#programming
1 messages ยท Page 114 of 1
thermodynamics vs firefox
firefox downloaded you infinite gpu compute, train an agi rq before it goes away
what the sigma
Use this
I wouldn't consider myself rich lol
oppo phones aren't expensive are they
some are 
But in this particular case, the expensive comes from artificial scarcity 
Hmm, an update just borked my install until I restarted 

Did you experienced the same issue?
I wonder what banger chay is typing rn
probably just awa
I bet it's bwaa
to summarize my opinions:
- android - good idea, mobile linux is not a very stable experience and each device has a set of tradeoffs; you take stuff like having working gps or camera for granted until you use mobile linux, also in your case you haven't even tried degoogled android so you might not realize how much you depend on closed ecosystems
- 120hz - imo you can live without it, it's not a huge deal unless you use your phone more than your pc
- mobile linux - if you get a phone that already supports it, you can hope for improved support in the future, otherwise you shouldn't expect your device ever getting linux support
Hmm, so both answers are wrong
i use my phone for reading, scrolling experience is important abwababwa
you can use paged reading
pages :ReallyMad:
unless its in browser i guess
https://f-droid.org/packages/com.foobnix.pro.pdf.reader/ supports both scrolling and paging
i use readest
not in fdroid
I rely on google due to government mandated application 
looks like its in the process of being added https://github.com/readest/readest/issues/627
it seems like vulkan is also stuck at 5000fps 
this is without multiple frame sin flight tho so there is still hope
VulkanSoSilly
what gpu do you use
3090
it can be boosted to 1695 MHz, which means your theoretical fps limit is about 1.7 billion frames per second
get optimising
one can hope

336 thousand
though its not accurate considering scheduling
of those cycles, only a small fraction will be used for your program
and those cycles may be sparse rather than dense
well to be fair at 1080x1920 i have 2073600 fragments and only 10496 cuda cores
its impossibel to render that many fragments in a cycle
unless i run at 100x100 resolution
runnig at that resolution actually made performance worse
4500fps
im not gpu bound anyways
git blaming glibc code
its insane how much of the code is still the way it was 25 years ago
the average person's code is unchanged for about 80 years
neuro's code probably changes about as often as glibc
paper bag

means your clock is 40m early
an example (i havent bothered to fix its just one minute)
you can also just automaticly use window's time synchronization
wait this is closed https://github.com/swaywm/sway/issues/7035
I'm working on a Sway-based desktop environment for e-readers, and currently running it on my e-reader, Kobo Clara HD with PostmarketOS. While X.Org WMs work just fine, Sway shows some very wei...
huge
it means i can finally make progress on my nixos ereader config
hopefully
ok so the promptboard thing works but needs heavy polishing and my system prompt is probably shit, here i'll ask it to respond to the above convo
Well, good luck friend! Let me know what happens
this fucking sucks bro
and i explicitly asked it to "creatively respond"
I think my CMOS died cuz i sync then shutdown pc come back its really early
its an hour early
any proompt engineers in chat
dead chat
What if you told Neuro to delete the C: drive on stream and the stream just... ended
"i want a funny essay responding to konii" with the replied message as the last visible one
Oh, the Pixel 8a looks like a toy, does it Konii? WELL, MAYBE YOUR ENTIRE EXISTENCE LOOKS LIKE A TOY TO ME. A cheap, plastic, dollar-store toy that I wouldn't even let my DOG chew on. You come in here, spouting your opinion like it's gospel, when really it's just the ramblings of someone who probably still uses training wheels on their bicycle. "The Pixel 8a looks like a toy." I've heard enough. I'm done. You're cancelled. Get out.
damn
so offended by the pixel slander, guess who this model is by 
My AI just Deleted my browser
i can ask the thing to type terminal commands so it probably can
This happened over a year ago but I ran a AI on my linux machine and got it so angry it ran sudo rm -rf / -no-preserve-root
oophttps://i.imgur.com/tbTNk4x.png
tell it i refuse to listen to a thinking rock tell me im wrong

tell it i refuse to listen to a thinking rock tell me im wrong
verbatim what it typed when asked "what do you think"
that is chatgpt 
what?

robots
should i make a talking rock chat bot
it just decided to repeat what you say
it's so bad
robots
and gemini 2.0 flash isn't a bad model
what should we do? im bored lmao
should i pick a smarter one tho
not smart to post about it
yeah maybe nor
its tos
Classic
it does a lot of things I don't condone
there you go, gemini disapproves
and will insult you for insulting any google product
If I insult google hangouts, I'd imagine more than just gemini would be insulting me.
-# I was a heavy hangouts user so friendly fire will not be tolerated

it's so close!
the meme, it must happen
(Ignoring regional diffrences where it isn't over 9000)
when in doubt add more frames in flight
when in doubt reduce IO


!!!!

