|
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. |
|||
|
00:00
reportable6 left
00:02
reportable6 joined
02:26
squashable6 left
02:28
squashable6 joined
|
|||
| MasterDuke | ugh, doing things with the CI pipelines always seems like such a slog | 04:51 | |
|
06:00
reportable6 left,
reportable6 joined
07:04
squashable6 left
07:06
squashable6 joined
07:32
samcv left,
samcv joined
08:38
japhb left
09:10
japhb joined
10:46
sena_kun joined
12:00
reportable6 left
12:03
reportable6 joined
14:39
epony left
|
|||
| timo1 | big sigh, thought i found an amazing optimization and the improved code barely runs | 16:23 | |
| i was hoping for thousands | 16:25 | ||
| 1.8k times during core setting compilation | 16:28 | ||
| 0 times during raku empty program compilation | 16:29 | ||
| lizmat | I'll take core setting compilation improvements any day :-) | 16:41 | |
| timo1 | uptime | 17:01 | |
| can you reliably measure improvements? | 17:05 | ||
| i found a few more calls to the optimized thing between the sofa cushions | |||
| now i got to 8460 times during core setting compilation | 17:06 | ||
| 13k | 17:14 | ||
| Geth | MoarVM/specialcase_genarate_dispatch_single_arg: 24a49c1f9a | (Timo Paulssen)++ | 2 files create single-argument versions of dispatch_* op generators this lets us avoid creating an nqp::list with just one argument in some spots in both nqp and rakudo, and these simpler versions end up being called about 13k times during core setting compilation |
17:20 | |
| nine | Does that make a measurable difference? | 17:21 | |
| timo1 | i am on a machine where measuring reliably is very much not easy | ||
| i've pushed a branch for all three repos, if lizmat wants to, or if you want to do some actual measurements, that would be lovely | 17:23 | ||
| i'm thinking this may cut one or two gc runs off of core setting compile, too? | |||
| i have to go afk now, so can't do my own measurements for this reason also :( | 17:24 | ||
| nine | jnthn: I may have hit a RakuAST design issue that I can't solve on my own for the first time. | 17:59 | |
|
18:00
reportable6 left
18:01
reportable6 joined
|
|||
| timo1 | nine: you think the difference should be negligible? | 18:08 | |
| i guess since it's in nqp code rather than raku code the for loop is a little cheaper, and a nqp::list is cheaper than a List or Array | |||
| Woodi | timo1: is it good ? other have ', $op0, str $op1,': dispatch_v1', sub ($frame, str $op0, $op1, $arg-index) { | 18:36 | |
| btw. in what C version is Moar written ? C99, C11 ? | 18:37 | ||
| nine | C99 | 18:38 | |
| Woodi | ah, k | 18:40 | |
| nine | MSVC is still the limiting factor. I'm glad we could at least move away from C89 | 18:42 | |
| Woodi | MS still develops MS C compiler or just C++ ? but C99 is nice, not magical | 18:46 | |
| and on what cpus Moar run ? x86, amd, power somethink other ? | 18:47 | ||
| s/amd/arm/ | 18:48 | ||
| btw. Gentoo moved to clang | |||
| nine | At least x86_64, i586, aarch64, armv7l, ppc64, ppc64le, s390x and riscv64. I have also run it on MIPS, but not recently | 18:53 | |
|
19:52
sena_kun left,
sena_kun joined
20:06
epony joined
|
|||
| japhb | That's actually better than I was expecting. | 22:24 | |
|
22:52
japhb left
22:58
japhb joined
23:09
Voldenet left
23:10
Voldenet joined
|
|||