Perl 6 - the future is here, just unevenly distributed

IRC log for #moarvm, 2014-05-18

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
06:23 woolfy joined #moarvm
07:20 woolfy joined #moarvm
07:23 woolfy left #moarvm
07:24 woolfy1 joined #moarvm
07:24 woolfy1 left #moarvm
07:59 lizmat joined #moarvm
08:08 FROGGS joined #moarvm
08:29 vendethiel joined #moarvm
08:33 woolfy joined #moarvm
08:38 brrt joined #moarvm
10:00 dalek MoarVM: 7785f67 | nicholas++ | src/6model/sc.h:
10:00 dalek MoarVM: Merge the bodies of MVM_sc_get_{obj,stable}_sc and *_set_*
10:00 dalek MoarVM:
10:00 dalek MoarVM: The two *_get_* functions call MVM_sc_get_collectable_sc, and the two
10:00 dalek MoarVM: *_set_* functions call MVM_sc_set_collectable_sc.
10:01 dalek joined #moarvm
10:03 nwc10 :-)
10:05 FROGGS nwc10++ # killing dalek
10:06 nwc10 ammo doesn't kill bots, people kill bots
10:06 nwc10 so
10:06 nwc10 jnthn++
10:07 jnthn I let the patch restore the thing I took out 'cus it SEGV'd the Panda build on Linux
10:07 jnthn It still does.
10:09 nwc10 :-/
10:11 nwc10 jnthn: I'm still playing with the 24:8 split - right now it seems to be slower
10:12 nwc10 which is wrong! as it should reduce memory allocation by about 256K
10:12 nwc10 so going to look at the generated code
10:12 dalek MoarVM: 4befa95 | jnthn++ | src/6model/serialization.c:
10:12 dalek MoarVM: Back out index-setting in deserialize again.
10:12 dalek MoarVM:
10:12 dalek MoarVM: This breaks the Panda build on some platforms.
10:12 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/4befa95214
10:12 nwc10 but also planning to use pahole to find all the gaps in structures, and where possible re-order to reduce memory usage
10:13 nwc10 and see if (a) that doesn't look crazy stupid unmaintainable
10:13 nwc10 (b) helps
10:13 nwc10 second-order thing - pahole neatly points out cache lines
10:13 * jnthn hopes for a hole lot of little improvements from it :)
10:13 nwc10 and one structure has a bunch of mutexes which are close enough together that some share cache lines
10:14 * nwc10 groans
10:14 jnthn ugh, yes
10:14 nwc10 reply pun failure
10:47 brrt joined #moarvm
11:19 * masak .oO( https://soundcloud.com/pomdeterrific/pomdeter-call-me-a-hole )
11:20 jnthn .oO( you're a hole )
12:19 brrt joined #moarvm
12:45 dalek MoarVM: 9dacb45 | jonathan++ | src/ (2 files):
12:45 dalek MoarVM: Specialize getattr_[ins] on P6opaque.
12:45 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/9dacb4595a
12:47 jnthn With that, time spent in get_attribute in CORE.setting is done from 4.23% down to 3.96%
12:47 jnthn Now for the object case...
13:27 brrt joined #moarvm
13:28 woolfy1 joined #moarvm
13:51 btyler joined #moarvm
14:05 btyler jnthn: not sure if this is helpful information, but Moar at 4befa95 was ok with panda prior to rebuilding nqp and rakudo
14:07 jnthn btyler: Yeah, decided we'll put that off
14:07 jnthn btyler: Until after next release
14:09 btyler jnthn: alrighty, sounds good
14:25 zakharyas joined #moarvm
14:28 dalek MoarVM: b31ac96 | (Tobias Leich)++ | src/math/bigintops.c:
14:28 dalek MoarVM: a bigint is big when it does not fit in
14:28 dalek MoarVM:
14:28 dalek MoarVM: On 64bit platforms the DIGIT_BIT is 60, so when we want to store something bigger, we need a
14:28 dalek MoarVM: bigint. The condition before this patch was always false.
14:28 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/b31ac9626a
14:30 ggoebel111115 joined #moarvm
15:15 masak nwc10++ # speeding up moar by aligning bits
15:47 dalek MoarVM: 187eedf | jonathan++ | src/ (2 files):
15:47 dalek MoarVM: Optimize object attribute accesses in spesh.
15:47 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/187eedf467
15:48 jnthn Used to spend 4.23% of CORE.setting build in get_attribute; now it's 1.85%.
