroyuki_y joined 01:03 abqar joined 01:06 cognominal joined
Whiteknight hello bacek 01:19
cotto_w0rk anyone feel like fixing headerizer? 01:22
chromatic Should be simple; I can poke at it in 90 minutes if no one beats me.
01:23 particle joined
Whiteknight oh no...must rush to try and beat chromatic to fix headerizer... 01:25
01:47 theory joined 01:49 kid51 joined 02:02 sjn joined 02:14 ash_ joined
bacek_at_work Whiteknight, can you rush to fix gc_encapsulate? :) 02:26
02:26 chromatic joined
Whiteknight what needs fixin? 02:30
building now 02:31
dukeleto hola 02:32
purl what's up, dukeleto.
Whiteknight hello duke 02:33
bacek_at_work Whiteknight, everything! 02:36
purl rumour has it whiteknight, everything is building now, but when t/op/annotate.t runs, the above happens
cotto_w0rk botsnack
purl thanks cotto_w0rk :)
Whiteknight gdbing now 02:37
dukeleto Whiteknight: howdy, fine sir
dukeleto is playing with new embedded hardware and researching how to get parrot on it 02:38
Whiteknight i'm going to have to read through this diff 02:45
damnit, no clean diff 02:50
02:54 dalek joined
kid51 heard a talk tonight on systems programming on embedded hardware 02:56
02:57 KatrinaTheLamia joined
Whiteknight bacek: Nothing I can do tonight. More looking tomorrow 03:07
bacek: one thing I noticed is that we have an Object getting collected prematurely. Suggests that either a CallContext isn't getting marked completely or another Object isn't getting marked properly 03:08
anyway, Whiteknight out! 03:09
03:23 dalek joined 03:37 cxreg left, cxreg joined 03:39 dalek joined
bacek_at_work slacker... 03:47
purl hmmm... slacker is like lazy bum
eternaleye cotto_w0rk: Trac is pretty darn easy. Set it up on my laptop once just for kicks, spent more time learnign how apache config files work than setting up trac itself. 04:08
04:20 eternaleye joined 04:32 chromatic joined 04:39 Austin_away joined
Austin Well, a segfault is nice. 05:42
cotto a segfault? 05:52
purl it has been said that a segfault is xkcd.com/371/
cotto a segfault is also nice 05:53
purl okay, cotto.
Austin cotto: Yeah, 2.0.0
cotto I guess it may not be so nice then.
Austin Heh. I'm trying to debug by dumping arguments. Naturally, dumping arguments causes a segfault. :( 05:54
cotto d'oh
Austin Heh. I'm trying to debug by dumping arguments. Naturally, dumping arguments causes a segfault. :( 05:59
Whoops. Lost track of the current window.
Hoo hah. 06:05
It segfaults in scalar concat called on a HashIteratorKey 06:06
The Hash in question is an automatically created :named:slurpy parameter hash. :( 06:07
07:03 bacek joined
Austin Significantly, if I set the wayback machine to 1.7, I get a segfault from an exception raised by PObj_is_PMC_TEST(sig_pmc), whatever *that* means. 07:16
07:28 fperrad joined
cotto Is it just me or is this line not sane: trac.parrot.org/parrot/browser/trun...rge.c#L616 07:34
It seems to have been added during a merge. 07:36
Austin Why do you think it's not sane?
cotto because it's adding an unrelated value to a char* 07:37
Austin Knowing nothing at all about the code... 07:38
It looks like mapping is allocated empty, then copied from somewhere.
The offset is bumped by line numbers, but I don't know what that signifies. Is offset a line-number type? 07:39
The filename is bumped by the const_start value of what looks like the current input - that is, the same data structure mapping was copied out of. 07:40
cotto It's odd. 07:43
I guess it makes sense though.
Austin There's lots of other places in the code where inputs[...]->const_start gets added. 07:44
I assume the filename is an offset of some kind.
So const_start is where the constant data for that segment starts, which would have been relatively 0 before the merge began.. 07:45
cotto In most cases it's clear why const_start is used.
