github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt`` \o 08:48
nwc10 o/ 08:48
brrt wonders whether nwc10 has a greet-bot secretl 08:49
nwc10 none that I'm aware of 08:51
brrt well, since you're always here 09:03
jnthn He *is* the greet-bot! :P
nwc10 :-) 09:05
brrt: I think it's just that I'm at my desk most times that you appear on channel
Geth MoarVM/more-pea: ebb455d059 | (Jonathan Worthington)++ | src/spesh/pea.c
Basic handling of attribute viv

Eventually it'd be good to do away with it, but for now we'll do the basic handling of it in the EA.		 12:16
Geth MoarVM/more-pea: eaa7f157e4 | (Jonathan Worthington)++ | src/spesh/optimize.c
Extra DU chain check after PEA phase

To help understand if that's where the problems came from.		 12:34
MoarVM/more-pea: 55d254cf74 | (Jonathan Worthington)++ | src/spesh/pea.c
Make sure new PEA transforms fix up DU chains
jnthn nwc10: It probably gets a decent bit further after ^ :) 12:43
nwc10 I didn't check *that* too carefully. It still "finishes" with a DU error report 12:43
jnthn Huh, not here 12:44
nwc10 MVM_SPESH_BLOCKING=1
MVM_SPESH_NODELAY=1
MVM_SPESH_GUARD_DEBUG=1
says env 12:45
and #define MVM_SPESH_CHECK_DU 1
jnthn Aha, yeah, managed to get it with that combo. Hmm. 12:46
nwc10 but it wins, not ASAN
jnthn Not sure the last one should matter but thought I'd done the previous two
bah, I typo'd one of them :P
hmm, it's sticking the materialization in a slightly naughty place 12:47
Though not sure why that'd make the DU chain wrong 12:48
Geth MoarVM/more-pea: 92d21cdc58 | (Jonathan Worthington)++ | src/spesh/pea.c
Don't break arg instruction sequence

When we insert a materialization, make sure it goes prior to the prepargs.		 12:56
jnthn Ohhh...ouch 12:57
It materializes the object in two different branches 12:58
Thus writing it on both sides into the very same register 13:00
Same version, even.
I wonder if that's even safe. Hm. 13:02
Ah, it is, I think. But how on earth to deal with the SSA versioning. 13:03
nine r 13:13
r 13:14
Geth MoarVM/more-pea: 28ee56669d | (Jonathan Worthington)++ | src/spesh/optimize.c
Move PEA phase earlier

We'll be moving all of the bigint lowering into PEA, so we need to do the unused log guard elimination *after* it. Further, PEA removes some guards, so it makes sense to clean up deopt after it. Caveat: we might produce more deopt point data that we never use during PEA, but I think that's mostly a small waste of memory/analysis time rather than something that can lead to lower quality code, whereas the way around we had it before could obviously reduce code quality.		 13:25
jnthn Hmm, first time I've seen the point benchmark run in under half of the Perl 5 version (even without factoring out the Perl 6 startup time overhead). :)
Gotta do other stuff; bbl 13:27
nine That's the object creation benchmark? 13:34
brrt .ask samcv what do you want to do with PR #1072? 13:37
yoleaux brrt: I'll pass your message to samcv.
jnthn nine: Yes 14:56
timotimo that sounds nice 14:58
jnthn Well, I think we can be 4x-5x faster with time/effort :) 15:01
But I should really tidy up the pieces I'm on with rather than chase that :P
timotimo aye 15:14
dogbert17 jnthn: Is 30+ spectest fails the expected number currently on the more-pea branch? 18:37
timotimo i could imagine it 18:39
dogbert17 then I can relax, no need to report SEGV's etc 18:40
libuv 1.29 is out, github.com/libuv/libuv/blob/v1.x/ChangeLog 18:47
lizmat seems to have some interesting fixes 19:37
dogbert17 I saw some Mac related stuff in there 19:38
jnthn dogbert17: I didn't even dare try spectest yet ;) 20:41
yoleaux 17:46Z <vrurg> jnthn: Invitation accepted. Thank you!
samcv .tell brrt ik hub gereageerd github.com/MoarVM/MoarVM/pull/1072...-492822850 21:04
yoleaux 13:37Z <brrt> samcv: what do you want to do with PR #1072?
samcv: I'll pass your message to brrt.

Powered by Cro and the Raku® Programming Language.

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

Thank you!