|
Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes. Set by lizmat on 24 May 2021. |
|||
| Geth | MoarVM/main: 5b48ae1e53 | (Daniel Green)++ | src/6model/reprs/NFA.c Remove some unused code Looks like it was missed in 6941cada41a2ae2f32eb6ae179d41053468304f1. |
00:16 | |
|
00:17
MasterDuke joined
|
|||
| Geth | MoarVM/main: f2220d93d9 | (Daniel Green)++ | src/core/coerce.h Increase the size of the int-to-str cache to 65 This means we will cache 64, which was coerced ~42k times during a Rakudo build. A micro-benchmark (`my int $i = 64; my $s; $s = ~$i for ^10_000_000; say now - INIT now; say $s`) decreased from ~0.50s to ~0.22s. |
00:26 | |
|
00:44
japhb left
00:45
japhb joined
01:31
MasterDuke left
|
|||
| Woodi | s/64/65/ is funny :) make me thinking: do calculating in-memory tables (compile/build time) of every possible conversions for 1..100 (and maybe other things) would speedup something ;) | 08:54 | |
| but maybe it's just profiled compilation subset | 08:55 | ||
|
09:17
lizmat_ joined
10:19
lizmat__ joined
10:22
lizmat_ left
12:41
lizmat_ joined,
lizmat__ left
13:05
lizmat_ left,
lizmat_ joined
13:10
lizmat_ left
13:11
lizmat_ joined
13:51
lizmat_ left,
lizmat_ joined
14:08
lizmat_ left
17:42
MasterDuke joined
18:03
kjp left
|
|||
| MasterDuke | jnthn: since i've only had minimal success speeding up runAlt on the jvm backend, i started looking at the script that gets run during that`+++ Generating gen/<backend>/ast.nqp` stage | 18:15 | |
| even for the moarvm backend 90% of the time is spent in run_alt | 18:16 | ||
| parsing ~25k lines in 35 files in 12s | 18:17 | ||
| github.com/rakudo/rakudo/blob/main...qp#L71-L87 is where in the script it spends the most time | |||
| do you have any quick-and-easy suggestions to speed that (the script in its entirety or just that part) up? | 18:18 | ||
| nine | lizmat: binding attributes is not atomic after all. Since attribute objects are inlined into their parents when you bind to an attribute, the full object gets copied which leaves a window open for race conditions. | 19:32 | |
| lizmat | argh... | ||
| nine | Binding to locals or lexicals is atomic as that's just updating a pointer | 19:33 | |
|
22:13
kjp joined
|
|||