github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
timotimo will this become the "free things in an extra thread" mechanism later, jnthn? 00:04
jnthn No, I don't think it'll ever be that heavily used 00:11
That'd be doing the nursery sweep asyncronously
Though that needs careful checking we can really get away with it :) 00:12
(we'd need to audit all the gc_free methods for assumptions)
That mechanism is just for when you'd use the FSA free at safepoint but didn't allocate with the FSA :) 00:13
timotimo OK 00:20
jnthn 'night 00:24
timotimo gnite jnthn
brrt \o 05:48
pmurias: I was under the impression that it had to be GPL fullstop 05:50
but anyway, i'm not so convinced that'd be actually cheaper than writing an ELF file 05:53
brrt btw, the simplest thing I can do for supporting jump-on-overflow in the expression JIT, is just to have it take a value and yield a flag 06:44
that way, we can do it after any value-yielding expression
it is also most portable. the downside is that we'd have to rely on subexpression elimination to eliminate the redundant ADD 06:45
but the good thing about it is, that on platforms that would not support an overflow flag (no idea if these are popular), this could still be made to work 06:46
whereas if we treat overflow as a 'trap' or conditional statement by itself we could not
brrt on the other hand, actually writing ELF files, on linux, isn't so hard.... 08:31
japhb testing 17:17
timotimo success
jnthn FAIL
:P
timotimo race'd
japhb It was a success. Needed to see if voice was winning over restrictions. 17:18
(Partially because I noticed that after the system message about restricting new connections, #moarvm and #perl6-dev got really quiet.)
Geth MoarVM: ugexe++ created pull request #948:
Update libuv to 1.23.0		 22:24
MoarVM/postrelease-opts: eb6110f8b8 | (Jonathan Worthington)++ | 3 files
Correct tracking of log guards at PHI nodes

Previously, we only had them on the facts of the PHI node itself. However, those facts could then be copied and relied on elsewhere. With this change, we now keep a list of the log guards that we depend on for each fact, and at a PHI node we copy all the input lists into a target list. We then, when we depend on or copy facts, just assign the input log guards list and count (which is safe since it's immutable, and we use the region allocator so there's no memory management concerns).		 22:25
timotimo oh, so we were not marking facts as "relied upon" when they went through a PHI, and that caused us to leave out guards that were necessary? 22:27
jnthn The original code set the merged log guards flag on a PHI node 22:28
But then the code that marked the facts as used expected that any node with that flag was a PHI node
This caused crashes when it wasn't a PHI node, so I stopped copying it 22:29
But that was also wrong
So now we just keep a list of the log guards that a given fact entry depend on
Merging them at a PHI node
timotimo that sounds sensible
jnthn And then we can carry them as far forward as we need, including through further PHIs 22:30
It's possible that we can carry a fact derived from a log guard through multiple PHI nodes before we reach the point something actually uses it
With this we can track it all the way back
timotimo good
jnthn CORE.setting on my home machine on that branch is now a reliable SEGV rather than sometimes another error...progress? :P 22:31
timotimo \o/
[Coke] reliably exploding sounds good. 22:32
jnthn Well, in 2 runs...then I did it with MVM_JIT_DISABLE=1 and it worked, though I'm not entirely sure if that's legit :) 22:34
But I really want a test case that isn't CORE.setting...
Alas, not yet. I thought there was one in test, but it was in the test for a bug I fixed yesterday in MoarVM master 22:40
Not much else in master that looks like it could help, alas 22:41
Randy12 | \ l /-----~~" / 22:49
| Y \ /
| | "x______.^
| | \ |
| j / |
| |; ; ; ; ; ; / ,--.........,
| |; ; ; ; ;/ ; .' -='.
| | ; ; ; ; / / .\ '
| | ; ; /` .\ _,==" \ .'
| \; ; ; .'. _ ,_'\.\~" //`. \ .'
| | ; .___~' \ \- | | /,\ ` \ ..'
| ~ ; ; ;/ =="'' |`| | | =="''\.==''
| ~ /; ;/="" |`| |`| ==="`
| ~..==` \\ |`| / /=="`
| ~` ~ /,\ / /= )")
| ~ ~~ _')")
| ~ ~ _,=~";`
| ~ =~"|; ;| Homerbird
| ~ ~ | ; | =========
| ~ ~ |;|\ |
| |/ \|
jnthn bah, I try to kick it and ChanServ netsplits before I can get ops 22:50
timotimo it seems to have finished anyway 22:51
jnthn Indeed
Randy indeed.
:P
jnthn Ah, got some spectest blowup now 22:55
One of them a SEGV in mp_mul
And only happens when the JIT is enabled
Wonder what I messed up...
jnthn Don't see the problem, alas 23:16
jnthn leaves it for another day 23:21
jnthn OK, so I failed to stop working on it but did figure it out and it was silly 23:58

Powered by Cro and the Raku® Programming Language.

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

Thank you!