Austin But considering that my segfault is in code that has been pbc_merged (and similar code, not merged, does not segfault) I encourage you to eliminate bugs in there. 07:46
cotto yeah. Don't use pbc_merge right now.
Austin LOL. I'm counting on it. 07:47
cotto It's definitely broken.
a workaround is to .include the files you need and compile that to a pbc with parrot
Austin Yeah.
That's what I don't want to do.
cotto My goal right now is to get pbc_merge to the point where it can become a part of Parrot's build process. 07:48
Once it's there, it'll be safe for you to use too.
Austin Is there some kind of -Dname=value supported by pir?
cotto only if you want to add it 07:49
Austin laugh
cotto (not that it'd be a bad idea, but I'm pretty sure that's not what you're hoping for) 07:50
Austin Much of the point of Kakapo is to provide a run-time library for nqp code, etc.
Which requires "library-ness", which I interpret to mean "link against this separate binary".
I'm okay with loading bytecode, but I hate the idea that every program has to have a .pir top level. 07:51
cotto which won't be doable until pbc_merge doesn't cause mysterious and wonderful failures
cotto gets back to work
Austin Well, get back to work, then.
:0
I'm glad you were here, cotto. Now I can go to bed with a clear conscience. 07:53
:)
Niterz.
cotto .o0( It's nice to see that the problem I'm working on will forestall someone's eventual insanity. ) 08:00
dalek rrot: r43714 | cotto++ | trunk/src/pbc_merge.c:
[pbc_merge] Use arrays to map input constants to their location in the output file.
 08:03
08:07 jan joined 08:51 ZeroForce joined 08:58 chromatic joined 09:03 payload joined 09:04 payload joined 09:06 payload joined 09:08 riffraff joined 09:44 eternaleye joined
mikehh All tests PASS (pre/post-config, make corevm/make coretest, smoke (#32053), fulltest) at r43714 - Ubuntu 9.10 amd64 (gcc with --optimize) 09:53
cotto good times 09:57
purl DON'T OPEN IT! IT'LL CRASH AOL!
cotto opens it expectantly
10:20 payload joined 10:40 cosimo joined 11:00 mj41_ joined 11:27 adu joined, mj41_ joined 12:13 payload joined 12:23 bluescreen joined 12:25 bluescreen joined 12:27 payload1 joined 13:17 whiteknight joined
whiteknight good morning #parrot 13:29
13:35 daqing joined
daqing hi, everyone! parrot 2.0.0 compiles failed on Mac OS X 10.6 13:36
13:41 daqing joined
daqing Parrot 2.0.0 compiles failed on my Mac OS X 10.6 machine, look at pastie.org/809296 for screen output. I 13:43
My gcc version:
gist.github.com/c18d37eaba4da7c3bb44 13:44
anyone there?
moritz takes a look, but it not very familiar with mac 13:45
daqing: how did you configure parrot?
Infinoid the "\\c" at the end of the compilation command line looks suspicious 13:46
daqing perl Configure.pl
moritz Infinoid: aye
daqing Infinoid: yes, but i don't know how to fix it 13:47
maybe the "\\c" should just be "\\" ?
or "-c"?
Infinoid that's a linker line, so probably not -c
moritz but why is it using c++ as a linker? 13:48
I don't think it does that by default
Infinoid my parrot-fu is lacking, these days. but what's the \\c magic in Parrot::BuildUtil for? 13:49
daqing I looked at the generated Makefile, only found this line: ICU_SHARED := -lpthread -lm -L/opt/local/lib -licuuc -licudata -lpthread -lm \\c 13:50
was ended with "\\c"
Infinoid is there a \\c elsewhere in that file? 13:51
daqing no
is "\\c" a typo? 13:52
Infinoid I suspect it's a regular expression "control character" match, that was taken literally by mistake 13:53
the ICU_SHARED stuff in 2.0 is generated by config/auto/icu.pm
daqing however, parrot 2.0.0 compiles perfectly on my ubuntu virtual machine, so this is weird.