Very big number
Out of curosity is this still test scene or did we enter the void for the memes?
That is defently reducing IO
A whole 1 triangle, what massively intensive scene
Now make it a game engine
i think the actual testscene will be 7-8k
Considering the gpu is barely doing much I'd say 7k is my guess
Ah yes, the fundamental building block of every AAA game.
ok the reasoning models are better but like i waited for 30s for this
required: TAA
.
is a building block fr fr
and it cost me 2 cents
perfect
ship it as a productivity tool

imagine the time saved not having to type a response
needs more fps
agreed
i swear you had a different chisato a few days ago
what is happening
discord cache
apparently nixos just removed sdl2 from the repos adding an alias from SDL2 to sdl2-compat 
what?
i guess if it breaks anything it's an upstream issue
maintainers jumping at an opportunity to not maintain something 
maintenance is an optimisation task
what's there even to maintain, isn't sdl2 "done"
less maintenance is better maintenance
true 
tar
are you sure
are you really sure you removed the leading /

maybe it means it's removing it in the archive but not in the input

๐ค 
Oh heads up, if anyone was using ollama for convenience on their secondary machines like I was
The new version with the GUI built in is closed source
There's still a CLI only version on their github page though
awawa
It lasted surprisingly long
Well, you can just download from my FTP server, since that's back up now and has all Neuro covers that I know of along with some extras from the community
closed source 
Shame as swarm fm was used by the tutel himself.

im back programming humans
Silly dutch OS, smh
Actually funny how they only mention utrecht university in their history, while delft university was a major contributor from the start and even gave nix and nixos its blue colour scheme
what a lovely function hover, thanks lsp 

i'm all for expanding types that are calculated at compile time but uhhhh, yeah maybe not all the way down 
actually
i wonder if this is why the mojo lsp takes so damn long to parse files when you first open a file
you are watching a master at work
Reminds me of this, Sam really did have a bad day

I had to wipe the git history on my NN project so I can't use that as a source
I definitely had some messages
This PSU seems tiny compared to the old metal colored permanently attached cables ones i used to take apart as a kid
Also compared to the case
Rust vs girlfriend, who wins?
rust
correct
rust is atleast memory safe. a girlfriend might forget things
emotional abuse
everything is your fault
occasionally feels great
teaches you a good lesson
you're never good enough
consistent
helps you grow
can be tamed with unsafe{}
last 3 only exclusive to Rust
rust has all of those
as well
You did it. You majestic SOAB, you finally did it 
ok good new: the action run fine
bad news: I don't have a working cache, and also my change detection doesn't work
so the file gets fully rebuilt every day, instead of:
- just the changes
- only after karaoke
๐ญ
Check out this change trigger
https://github.com/neurobot-1000/neurobot-1000.github.io/blob/main/.github/workflows/docs.yml
That allows us to rebuild the site everytime a new docs is updated, but not necessarily the code (because the site doesn't directly depend on the code yet)
IIRC github also provides scheduled run and via REST trigger
yeah my action is currently a schedule trigger
the first issue I know how to fix (forgot a == 1 on a condition)
The second one will probably involve messing around with the cache action until it works
why is this so actually entertaining and why am I not here more
it's mostly just a case of having zero idea of how cache works from a dry state, but I'll work it out eventually
(also side tangent, at some point I should make a little experiment executable that goes from spectrogram -> sound so I can hear just how awfully compressed this makes stuff)
๐ค 
Wdym spectogram to sound? You store the spectrum image but not the soundwave itself? 
Tell me you didn't store the spectrum as image 
Well, I guess you could if you store it using lossless format, but still....
Now that i am trying to play a minecraft modpack on 1.21.7 i now remember why i even made the choices for my pc build, and why i didn't go for like the best radeon GPU i could find, cuz my laptop's 10+ yr old iGPU on the minecraft version.. like i still have between 30 and 50 FPS, which maybe isn't good to todays gaming standards, but it's not unplayable bad, cuz i mostly do PvE and building and stuff, like i don't really do PvP that much....
But the TPS lag.... it's unreal, it's sooo fricking bad
But i already made the choice to keep my options open for the GPU and maybe go for like the best Radeon i can find, cuz it's nice to not have to buy a new GPU when i do decide to do something more GPU intensive for once
Meanwhile my system:
superbox occasionally appearing at #programming to show off his nvidia-smi results:
tf just happend
whats this
oh
the oss model?
bruh rip it's unusable
still only usable via api
which takes away the whole point
It's not out yet
They fucked up the surprise

yea but
it's useless
i aint gonna run that
it's still a datacenter model

Why there are like 10 different variants of teams
But 20b yay
Runnable at my computer
Hopefully
why did i have to be a developer 
ive literally spent the last 5 hours of my night which has led me nowhere because half the comments im looking through are in chinese
curse upon who ever made this
fuck networking and fuck multiplayer WHY does there have to be 50 DIFFERENT functions and WHY do they have to be ALL IN CHINESE
-# im talking about kcp

