| timotimo | lizmat++ | 00:01 | |
|
00:27
dogbert11 joined
01:36
coverable6 joined,
bloatable6 joined,
committable6 joined,
evalable6 joined,
quotable6 joined,
bisectable6 joined,
greppable6 joined,
benchable6 joined,
unicodable6 joined,
statisfiable6 joined
01:52
ilbot3 joined
02:19
brimonk joined
02:25
ingy joined
|
|||
| MasterDuke | .tell jnthn aliased named parameters broke with a recent MoarVM commit rt.perl.org/rt3/Public/Bug/Display...?id=131857 | 02:51 | |
| yoleaux | MasterDuke: I'll pass your message to jnthn. | ||
|
03:33
pharv joined
04:14
pharv joined
05:21
pharv_ joined
06:19
brrt joined
|
|||
| brrt | lizmat++ weekly | 06:20 | |
|
06:32
lizmat joined
07:02
lizmat joined
07:11
pharv_ joined
07:23
edehont joined
07:42
zakharyas joined
07:58
robertle joined
08:42
edehont joined
08:47
Ven joined
|
|||
| jnthn | moarning o/ | 08:52 | |
| yoleaux | 02:51Z <MasterDuke> jnthn: aliased named parameters broke with a recent MoarVM commit rt.perl.org/rt3/Public/Bug/Display...?id=131857 | ||
| jnthn | Well, then they shoulda been better spectested... | 08:53 | |
|
08:58
Ven_ joined
09:03
pharv_ joined
|
|||
| Geth | MoarVM: 7e841351cd | (Jonathan Worthington)++ | 3 files Clean up sim stack when thread context is freed. Fixes a (small) leak introduced when we started keeping these around. |
09:12 | |
|
09:20
brrt joined
09:27
travis-ci joined
|
|||
| travis-ci | MoarVM build passed. Jonathan Worthington 'Clean up sim stack when thread context is freed. | 09:27 | |
| travis-ci.org/MoarVM/MoarVM/builds/262159134 github.com/MoarVM/MoarVM/compare/1...841351cd57 | |||
|
09:27
travis-ci left
09:55
zakharyas joined
10:27
Ven joined
|
|||
| brrt | i hae to say, the distinction between the range operator and the sequence operator is subtle | 10:46 | |
| jnthn | . | ||
| Zoffix | .oO( Subtle is the sequence operator. Malicious it's not ) |
10:53 | |
| jnthn | I'm still trying to work out if the missing letter brrt's comment is a t or a v :P | ||
| Zoffix | :D | 10:54 | |
| jnthn | *in :P | ||
| brrt | which comment? | 10:55 | |
| oh | |||
| haha | |||
| *have | |||
| Zoffix | brrt: what sort of distinction? The way they look? | ||
| brrt | yeah, the way that, say | 10:57 | |
| my @a = 1, 2, 4 .. * is very different from my @b = 1, 2, 4 … *; | |||
| Zoffix | :) | 10:59 | |
|
11:06
lizmat joined
11:31
edehont joined
11:36
colomon joined
11:40
Ven joined
12:55
lizmat joined
13:07
edehont joined
|
|||
| Geth | MoarVM: 98ff03964d | (Jonathan Worthington)++ | src/spesh/graph.c Further harden against removed handlers. |
13:17 | |
|
13:23
eater joined
|
|||
| Geth | MoarVM: a5a0e82c0d | (Jonathan Worthington)++ | 6 files Add spesh ops for specualtive call optimization. |
13:25 | |
| MoarVM: a66c580994 | (Jonathan Worthington)++ | src/spesh/optimize.c Factor out offset resolution to avoid dupe. |
|||
| MoarVM/even-moar-jit: 70c0afae64 | (Bart Wiegmans)++ | 5 files Extract bitmap data structure from register allocator A bitmap is generally useful. I've also replaced various somewhat obscure bitmap-manipulation expressions with the more obviously correct get-and-set calls. |
13:47 | ||
|
13:47
lizmat joined
|
|||
| jnthn has speculative call opt working-ish | 14:04 | ||
| It causes a couple of bizzare failures | 14:05 | ||
|
14:16
lizmat joined
|
|||
| brrt | jnthn: re the proliferation of special-jitcode-branches | 14:20 | |
| this is true. arguably, we might also consider 'fully overwriting' the spesh cands' own data | 14:21 | ||
| in essence, dropping the created speshed non-jitted code | |||
| jnthn | Yeah, we could certainly consider that | 14:22 | |
| brrt | (in which case, why create it at all? but i recall that in the past being more difficult than not, and it might be wasteful) | ||
| timotimo | we can skip the spesh code gen stage and directly feed the jit? | 14:23 | |
