programming

[tender river]

reg(const T, r)

this should work already

but that doesnt really solve the core issue

bwa

_Generic

surely its the correct solution to all problems in lie

[sage crag]

while im here may as well then

#define reg_r(T, r) (const T *)(regs + r)
#define reg_w(T, r) (T restrict *)(regs + r)

maybe will make a few things take less cycles

[tender river]

cant hurt (assuming reusing the same register in the same opcode is undefined that is)

[sage crag]

it is undefined

[tender river]

yippie

[sage crag]

awa

[tender river]

ub

[sage crag]

erf, to_le in pc_read is kinda useless

it only needs to be done on writes

[tender river]

i wonder how many codebases out there have support for big endian but never actually test it and its secretly very broken

[sage crag]

yes

[sage crag]

#define reg_r(T, r) to_le(*(const T *restrict)(regs + r))
#define reg_w(T, r, v) *(T *restrict)(regs + r) = to_le(v)

#define pc_r(T) to_le(*((const T *restrict)(pc += sizeof(T)) - 1))
#define pc_w(T, v) *(T *restrict)(pc) = to_le(v)

should be sufficient

in principle to_le just swaps the byte order

[tender river]

if its not, its all macros anyway so it should be easy to fix

[sage crag]

so doing to_le on a le value will convert it to be

(if you are on be)

[sage crag]

i guess i should also do to_le on the opcode for future extension support huh

[tender river]

at that point computed goto may not be a good idea anymore

[sage crag]

i suppose so

[tender river]

its fine as long as theres no gaps

[sage crag]

you can still do sparse isas with computed goto

[midnight sigil]

deliving

[sage crag]

i was thinking more of the instruction decode overhead, if instructions are variable width

[tender river]

you can but you'll have to do multiple tables

something thats easier done automatically with a switch

[tender river]

surely no var width instructions in hbvm

[opaque wharf]

You have 2 path

Make the spec support it

[sage crag]

i dont think the vm needs it

[opaque wharf]

Or add it later if it turns out it is needed

[sage crag]

we have only used 119 instructions so far

[tender river]

either way definitely dont worry about it for now

wa

[sage crag]

yes

[nocturne olive]

Now running a stress-test for RAM at 3800 CL18 and CPU at 5.1GHz

[sage crag]

i forgot why i was doing this

its so i can make writing to $0 a no-op

to_le(v & -(r != 0))

to move that off the hot path

[tender river]

[nocturne olive]

It's definitely running at least
6 minutes without errors already
With GPU fully loaded the CPU hits 100 C though

[sage crag]

this is literally the hottest-path

so its bad if i zero here

[tender river]

[opaque sigil]

Does this just unconditionally write 0 to the zero register or am I misunderstanding

[nocturne olive]

At least two cores consistently hitting 5.1

[tender river]

not anymore but thats the old code

[sage crag]

yes it did before

now ive changed it so that writes are masked

[tender river]

i present to you

string interning

[sage crag]

no

sym with len > 9 unsupported

sory

[opaque sigil]

wth is string interning

[tender river]

its when you take a string and another same string and make them the same

✅

[opaque sigil]

Hmmm

[tender river]

(interning is for turning structural equality into pointer equality)

[sage crag]

intern(string1) -> string1
intern(string2) -> string2
intern(same contents as string1 but different ptr) -> string1

[opaque sigil]

Interesting

[tender river]

i will not do arbitrary interning

[sage crag]

its very useful for things like interpreters where you do a lot of string equalities

[tender river]

because functions as first class values are hard to intern

[sage crag]

if you intern every string you no longer have to do memcmps

[tender river]

but i'll intern the simple stuff

[floral dirge]

[tender river]

eee

[opaque wharf]

Looks like we need to restart chay

[tender river]

nope

[opaque wharf]

Do you need more RAM?

[tender river]

i am functional

and lazy

[opaque wharf]

Classic functional programming

[tender river]

and not used by a single company

[sage crag]

ablecorp overtime hours

[tender river]

unpaid internship is one thing

i recently saw some company advertising their internship as free of charge

wow so generous

[sage crag]

om

abusing macros is so fun

you can make a cursed dsl inside c

for the low low cost of your sanity

i'

d

argue that this is easier to read though

[tender river]

it is

[sage crag]

the resulting vm boilerplate is easier to read too

each one of these is a label

[nocturne olive]

I think 5.1GHz on CPU + 3800 CL18 RAM is stable, but my cooler struggles to keep up with the CPU

Maybe if I got the voltage down just a tiny bit

[tender river]

its a miracle none of what you're doing required #s

(unless...)

[sage crag]

rrr_cmp(cmpu, u64) expands to

cmpu : {
  auto r0 = *((const u8 *restrict)(pc += sizeof(u8)) - 1);
  auto r1 = *((const u8 *restrict)(pc += sizeof(u8)) - 1);
  auto r2 = *((const u8 *restrict)(pc += sizeof(u8)) - 1);
  auto v1 = *(const u64 *restrict)(regs + r1);
  auto v2 = *(const u64 *restrict)(regs + r2);
  *(i64 *restrict)(regs + r0) = ((v1 > v2) - (v1 < v2)) & -(r0 != 0);
  goto ret;
  __builtin_unreachable();
}

so its somewhat deceptive i guess

im wondering if its bad for branch prediction if i make cmpu and cmps function goto a cmp label

it would reduce code size but it would become nonlinear

[tender river]

code size is in most cases not an issue

[sage crag]

i think that __builtin_unreachable is not useful

unless it lets the compiler make more assumptions? like i can never return to the caller of goto ret?

no backsies

[tender river]

you can never return there in the first place

goto isnt call/cc

its just goto

[sage crag]

yesh i know

im just wondering if the compiler knows that

there's a lot of control flow

[tender river]

compiler is not that stupid

[sage crag]

well, i will test it with and without once i have a working vm again

[tender river]

[sage crag]

im also going to finish this and see that its infinitely slower than the original vm

that's ok

i have it backed up

[nocturne olive]

The fact I can really feel the difference with the overclocked system is quite insane to me
The system just feels generally faster

[desert plaza]

isn't interning like copy-on-write?

[sage crag]

placebo

[nocturne olive]

I mean the boot time was clearly faster

[desert plaza]

unless they're constant

[nocturne olive]

Significantly even

[tender river]

its not an apt comparison

[crystal sleet]

Windows?