Infinoid Does your mac have libicu installed? Does your ubuntu have libicu installed? 13:54
(I think this can be seen in the configure output.) 13:55
daqing auto::icu - Is ICU installed....................................yes. 13:57
yes
14:01 payload joined
daqing I just found out that the Make file generated on ubuntu had no "\\c" character on that line.. 14:04
Infinoid yeah, doesn't happen on my gentoo box either 14:10
adu hi all 14:15
Infinoid daqing: Any chance you can nopaste the output of "perl Configure.pl --verbose"?
14:15 d4l3k_ joined
daqing wait a moment.. 14:15
OK, I know the reason... 14:19
I run "icu-config --ldflags" and get: -lpthread -lm -L/opt/local/lib -licui18n -licuuc -licudata -lpthread -lm \\c
Infinoid oops :)
daqing thank you guys 14:20
Infinoid daqing++ 14:21
14:23 cognominal joined 14:30 iblechbot joined 14:37 particle joined
Coke the \\c bug is one that has been biting people on OSX on and off. 14:43
for some reason the generated makefile ends up with them.
moritz maybe we should manually remove it from the output of icu-config 14:44
Coke oh, it's from Icu?
yay.
moritz according to irclog.perlgeek.de/parrot/2010-02-04#i_1956595 yes 14:45
Coke yes, check out config/auto/icu.pm, presumably there's a good spot there to strip out the \\\\c
moritz there is indeed 14:49
redbrain hey guys have you done much more on jit support? I've been playing with llvm jit seems one would have to write a llvm jit driver in C++ then just have it callable from c code 14:52
dalek rrot: r43715 | moritz++ | trunk/config/auto/icu.pm:
[config] remove \\c that is sometimes returned from icu-config on MacOS
 15:12
15:22 particle joined 15:30 Psyche^ joined
whiteknight If I add a new project to the Modules page on the Tracwiki, it should get picked up by dalek? 15:39
15:40 bubaflub joined
Coke redbrain: I don't think anyone has touched llvm integration since we decided "yay, that's nice." 15:42
Infinoid whiteknight: yes, assuming dalek knows how to find and parse its rss feed 15:44
whiteknight yeah, it's a standard github project
Infinoid cool, should work. May take dalek a couple hours to notice it tho
dalek tracwiki: v5 | whiteknight++ | Modules
tracwiki: trac.parrot.org/parrot/wiki/Modules...ction=diff
tracwiki: v6 | whiteknight++ | Modules
tracwiki: PLA is in use by Matrixy and is mostly stable for now. Bump to Beta stage
tracwiki: trac.parrot.org/parrot/wiki/Modules...ction=diff
whiteknight Infinoid: how often does dalek scrape that projects page?
"Modules" page* 15:45
Infinoid it polls once per hour, I think 15:46
NotFound Impatience the Dark Side is ;) 15:47
15:47 Andy joined 15:48 Andy joined
dalek nxed: r401 | julian.notfound++ | trunk/winxedst1.winxed:
tiny optimization of const statements in stage 1
 15:57
redbrain Coke: yeah its alot of work i think but its interesting :) 16:03
16:08 ruoso joined
darbelo dukeleto: ping 16:12
16:19 davidfetter joined 16:27 hercynium joined 16:31 allison joined 16:33 cotto_working joined 16:46 PacoLinux joined
darbelo whiteknight: ping 16:50
whiteknight darbelo: pong 16:51
darbelo I tried to build parrot-linear-algebra and it failed with sh: ./probe: not found 16:52
whiteknight hmm... weird
darbelo Is the repo missing a file?