| brrt | that would be ideal, but i recall the spesh code gen stage does some essential setting-up | ||
| timotimo | mhm | 14:24 | |
| dogbert17 | jnthn: do you get test failures in t/spec/S32-io/IO-Socket-INET.t | 14:25 | |
| dogbert17 or is a recent rakudo commit to blame ... | 14:26 | ||
| jnthn | No, that's clean for me even under spesh stressing | ||
| dogbert17 | then rakudo might be the guilty party | 14:28 | |
| jnthn | Joy. An optimization works in the first 14,283 places it's performed. Then the 14,284th time something goes rotten. | 14:32 | |
| timotimo | cool, that can be our magic number for when to stop doing teh opt :) | 14:33 | |
|
14:41
lizmat joined
|
|||
| brrt | (what is different the 14284th time?) | 14:42 | |
| jnthn | That's what I'm trying to work out | 14:44 | |
| I fear it's memory corruption of some kind, though valgrind makes it work rather than complaining... | 15:21 | ||
| timotimo | hate when it does that | ||
| jnthn | Yeah, I'm getting non-deterministic behavior | 15:25 | |
| It works sometimes, not others | 15:26 | ||
| timotimo | i wonder if i should add more redzone support all over moarvm for the --valgrind flag | ||
|
15:27
Ven joined
15:46
brrt joined
|
|||
| jnthn | oh wow | 16:05 | |
| Well this was a wild goose chase | |||
| So it looks like the mere presence of a new op (resolvecode) caused the failure | |||
| Said new op prevents JIT compilation 'cus I don't JIT it yet | 16:06 | ||
| So what if I just don't insert it and then disable JIT? | |||
| Same sorts of weird failure | |||
| timotimo | ooooh that's really strange | 16:07 | |
| jnthn | So somewhere we get something wrong in the interp that the JIT gets right?! | ||
| timotimo | i know we moved some things from interp.c into functions so the jit could use them | 16:08 | |
| but that's the opposite direction | |||
| jnthn | Yeah | ||
| timotimo | hm, jit bisect doesn't help here, right? | 16:09 | |
| jnthn | Don't think so | 16:10 | |
| I mean, it's not a JI | |||
| JIT bug | |||
| It's an interp bug | |||
| Question is if it's recently introduced | |||
| timotimo | right, we need the opposite of "go until n and then stop" | 16:11 | |
| Geth | MoarVM: 54ee129e4f | (Jonathan Worthington)++ | src/6model/reprs/MVMMultiCache.c Fix potential read over end of args buffer. |
16:12 | |
| MoarVM: 576284f130 | (Jonathan Worthington)++ | src/6model/reprs/MVMStaticFrame.c Avoid passing NULL to memcpy. |
|||
| MoarVM: 3bb694c67f | (Jonathan Worthington)++ | src/spesh/dump.c Handle missing type in spesh dump. |
|||
| jnthn | Aww, no, it's not my local changes | 16:13 | |
| Geth | MoarVM/speculative-calls: 6b3aee3fad | (Jonathan Worthington)++ | 3 files [WIP] Speculative calls |
16:15 | |
|
16:16
zakharyas joined
|
|||
| jnthn | And wasn't anything I did today | 16:18 | |
|
16:19
dogbert11 joined
|
|||
| jnthn | Or yesterday | 16:19 | |
| So this goes back quite a while | |||
| I should go and work on dinner but | 16:20 | ||
| t/spec/S09-typed-arrays/native-num.rakudo.moar | |||
| Run it with MVM_JIT_DISABLE=1 | |||
| To see the explosion | |||
| oh and | 16:21 | ||
| also stressing | |||
| plus turn off inline just to rule it out | |||
| MVM_JIT_DISABLE=1 MVM_SPESH_INLINE_DISABLE=1 MVM_SPESH_BLOCKING=1 MVM_SPESH_NODELAY=1 ./perl6-m -Ilib t/spec/S09-typed-arrays/native-num.rakudo.moar | |||
| Is the complete failing command I'm seeing | |||
| Which works if MVM_JIT_DISABLE=1 is removed | |||
| ok, 388a769b461c10d7d4ee901409288f0122a6d2a4 is good even with JIT | 16:27 | ||
| So it's recent | |||
| Will bissect tomorrow | |||
|
16:57
colomon joined
17:55
Ven joined
18:32
Ven_ joined
18:54
Ven joined
19:58
zakharyas joined
20:01
Ven joined
21:24
lizmat joined
21:34
colomon joined
21:38
markmont joined
21:42
MasterDuke joined
22:45
geekosaur joined
23:04
dogbert17_ joined
|
|||