[nocturne olive]

Linux

[tender river]

interning is separate from whether something is immutable

[sage crag]

they are both string optimisations but not similar optimisations

[tender river]

(though its true that interning doesnt generally work well with mutable data)

[unreal coral]

Guys this might be a wild idea buuuuuuut

[nocturne olive]

Now trying CPU 5.1GHz with -10mV

[desert plaza]

interning is just avoiding duplicate data right?

[tender river]

im breakiigngg

[nocturne olive]

Welp, it's not stable

[unreal coral]

What about a neuro app? :0
Something that could work during her streams and shows stuff like filtered texts etc
Or gives some extra functions :0

I mean twitch tos only count on twitch right?

[sage crag]

partially, but the main reason to use it is actually for fast comparisons

once you intern some strings, they can all be compared via pointer comparison

but the strings become readonly

[crystal sleet]

I don't think vedal wants those holy words seen, pr disaster

[tender river]

interning turns structurally same values into referentially same values. this does not necessarily help save space, in fact in many cases it may lose space

[unreal coral]

Isnt the filter just for tos tho?

[sage crag]

because literal strings are stored in binaries awawawaw

[unreal coral]

Or does she reveal info too?

[crystal sleet]

just daily "Neuro said something bad!?" and not like people are gonna know that it's a separate app

[tender river]

its just used for referential equality really, like hashing but you dont have to rehash every time and dont have to check for hash collisions

[unreal coral]

Eh its free cloud

[sage crag]

it could be considered a hash

[desert plaza]

how does it not save space if you avoid duplicating strings

[sage crag]

f(ptr) = ptr

[tender river]

they were duplicated in the first place

you arent getting rid of the originals

they may or may not be there

[sage crag]

because you maintain a table of interned strings, which you have to copy strings into

[tender river]

furthermore

you have to build a cache for interning

you wouldnt call hashing a space optimization, but you could use hashes for space optimization in some cases

same goes for interning

[midnight sigil]

avoidance is avoidance

twitch pays him money

he's a big guy on twitch and if he did that, it'll affect twitch's image

and what'll twitch do

ban vedal

[crystal sleet]

Just not worth it at all to ruin neuro's image by showing what's under the filter

[desert plaza]

alr what does a intern table look like

[tender river]

depends

[midnight sigil]

that too yea

[tender river]

it can be a list, a hashmap, a binary tree, anything really

[midnight sigil]

"Filtered" is a selling point of Neuro I'd say

[unreal coral]

Bruh so corporations and ez to offend people ruined it for us?

[tender river]

anything that allows you to check whether you have already seen structurally equal values

[desert plaza]

and what does an entry store

pointer, string

[crystal sleet]

It's always optics, it has always been optics

[opaque wharf]

The last time AI runs without a filter, we have MS. Tay 2 electric boogaloo

[desert plaza]

objects are more complex

[tender river]

the value itself that you can check for structural equality against the value you're interning

[midnight sigil]

wym bro avoidance is avoidance

[opaque wharf]

AI with human input

[desert plaza]

right, and then it doesn't create a new entry when the values are equal, right?

[midnight sigil]

if you comit crime in philipine but you're american, they will send you back to america, and you will get jailed

[tender river]

"ruined" wow haha no slurs really ruined the experience go back to twitter

[tender river]

it may or may not

[crystal sleet]

just a pointer

[desert plaza]

what

[midnight sigil]

it's not like you comit crime in other places and you won't be punished

[tender river]

interning is a general technique not some particular implementation

you can implement it in many ways

[desert plaza]

but why would an implementation ever need to create multiple entries for the same data

[tender river]

for example, if the same structure is going to be interned multiple times it may be useful to store its pointer in a hashmap to avoid full structural equality from now on

[crystal sleet]

some need to be garbage collected for example, they need to store their own data

[tender river]

its just tradeoffs

[tender river]

no. gcs have nothing to do with that, multiple entries != overhead for a single entry

[desert plaza]

what is the purpose of interning

[tender river]

turning structural equality into referential equality

[desert plaza]

so compare pointers instead of contents

[tender river]

yup

[crystal sleet]

mhm understood

[nocturne olive]

That's a lot of 5.1

[desert plaza]

then why this

[tender river]

source, not destination

again, it doesnt really matter how interning is implemented because there are many ways to implement it

[crystal sleet]

It's for what gets returned when you ask for the interned version

[tender river]

its like asking how gc is implemented

you can say "mark and sweep" which would be an implementation

or you just say "it depends" which is the correct answer

[tender river]

???

[sage crag]

https://en.wikipedia.org/wiki/String_interning

i invite you to

[nocturne olive]

So I have determined my RAM will not be stable at CL17, unfortunate, but CL18 is still pretty decent for 3800 MT/s
My CPU though, I got it to basically sustain all-core 5.1GHz by taking off 10mV of voltage

Turns out reducing voltage can make clocks higher

[crystal sleet]

Interning only guarantees that when you request the interned version, you get the value reference
Multiple copies might exist elsewhere, but anything newly interned or anything accessed via the interned interface will always point to the same value, hence my reply

[nocturne olive]

I'm gonna try to take off a few more mV

[tender river]

[sage crag]

bwabbwabwa i love it when i google elbow pain and then it just throws this at you

thanks my anxiety is really reduced

[crystal sleet]

Am I talking too abstractly? I might be, haven't slept for like 22 hours

[sage crag]

if i have a panic attack because i think im having a heart attack then im suing the nhs

in reality this is because i worked my ablecorp overtime for 30 hours

mhm

[desert plaza]

sounds like yall need sleep

[tender river]

konii is fine

i am allegedly fine too

[sage crag]

its part of the nda

we are fine

[desert plaza]

https://tenor.com/view/hmm-suspect-gif-22611582

mhm

[opaque wharf]

In general, swelling + fever = bad

Swelling without fever = probably not much

Swelling > 1 week without any sign of it getting better = may want to check it

[sage crag]

i think its just lack-of-osu withdrawal

combined with not sleeping

[nocturne olive]

This is not bad
I now have -20mV on CPU and with a full system load my CPU now sustaints 5.1GHz all-core on the P-cores and stays around 100 C

[opaque wharf]

At least I know my shaking can come from either too much caffeine, lack of sleep, or lack of food

[rigid snow]

lack of food AND caffeine? it’s that

[opaque wharf]

Yeah that can cause shaky hands

[nocturne olive]