Or should I stay away from the setup.pir file? 16:53
whiteknight how did you try to build it?
setup.pir is the way to do i
it
darbelo "parrot setup.pir"
16:53 cotto_working joined
whiteknight you're on FreeBSD, right? 16:53
darbelo OpenBSD 16:54
whiteknight okay, you probably need to add configuation settings o system_linker_settings in setup.pir
16:54 theory joined
darbelo If you point me in the right direction I can probably do that. 16:55
whiteknight does OpenBSD have ldd?
darbelo Yep. 16:56
whiteknight okay, so you just need to figure out the install path where cblas or atlas is on your machine
/usr/lib/libblas.so or /usr/lib/atlas/libcblas.so are the paths i have see in linuxland 16:57
darbelo And incorporate to system_linker_settings?
whiteknight exactly 16:58
it should be straight-forward, I just wasn't sure what "$S0 = sysinfo 4" is on a BSD system
darbelo Oh, crap. I didn't have cblas on this box. 17:00
darbelo facepalms 17:01
Still, I don't know where that reference to 'probe' came from.
whiteknight yeah, that's a weird one
17:02 plobsing joined
darbelo And, we should move the place where the lib-detection takes place. 17:02
17:03 particle joined
darbelo Right now that happens even if I try parrot "setup.pir update" 17:03
eh "parrot setup.pir update"
plobsing am I correct in thinking that RPA shift/unshift are going to suck for the forseable future? 17:06
whiteknight plobsing: yes
plobsing: but take a look at github.com/Whiteknight/parrot-data-structures/
started a new project with higher-performance single-purpose PMCs 17:07
plobsing whiteknight: then we should find all places in core that use RPA and make sure they don't shift/unshift
I just did it for freeze/thaw in r43716.
whiteknight plobsing: that is probably a good idea
NotFound Emmmm... main args 17:10
whiteknight plobsing: freeze/thaw use an ImageIO PMC, not an RPA 17:11
NotFound Any program that uses GetOpts use shift.
darbelo That has an RPA inside.
whiteknight darbelo: it doesn't need to. It's encapsulated, we could fill it with whatever we want, eventually
NotFound You shift argv before passing it to GetOpts 17:12
darbelo whiteknight: It has an RPA now ;)
plobsing whiteknight: it doesn't make sense to have ImageIO manage it's own list when we have a type whose responsibility is to manage lists
whiteknight plobsing: we have a type whose responsibility is to provide maximum flexibilities in managing all sorts of lists
darbelo But, in my opinion, ImageIO has one list too many. 17:13
whiteknight when ImageIO only needs a very specific type of list
plobsing whiteknight: last time we wanted tunable lists like that, we wound up with src/list.c
darbelo: which is the one list too many? 17:14
darbelo: are you talking about merging todo and id_list?
whiteknight plobsing: list.c was the wrong approach, but was not born from a bad idea
sparse arrays do have some benefit, and arrays implemented as linked-lists have benefit
plobsing darbelo: I've considered that. The visit loop would have to change and would likely not be compatible with the freezing logic. 17:15
darbelo: if you separated the freezing and the thawing, you could do that
whiteknight improved shift/unshift/push/pop performance in exchange for worse direct indexing and higher memory requirements
plobsing whiteknight: my main point is that an external library doesn't help core usages 17:17
darbelo plobsing: I've been thinking about that. And separating freeze from thaw looks like a win all around.
But doing that right probably means adding a vtable. 17:19
dalek rrot: r43716 | plobsing++ | trunk/src/pmc/imageio.pmc:
use push/pop with imageio todo list instead of shift/unshift.
 17:21
Coke 100 what? 17:22
purl 100 is the reference
plobsing 100 times
Coke ah. 100x would have been more clear to me.
plobsing 17s becomes 0.12s for me
Coke might be worth adding a note to RPA's pod docs about that. "if you're using this as a stack/queue/blah, prefer <foo> over <bar>" 17:23
NotFound I like the C++ approach: you have vector, deque and list, choose what's better to each task. 17:26
darbelo NotFound++ 17:27
Coke huh. I'd never heard of deque before.
Infinoid queue? 17:28
purl queue is FIFO
NotFound You almos never use it directly, but is the default container for stack and queue.
Supposedly it has not bad direct access and not bad pop and shift. 17:29
In practice, check your standard library implementation. 17:30
Coke certainly that 100x speedup is a good reason to either 1) fix RPA or 2) split it out. 17:31
whiteknight plobsing: the external library does help if it is really good and we can make the case to merge it into trunk eventually 17:34
Coke certainly easier to prototype stuff out of core. 17:35
whiteknight exactly
Coke I wonder how much of a speed up partcl could get if it were smarter about its RPA usages.