15:48 jnthn And think there's some more win to be had yet.
15:52 tadzik :o
15:52 tadzik jnthn++
16:02 lizmat joined #moarvm
16:03 woolfy joined #moarvm
16:07 tgt joined #moarvm
16:13 brrt plz checkout my blog again :-) http://brrt-to-the-future.blogspot.nl/2014/05/moarvm-as-machine.html
16:14 jnthn ooh, new post
16:14 FROGGS joined #moarvm
16:15 brrt yes, new post
16:15 brrt :-)
16:15 * brrt dinner
16:15 moritz brrt++
16:15 brrt left #moarvm
16:15 moritz brrt: iirc one of the reasons for making moarvm a stack machine is that it makes some codegen easier
16:15 moritz ... to slow
16:15 masak ironic that the blog post requires a JavaScript VM to read ;)
16:16 moritz aye, blogspot--
16:16 jnthn uh, moarvam ain't a stack machine
16:16 moritz s/stack/register/
16:16 moritz argl
16:17 tadzik br<TAB><TAB>
16:17 moritz 18:15 -!- brrt [~bart@195-241-164-190.ip.telfort.nl] has left #moarvm []
16:17 tadzik yeah
16:17 tadzik is there a .tell bot here?
16:18 tadzik no. Oh well, I'll wait :)
16:20 lizmat .tell
16:20 lizmat :-(
16:26 masak brrt++ # nice post
16:27 jnthn One other reason cheap deopt is desirable is it makes it uncostly to speculate lots
16:39 dalek MoarVM: ef5657e | jonathan++ | src/ (2 files):
16:39 dalek MoarVM: Spesh indirect form of attr ops with known names.
16:39 dalek MoarVM: review: https://github.com/MoarVM/MoarVM/commit/ef5657e713
16:40 woolfy left #moarvm
17:10 lizmat joined #moarvm
17:10 colomon joined #moarvm
17:34 nwc10 jnthn: I can haz SEGVs
17:35 nwc10 eg ./perl6-m t/spec/S05-transliteration/with-closure.rakudo.moar
17:35 nwc10 goes away with export MVM_SPESH_DISABLE=1
17:55 brrt joined #moarvm
17:58 * brrt is back
17:58 brrt did is say moarvm is a stack machine?
17:58 brrt :-o
18:01 brrt tadzik?
18:29 tadzik brrt: oh
18:29 tadzik freqeently in your blog post :)
18:29 brrt hi :-)
18:29 brrt really? that is embarassing
18:29 brrt where
18:29 tadzik it's a hyperlink
18:30 tadzik somewhere at the beginning iirc
18:31 tadzik also, I really liked it :)
18:32 brrt thanks :-)
18:32 brrt oh i've found it
18:39 nwc10 jnthn: 4 spectests SEGV with spesh. Probably all the same bug.http://paste.scsys.co.uk/377252
18:45 nwc10 jnthn: culprit is ef5657e713e8c7a92753614cf1ea9e2bbcf3e0e5 --    Spesh indirect form of attr ops with known names.
18:49 nwc10 jnthn: may not be obvious from that nopaste - 4 failures are t/spec/S05-transliteration/with-closure.rakudo.moar t/spec/S05-transliteration/trans.rakudo.moar t/spec/integration/advent2009-day09.t t/spec/integration/real-strings.t
20:12 * brrt is off
20:12 brrt left #moarvm
20:29 timotimo brrt: if you read this in the backlog, you may want to change the wording of the sentence about add_i to indicate that the first operand is the target, as is the case with all operations i can think of right nw
22:03 lizmat joined #moarvm
22:14 camelia joined #moarvm
22:18 lizmat joined #moarvm
22:19 sergot_ joined #moarvm
22:19 nwc10_ joined #moarvm
22:19 dalek joined #moarvm
22:19 cognominal joined #moarvm
22:19 Util joined #moarvm
22:19 tadzik joined #moarvm
22:19 64MAAITRZ joined #moarvm
22:19 TimToady joined #moarvm
22:19 tokuhirom joined #moarvm
22:19 harrow joined #moarvm
22:19 flussence joined #moarvm
22:19 rurban_ joined #moarvm
22:19 timotimo joined #moarvm
22:19 bcode joined #moarvm
22:19 bonsaikitten joined #moarvm
22:19 zacts joined #moarvm
22:19 17SAADFTQ joined #moarvm
22:19 nwc10 joined #moarvm
22:20 zacts_ joined #moarvm
22:21 FROGGS joined #moarvm
22:24 woolfy joined #moarvm
23:01 lizmat joined #moarvm
23:08 woolfy joined #moarvm
23:11 woolfy joined #moarvm
23:12 woolfy joined #moarvm
23:21 xiaomiao joined #moarvm
23:27 woolfy joined #moarvm
23:47 woolfy left #moarvm

| Channels | #moarvm index | Today | | Search | Google Search | Plain-Text | summary