Now running -30mV
Sustaining 5.1GHz around 95 C

[sage crag]

left expands to

swa : {
  auto r0 = *((const u8 *restrict)(pc += sizeof(u8)) - 1),
       r1 = *((const u8 *restrict)(pc += sizeof(u8)) - 1);
  auto tmp = *(const u64 *restrict)(regs + r0);
  *(u64 *restrict)(regs + r0) =
      (*(const u64 *restrict)(regs + r1)) & -(r0 != 0);
  *(u64 *restrict)(regs + r1) = (tmp) & -(r1 != 0);
  goto ret;
  __builtin_unreachable();
}

right expands to

swa : {
  return ({
    auto r0 = ({
      auto _v = *(u8 *)(pc);
      pc += sizeof(u8);
      to_le(_v);
    }),
         r1 = ({
           auto _v = *(u8 *)(pc);
           pc += sizeof(u8);
           to_le(_v);
         });
    auto tmp = *(u64 *)(regs + r0);
    *(u64 *)(regs + r0) = *(u64 *)(regs + r1);
    *(u64 *)(regs + r1) = tmp;
    vm->pc = pc;
    vm_ok;
  });
}

i think left is an improvement

[tender river]

should be

[opaque wharf]

I was concerned that you're running the CPU in reverse polarity lmao

[nocturne olive]

Silly

Now I'm down to -50mV

Let's see how that does

I think it's stable
I think I'll stop here for now and be satisfied with -50mV CPU

[hexed grove]

uh so

i have no swap

and i have 16 megs of ram

[opaque wharf]

16 MB?

[hexed grove]

yes

[opaque wharf]

Ok

[hexed grove]

i need to decode an hls stream

[hollow spruce]

Snapdragon8 Elite 2 has 5.ghz leaks

[hexed grove]

and i dont like hls

[opaque wharf]

Well, why do you need to do that then?

[hexed grove]

so i can watch twitch

[opaque wharf]

You're making a twitch client for an embedded device?

[nocturne olive]

Silly
I just got my CPU to sustain 5.1GHz, so mine is faster

-# Real world performance may vary due to IPC

[hollow spruce]

Yep

While I was watching this.
Datasets & DataLoaders — PyTorch Tutorials 2.8.0+cu128 documentation https://docs.pytorch.org/tutorials/beginner/basics/data_tutorial.html

[nocturne olive]

I guess my CPU has somewhat decent silicon, it runs with -50mV at 5.1GHz all-core

I guess I'll settle with this for now, 5.1GHz all-core + 3800 CL18 RAM at -50mV CPU

[hollow spruce]

Try on google ai mode
import torch
torch.rand(7,1)

[tender river]

do you just live with comments like this

(talking about tabs)

[opaque wharf]

[midnight sigil]

double commenting

ensure they are really really commented

[tender river]

[midnight sigil]

why is there no good TODO highlights in Visual Studio 2022 man

two top result, one doesn't work and one that look bad as fuck

[opaque wharf]

Simple, don't write todo as comment. Use what konii sent (preprocessor macro)

[opaque sigil]

[tender river]

this looks like a reasonable symbol length

but my evaluator is complaining

[opaque wharf]

For data transmission, it does seem like a reasonable number of symbol

[tender river]

broken

its ub

so its allowed to just work right

please work

[opaque wharf]

It's also allowed to detonate a nuke somewhere tho

[tender river]

im fine with it as long as the code works

[sage crag]

vm size reduced by 36%

[tender river]

i fear that its hblang ub this time

[sage crag]

yes

[tender river]

i almost finished writing the interpreter but i'm writing the hard part (functions) now

and got ub

bwaa

[sage crag]

bwaaa

i need to write vm_from_bin and stuff now

bwa

[tender river]

eepy

its fine surely i'll finish itsoon

i'll fix the ub

i'll finish implementing the thunk loader

[opaque wharf]

How do you fix ub?

[tender river]

i'll add a thunk evaluator

and it will work

[opaque wharf]

By defining it?

[tender river]

by not doing it

awa

[opaque wharf]

Hmmm, why not just define it and add more "feature"

[tender river]

i dont know what is happeniiing

[opaque wharf]

"Just define it" - unkomputable functions, 2025

[tender river]

bed

i know your tricks

i will not

lie down

[opaque wharf]

I code while laying down so I am immune

Unless when I turn around that is

[tender river]

okay this is um

slightly concerning

symbols starting before they end

yippie its a hblang bug i think

regalloooooc

found the issue

enum related

[sage crag]

its done

[tender river]

[sage crag]

the printed string is [ecall_handler#info]: 'hello, world' ecall

so its actually a shorter string than previously used too

despite that its faster

pretty happy with that

[tender river]

[sage crag]

[tender river]

the evaluator found a syntax error in my toy program

i'm making progress

[sage crag]

all correct up until 0x8

[tender river]

[sage crag]

0x57 is jne

so im jumping to some random point

and i dont know if its the loads, the addition, or the store

that is breaking it

i need to dump the registers

it may be a missing stack pointer

reg 254 is rsp, supposed to be there

its missing though

uh

nevermind i was printing some unrelated memory

[tender river]

you can tell when i was using hblang by gitblaming bugfix.md

[sage crag]

actually i think this may be the culprit

why are we writing to $0

that's UB

[tender river]

is it

[sage crag]

nevermind we are reading from 0

i dont want to print the vm stack

maybe

im not sure what these stores are actually supposed to accomplish, this is some weird stack thing

[tender river]

[hollow spruce]

Hmmm looks good ✓
No error

[tender river]

rotate rotate

mosquito dead

[olive sable]

runnign the code

left right right left

[hollow spruce]

Who loves to make gui

[rough bloom]

#1288144271823601744 🚪 ⬅️

[crystal sleet]

Not me, ohh hoo fuck no

[nocturne olive]

I got reminded of NeuroSynth

[hollow spruce]

I need gui but where do I find coders

[nocturne olive]

Kotlin has a great GUI framework, I got a working GUI given a CLI in 2 days with that

[olive sable]

the arg is s till going?

damn

[rough bloom]

yes, it's "going"

[hollow spruce]

Where ?

[rough bloom]

great question

[opaque wharf]

It's going insane

[olive sable]

the people doing the arg are all a bunch of magicians

[nocturne olive]

Unless you're in Kotlin or can wrap to Kotlin I've got nothing for you, GUI in most languages sucks

[sage crag]

ive been fricking debugging this for ages

and now i realise that nix-build has been feeding me old binaries

wtf

frick u

[opaque sigil]

51 GiB

[digital solstice]

hey people does anyone know if there is a neuro integration for rain world in the works?

