|
00:01
colomon joined
02:29
colomon joined
02:48
ilbot3 joined
02:52
Util_ joined,
avar joined
03:07
colomon joined
05:43
vendethiel joined
07:12
FROGGS joined
07:20
domidumont joined
07:26
domidumont joined
07:51
kjs_ joined
07:59
zakharyas joined
09:08
virtualsue joined
10:28
virtualsue joined
10:51
donaldh joined
11:02
leont joined
12:18
kjs_ joined
12:22
brrt joined
|
|||
| brrt | quite possibly too much to ask, but can anybody recall why i wanted to put args with the tile | 12:31 | |
| if not, i'll read the logs and figure out my thought process myself | |||
| timotimo | i can't recall :( | 12:32 | |
| brrt | neither can i, but it was important, and it was a good solution, so i'm wondering for what problem | ||
| timotimo | mhm | 12:33 | |
| brrt | here it is: irclog.perlgeek.de/moarvm/2015-12-16#i_11725785 | 12:38 | |
| the point is that we can make pseudotiles that take arguments; those arguments are not stuffed into the tree, hence, we need some other space | 12:39 | ||
| timotimo | ah | ||
| pseudotiles is a thing i forgot about completely | |||
| brrt | we can therefore reuse the same space for both the pseudotiles and the regular tiles | ||
| timotimo | even though i argued for that solution at one point, iirc | ||
| brrt | i think you did yes | ||
|
13:08
brrt joined
14:18
kjs_ joined
14:25
kjs_ joined
14:39
brrt joined
|
|||
| brrt | ohhai #moarvm. i've just found another complexity-balloon | 14:39 | |
| in the jit | |||
| although, fortunately, it can be handled *after* tiler linearissation | 14:40 | ||
| i want to add offline register allocation, so i can do stuff like backpropagating to the desired register (e.g. when values have to live in RDI, RSI, RCX eetc because they'll be passed as C call arugments) | 14:41 | ||
| currently register locations are held in a per-node 'value' struct | |||
| however, the values computed in a node may change location as it is being moved about | |||
| so, if we attach the value struct to the tiles they are used in, they'll change underneath the tiles | 14:43 | ||
| so, they'll need to be copied | 14:44 | ||
| that makes things simpler again, but its yet another thing to take into account | |||
| it's funny how all these steps become individually simpler but the amount of steps increases perhaps even faster | 14:45 | ||
| [Coke] | just did a git pull --rebase in moarvm and got: | 14:50 | |
| Fetching submodule 3rdparty/libuv | |||
| fatal: unable to access 'github.com/joyent/libuv.git/': Could not resolve host: github.com | |||
| ... and I didn't read the error message, nevermind. | |||
| brrt | what :-o | 14:54 | |
| anyway, perl Configure.pl should do a git submodule sync and git submodule update | |||
| and that should get you the right url and checkout of libuv | |||
| geekosaur | ever used travis-ci? it fails builds because it can't pull from github all the time. no idea who's at fault where... | 14:56 | |
| brrt | haven't, no | 14:59 | |
| oh, great news, i've also just added a hash table and a linked list to the register allocator | 15:00 | ||
| timotimo | luckily we already have a library for hash tables in moar :) | 15:01 | |
| brrt | true enough | ||
| but maybe a simple table is simpler to write by hand | 15:02 | ||
| i.e. i can do lots of simplifications because i know the purpose :-) | |||
| [Coke] | brrt: the problem is that I'm behind a firewall with crippled DNS, and I didn't configure my shell properly beforehand. | 15:04 | |
|
15:06
FROGGS joined
|
|||
| brrt | right, i didn't know that | 15:08 | |
| fwiw, gist.github.com/bdw/33b977e07e7340f110b8 provides some hints on the tiler linearisation plan | 15:28 | ||
| (and by extension, on how the jit will proceed) | 15:29 | ||
| dalek | Heuristic branch merge: pushed 57 commits to MoarVM/even-moar-jit by bdw | 15:32 | |
| [Coke] | brrt: tilingare ? | 15:52 | |
| oh. much more obvious in this font what that is. | |||
|
17:40
domidumont joined
18:07
Peter__R joined
18:50
vendethiel joined
19:09
tadzik joined
19:13
kjs_ joined
20:14
zakharyas joined
21:08
Ven joined
22:17
kjs_ joined
|
|||