|
š¦ Welcome to the IRC channel of the core developers of the Raku Programming Language (raku.org #rakulang). This channel is logged for the purpose of history keeping about its development | evalbot usage: 'm: say 3;' or /msg camelia m: ... | Logs available at irclogs.raku.org/raku-dev/live.html | For MoarVM see #moarvm Set by lizmat on 8 June 2022. |
|||
|
02:22
japhb left
02:41
japhb joined
02:45
japhb left
02:46
japhb joined
03:20
japhb left
03:34
japhb joined
08:59
sena_kun joined
|
|||
| Voldenet | MasterDuke: I couldn't apply exactly the same patch directly, because of fastutil collections used, but 0x0.st/H7PR.diff works - with this memory goes up to 2.1GB and is painfully slow (10+ minutes so far, no end in sight) | 09:38 | |
| I'm hardly believing what I see so I'll rerun main branch in a while, just to make sure the memory usage used to be lower | 09:40 | ||
| So, I've re-tested things on main branch and the memory usage isn't higher at all, the difference was amount of memory used by VM | 10:52 | ||
| both runs were done with AggressiveHeap, but first raku-ast-compiler ran with lower memory (which wasn't enough for v6c) | 10:53 | ||
| since I got max value for fastutil branch (RSS 2195MB) I'll compare with maximum from main branch now | 10:59 | ||
| AggressiveHeap will make comparisons a bit more difficult, but the resoult should be meaningful either way | 11:03 | ||
| though honestly, fastutil branch is many times faster | 11:04 | ||
| I confirm that when using the same amount of memory on VM, the memory usage is about the same | 12:15 | ||
| except it's a few times slower | |||
| on main | 12:16 | ||
| lizmat | which one is a few times slower ? | ||
| Voldenet | main, I could do exact timings by modifying nqp-j | 12:18 | |
| rough estimate is fastutil being 3x faster | 12:21 | ||
| at least | |||
| reading these plots isn't really meaningful way of measuring anything either way | 12:25 | ||
|
12:54
sena_kun left
|
|||
| MasterDuke | Voldenet: when you say "VM", do you mean the jvm or the vm in which you're running these experiments? so the rough conclusion is that memory used while building nqp (and/or rakudo?) on main and the fastutil branch is the same, but fastutil branch is faster? | 13:26 | |
| Geth | rakudo/main: bef608a8ff | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Move Binder methods closer to where they operate on Also move variables not needed on the JVM into the "not compiled on JVM" area |
13:27 | |
| rakudo/main: 4b124a9852 | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Streamline container descriptors Mostly esthetics, with ContainerDescriptor.new now bypassing the BUILDPLAN altogether |
|||
| Voldenet | By saying VM i mean the VM I'm running experiments in and yes, that conclusion is correct | 13:28 | |
| MasterDuke | just successfully built rakudo with the jvm backend at 2fa69e878 | ||
| linkable6 | (2024-02-17) github.com/rakudo/rakudo/commit/2fa69e8789 Bump NQP to get latest compose check tweaks | ||
| MasterDuke | Voldenet: cool, thanks | 13:29 | |
| i had hoped it would also reduce memory usage, but i'll take just faster | 13:32 | ||
| Voldenet: oh, and i didn't quite follow what you tested re that one change and its effect. main, main + that change, and fastutil (which includes that change)? | 13:35 | ||
| Voldenet | Yes: main 1h+, with change: 23m+, with fastutil: 8m+ | 13:36 | |
| MasterDuke | nice! | ||
| Voldenet | Btw, the problem with this memory measurement is that RSS is only the total jvm-allocated size | ||
| I could probably do some inspection on `jcmd pid GC.heap_info` | |||
| iirc it forces GC as well | 13:37 | ||
| MasterDuke | yeah, might need to look at a profile to see if we can reduce the size of the heap we configure to run with | ||
| Voldenet | or GC.class_histogram, but there's entries like those | 13:38 | |
| > 1: 29383 3055832 __P6opaque__28 | |||
| > num #instances #bytes class name (module) | 13:39 | ||
| MasterDuke | i'm also going to try to reduce the size of the fastutil jar. they have some script to find what classes you actually use and create a minimized jar, which hopefully will be quite a bit smaller than even the "fastutil-core" jar it's using now | 13:43 | |
| also we're using jna-platform 4.5.0, but they're now at 5.14.0, might be some improvements there | 13:47 | ||
| also we're using asm 4.1, butĀ they're now at 9.6 | 13:53 | ||
| afk for a while, hopefully will get to play around some more this afternoon or evening | 13:56 | ||
| Geth | rakudo/main: 1491f82228 | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Streamline Binder.trial_bind |
14:13 | |
|
18:17
sena_kun joined
|
|||
| Geth | rakudo/main: 8e863c81f5 | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Two additional micro-opts in Bindner.trial_bind |
20:39 | |
| rakudo/main: 7dee053075 | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Streamline creation of several classes in bootstrap Mostly esthetic: - Mu - Any - Cool - Attribute - Scalar |
|||
| rakudo/main: 375393fa50 | (Elizabeth Mattijsen)++ | src/Perl6/bootstrap.c/BOOTSTRAP.nqp Streamline ScalarVAR and xxxRef classes |
20:49 | ||
|
22:40
vrurg_ is now known as vrurg
23:02
sena_kun left
23:04
sena_kun joined
23:06
sena_kun left
|
|||