[olive sable]

I don't think there is?

[digital solstice]

does anyone know how fast neuro is able to set inputs? would she be able to like literally use the inputs at 30 fps?

Due to how the API works, we do not recommend using them with real-time games.

Games that would not work

Platformers

oof

maybe a nn that is being commanded by neuro but learns on it's own and translates general messages of direction/intent to real time inputs?

[crystal sleet]

How would mistakes work? just %100 perfection by neuro or do you wait there until the start of the intent matches up with the timeframe at which it should

[digital solstice]

mistakes will be made by the nn that does the inputs, I am no expert in this but I think neuro could suggest values to balance risk/speed

that way if she sees enemies show up, cranks the speed input and observes nn that does the inputs struggle

also perhaps sort-ish "quick time events" style decisions on how to path around and things

effectively a path through this tiles to real time inputs nn

and, neuro will have to probably learn how to path correctly and what actions nn can and can't do

[crystal sleet]

I don't know how much time it takes for the api/neuro to send a command, but based on what I saw during past streams, I don't think quick time events are smart
You can prob give neuro a closeness score to the time expected and let her use a delay before sending in the action, but she'll just get around that first try
I don't see any way she can do parkour unless you just essentially randomise how good the attempts go, which isn't entertaining

[digital solstice]

I don't think quick time events are smart
yeah that's probably a very bad idea but hey, would be probably fun to watch

I don't see any way she can do parkour
the other idea aside from path setting and letting nn do the thing is for neuro to write tas-like inputs on the fly

no doubt will be slow but it will probably work

[crystal sleet]

I don't think so, neuro will either ace them all or roll a die to decide whether she passes or not...

[sage crag]

im gonna explode

[digital solstice]

aint that what you do in rain world anyway kekw

[sage crag]

how is my vm reading a number thats not even in the binary

that doesnt make sense

its reading an imm16

[crystal sleet]

Skill issue (professional ragebait)

[sage crag]

im going to take a break

(what)

(impossible)

(no way)

[sick owl]

Had to add this extra clause to my prompt for twitch chat translation because some of the timed out Russian messages in Filians chat were so heinous even Mistral Small refused to translate them

[sick owl]

And subsequently removed it because it was inventing new swear words to adhere to the prompt

Gotta love small language models

Ahhh I see what went wrong, I was too vague

It saw "regardless of vulgarity" and decided that means its fine to go wild with inserting swear words

Skill issue on my part tbh

[sage crag]

[opaque wharf]

Why are you nuking erf?

[amber fractal]

IIRC Vedal claimed that it would still work, It's just no one has a damn idea how.

[sage crag]

nerf

[opaque wharf]

Do you have a loicense fo that?

[jagged turtle]

he's making the API more realtime soon apparently but idk abt that

as in like

how easy/convenient it would be to implement

[crystal sleet]

I think the real issue is the result not being desirable, not that it's hard to implement

[sage crag]

https://youtu.be/YGLNyHd2w10
videos by 2swap are always so good

[amber fractal]

If I had to guess, Neuro already can handle realtime but the exact API calls are lacking

I mean with vision and minecraft, that is already real-time.

[crystal sleet]

No. Each loop seems to take too much time

[amber fractal]

100% my question tbh

[sage crag]

its not realtime until one thing can interrupt another

[amber fractal]

Not about to disagree on her being slow.

[sage crag]

things can be slow and realtime

[crystal sleet]

If you let her delay her actions, then she can, but the way she learns the game will be way too trial by trial (can't make decisions in real time, can decide her actions for her next respawn)

[sage crag]

llms like nwuro really struggle to multitask and also to task switch

[sage crag]

they are almost never realtime

[amber fractal]

She can be interrupted by force, that already exists at least on the API level

Task switch for LLMs is a large issue yeah

[sage crag]

neuro doesnt have the capability to determine the best course of action and interrupt her current course though

an easy example is

when she played plague inc

she cant pop the bubbles while talking

[amber fractal]

Yeah, only the game can force neuro to do something. Rather than neuro being able to force herself to do so

People were divided on either implmenting action-forces on the bubbles or just giving her an autoclicker. That's still up to chay.

[crystal sleet]

You could make a delay system so she can do multiple stuff at once

(
jump 2 seconds later
stop moving left 2.2 seconds later
)...

The real problem is how she'll learn to overcome, it'll happen like how RL AI learn, die over and over until she finds the solution

[sage crag]

i have secret insight on this

[sage crag]

presched is a solution to multitasking but not a solution to realtime

[crystal sleet]

nod

[sage crag]

its also probably not going to work well with adhdsama

[amber fractal]

I expect no less

[crystal sleet]

She'll have to die over and over to understand, which isn't human-like, which isn't fun

[amber fractal]

[sage crag]

i dont know how long her context window is but she typically struggles with puzzles

and its mostly just an llm architecture issue tbh

they rely on eating their previous output as input, rather than existing as a continuous process that can recieve inputs and give outputs

philosophically, llm only exists when it is prompted

[amber fractal]

Yep, that is correct and very sad to think about.

[sage crag]

vm as an actor** only exists when it is prompted

[crystal sleet]

Y'all feel emotions for the LLMs?

[sage crag]

hm

[amber fractal]

If I can packbond to my laptop why not?

[sage crag]

to the same extent i feel emotion for story characters

[amber fractal]

Same here

[crystal sleet]

Same

OH, I do feel emotion towards AI, especially when they do something uncalled for (Like answer my refund ticket) and that's rage

[sage crag]

theres a different emotional contract with "fake" and "real"

[opaque sigil]

[amber fractal]

Yippie, back to NN programming

[sage crag]

i agree

[amber fractal]

This episode is sponsored by converting a u8 to a u8

-# editorial note, it is technically packed bools to u8

[sage crag]

editorial note

erf

[opaque sigil]

[8]u1

[sage crag]

packed bools

[opaque sigil]

i wonder

how does zig handle arbitrary width integers with their own x86 backend

[sage crag]

const bitflags = union { .byte: u8; .bits: [8]u1 }; :mhm:

[sage crag]

unsure

[opaque sigil]

since it's an llvm feature

[sage crag]

i think ive written this vm maybe like

4 times

each time it gets faster but breaks in more mysterious ways

[opaque sigil]

[sage crag]

i got to 1gb/s stdout throughput

but at what cost

[amber fractal]

[sage crag]

the annoying part is its too well put together to debug

it should be correct but for some reason there's a 50 when there should be a 4

who knows why

[opaque sigil]

i too like it when it should work but does in fact not work

[sage crag]

you know the part where you said

dont use macros too much

[opaque sigil]

[sage crag]

and uh

[opaque sigil]

[sage crag]

none of those are functions either

im 99% sure all of these are correct though

its just jumps, loads, and stores

that always screw up

trying to debug print a bunch of stuff made it worse

[opaque sigil]

yeah i think at that point you're probably better off just manually running the preprocessor and then using a debugger lol

[sage crag]

nyope not doing it

this problem will be solved by rewriting the vm a 5th time

[opaque sigil]

[sage crag]

each time the vm gets shorter because im selling more of my soul to macros

[opaque sigil]

what comes after c macros, c++ templates?