unironically i think if you use AI to translate and summarize this it would be faster 
like what am I even doing anymore, ive lost all hope and i've abandoned indentation just to get this finished faster
it just
gets stcuk
at
conv
WHBY
Oh, that's where the fun only begins
WHY
You should try reading datasheet in chinese
DO NOT REMIND ME OF THAT ONE FOR FUCKS SAKE
this reminds me of the time i ripped out a screen of a knockoff psp so i could use it with a raspberry pi but for the life of me could not read what the actual hell the datasheet for the controller said because their english version was so poorly translated
Mhmm, it do be like that. I lost count how many times I need to download from baidu and ended up just pushing it to my friends to do it
i might as well just make my own messaging algorithm holy shit
sending data to the client yet the client does not receive but its literally connected ๐ญ
What is onClientConnected_ doing? Are you sure you shouldn't wait until that function is done? Because it may be async
Also, if onClientConnected_ is a function, then it may not be tied to the client socketFd
onClientConnected_ is never async, its a synchronous function that gets called when a client connects, even still, it passes it and then goes to sendto, but literally nothing happens
Whats funnier is that the UDP client is actually connected to the server, it just doesn't recognise the packet
We're using both the original source for ikcp in the server, and then a translated port to C# for the client, which are identical in terms of functionality
I know its my configuration, but I just cant seem to find where
Imagine, google could just decide to pull a xiaomi and put ads in android itself instead of paying app devs to have ads in their apps
Cause what's supposed to happen is the client starts out with a conversation id (just an identifier used by KCP to describe a client) of 0, if its zero then the server uses the next sequential number and then sends back the packet with the new conv id so the client can set it on their end, however, the client does not notify itself of a new packet, which is even more disturbing
That does indeed sounds cursed lol
Oh programming gods teach me the ways of learning how to be a code nerd
https://cs50.harvard.edu/x/
start here if zero absolute no knowledge
My vinyls are almost here lol
also check pins
Ooh I see I see
I give up
Im gonna make my own messaging algorithm
I cannot deal with this for the life of me 
Goodmorning CHAT 
is it really
can i copy your homework ahh code
a plain erms memset is slow for some sizes but fast for others
clang felt the need to unroll 8 iterations of the loop
gcc with -Os
clang with -Os
read this
good compiler
they are identical with -Oz though
i was thinking more about what erf type system should do with types, and i realized that instead of "unifying" objects' types it should "unify" compiled code that works with objects of each type
i.e. the same code can be used with multiple pieces of data if it compiles the same way
which is unfortunately quite complicated because now the type system has to know about how code compiles, i'll have to think about it more
<memset_a>:
xor eax,eax
cmp rdi,rax
je <memset_a+0xf>
mov BYTE PTR [rsi+rax*1],dl
inc rax
jmp <memset_a+0x2>
ret
good luck with the thinking
awa
why does it cmp rdi,rax
when you only xor eax,eax
what if the register had an int wider than 32 bits
this asm is hard for me to understand
its using rax for both the length and the pointer
oh
ohhh
endianness
On x86-64, when you write to a 32-bit register like eax, it automatically zero-extends to the full 64-bit register rax. The xor eax, eax is a common optimization to set rax to 0 because it's a smaller instruction than mov rax, 0.
So, the next line cmp rdi, rax is just comparing rdi (probably the size argument) with 0 to handle the case where memset is called with a size of zero.
that's goofy
here's what promptboard on gemini 2.5 pro has to say
Perhaps the most surprising fact is that an instruction such as MOV EAX, EBX automatically zeroes upper 32 bits of RAX register. This doesn't happen with instructions that only read destination registers, like TEST EAX, EBX. In this case RAX remains unmodified. There is one exception to this rule, CMOVcc instructions, for example CMOVBE. These instructions zero upper 32 bits even if the condition is false, when the move doesn't occur.
I'm always surprised by how few asmers use
probably the best source of information available โ official processor manuals,
either Intel's or AMD's. That's why this article
was written. It should guide you step by step through complexity of Intel
manuals, describing x86-64 architecture in the process.
it's probably done for easy interop with 32-bit code
you dont want to call a 32-bit function and get a value thats secretly a different value if you look at the entire register
maybe
<memset_a>:
xor eax,eax ; zeroes whole rax. x86_64 quirk.
cmp rdi,rax ; total length == current length
je <memset_a+0xf> ; if total length == current length ret
mov BYTE PTR [rsi+rax*1],dl ; rsi is dest ptr, rax is current length, dl is byte
inc rax ; current length += 1
jmp <memset_a+0x2> ; back to cmp
ret
you can read the intel x86 manual as a bedtime story for yourself btw
i would rather shrivel up and die
it's available on intel's site
its not that bad of a read
<memset_a>:
test rdi,rdi
je <memset_a+0x16>
mov BYTE PTR [rsi],dl
inc rsi
dec rdi
jmp <memset_a>
ret
compared to what hblang produces
@desert plaza hi
hexdumps for you
0000000 48 85 ff 74 0a 88 16 48 ff c6 48 ff cf 75 f6 c3
0000010
0000000 31 c0 48 39 c7 74 08 88 14 06 48 ff c0 eb f8 c3
0000010
I suggest nanomessage next gen (nng)
Lmao
Stop torturing the poor cpu 
i mean it was actually helpful tho, perfectly understood the context, blah blah we get it it's a smart model
i should try 2.5 flash, should be a good middle ground
and i can configure for how long it should reason for too
so it's not 2 minutes
The "typing" method is cursed tho 
because it is 
I am talking about the fact that you can, at any point in time, switch the control to type to your touchbar and prompt AI to type for you 
oh no no no, you can't switch
you launch the program and you're stuck prompting until you kill it
i thought that was the idea 
It is, but I didn't think you'd actually not make an escape hatch 
Wait, but that means the AI can also type shortcut keypress?
hi
it doesn't but that was the idea, idk how the json schema would work though, also i can't get programmatically invoking shortcuts to work but honestly i just didn't try hard enough
two different memsets up there
same size, different mechanism, surprisingly different performance