(ditto nqp-rx)
NotFound "This may seem like a joke, but it's not. The US Patent and Trademark Office will not accept patent filings faxed in if they arrive upside down. That's right, the home of innovation of the federal government is incapable of rotating an incoming fax file, whether electronically or on paper." 17:39
Ups, wrong channel.
Coke ... still, funny. 17:44
plobsing whiteknight: it all depends on when eventually is? 2.3? 3.0? my grand children might see it? 17:46
whiteknight true
17:49 ruoso joined 17:53 cognominal joined
plobsing what's the best way to get a stacktrace of everywhere RPA.shift or RPA.unshift is called in the testsuite? 17:58
Coke ... a stacktrace?
plobsing I'm looking for internal usages. I don't really care about calls from PIR, so I plan to filter out calls from src/ops/* 17:59
Coke if you don't care about PIR or PASm, the number of times it's invoked should be pretty miniscule.
(explicitly) 18:00
plobsing this morning it was being called in freeze/thaw. I'm looking for similar cases. 18:01
low hanging fruit
Coke I don't know an easy way to do that, no.
If I were looking for someone, I'd probably ping chromatic or cotto.
dngor Anyone still have svn projects they'd like to convert to git? 18:07
cotto_working Parrot. 18:08
cotto_working ducks
Coke wonders why opensolaris.org's website looks awful in every browser except firefox. 18:09
dngor Parrot... ducks... what next? Flamingos?
szbalint crocoduck
dngor I've got snerp vortex working on a single repository (2700 commits) reasonably well (diff -r vs. svn checkout is identical) and somewhat fast (800 seconds to convert). 18:10
It's only a single test case, though.
szbalint will probably convert $work repo to git this month 18:11
dngor If people have some idle time to evaluate it, I'd love some bug reports.
dukeleto 'ello 18:14
cotto_working hi dukeleto 18:15
dukeleto are we talking about converting to git again?
dngor Oh, you know what, I'm in the wrong channel. I recently re-sorted my windows. 18:16
Epic ww fail. Sorry.
dukeleto dngor: :)
szbalint pretty funky :) 18:17
dngor wanders one window over to #perl.
szbalint anyone attending FOSDEM from here this weekend btw?
18:19 joeri joined 18:20 ruoso joined 18:23 payload joined
whiteknight plobsing: locally, modify the VTABLE_shift() macro to print __FILE__ and __LINE__ to stderr 18:45
plobsing whiteknight: that's too coarse, that gets all shifts. I could just grep for VTABLE_shift, but there are legitimate places to shift. 18:46
whiteknight plobsing: how many types do you suspect are using shift/unshift? 18:47
cotto_working You could further modify it to only print based on vtable->base_type.
darbelo dukeleto: ping 18:48
whiteknight #define VTABLE_shift(i, p) (fprintf(stderr, ((p->vtable->type == enum_class_ResizablePMCArray)?("%s, %d"):("")), __FILE__, __LINE__), p->vtable->shift(i, p));
plobsing cotto_working++ whiteknight++
whiteknight It's been a long time since I used that comma operator, so I might have put things in the wrong order 18:49
18:57 kurahaupo joined
Coke purl? 19:19
purl yes, Coke?
19:20 mikehh joined 19:28 hercynium joined
dukeleto darbelo: pong 19:47
szbalint: davidfetter will be at FOSDEM 19:48
darbelo dukeleto: I'm thinking of starting a branch to improve our cross-compilation capabilities? Would you be willing to be a guinea pig? 19:54
dukeleto darbelo: yes 19:55
darbelo++ 19:57
darbelo: i am doing research on my new BUG toy
darbelo: how much do you know about OpenEmbedded?
darbelo dukeleto: Not much. Skimming their site now. 20:03
20:04 chromatic joined, kurahaupo1 joined
szbalint dukeleto: oh, he's even giving a postgres talk, excellent :) 20:11
20:14 clinton joined
dukeleto darbelo: i have an embedded linux device that runs PokyLinux, which is a distro of OpenEmbedded 20:23
Coke Anyone here using Padre? 20:24
(for working on parrot?)
