| diakopter | timotimo: there used to be an instruction that encoded that, but (ironically, perhaps?) it might be optimized away | 01:24 | |
|
01:48
ilbot3 joined
06:50
geekosaur joined,
domidumont joined
06:55
domidumont joined
07:33
zakharyas joined
|
|||
| jnthn | timotimo: On callee side, you can look at the callsite to validate it | 07:41 | |
| dalek | arVM: ca1a21a | jnthn++ | src/core/frame.c: Fix fragility in VMNulling work registers. It could potentially use the spesh num_local_types when not using the spesh local_types. |
08:15 | |
| jnthn | Unfortunately, this (well, actually the original patch that it makes more robust) accounts for a huge slowdown | 08:18 | |
| Well, maybe not huge, but 10s or so on the Rakudo build | 08:21 | ||
|
09:40
zakharyas joined
09:41
ingy1 joined
09:42
ilmari_ joined
10:29
Unavowed joined
10:34
domidumont joined
11:35
ggoebel113 joined
11:36
vendethiel joined
11:43
psch joined
12:23
lizmat joined
13:37
orbus_ joined
13:42
domidumont joined
|
|||
| timotimo | that's bad :| | 13:57 | |
| jnthn | Yeah, though we'll figure something out on it | 13:58 | |
| Today I've been digging into the long-awaited invocation/frames re-design. | |||
| timotimo | sounds exciting :) | ||
| jnthn | Yeah...I think I can save us a decent bit | 13:59 | |
| Along various axes | |||
| And make that refs_frames thing go away | |||
|
14:09
mtj_ joined
|
|||
| nwc10 | and a pony? | 14:15 | |
| timotimo | refs_pony = 1? | ||
|
14:18
zakharyas joined
|
|||
| jnthn | Well, it maybe prepares for the way for a pony ;) | 14:19 | |
|
14:31
domidumont joined
15:33
ilmari joined
15:35
zakharyas joined
15:41
ggoebel113 joined
15:47
geekosaur joined
15:50
geekosaur joined
16:05
domidumont joined
16:07
domidumont joined
16:30
zakharyas joined
16:49
pyrimidine joined
|
|||
| dalek | arVM/reframe: 9103b61 | jnthn++ | src/6model/6model.h: Add collectable flag for frames. |
16:53 | |
| arVM/reframe: 3daded3 | jnthn++ | src/core/frame.h: Give MVMFrame an MVMCollectable header. |
|||
| arVM/reframe: 95c87b2 | jnthn++ | src/gc/allocation. (2 files): Add function for GC-allocating a heap frame. |
|||
| arVM/reframe: 1ce7271 | jnthn++ | src/core/frame. (2 files): Pull frame destroy out of reference dec. |
|||
| arVM/reframe: d2591ad | jnthn++ | src/ (19 files): [WIP] Switch frames to be collectables. Danger: highly explosive branch; very unlikley to build NQP, though passes all but 2 of the NQP test suite. Pushed simply for backup. The eventual goal is to manage frames that are not referenced at all by other collectables in their life time using a traditional-style call stack, and promote them to the heap as normal GC-able objects upon escape. As a stepping stone, we'll allocate all frames on the heap, to help uncover missing mark bugs and the like, as well as to break the big change into steps between which stuff should entirely da21d2f | jnthn++ | src/ (3 files): Add some of the missing write barriers. |
17:15 | ||
| jnthn | dinner :) & | ||
| (btw, I've a list here of stuff to fix still on that branch, plus expect to rebase it further locally) | 17:16 | ||
| Also, I doubt this branch will merge until the stack part is done, as my guess would be that it'll represent a sizable regression. | 17:17 | ||
| (Performance wise) | |||
| Until that is in place. | |||
| Really dinner. & | |||
| timotimo | <3 | 17:18 | |
| have a good food! | |||
|
17:27
lizmat joined
|
|||
| nwc10 | jnthn: most recent commit on that branch SEGVs in NQP build (looks like a NULL pointer dereference) | 17:48 | |
| building previous commit to see how far I get | |||
|
18:05
zakharyas joined
|
|||
| jnthn | nwc10: It'd be a near micracle if it worked, given there's still missing write barriers to add and some places it's not marking frames at all 'cus it got away with it before | 18:39 | |
| (HEAD, that is) | 18:40 | ||
| The commits leading up to HEAD shoulda been benign refactors though | |||
|
18:40
leedo joined
|
|||
| nwc10 | ASAN "likes" the commit before HEAD | 18:42 | |
|
18:43
cognominal joined
19:07
patrickz joined
|
|||
| arVM/reframe: 17adaa4 | jnthn++ | src/ (3 files): Fix some missing frame marking. |
|||
| arVM/reframe: 6f61939 | jnthn++ | src/6model/ (44 files): Remove now-unused refs_frames. |
|||
| jnthn | Innerestin'. If you disable spesh it passes all the NQP tests now | 19:35 | |
| heh, and if you enable spesh but not inlining you get more fail | |||
| I've a horrible feeling this trail might lead to deopt.c | 19:37 | ||
| nwc10 | doesn't build Rakudo, though | 19:46 | |
| failed to load library 'dynext/libperl6_ops_moar.so' at gen/moar/m-BOOTSTRAP.nqp:1066 (<ephemeral file>:) | |||
| ... | |||
| however, that's the failure. not ASAN | |||
|
19:57
orbus_ joined,
arnsholt joined,
_longines joined,
jnthn joined,
khagan_ joined
20:04
_longines_ joined
20:09
orbus_ joined,
arnsholt joined,
jnthn joined,
khagan_ joined
20:17
jnthn joined,
arnsholt joined,
khagan joined
20:26
orbus_ joined
20:39
Util_ joined
20:40
arnsholt_ joined
20:44
harrow` joined
20:53
ingy joined
21:41
Util joined
22:11
diakopter joined
|
|||