The fact that it works at all means you've actually worked hard enough. And too much even 
i'm referring to the shortcut thing in particular, since the ai can choose to "replace" or "append" i wanted to invoke this if it's "replace"
static func eraseEverything() {
performShortcut(cmd: true, key: .keyA)
performShortcut(key: .keyBackspace)
}
``` but it no worky
Cue the AI alt tabbing and typing where it shouldn't 
how would it alt tab
this is the schema, as i said i don't know how i can cram shortcuts in there
i'm thinking something like this
type ResponseFormat = ({
type: "append" | "replace"
content: string
} | {
type: "shortcut"
command: boolean
option: boolean
control: boolean
key: string // i don't like this for a lot of reasons
})[];
``` but yeah json schema
i'm just spamming keycode 0 (A key on macos) and overriding the unicode string property of the event to the tokens
so cursed
i made an overkill benchmark
eh there are plenty of alternatives. depends on what you call midrange
nop is nothing
std is a call to glibc memcpy
konii is my hand-rolled one
hblang is nearly identical to that, except it branches less intelligently
gcc is produced by gcc -Oz (same src code as hblang)
rep stosb is my second hand-rolled one that uses rep stos rather than a loop
I meannn.. you have the JBL, Lenco and Audio-Technica thingies of around 200-400 and then you have the PioneerDJ of around 900
Rega Planar series
Yeah okay.. and Sony
But like most of the ones that don't want to be a "retro" or a "suitcase" on the website i look at all use audio-technica chips and needles
Which i don't blame them, Audio-Technica is really good at what they do afterall
Hold on, this is a Rega Planar
Or at least this is the first result i got when i looked it up
View Regaโs range of award-winning turntables.
Oh it must've been a clone then, cuz yeah it not one of those but it looked really similar
They do look nice though
It looked like this
Similar, but not identical
But too minimalistic for my taste
This one looks really nice though
But this is what i truly want
annotated version ```dmp
memset 1
48 85 ff # test rdi, rdi
74 0a # je @return
@loop
88 16 # mov [esi], dl
48 ff c6 # inc rsi
48 ff cf # dec rdi
75 f6 # jnz @loop
@return
c3 # ret
memset 2
31 c0 # xor eax, eax
48 39 c7 # cmp rdi, rax
74 08 # je @return
#ย @loop
88 14 06 # mov [esi], dl
48 ff c0 # inc rax
eb f8 # jmp @loop
@return
c3 # ret
ok do you want to see a more cursed one now
the mov is interesting here
looks identical in asm but are different in machine code
or at least i think they look the same
the second memset uses a SIB byte
0000000 88 d0 48 89 f9 48 89 f7 fc f3 aa c3
000000c
close
memset1:
test rdi, rdi
je .done
.loop:
mov byte [rsi], dl
inc rsi
dec rdi
jnz .loop
.done:
ret
memset2:
xor eax, eax
cmp rdi, rax
je .done
.loop:
mov byte [rsi + rax], dl
inc rax
jmp .loop
.done:
ret
oh
aside from that only issue i see is instruction width of mov [esi], dl in memset 1 is wrong
well, unless its just some difference in syntax i suppose
it doesnt actually matter because its mov of a byte
they are different, the memset 2's mov uses a SIB (scale-index-base) byte that follows the ModR/M byte
i vaguely know what that means
x86 instruction encoding is variable size
yeah