dukeleto szbalint: yes, and he will be telling people about PL/Parrot as well
szbalint excellent :) 20:29
20:30 davidfetter joined
dalek tracwiki: v24 | kurahaupo++ | ArrayTasklist 20:39
tracwiki: trac.parrot.org/parrot/wiki/ArrayTa...ction=diff
20:40 bluescreen joined
dalek rrot-linear-algebra: a050c24 | Whiteknight++ | setup.pir:
editor auto-removed some whitespace
 20:43
Coke array tasklist shows "depreation of array" as a task item; already ripped out. 20:47
chromatic Parrot VM: Can't stat ext/nqp-rx/nqp-rx.pbc, code 2. 20:54
whiteknight oh that looks wonderful 20:55
nqp-rx didn't get deleted from the repo, right?
chromatic No Makefile in that directory either.
... which isn't a problem. 20:57
Maybe I should use parrot-nqp.pbc instead. 20:58
whiteknight does the new nqp build in the repo, or are the .pbc files just dumpd there?
Tene whiteknight: build in the repo. 20:59
whiteknight ok 21:02
chromatic NQP generated code does a fair amount of RPA unshifting for exception handlers.
It doesn't lead to array resizing though; one RPA and one EH per invocation, so far. 21:04
cotto_work easy karma if someone feels like ripping out the malloc return value checks from src/pbc_merge.c 21:30
The malloc functions already handle failed allocations. 21:31
chromatic I have a fresh master checkout.
21:33 bluescreen joined
chromatic make: *** No rule to make target `include/parrot/list.h', needed by `src/string/api.o'. Stop. 21:36
Hm.
Coke wasn't that ripped out recently? checking... 21:37
cotto_work The list code was removed recently with the Array excision iirc.
Coke You would think someone would have noticed that by now, though. 21:38
cotto_work You'd think.
chromatic It's strange.
Let me reconfigure. 21:39
dalek rrot: r43717 | chromatic++ | trunk/src/gc (3 files):
[GC] Don't run a full mark and sweep when a constant pool is out of free

benchmarks.		 21:40
Coke WFM with a fresh checkout.
purl benchmarks are like college lab reports....often fudged to achieve the desired results
Coke you're using git-svn, yes?
chromatic Yes, like God and queen intended.
Coke no, they meant for git. 21:41
git-svn is evil. =-)
chromatic I'm no royalist.
God and Queen (without Paul Rodgers) want me to use git. 21:42
The monarchy... well, they're pragmatic. That's the difference between established positions of power and artists who play the piano and sing high.
A reconfigure and rebuild fixed things. La plus ca change... 21:43
dukeleto did someone say git? 21:44
and yes, git-svn is evil
i have found out that my embedded hardware is a "Freescale i.MX31 ARM-based processor" 21:45
darbelo dukeleto: /me thinks he know how to enable cross-building. 21:46
dukeleto darbelo: go ONNNNNNNN
darbelo dukeleto: does your new hardware have perl ported to it?
dukeleto darbelo: yes
darbelo Then we just have to convince Configure to ingore what the perl used to run is tells it and to feed it data from the other perl. 21:48
dukeleto darbelo: interesting
darbelo: quite a nice little trick
darbelo That and a medium sized hints file should give us a makefile that can cross build most of the c code.