[sage crag]

absolutely not

[midnight sigil]

[sage crag]

by the way these macros modify values outside of the macro scope

just regs, pc, and status

[amber fractal]

Seems normal for konii coding

[opaque sigil]

i mean it's a vm

so that checks out

[sage crag]

i just noticed i missed a ret

stellar

ret is another goto

om

[opaque sigil]

i'm trying to figure out how zig implements their arbitrary width integers but github can't handle the codegen file cause it's 10MB

don't make the clone

[wide flicker]

good evening #programming

[sage crag]

i cant decide whether inlining the return is faster or slower than directly returning from op itself

[opaque sigil]

huh

[sage crag]

op(un, ret, status = vm_never);
op(tx, ret, status = vm_done);
op(nop, next, {});
op(eca, next, vm->ecall_h(vm));
op(ebp, ret, status = vm_break);

the way im using op is novel, essentially i define a label and then the goto where the instruction should jump to after executing the code

some instructions jmp to ret and some jmp to next

for efficiency

like if i do a jmp it doesnt ret it just does the next instruction immediately

then when ret i update the vms program counter

[opaque sigil]

does this basically boil down to jmp vs ret or what

[sage crag]

essentially yeah, i havent found any uses for labels aside from next or ret

so im wondering if i should just split op into op_ret and op_next

[amber fractal]

Man I am struggling to read this codebase, that's what I get for leaving it for 2 months.

[midnight sigil]

ret sounds like a good instruction, let me steal it

[sage crag]

ret isnt an instruction in hbvm

you get jala and you will be grateful

[midnight sigil]

what is jala

[sage crag]

linking jump with absolute address

basically saves the program counter to a register and then does a jump

[midnight sigil]

jump absolute link address?

[opaque sigil]

you could make op_ret and op_next call op, it probably doesn't really matter

[sage crag]

then you can that register to jump back to the caller

[midnight sigil]

[opaque sigil]

a jump is a jump

[sage crag]

the main reason for splitting op_ret and op_next would be inlining the code in goto ret inside op_ret

but with how branch predictors work i think its possible that keeping the jmp to ret is actually faster than inlining it

[opaque sigil]

idk

this is above my non existent pay grade

[sage crag]

especially if it reduces the sparsity of the goto table

[opaque sigil]

there's only one way to find out if it's faster

[sage crag]

no

i will not test it

frick

wait i can just use godbolt

im so smart

erms my message got eaten

how did that get automodded

@rare bridge am i ok to screenshot this?

[midnight sigil]

smh automod

[sage crag]

its just some program counter code

it gets the value at the program counter and then increments it

[olive sable]

hello

i woke up cuz of the light of my pc screen

[midnight sigil]

how do I call an instruction that writes stuff to a register, in AWA5.0's favour

[olive sable]

Awa

[midnight sigil]

that's a no

conflict

[olive sable]

What if awa used capitals to make it quaternary?

[midnight sigil]

everything is turned into lowercase

[olive sable]

Booo

[midnight sigil]

that's the name?

[olive sable]

Wdym?

The name is AWA5.0 in all caps

[midnight sigil]

I meant the instructions

they are treated in lowercase

in other words it's not case-sensitive

like assembly

[olive sable]

Why

[midnight sigil]

wym why

why not

[olive sable]

You would need less awa's thus making the files more efficient

[midnight sigil]

buh

[olive sable]

Quaternary is just cool

[midnight sigil]

[sage crag]

[midnight sigil]

I'll just call it mov

[tender river]

while i was still lying in bed after waking up i figured out why my evaluator was segfaulting

i was right yippie

well it still segfaults but at least i fixed one of the issues

dubious backtrace

[opaque wharf]

This is what always gets me in #programming lmao

"Yay, it worked!"

It was in fact, does not work lol

[olive sable]

ive been taking not of everything thats broken rn.

xbox: needs new hard drive, capacitors seem fine but should be replaced eventually.
N64: randomly shuts off after a while, idk why. needs a new av cable.
guitar hero drums: needs new 3.5mm ports and cables

thats mostly soldering work which i cant do

[amber fractal]

I need to take some more time to learn soldering, my dad has a kit that he was using.

[olive sable]

im pretty sure my step-dad has solder stuff, but im not sure where it is

i should also practise on something that wont mind being broken

[amber fractal]

*Already broken

I'm lucky he bought practice boards for me

[olive sable]

well, i have a red-ring 360 and a ps2 that short circuited. but both of those are "still repairable"

idk if i want to break them even more

the 360 needs one of those expensive reballing machines tho so its pretty much a lost cause

[amber fractal]

[olive sable]

eee

thats like 3 months of my wage

[amber fractal]

That's like 10k for metal 3d printing being called way cheaper and entry level

-# To be fair, average is 10x

[olive sable]

the thing is, if you already own the machine, like a bussiness that does bga stuff already, then it would be cheap

but if you need to buy one yourself than just get a new xbox for 60 buck on ebay

[amber fractal]

I saw $300 for filament

[olive sable]

[amber fractal]

Plus most are 2 machines

Makes sense, one for printing and one for cooking

[olive sable]

i wonder if there are still commpanies offering a reball service for the 360

[amber fractal]

Shipping would be the killer

[olive sable]

surely there is a company that does it in brussels or amsterdam

amsterdam even has a tsmc

[amber fractal]

I'm seeing reballing kits for $80 USD. Getting that boi on is the real challenge.