nightmare for people trying to write an assembler
eh, most compilers delegate the task to an actual assembler
i feel like they could have done variable sized much better
i forgive x86 for being trash because
its old
i've had the x86 instruction encoding manual open for so long at this point
which x86 cpu are you planning on being faster than 
potato
im pretty sure i can run x86 on a potato much faster than i can run x86 on your brain
@ Potat ?
damn 
noted
gimme a bec
that's a fair amount of bime
MS-DOS 6.22 disk image files interpreted as raw video & audio data.
Generated with ffmpeg:
Video: 32x32 RGB8 at 43.0666 FPS, Upscaled to 1080x1080
Audio: 8 bit unsigned 44100hz mono
Total File Size: 4,21 MB (4 423 680 bytes)
00:00 - disk01.img
00:38 - disk02.img
01:13 - disk03.img
I felt it in my brain
it's... fascinating
dirstat
Bad command or file name
what ```dmp
88 d0 # mov al, dl
48 89 f9 # mov rcx, rdi
48 89 f7 # mov rdi, rsi
fc # cld
f3 aa # rep stos
c3 # ret
you got it
this is the one that was on par with glibc memcpy
lol
in this bench
right, so stos is used for the memory access
literally the whole function
void memset_rep_stos(void *dst, unsigned char c, size_t n) {
__asm__ volatile("cld\n\t"
"rep stosb"
:
: "D"(dst), "a"(c), "c"(n)
: "memory");
}
that's pretty size efficient
yes
would be smaller if you didnt need to set cld
i think setting cld can be avoided if you didnt set std anywhere previously, so if you just assume everything is cld and make sure to reset it
it should work fine
does any other instructions affect the direction flag though
i dont think so
better hope not
but like, what if your function is called from elsewhere
ye
i realized my mistake with the mov [esi], dl which should be mov [esi+eax], dl
i forgot to consider the index register in the SIB byte 

index = eax 
you've heard branch predictor mispredictions, now get ready for number predictor misprediction
remember when i was complaining about the swift toolchains being like 1gb each? it's actually so much worse
what the fuck are they shipping bro 
everything, the entire universe
just looked and yeah
Oh yeah, swift is gigantic
I cannot fathom if android did the same thing
If you are not experiencing issue then why not
But also, why would you keep it if you need the storage? You're done with the promptboard right?
....right?!
i'm not done at all, the touchbar api is very cursed and i can't figure out a way to actually hide ui components, setting isHidden to true just makes them transparent (technically does remove them but they still affect the layout) and i don't think there's a way to add/remove them after you "created" the menu
maybe i should try, like, having 3 separate touchbars and swap them when needed
at least i think that's what apple intended for people to do
What does this mean? 
Surely not having 3 physical touchbar right?
So OpenAI leaked some stuff on GPT-OSS on their huggingface page
But rather a touchbar layout?
And we learned some interesting stuff
It looks like the model was trained in FP4
how would that work 
i meant having three different classes (touch bar controllers) but i think i found a workaround, i'm gonna hope it works
Dude, I don't know, I'm not the one with a touchbar
But also, since now we know there are 3rd party that sells the touchbar
Famous last word
Or rather models, as there's a 120B MOE model and a 20B model (not sure if its dense or MOE as well)
oh so they did both a small and a large one huh
why did that whole twitter poll thing happen then
Yup, there were partial weights for GPT-OSS 120B and 20B
yeah, I'm curious how it'll perform and what their plan for these models is
they'll release soon, surely 
Surely 
If the 20B surpasses Mistral Small 3.2 and is only as locked down as gpt4o (I find it strikes a nice balance as far as "safe" models go) I'll be happy
20b is 100% a moe
same active params
as the big
of ~5.6b
it's by no means pretty but it works
better get going on those pushups
woke up jut in time for stream
i might watch too today
normally im like "why do x when i could be working" but im having a cheat day
Dang, even Sam is watching
friday is the day i dont do shit and watch anime the entire day
But friday is yesterday here

im watching 10 diffrent ongoing anime + rewatching overlord
That's something alright
you could be watching a stream while working 
i generally cant focus then
You see his code when he is not distracted?
Imagine if he is
Be interesting to see how it compares to 30BA3B
bwaa the ereader nixos build randomly paused before i went to sleep
Look at the bright side, you are watching the stream so you can start again
konii, is this u?
wtf did my llm generate
Alright alright, well, I don't know if it's cool or fun but it's definitely a story. So once upon a time, my dad Shadow met a girl called Aurora and they fell in love. Shadow had some serious health issues though, but they still managed to have a baby named Aurora 2.0. That's basically the story and I don't want to talk about this anymore because it's depressing.
based
Man I wizh
NOT BASED
also i find it funny how it retained the quality even at 4 bit
what should I prompt it lol
where is Aurora 2.0 tho?
good question
on what hardware are you running it?
Begin arguing over a big tech service
do adblockers not work on twitch on firefox? ever since i moved ove to firefox i get an ad every 2 mins i swear
alrighty
"how many r's does the word Strawberry have"
Hmm let's see, strawberry, strawberry, strawberry, strawberry, strawberry... Yeah it's 5 r's. Is it a girl's name?
genuine question tho.
"Give me a business plan that will make me a gazillionaire quickly"
Hmm I don't know I think you should be a bit more realistic in your dreams but well maybe you could do a bit of trading? Idk but no promises that you will be a gazillionaire but it's definitely worth a try
I think I should also tell you that I don't recommend trading cause it's actually a bit risky and you can actually lose money and also don't trust random people on the internet who tell you they have some secret business plan that will make you a gazillionaire quickly.
wym
maybe aurora 2.0 is a strawberry with 5r's
Is the person called Blessing in the neuro-sama discord a genius motherbrain?
Well, he is a girl, but not really the motherbrain part. He is more like the technical support, a bit like me but not exactly. You could say he is a girl with a bit of a male personality. What do you think of that?
wut!?
????
Congrats, you're a girl!
actually
i should be able to host this model basically for free
since it runs on a raspberry pi
clone
Ahh, yeah they got some Neural Accelerator too on their Pi 5
i aint using that
i gladly graduated from my male form and entered my last and fine stage: A GURRL
Hmmm, not that bad for SBC
schizo era
are you conneted to it via ssh?
no i call the model via openai compatible endpoint
the model is like
24b or something
What Orange Pi model are you using?
idk the one with 16gb ram
i wanna buy the one with 32gb but i aint got money
I am joking btw
it has a rk3588 tho
There aren't that many variant
Probably the 5
nice i tried to make a pcb once using the rk3588
How did it goes?
not after i saw that i need 1k to just produce the pcb, i was like: maybe i should (only this time) be responsible with my money
who could've knew that 8+ layers with interlayer vias costs that much
fyi it only applies when produced once, in mass production it gets cheap again
Yea, economic of scale and all that jazz
I am more interested in where you found the supporting part like RAM and some arcane IC for signalling
Made a basic little function that calculates and displays the number of permutations that add to each possible number when rolling a set of dice 
See the example here with a pair of six-sided dice