I guess we might be alble to get as far as miniparrot on that. 21:49
dukeleto this boot loader gets a prize for having a cool name: en.wikipedia.org/wiki/Das_U-Boot
darbelo: sounds like a plan. do you want to make a branch for it?
cotto_work chromatic++
chromatic Where'd dalek go?
cotto_work bot vacation 21:50
chromatic bot trek
cotto_work probably to 01
dukeleto botbattical
chromatic walkabot 21:51
darbelo dukeleto: I'll bran myself when I get some code written. 21:53
s/bran/branch/
dukeleto darbelo: sounds good! let me know if you need testing 21:56
dalek rrot: r43718 | chromatic++ | trunk/src/pbc_merge.c:
[pbc_merge] Removed malloc() checking, because libparrot's malloc() wrappers
GeJ Good morning everyone. 22:03
dukeleto GeJ: o hai 22:08
22:22 Whiteknight joined
Whiteknight I would like to write a parrot plugin for xchat\\ 22:23
darbelo I don't use xchat, but will offer moral support to the endeavor ;) 22:25
22:29 elmex joined
dalek nxed: r402 | julian.notfound++ | trunk/winxedst1.winxed:
time and floattime predefs in stage 1, quick fix for float * operator
 22:33
22:36 patspam joined
dalek nxed: r403 | julian.notfound++ | trunk/examples/opengl.winxed:
opengl example
 22:38
japhb NotFound++ # OpenGL examples, w00t! 22:50
dalek tracwiki: v7 | cotto++ | Modules
tracwiki: typo fix
tracwiki: trac.parrot.org/parrot/wiki/Modules...ction=diff
NotFound japhb: just one more incarnation of the infamous triangle ;)
japhb is considering porting his old Perl 5 Doom Classic level viewer to Parrot. I was thinking PIR or NQP, but maybe Winxed is dying for some OpenGL hackery .... 22:51
NotFound I'll support such noble goal 22:52
japhb is currently reading a code review of Carmack's recent "official" port of Doom Classic to the iPhone. 22:53
NotFound Looks like Doom is on the wave right now, there is even an iphone port.
japhb :-)
dukeleto wowzers
NotFound japhb: I've readed it this morning.
japhb loves the fact that Carmack has been respecting the spirit of his GPL release by starting all his recent iPhone ports from community releases (iPhone Doom Classic is I believe based on prBoom, for example, and iPhone Wolfenstein 3D is based on Wolf Redux), and then re-releasing the upgraded source .... 22:55
22:56 joeri left
tewk What can't RPA just have a start pos and a length field and be a circular buffer that only moves data when it overflows, That should support shift, unshift, pop and push in constant time. 22:56
What does perl5 do internally? its fast right? 22:57
Whiteknight circular buffers can't be easily resized 22:59
japhb Whiteknight: why not?
tewk Whiteknight, sure they can, its just two memcpys
Whiteknight "easy"
that's worse then current wit one memcpy 23:00
japhb A little pointer fiddling and an extra memcpy do not argue against a vastly faster implementation.
And the memcpy is amortized well if you do exponential resizing.
(or heck, any decent resizing scheme) 23:01
Plus, you can save a memcpy invocation by simply checking whether one of them is zero length.
tewk two memcopies are performance bounded by total memory moved not by two vs one function calls
japhb What tewk said
Whiteknight memcpy tends to be higy optimized. biggest part of it is the call and setup 23:02
japhb Speaking of performance improvements,
Whiteknight fewer calls == less overhead == better performance
tewk no the malloc before the memcopies is the bottle neck
Whiteknight, thats not true 23:03
japhb idly wonders if there's a way to reserve a register for interp. Might be a win on 16+ register arches.
mikehh All tests PASS (pre/post-config, make corevm/make coretest, smoke (#32058), fulltest) at r43718 - Ubuntu 9.10 amd64 (g++ with --optimize)
darbelo japhb: There is no portable way to do anything to registers ;) 23:04
chromatic Any decent compiler should already realize that where it matters.
tewk performance on modern processors is almost always a function of cache misses not ops.
japhb Whiteknight, I agree with generally trying to save calls, but in this case as I said most of the time one of the memcpys can be avoided anyway.
chromatic, I disbelieve that "any decent compiler" will notice that we use a param called 'interp' freaking everywhere and realize it ought to be a register, and stop thrashing the stack with it. 23:05
Whiteknight can't be avoided in a threade app
darbelo japhb: it's always called 'interp' but it's not allways the same interp ;) 23:06
Whiteknight true, especially in interp cloning routines
japhb Whiteknight, there's no reason that reserving a register would be a problem with a threaded app, because the thread switch will save and restore all registers. It *would* be a problem if it was a memory global. 23:07
Whiteknight what we could do to save effort is set the compiler to use fastcall for calls by default.
japhb Interp cloning would be a special case.