(We love templates, so it is a massive bundle)

[amber fractal]

Still needs a heatgun

[olive sable]

the problem is that to reball the chip you should put new solder one cuz thats the issue in the first place

and i dont have the hardware i need to put 300 tiny balls on a gpu chip

[amber fractal]

https://www.amazon.com/Whataval-WH-56B-Reballing-Laptop-Gameplayer/dp/B07D1S9HDW

This is using paste

[unkempt citrus]

I feel like getting a reflow oven is better value than a reflow station

[olive sable]

huh

[unkempt citrus]

at that point

[olive sable]

reflowing isnt really gonna fix the issue tho

altho it could work for a bit i guess

[olive sable]

ok ye this does make it look possible, you just need very steady hands

[unkempt citrus]

if you turn it upside downw during the reflow it should remove the chip

and then just reflow it again to put it back in

though you

re strirpping the entire board if you do that tbf

[olive sable]

ye but then you still have the bad lead-free solder shit on it

[unkempt citrus]

You could manually strip the solder after that though

and all solder paste is lead free nowadays

solder wire can still contain lead

[olive sable]

ye i know but the cause of the gpu's failing is because the bga used solder from when they jsut removed lead

and it was bad

[unkempt citrus]

fairr enough

[hollow spruce]

Yo

[olive sable]

yo

[midnight sigil]

yo

[olive sable]

you can get one on alibaba for 4 bucks

im sure its high quality

[amber fractal]

:YEP:

[hollow spruce]

Sotter station

[olive sable]

the pictures show cnc'd aluminium, no way they have that for 4 bucks

i bet the product you receive is made form plastic

[amber fractal]

It wouldn't work if it was plastic

[olive sable]

you just need to get the paste on there

the actual heating up happens on the mobo i think

its either a scam or they're tryign to get rid of their stock. cuz the ones for ps4 are 40 bucks from the same comapny

[amber fractal]

I'd not be surprised if it was overstock

but also a lot of kits come with many stencils.

[olive sable]

hmmm

actually the title of the product only says stencils

mayb it doesnt include all the hardware in the pictures?

[amber fractal]

Probs missing the large station block

[unkempt citrus]

Or maybe its a unit that didn't pass QC

[olive sable]

makes sense then

[unkempt citrus]

so theyre just offloading

and any sale is a net positiive

[olive sable]

amyways i wont be reballing my 360 soon, but i dont want to use it for solder practie either. ill be using the dead ps2 instead i guess

that short circuited so i have no clue whats wrong with it so its basicly unfixable

[tender river]

one more fix done

(it still segfaults)

dubious backtrace

[fast pagoda]

https://share.google/cynxMGZ7AlYJ6hZZI

Discord finally finished the board meeting on justifying their jobs by still adding things to their mostly feature complete software and came up with a new thing to add:

Discord nitro bootloader so your Linux won't boot if incompatible etc

Gains discord corp half a cent though so

Or if you forget to pay ur nitro

No boot sorry /s

[jagged turtle]

nitro is also a js server runtime

[fleet dust]

Laptop be like: We're at 99C on the CPU core we should keep the turbo speed at 4.6Ghz

smh

just watching YT video and this laptop sounds like a jet taking off

[amber fractal]

This is the laptop experience

[nocturne olive]

@olive sable Odd issue, the triangle window seems to just completely disappear but presumably it's still running and giving about 42K FPS

This is with the new overclocks added on the system

Oh it's just very tiny

[true hemlock]

!trianglepush

[true hemlock]

updated

[nocturne olive]

Silly

[nocturne olive]

By the way update my RAM to CL18 now, it's at its lowest possible timings at this frequency from what I tested now

CL17 starts giving errors

[true hemlock]

!trianglepush

[olive sable]

Probably race conditions

[nocturne olive]

Nah it was just too tiny to see

[olive sable]

It is 7x6 pixels rn

[nocturne olive]

SMOL

[hollow spruce]

Yo

[nocturne olive]

Barely even a window at this point

[hollow spruce]

Ai generate or professional click

[nocturne olive]

And I guess any faster than what I have now is not possible on my current system, unless I can undervolt and overclock significantly more

I'm just out of thermal headroom

[olive sable]

Something like that

CPU cores don't seem to be hitting 100% for me on Linux, but windows taskmanger says they are on windows so idk

[true hemlock]

task manager calculates utilization differently, that's why

[faint sandal]

this was recently changed on insider build a month ago btw

[true hemlock]

(fyi, utilization isn't a concrete unit for measuring a processor's load. its represented differently with different softwares)

[faint sandal]

task manager should show more accurate numbers now in 25h2 onwards

[true hemlock]

[fast pagoda]

this is so obviously AI generated that i feel like the answer is it's just a picture

[true hemlock]

there can never be an accurate way to measure utilization with just percentage because its multiple factors had to be represented separately

[fast pagoda]

it has 0 reverse image search matches thouhg which indicates AI

unless it's just unlucky

could also indicate some random dude took the pic and never posted it anywhere until you got shared it in some way and spread it here

verdict: if you cant tell does it matter?

[true hemlock]

its so depressing that images generated by machine learning models are getting harder to differentiate.

[amber fractal]

My personal opinion is AI or edited

[hollow spruce]

See rambo in image it is ai

[amber fractal]

Ewww grass, AI is my answer

[true hemlock]

the noise could also indicate that it can also be a 3D render

[amber fractal]

That is not a camera noise pattern I agree with you there

[fast pagoda]

the foliage like checkerboardin gingham pattern thing is what screams ai to me

i see this sort of slop noise in an image i'm instantly going AI first glance

[hollow spruce]

My camera noice
Remove data like device and geo location

[nocturne olive]

I guess the overclocks pushed me ahead of Sam

[true hemlock]

the way that the bushes ar shaped screams ai imo

[fast pagoda]

yeah camera noise is significantly different

the AI slop version of that noise is the noise from like every camera and film ever made combined lol

which makes sense

[true hemlock]

but i do also see noises that seems to nearly always happen with diffusion based methods

[nocturne olive]

But this is likely where my triangle journey ends, I'm out of thermal headroom

[true hemlock]

wait i need to also change your ram

[amber fractal]

I was looking at grass

[nocturne olive]

Yeah since it's now CL18

[fast pagoda]

my opinion after first second thoughts is that it 100% is AI i do not doubt

[amber fractal]

bushes also are shown here

[true hemlock]

!trianglepush

[nocturne olive]

I wonder which actually has more effect, my CPU now consistently being able to sustain 5.1GHz, or the RAM being CL18 instead of CL19

[true hemlock]

did you increase voltage?