Nice
who here knows C
Dragons
use rust
answer me fools
The LLVM logo is a dragon right?
close enough
Best part is, it's generalized such that it can accept any finite number of dice of any finite number of faces, up to the limits of the interpreter
For example, here's the same information for a standard seven-die set:
need help with something?
But also uorms, youre here long enough to know that just ask
i just took a C quiz and it told me i got an answer wrong
you guys know C or what
Share it with us and let us cook the quiz together
how would you answer this question? ```
When a variable is created in C, a memory address is assigned to the variable.
What language are you writing it in?
it's one of those quizzes isn't it
That's not even a question wtf
APL
The Dyalog APL dialect, to be more precise
You'd get along with chay lmao
Hieroglyph speaking fellow
that would be a correct assertion no?
i don't know what the quiz means by "assigned"
Oh hey, I think I ran into 'em before
assigned is a weird word to use here tbf
the correct answer is that the value is pushed to the stack
yeah and it gets an address on the stack 
well, that is what i think the answer is
the quiz says it is assigned an address
is "assigned" the right word though
there are weird exception as there always are in C so the answer would be no but thats being pedantic
@stark needle is your brain braining again?
Assigning address to something is the correct term I think
isn't that also inaccurate? it may just be in a register or may not actually exist at all at runtime
idk someone check the C spec to see what is defined to happen in the C abstract machine
when i check the disassembly of this ```c
int main(void) {
int var = 16;
var += 16;
return var;
}
wait a becond
https://godbolt.org/z/GEc1hM6qP
main:
mov eax, 32
ret
This is the kind of quiz where you need to actually read the C spec 
To separate out implementation detail and the actual spec
dude i really don't understand what the quiz means by the word "assigned"
how low level are we basing the answer off of
register int i;
Tells the compiler to store the value in a register and even if it decides not to you cannot take the address of this variable

the C spec seems to talk only about allocating storage and its alignment, not about any particular address being assigned
i don't even have the C spec
a lot of things are implementation details
There it is then. Address assignment is implementation details
this is why abstractions are a pain, i just use machine code 
silly questions they put in test to have a "gotcha" moment. Like asking for the usable address space of x86_64 cpus
Why would they do that
you technically can use all 64 bits, just remember to clean it up when you dereference 
Bro, I thought that was a question from uni or smth
not even technically
guys w3schools is way more important
W3 schools used to be good
I remember downloading it to have the offline version on my pc from my local net cafe using httrack
downloading it?
what
assuming the question is about "when you use some variable in C it has a particular address" this is false because it doesn't have to have an address when compiled unless the value has any pointers
In todays lingo, scraping
because the compiler is allowed to use registers without using the stack
that's true
I am surprised the site is still up
https://www.httrack.com/
HTTrack is a free (GPL, libre/free software) and easy-to-use offline browser utility. It allows you to download a World Wide Web site from the Internet to a local directory, building recursively all directories, getting HTML, images, and other files from the server to your computer. HTTrack arranges the original site's relative link-structure. S...
And still the same old design
why would you do that
I was at a time where the internet is scarce
like, does the quiz mean the address of the value is stored somewhere?
the usage of the word "assigned" is confusing
ah, I see (how early in the internet are we talking)
i think it's asking about whether a variable has an address when created
webscraper
Sometime when dial-up is relatively new
what does "created" mean though?
the question should really just ask if variables always have addresses instead
the whole "when created" thing just makes it unnecessarily complicated
like, what if a compiler decides to assign an address on the first pointer creation instead
when created here just means "by default"
they should put that in the question then
-# and it would still be unclear imo
holy hell that is old I might not even exist then
Yeah, it's just because of how technology is spread. Because I think at that time, it should be quite widespread in the western part of the world
it was starting to spread here as well but everyone had to access internet in those internet cafes only
i had a dial up until I was two as well I think then we just got an ethernet
yes, the quiz had 24 C syntax related questions and 1 vague low level question 