Whiteknight but then we would need to add _declspec(cdecl) to PARROT_EXPORT 23:08
that would save on stack thrashing
chromatic If a compiler can't recognize that the first parameter passed to a function gets used as the first parameter for every sub function call, it's not a decent compiler.
tewk x86_64 calling convetions already pass it in a register all the time. Just wait for 32bit to die.
japhb chromatic, fair enough ... but we're talking about gcc (which fails to recognize all sorts of things) and msvc (which still can't manage to update to a C spec over a decade old) 23:09
Whiteknight tewk: when's the heat death of the universe again?
chromatic I don't recall calling GCC or MSVC particularly "decent", no....
Whiteknight I like ICC, good optimizations (on some platforms)
I'm also liking clang when I use it 23:10
japhb 32-bit never dies, it just moves downmarket (and gains orders of magnitude more devices, sigh)
tewk: ooh, I hadn't read a calling conventions spec since the dark ages. Didn't realize x86_64 did the "first 4 params in registers" optimization. Very nice. 23:12
Whiteknight I'm surprised I can't find any 16-bit PCs at best buy
darbelo dukeleto: nopaste.snit.ch/19478 23:13
chromatic That's because they haven't upgraded their sales droids from 8-bit.
japhb They're all over in the "truly embedded" space ... but at this point I think a million transistors (a 486) is so freaking small that you're completely swamped by packaging costs, so why not go 32-bit?
Hell, a modern Intel chip contains a 486-equivalent embedded in it just to compute power saving states. Cause when a million transistors is .1% of the die, why not? 23:14
darbelo japhb: Actually, that's what happening. The embedded market is abandoning 8-bit and 16-bit and going to ARM this days.
japhb darbelo, nodnod. I knew that, I just worded my comment poorly. 23:15
chromatic I wonder if we could shove __attribute__((fastcall)) in the PARROT_INTERP macro.
No, that's not where it goes.
japhb I still have a few freaky friends in my Perl Mongers group (dominated by EE types) that still do PIC-controlled devices, but I think they just do it for the retro fun. 23:16
darbelo japhb: Well, I was part of an embedded-control project two years back that was built around an 8-bit uC. Not for the retro fun, but becouse of the costs. 23:17
tewk Also I'm guessing the microcode on x86 32bit cpus tears apart stack parameter passing instructions and highly optimizes them, but I may be wrong.
japhb darbelo, reduction in board traces and layers? 23:18
darbelo Not really, it was a thorougly dumb device in a small control loop with no need for speed. And most of the sensors were off-board. 23:20
The most coplicated part in it was a small 64x128 pixel screen. 23:21
japhb has visions of 8-by-8 sprite blocks
chromatic We could try regparm(3). 23:22
darbelo You can't go cheaper than a 8051-alike with those constraints.
japhb chromatic, not in my manpages ... what OS? 23:23
japhb wonders if Debian decided not to include one of the base manpages again
chromatic GCC should have it since at least 3.0. 23:24
It's a function attribute, not a syscall. 23:25
japhb ah. The (3) confused me. 23:28
chromatic Yeah, I realized that as soon as you asked.
It segfaults nicely, at least. That indicates that *something* is working. 23:34
dalek rrot: r43719 | darbelo++ | trunk (3 files):
Make the used hints file overridable on the Configure.pl command line.
chromatic Hm, that naive patch made things slower. 23:41
japhb Maybe gcc is defaulting to doing something similar already? 23:45
23:47 plobsing joined
cotto_work Whiteknight, your blog post has an unusually entertaining typo: "PriorityPMCQueue: This is a queue type where each element has an ass" 23:53
Whiteknight I copy+pasted that from wikipedia :) 23:54
cotto_work That doesn't make it true.
Whiteknight lies! 23:55
cotto_work I know of very few queue ATDs that have asses.

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!