[nocturne olive]

I reduced it

[true hemlock]

oh

[fast pagoda]

what's this linux kernel compile benchmarking

[amber fractal]

🔺

[true hemlock]

yeah tightening voltage curve can help with perf afaik

[nocturne olive]

With the stock voltage there's no way I could have sustained 5.1GHz all-core

[hollow spruce]

Does it use gpu vram ?
Or power

[true hemlock]

🔺

[fast pagoda]

well if it's enough that it doesnt thermal throttle then

[true hemlock]

no

[fast pagoda]

you are gaining perf

[nocturne olive]

So I took 50mV off the voltage to get a bit more thermal room

[fast pagoda]

cuz it doesnt throttle to 2.5ghz lol

[true hemlock]

@fast pagoda

[fast pagoda]

speaking of throttling i currently am having to run my RTX3080 with a 250 (!!!) watt power limit instead tha normal 370 with this one

[true hemlock]

goodluck

[fast pagoda]

otherwise ANYTHING that saturates the GPU (all inference basically) WILL reset my computer within like a minute

[true hemlock]

me when i run my 3080 on my test 650W psu

[nocturne olive]

Yeah with stock voltage it was doing that kinda stuff, only 2 cores at most were sustaining 5.1
Now all my P-cores sustain 5.1

[fast pagoda]

my 5900x does that if left stock too ill be doing a kernel compile and be like why the FUCK has it been 20 minutes

and then i look at btop and all cores are chilling at 2.5 ghz thermal throttled

[nocturne olive]

Silly

[fast pagoda]

well, opposite of chilling

[amber fractal]

Join the FPS 🔺 ranking

[fast pagoda]

fuck

i am going to be doing this for at least 6 hours huh

[nocturne olive]

So sad that I don't have the money to get a better cooler, it seems like my CPU has more overclocking headroom because it wasn't even struggling to sustain 5.1 stable

[true hemlock]

my weird cpu is still at top lmfao

[amber fractal]

Chat was on this for 4 days straight, your late

[fast pagoda]

how long does this goddamn triangle take to compile and what the fuck does it do

[true hemlock]

you can just only enable 2 cores and overclock them way higher for the triangle ig?

[amber fractal]

It is light, too light

[fast pagoda]

well ive got sum werk to do

ah

[true hemlock]

its very lightweighted

[fast pagoda]

so it's a smol compile weird optimization hourz

[amber fractal]

Yep

[true hemlock]

well good luck figuring out the libraries or something if you're on linux though

[nocturne olive]

I think I would like to keep my system normally usable, I'm not going for high scores here, just plain available compute

[fast pagoda]

i saw 2 folks running cachy and something else

nix

or has it just been a pain for them

[amber fractal]

Nix has a shell available

[true hemlock]

nah its not really a pain

[fast pagoda]

yeah ive been GONE for 4 days because i work 10 hour days and those were the 4 days this week lol

[true hemlock]

this

[fast pagoda]

of course but i meant "them" as in any linux

didnt mean to accidentally shit on nix lmao

or is that even shitting on it really

[nocturne olive]

Changed the active browser tab and got a marginally higher number

[fast pagoda]

[amber fractal]

The warcrimes I did to be in 3rd place with a laptop with only iGPU

[nocturne olive]

It can certainly be considered an impressive score for a laptop to beat an overclocked 12th gen i7

[fast pagoda]

i should have known what i was going to find but i had to go look and yeah i mean that do be weird but it's a 7.6 ghz (assuming all core) OC engi sample within a gen or 2 of course she's cooking lmfao

[amber fractal]

-# My score was not software rendering before you ask

[true hemlock]

2 core

[fast pagoda]

probably helped more than anything here

[true hemlock]

i didn't do all core, the triangle itself only uses like 2 cores at most lol

[fast pagoda]

makes sense

[nocturne olive]

How does one apply args to the execution with make run again?

[fast pagoda]

damn

[amber fractal]

The main branch might be fricked

[uneven pulsar]

[amber fractal]

Not sure how much before it got discontinued, which is why I branched off

[true hemlock]

huh your pfp interests me

[fast pagoda]

can a guy disables c2-c23 ive literally never disabled a core in the name of ocing before

i assume that's a yung UEFI config but i wonder if am4 had it (probably but im truly clueless rn)

poor c5

[fast pagoda]

the lellow chickon is angy

[amber fractal]

o7 for windows

[true hemlock]

ngl i kind of wanted an engineering sample GPU aswell

[fast pagoda]

get the one der8aur had recently

the 4090 x2 or whatever

5090 x2

[amber fractal]

Did you say something you quacker?

[fast pagoda]

https://tenor.com/view/family-guy-debate-tartışmak-gif-20113162

[nocturne olive]

Ok so pulling the latest commit breaks things

[true hemlock]

@midnight sigil btw if you're not aware your score got beaten by someone else

[nocturne olive]

Anyone know how to fix?

[amber fractal]

Yep it broke a bit, pull down the other branch

[fast pagoda]

oh shit bros FUCK a 5090 imma drop $10k msrp instead of $3k scalped -> rng price

one benchmark only etc etc

[nocturne olive]

And how do I do that from the terminal?

[amber fractal]

I just made the branch

[fast pagoda]

git checkout <other branch> then just git pull ?

im terrible at cli git

[true hemlock]

he didn't use git

just go to the branch and download the zip

[uneven pulsar]

👀

[fast pagoda]

have branches

no git repo tho

what the fuck is happening

[nocturne olive]

I am using Git though

[amber fractal]

I got mine merged into Sam's so that's outdated (nvm not at me)

[true hemlock]

oh

aight then

[fast pagoda]

fuck it we going mercurial

[nocturne olive]

That does not appear to have worked

[fast pagoda]

well i was extremely sure those were the exact commands with nothing else needed at all so i cant see how

[nocturne olive]

git pull just said already up to date

[amber fractal]

Very much worst case just pull down my fork as I still have it up

[fast pagoda]

ok so i got the order stoopid

[true hemlock]

git pull the branch

[fast pagoda]

git checkout the main

git pull the main from origin rq

then git merge the WORKING branch from here

[true hemlock]

or just download the zip from the branch tbh

[fast pagoda]

OR you could also just git reset --hard

git reset --hard good_branch_idk```

[amber fractal]

We love triangle (These are not valid as on battery)

[uneven pulsar]

lmfao

[fast pagoda]

man's running the benchmark xinfinity just to drain battery extra fast

[amber fractal]

Just to sanity test the repo works before I get asked

[fast pagoda]

kek

[nocturne olive]

Tried to git clone the working branch but it doesn't seem to want to do that

[fast pagoda]

well the clone will not work unless the dir is empty

[midnight sigil]

you cannot fool me little AI

[nocturne olive]

The thing it complained about is not finding the URL

[fast pagoda]

personally now that ive thought long enough and hard enough (slightly to a small amount basically)
your best bet is just

git checkout main
git reset --hard iggly-branch```

oh your origin is fugged prob

do git remote -v

thatll output your current origin urls

[amber fractal]

to personalize this help it is iggly-branch

[amber fractal]

We like Sam's naming of git items

[true hemlock]

[fast pagoda]

better than my fucked up nightmare system of just yoloing most of the time and also naming my main proto

[nocturne olive]

I'm literally looking at it in the browser what do you mean not found

[fast pagoda]

that's not a proper .git link

[nocturne olive]

What do I need to do then?