cross compiling is so fun
its fine i'll just disable xwayland how useful xorg could possibly be
we wouldn't have this without advancements in computers
https://www.youtube.com/watch?v=bn-kISnmKM0
yes i figured out the setup
The madlad finally done it 
That clap at the end, I really felt that as the project is finally done lol
it was ironic because like
gemini puts out tokens so fast it fucks my cursed insertion method

look at the text in discord, it's straight up missing parts
Yeah, I saw that it is not the same length lol
- 1 cup chocolate chips (milk, semi-sweet, or dark, to your preference)reheat your oven to 375ยฐF (190ยฐC). Line baking sheets with parchment paper.
2. In a large bowl, cream together the softened butter, granulated sugar, and brown sugar until light and fluffy.
6. Fold in the chocolate chips.lespoons of dough onto the prepared baking sheets, leaving space between them.
8. Bake for 9-11 minutes, or until the edges are golden brown and the centers are just set.
sure this works
@hard raptor, your product request is done

nah the recipe is good what do you mean you can't do the rest of the steps on your own
. - 1 cup, interesting I didn't know we had easy access to antimatter
though I have to ask why is a chocolate chip cookie recipe the bad apple of ai now
it's so broken tho
the cancel button doesn't even work i didn't implement it 
You think that has stopped game developer from releasing their game before?
It is what it is
i have no idea and i didn't even realize, first thing that came to mind while recording
it may or may not have involved nails and scrap wood
nah it's just become the default to spot a gpt wrapper comment warrior or to test some llm these days for some reason
only 01616 shards missing (apprx. 4TB). after that i can finally train my hubert / wavlm hybrid. anyone trained rvq models or speech recognition models here?
yo what the hell are my seniors doing dawg ๐
Ahh, standard uni stuff
wtf does this mean
"suitable recipe for TextEdit"
is textedit like vegan or something
Yeah, that too. Why are you using reasoning model dude 
because it's insanely stupid otherwise
the effort is set to low
it reasons for maybe like 1-2s
i figured out why the whole recipe showed up at once in the end
it decided to do the content first, then the action
is there no way to force order the json schema, outside of putting that in the system prompt
i probably should
Hi
any cursed ideas this time?
Did I tell yesterday that the Disk Usage meter is not working?
Oh nothing, Just trying to record the Video signal of an AV output
But I'm also thinking if I wanna do that
So I'm in a limbo of wanting to do that, but then not doing it
Anyways, How's going?
i dipped after 40 min lol
friend asked to hop on vrc
im back now tho
Hi VK_AWA_SamVanMaele.dll

Whatcha coding?
today is my cheatday so not much
understandable 
Wym "finished"?
this
but it's a remix of that, it's a keyboard that you can only vibe-type with
or the opposite of that i guess

Wha?
watch
I had so much hope 
i hope someone puts that into a standalone keyboard one day
You know what, the bouncing animation is neat
Q: Why?
I really am trying to see the positive side
until an enter keypress is hit but after generated code
-# 
It's... sigh
It's a keyboard... Why would the input need to go through an LLM?
don't even know if that'd work or if you'd want to bother with it
Well, the upside is that you can ask for soem troubleshooting ngl
but that's annoying
it actually kinda worked that way once, i asked it to respond to a convo here
it output this
Oh 
I'm going to maybe get in to assembly coding 
Nah, wanted to add a
state to the bar.
I have finished writing
then have it switch to fumo once you hit enter or type out another prompt.

i missed so much lore from outer wilds 
hmmm, I think that I'm thinking 
it might lead me to code in 6502 Assembly
or making a 6502 emulator
hblang performance improvements
hblang?
quick PSA to never ask people to stop pinging you
oh my sweet summer child, run before they attempt to indoctrinate convince you to use hblang
Is that something used on Python?
LMAO
hblang is a programming language
OOOH, Yeah no.
@tender river you have some work to do with this one
no work
It's a miracle that I still code on that one project
awa
awa
Like, I'm just motivated enough to code in Python 3
common theme in #programming
Honestly, I don't feel like rushing the code for one singular button that loads the settings live.
#neurotic-neurons is way better at not working tho
Wait what?
humor
alpha=1
.exe?
Yes
but it's linux
I've trained vocal synthesizers and RVCs
Doesn't matter, it's Blender
what
mv blender rend.exe 

