|
00:09
colomon joined
05:08
TEttinger joined
08:47
rurban joined
10:23
brrt joined
|
|||
| brrt | \o | 10:24 | |
| nwc10 | o/ | ||
| brrt | ehm | 10:25 | |
| hmm | |||
| table generation looks like it'll take O(m*n^2) iterations | 10:26 | ||
| n will at least be 100, 200 | |||
| so that... sucks, i think | |||
| but it's probably worth it to do it at preprocessing time rather than at runtime | 10:27 | ||
| maybe any of you know of a cheap way to represent a sparse 3-dimensional array | |||
| or a cheap way to test filter impossible combinations | 10:28 | ||
| (e.g. (add reg (const)) can be combined with (add reg reg) but not with (add reg (load mem)) | |||
| jnthn | brrt: Tree? | 10:29 | |
| brrt: Bloom filter? | |||
| brrt | i'm unfamiliar with a bloom filter | ||
| jnthn | (for the "represent" and "filter" respectively) | ||
| Oh, they're cool! | 10:30 | ||
| brrt | i thought about a tree, yes | ||
| jnthn | en.wikipedia.org/wiki/Bloom_filter | ||
| brrt | but i figure the key has to be at least three words long, and repeated | ||
| or the value should be the marker for another tree | |||
| let's read about it :-) (the bloom filter) | |||
| ok, i agree, that's pretty cool | 10:33 | ||
| not sure if necessarily what i want, but cool :-) | 10:34 | ||
| i think the problem to: i can combine rule a and rule b, if for all children c identified by i, a[i] can reduce to b[i] or b[i] can reduce to a[i] | 10:35 | ||
| e.g. (const) can reduce to reg, as can (load mem), but (const) cannot be reduced to (load mem) or the other way arround | 10:36 | ||
| brrt bbl | 10:37 | ||
| thanks for that link, though :-) | 10:38 | ||
|
10:52
lizmat joined
11:00
bonsaikitten joined,
hoelzro_ joined
11:11
timotimo joined
11:12
retupmoca joined
12:22
vendethiel joined
|
|||
| nebuchadnezzar | hello | 12:46 | |
| lizmat | nebuchadnezzar o/ | ||
| nebuchadnezzar | I'm preparing the 2015.07 Debian package which permit to enable mipsel arch using libffi (and possibly others too) | 12:47 | |
| lizmat | ++nebuchadnezzar | ||
| nebuchadnezzar | Is there any reason not to use libffi for all arch or should I limit the libffi case to thoses dyncall does not support? | 12:48 | |
| lizmat | nebuchadnezzar: jnthn FROGGS moritz brrt might know | 12:50 | |
| jnthn | I'd do the latter for now. | ||
| As dyncall is what most folks are working against, and so that codepath is much better tested | |||
| nebuchadnezzar | ok, I'll do that | 12:59 | |
| thanks | |||
| moritz | lizmat: fwiw I don't know all that much about MoarVM; I just hang out here :-) | 13:01 | |
| lizmat | moritz: but you have some dyncall / libffi knowledge ? | 13:16 | |
| moritz | lizmat: negative | 13:22 | |
| lizmat | moritz++ # nonetheless :-) | 13:23 | |
| moritz | :-) | 13:24 | |
| jnthn is glad moritz++ hangs out here :) | 13:26 | ||
|
13:53
brrt joined,
FROGGS__ joined
14:02
brrt left
14:57
rurban joined
15:13
ggoebel joined
15:55
TEttinger joined
17:08
zakharyas joined
|
|||
| timotimo is looking at the "suddenly decont" issue again a little bit | 17:13 | ||
| oh, look, there's a substr + eq in the compile_mastop method that can become an eqat instead | 17:16 | ||
| anyway. i put a debug print into the "if @deconts[$arg_num]" part of that method and it doesn't seem to trigger for the rpoblematic case i'm looking at | 17:17 | ||
| oh, aha, getting somewhere now | 17:21 | ||
| masak | timotimo++ | 17:49 | |
| timotimo | i'm not 100% sure how exactly code should look that references a local "into" a localref | 18:08 | |
|
18:08
brrt joined
|
|||
| timotimo | currently using a bind onto a localref will complain | 18:08 | |
| but maybe that's exactly what we want? | 18:09 | ||
| brrt | aw, yeah | 18:38 | |
| brrt has tiling working | |||
| the only trick is efficient translation to a C header file | 18:39 | ||
| timotimo | ooooh | ||
| tiling works? | |||
| oh, you probably mean just the preprocessor? | 18:40 | ||
| brrt | well, the preprocessor, but the tiling algorithm, too :-) | ||
| just implemented in perl rather than c | |||
| timotimo | ah | 18:42 | |
| neato :) | |||
| we're getting close to actual assembly code getting spit out? | |||
| it's food time locally | 18:44 | ||
| brrt | hmmm... yeah, conceptually a lot closer, at least | ||
| timotimo | buy maybe after that i'll look a bit more into the graph.c parsing thing | ||
| brrt | don't worry about it too much | 18:45 | |
| :-) | |||
| \o/ | 19:00 | ||
| dalek | arVM/even-moar-jit: 8be7d2a | brrt++ | / (3 files): Generate tiler tables in perl Add a testcase to check the suitability for tiling actual trees. The next step is the to represent this in some static data structure suitable for search in the VM. |
||
| brrt | this was *such* a brain killer | ||
| i'm wondering how it'll perform for a bigger grammar | 19:02 | ||
| not perfectly, but i suspect that's due to grammar inconsistency | 19:09 | ||
| dalek | arVM/even-moar-jit: 8761634 | brrt++ | tools/tiler-table-generator.pl: Ambiguous test case How do we deal with ambiguity? Well enough, it seems. |
19:18 | |
| brrt | ooh... it is dramatically slow though | 19:20 | |
| hmmm | 19:22 | ||
| well... | |||
| we'll see what we do about that | |||
|
19:23
zakharyas joined
|
|||
| brrt | worst case we can commit the stored preprocessed tiles | 19:24 | |
| and distribute them that way | |||
| brrt afk for now | 19:27 | ||
| or, wait afk in a few minutes | |||
|
19:32
brrt left
20:27
rurban joined
|
|||
| jnthn | brrt++ # JIT progress | 21:34 | |
| timotimo: iirc binding to localref is needed; I think we find to lexicalref to handle "is rw" native parameters | |||
| uh, s/find/bind/ | 21:38 | ||
| Apparently I made good use of The Bishop's Arms near my hotel :) | |||
| timotimo | ok 4 - a localref'd var can have a local ref'd thing bound to it and accessed (str) | 23:12 | |
| so the code had been right all along, but the test was b0rked? | |||
| i feel dumb :) | 23:13 | ||
| especially dumb for bothering jnthn about it so often | 23:15 | ||
|
23:21
harrow joined
|
|||