github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
lizmat github.com/wangyi-fudan/wyhash 14:08
according to rurban: now the fastest (and safe, if I understand this correctly): github.com/rurban/smhasher/ 14:09
brrt ohai #moarvm 14:48
someone reached out about the linking proposal, yay
timotimo that's gsoc, right? 15:19
brrt yes 15:25
timotimo very nice 15:26
so brrt, i was thinking. perhaps the exprjit should get a set of ops that do memory traversal (offset, load, index) but allow constant-folding at compile/optimize time. that way we wouldn't have to teach the jit what datastructures are fine to constant-fold access to while jitting 15:29
timotimo does that make any sense? 15:40
pmurias what's the state of this PR: github.com/MoarVM/MoarVM/pull/1049? 16:15
a nqp::uname would be an elegant way to go around the lack of bin/uname in browsers
lizmat pmurias: afaik it was waiting for the MoarVM release 16:21
brrt timotimo: that's pretty much the plan, yes :-) 16:35
timotimo oh, ok! 16:36
what's keeping us? :)
brrt the missing ancestor problem
I'll write a blog about it sometime
timotimo why can't we get compile-time evaluation without pulling in the optimization that causes that problem? 16:37
timotimo who wants to help me make zstd decompression work? well, it could actually have been the compression part that went wrong 18:51
BBIAB 18:52
masak brrt: I for one would love to read a blog post about the missing ancestor problem 19:35
timotimo so 20:00
still looking for volunteers to figure out what my zstd code does wrong :| 20:01
dogbert17 timotimo: what kind of help do you need? 20:18
timotimo i just made it work 20:21
there was a dumb off-by-one in my code :|
dogbert17 timotimo++ 20:22
timotimo anyway, i can now decompress some zstd data and if the compression finished get the "leftovers", which is important for my use case 20:23
next step will be to implement compression on moarvm's side, i.e. in C 20:25
dogbert17 and that code will be used by the heapdump analyzer? 20:27
timotimo well, it'll be used to write the snapshots
the part in the analyzer binds libzstd via nativecall
dogbert17 are you going to make a module of the libzstd bindings? 20:30
timotimo yep
dogbert17 cool
timotimo the code is actually horrifying right now
have any interest in helping? :P
dogbert17 to make the code less horrifying :) 20:31
timotimo yeah 20:35
dogbert17 btw, are the GPW videios (if any) online?
*videos
timotimo not yet, i don't think 20:36
nwc10 a1 20:58
1
le sigh.
VPN hates me
you can have a random VPN password too if you like :-) 20:59
307834
(now expired)
timotimo that does not fill the emptyness in my soul 21:01
i wonder if it's a bad idea to scroll through a few stores to see what kind of laptop you can get for relatively little money at the moment 21:06
tadzik not sure how relative is your relative, but I got an A485 a few days ago 21:07
timotimo wow enterprise grade security 21:09
how did you configure it?
tadzik hardware-wise? I think it's the cheapest model :P
timotimo but hopefully FHD rather than HD? 21:10
nwc10 timotimo: the fault you had with the laptop you had at GPW can't be fixed?
tadzik oh, so not the cheapest model :P
timotimo i haven't yet had the opportunity to dig into that more :|
tadzik but yes, fullhd
timotimo and hopefully more than 4 gigs of ram?!?
tadzik 20muf25epb is the model number I have 21:11
16G
timotimo yeah, thought so
tadzik damn, they make them worse than I thought then :P
it was the cheapest one from the retailer I used then P:
timotimo default is 500 gigs 7200rpm hard drive, but the others are all ssds 21:12
tadzik I have a 512G nvme ssd in this one 21:13
timotimo huh, the ips multitouch is cheaper than the ips anti-glare
but i'd really rather have anti-glare than multitouch?!
tadzik I have the anti-glare 21:14
timotimo mhm
tadzik maybe that's why it's more expensive, see :P
becuase it's more desirable
timotimo the cheapest cpu? they have ryzen 3 pro, ryzen 5 pro or ryzen 7 pro
tadzik model name: AMD Ryzen 5 PRO 2500U w/ Radeon Vega Mobile Gfx
timotimo the way i configured it, it'd be USD970 21:15
but an ssd would be a nice-to-have, too
BBL, gotta cook 21:16
brrt timotimo: because hard 22:14
roughly; the bug exists because the expression IR is unordered
because it is unorded, value-based, and a DAG, an expression cannot be strictly assigned to a basic block.
Because an expression is not assigned to a basic block, we can't know when we're eliminating a value that is referenced by some other block, and so we can't maintain the 'ancestor value' property 22:15
we currently maintain that property by hand, as it were
humans write code which is correct
but the optimizer breaks that because it has no (local) way of checking whether the optimization is valid 22:16
timotimo i'm having kind of a hard time getting much of the expr graph stuff into my brain 22:22
do we have anything much like the spesh optimizer tracer? 22:24
which is the thing that dumps a spesh log every few steps of the optimizer
brrt no 22:39
we don't have that yet
it's kind of too late for me to wrap my head around anything
so....
I'll leave the bigger explanation for another day :-)