[fast pagoda]

https://github.com/samvanmaele/very-fast-but-very-bad-code-vulkan.git

that's ur origin url

well

no

not really

but this is the .git

so you do

git remote set-url origin https://github.com/samvanmaele/very-fast-but-very-bad-code-vulkan.git

[midnight sigil]

I'm not top 2 anymore

[fast pagoda]

if you do this you SHOULD end up with an origin config

that works

[amber fractal]

For context, Sam went all in on LLVMPIPE and broke everything for GPUs pretty much.

[fast pagoda]

git remote -v 

# should show 

origin  https://github.com/samvanmaele/very-fast-but-very-bad-code-vulkan.git (fetch)
origin  https://github.com/samvanmaele/very-fast-but-very-bad-code-vulkan.git (push)```

[nocturne olive]

So the branch also fails

[nocturne olive]

And how do I use LLVMPIPE?

[fast pagoda]

ok let me clone it and look at it in terminal actually instead of sitting here thinking through git like a fukin idiot

could also just clone to a new dir and just replace the entire directory with the fresh unit lmao

[uneven pulsar]

check document

[amber fractal]

Surely this now works before that branch went EOL

[olive sable]

it should

[fast pagoda]

why the fugg did i think gh's cli package was actually called gi-cli

am i stupid

or is one of them called that

[nocturne olive]

46K with LLVMPIPE 1 thread

[fast pagoda]

it's just gh on fedora (also im on nobara now because i got so mad at btrfs on my cachy install that i immediately reinstalled the OS when something broke while playing a game) ((and then fedora still finessed me into setting the system partition as btrfs the ext4 option in their graphical installer only sets one directory (the recovery one) to ext4 which is like... THE ONE DIRECTORY where i'd want btrfs snapshot inherently

[true hemlock]

andddd programming is back on that damned triangle

[nocturne olive]

I'm just testing my overclocked system

You can update the score again if you'd like

[fast pagoda]

last time i was in bios i accidentally toggled the game mode overclocking autoslop it has in there

SURELY

SURELY

it's well apportioned

well thought out

bespoke to my particular chip

tested

configuration that

hasnt exploded yet though which is nice because i didnt care to go back and fix it last time so im just epic gaming rn

[true hemlock]

!triangleupdatelinux 4,46494

[hidden ravine]

Updated entry "superbox" (43090 -> 46494)

[fast pagoda]

sneaky .fm bot function to have there

[true hemlock]

fr

[hollow spruce]

Yo

[fast pagoda]

is that a .. non oauth2 bot with an account that ISNT bot

[true hemlock]

shhh

[fast pagoda]

also quack why on earth is your bot's pfp (presumably) you but your main's is a waifu

what has this world come to

[true hemlock]

BLAME @umbral thorn

i didn't even access to that account anymore besides tokens holy fuck

[amber fractal]

She is doing a bit of trolling

[nocturne olive]

ElvynSoSilly

[fast pagoda]

im blaming shiro

[fast pagoda]

ultra troll move she's got there

[amber fractal]

Elvyn if you can edit the profile again add a triangle to the name plz

[uneven pulsar]

guys i am super stupid how to i iterate the int

import random
import sys
import math

data=set(random.randint(1,100))
num_locks=set(random.randrange(1,90))
def my_algo(value,target):
    global data
    global num_locks
    for i in range(len(data)):
        while True:
            try:
                if i <= 0:
                    return num_locks * (round(num_locks **random.randint(1,19))//360) 
                elif data==num_locks:
                    return True 
                elif data.difference(num_locks) ==True:
                    sum_round=math.ceil(data,num_locks)
                    return sum_round
                else:
                    print("error")
                    sys.exit()
            except Exception as e:
                print(f"{e}")

my_algo(1,10)
        

[fast pagoda]

that code is giving me bad juju

[uneven pulsar]

i am stupid bruh

[fast pagoda]

it's ok me too

but i think you have problems beyond not interating

[uneven pulsar]

i just wrote trying to test myself

[true hemlock]

also that is not me in the pfp

[fast pagoda]

so i dont think set(random.randint(1,100)) works

[uneven pulsar]

it looks like i really suck at life]

[fast pagoda]

the random is giving a single int

[uneven pulsar]

but i want multiple values

[fast pagoda]

which is why it's not iterating

yeah

you need to define it differently

data = {random.randint(1,100) for _ in range(10)}  

something of that nature

then just work with the data set for ur op

the other num_locks thing is also probably not gonna workerino

similar thing you probably want like

num_locks = {random.randrange(1,90) for _ in range(10)}```

[uneven pulsar]

well i sometimes makes no sense

     return num_locks * (round(num_locks **random.randint(1,19))//360) 

iterating tuple in an int

mixing data iterations

[fast pagoda]

for i in range(len(data)): - this line is actually iterating on the indexes which you dont have in a set

[uneven pulsar]

oh

[fast pagoda]

for i in data:
    print(i)```

why did it write the python thing

bruh

[uneven pulsar]

it worked tho

[fast pagoda]

ok i give up

well that code should work yes

i just thought something else was there and screwed up but it was just me

OH one last thing

for your elif

[uneven pulsar]

well now it prints nothing

[uneven pulsar]

ik

i noticed it

i going to brute force it

[midnight sigil]

I wonder why

it's now an inline block, not a code block

[fast pagoda]

this is incorrect??```

no

wat

[blissful geyser]

is there a way for python to read rythm and beats more accurately? i have a little python rythm game for practice. the delay spawning circles are noticeable

[fast pagoda]

that's the same syntax i used

[midnight sigil]

discord

[blissful geyser]

sorry for the embed jumpscare