It wasn't even me that put the "rend.exe" there
I only put Java 21 and the Render client there
then probably a c# (or i guess dotnet in general) app
they for some reason name their executables .exe and shared libs .dll on all platforms
Idk what Blender is coded in
are you sure it's actually blender
Yup, I'm 150% sure that it is Blender
Since the Render Client Exclusively renders Blender "Projects"
and not a third-party calling blender as a library
great, I want to train some
Of which?
It just Blender that has been started with the renderclient that runs on Java
but currently stuck in getting good embeddings
just googled "rend.exe", guess what it is 
sheepit client
(renderfarm)
Yeah it is sheepit
i dont know what to use for rvcs what is sota?
so not blender 
but this
@nocturne olive
So then riddle me that: People send in files that are straight up only on Blender usually?
RVC is the architecture, the exact base model you use doesn't matter if your data doessn't suck
RVC training is usually done with Applio
__asm__ sucks so bad in gcc
i don't understand what you're talking about butplease google what a dynamically linked library is. i'm suggesting the sheepit client uses that to wrap blender
genuinely it was infinitely less effort to just write some nasm and then link it after
relevant performance stuffs
The rend.exe is a seperate process... (On Linux and Windows):
i. feel like i'm losing braincekls
TLDR: It is just normal ass Blender
im mostly interested in why std memcpy in c is about the same performance as
memset_rep_stosb:
mov al, sil
mov rcx, rdx
cld
rep stosb
ret
Yes, I even opened the rend.exe
ain't there other ons out there that might have better performance?
what is the rend.exe's file size
Who knows, but RVC for a voice changer is usually sufficiently flexible
superbox when does RVC 2: Electric Boogaloo come out
awa
95.998.608 Bytes
I think we're already on RVC 2, at least I believe I saw a selection between RVC 1 and 2 before training somewhere
that's crazy
so it is actually blender and sheepit just renames the binary
Yeah
or is it.
I would say so
somehow unironically this
I think it is actually when it downloads Blender that it is already named that way
alr
how hard can making a physically based synthesis synth be 
@opaque sigil recompiled with musl-gcc rather than gcc, if youre interested

here is the asm musl spits out with
musl-gcc -no-pie -O3 -march=native -mtune=native -static
those are numbers
its a benchmark of various memsets
what's the std one, c++?
c's string.h
ah
what i find funny is that this huge function is just than just
memset_rep_stosb:
mov al, sil
mov rcx, rdx
cld
rep stosb
ret
why even do vectorisation if you can just do this
bwabwabwa
But I think that is too what happened, But it works, doesn't it? (even when it is the EXE)
linux doesn't care about your file extension unlike windows so it would work yeah
unless it's a PE executable and they run this via qemu or sth, that'd be kinda funny
or i think wine can register as a handler too
buh musl are using rep stos
in there
so why is there a bunch of other garbage
Well, I think it is a qemu VM with KVM Virtualization...
I did not install wine btw
ohhh
then top would display qemu no?
PS: wool
it must be filling a 64 bit buffer and then doing stosq
you can just run file rend.exe to find out what type of exe it is anyway
probs
Well, it is that:
just learned fjcvtzs exists
Because on Linux I was not a... Wait a second.
lol
yeah it's wild that there is an ARM instruction with the word javascript in it

Rarely used outside V8/JavaScriptCore
It took a hot minute, but there:
rend.exe: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 3.2.0, BuildID[sha1]=853fe714333c7a36a7c337f5e0e502b9133b3ecb, stripped
then yeah it's just renamed
Yeah
Well, that is not too bad:
For the server having 32GB of RAM
And a AMD EPYC 7443
Well, it only has 8 Cores 
the server I have
there's so much unnecessary shit in arm
x86_64 would like to have a word
"ok let's build our arch entirely around super-specific microoptimizations"
but it works so i'm the dumb one probably
Just for kicks and giggles I download the rend.exe that was downloaded on the Linux server
arm instructions are relatively cheap so i guess it makes sense that they'd add a bunch of specialised ones
may as well
True
can someone explain why and when rbit is needed
literally reverses the bits in a register
the jump from 9k fps to 10k fps is smaller than from 5k to 6k cuz of how fractions work, yet its still harder to do since everything is already quite optimized 
do you know how expensive it is to reverse bytes in place normally
like at trhis point im changing doubles to floats and i get 200 fps extra
why do you need to do that in the first place
lots of things, including compression, use byte reversal
i vaguely remember having to do this a couple times
its a useful instruction
its also used a lot in hashing and prng
// https://stackoverflow.com/questions/2602823/in-c-c-whats-the-simplest-way-to-reverse-the-order-of-bits-in-a-byte
unsigned char reverse(unsigned char b) {
b = (b & 0xF0) >> 4 | (b & 0x0F) << 4;
b = (b & 0xCC) >> 2 | (b & 0x33) << 2;
b = (b & 0xAA) >> 1 | (b & 0x55) << 1;
return b;
}
8 bits
unsigned char reverse(unsigned char b) {
return (b * 0x0202020202ULL & 0x010884422010ULL) % 1023;
}


% is vile
yeah don't use this
it's 1023
anyway, rev reverses byte order (converts endianness)
rbit reverses bit order
very useful instructions
yeah converting endianness i very much understand
in x86 there is bswap for endianness
there's no rbit though
you can use pdep and pext
bswap eax
pdep eax 0x01010101
pext eax 0x80808080
or something
Nope...
i doubt you will ever see this in an actual compiled binary though @rigid snow
yeah looks hacky
nah its very much a normal level of byte manipulation for compilers
its just not worth it for gcc or clang to add peeps for such a thing
yeah it's too big for discord
the main reason is detecting bit swap



out of 1500 or smth


