Foundations - Learning Graphics Programming with the Zig Programming language

[delicate delta]

made my camera and grid reusable

[delicate delta]

I like how I can just give a function a type T as an arg and then start calling functions on a variable of that T without having to define any kind of interfaces, just pure ducktyping

it is just automatically inferred at compile time what the constraints of T are based on how I end up using it

[verbal mason]

I don't like C++ all that much, but it just gives me the most control

[delicate delta]

you get a lot with C++, it's pretty robust I kind of wish I was writing C++ sometimes, like when I realized I had no good ecs options

tons of stable libraries

all the editors and debuggers work

all the books and websites use C++ examples

lots of people on the server can offer advice, although I can get that on the zig server

despite all the problems though I just really like zig

I've never once had a single footgun or gotcha moment in nearly a year of using it

[verbal mason]

I started writing my own ECS, it's actually not all that hard (at least for the basics)

ended up switching to flecs just to save time, cause it was fully featured, but for some simple stuff it's not more than like 1 week of work

[delicate delta]

yeah I am going to basically write something of a game engine a little bit as I go along

not all at once

I'll try to keep in mind how I would use an ECS and try to build something similar as I go so I don't get into too much of a mess, and once it performs badly it'll be hopefully isolated behind an API and I can change how that part works without having to rewrite everything

writing my own pseudo ECS or making zig bindings for flecs are basically equal amounts of work and if it came down I'd rather just write the zig bindings for flecs though

[delicate delta]

It took me like around 2 minutes to build a brand new scene with a grid and a move around look at camera

https://github.com/btipling/foundations/commit/86b58713ea6bf0503c5ab4421d1a4bc6537bd2fd

just that little amount of code there

also I caved and did it camel case :<

now I need to figure out what an infinite plane should look like

[delicate delta]

I couldn't help but keep working on my camera

it now has smooth deceleration, I think decent mouse sensitivity and behaves really nicely, and adjustable speed

it's pretty nice now, fully time step and using integration

[ivory verge]

: )

[delicate delta]

I added coderabbit to my repo and it gave me a nice diagram of my scene

[delicate delta]

not doing any math just yet

I need to reread the plane math

[delicate delta]

this isn't right, I'm just winging it instead of reading. I need sleep

[verbal mason]

idk what's supposed to happen but looks wrong

[delicate delta]

trying to just create arbitrary plane position and normal orientation

via the parameterized plane [n|d]

this is an infinite plane

but drawing it like that looked not great

it'll make more sense when it's right

I am going to put a little ball at an arbitrary 3D point and have it calculate the distance to the plane and which side it's on

and I'm also going to do an arbitrary plane reflection

so move and rotate a cube around and it will be reflected across the plane

the controls are wrong, it's possible to create a zero vector via cross product, I have to just reread the plane stuff

[delicate delta]

I need a little pointer thing to tell me which way a plane's facing

so I'm building a cylinder mesh and a cone mesh, and will combine them to create a little pointer object

not a cylinder yet, but it will be :O

not a cylinder yet, but looks kinda cool

also not a cylinder

[delicate delta]

yay

now I need to make a cone

[ivory verge]

i really enjoy watching this journey 🙂

[queen hinge]

Those colors are supposed to be the normals?
If so, doesn’t look right to me—the bases should have a single normal vector, not a gradient

[delicate delta]

There’s no base

[queen hinge]

OH

[delicate delta]

It’s just open at the ends

[queen hinge]

Yeah that makes a lot of sense

https://www.khronos.org/opengl/wiki/Calculating_a_Surface_Normal
Also this is my goto for surface normals
Though I’ll also often actually manually assign normals

[delicate delta]

I have circles I could use, this is just the cylinder sides

I calculate surface normals via tangents which I get by subtracting points of a triangle

I take the cross product of the tangents as the surface normal

Looks like that’s what that also says

[queen hinge]

Sounds like what that page is doing, yeah

I go for that page because it’s really easy to find it
So for that like one time every couple moths that I start a new rendering project I can just google it

[delicate delta]

I just sit next to a mountain of math books and use those:P

The point of this project is to build reusable code I can just give it three points and get all that for any future projects

Like all these shapes I can use in the future once I have built them

I am going to delete my cube now that I have my parallelepiped actually, it’s way better

Cubes suck tbh

[queen hinge]

I tend to prototype stuff in minecraft, so doing shared code is a bit annoying sometimes

Especially considering I tend to have like 1% of my code even being able to be shared

[delicate delta]

How is your project going?

[queen hinge]

My game engine?
On pause
Currently trying to make an LLM model

[delicate delta]

Oh nice

I use those for work

[queen hinge]

I got my LLM model to something that could loosely be considered as learning

Aka before training it generates random tokens and after training it generates <finetune_right_pad_id> over and over

[delicate delta]

That’s cool i

haven’t looked into that

[queen hinge]

Granted my dataset is also unrelated to my usage and I don’t think a suitable dataset exists

(Multiuser conversations is what I’m trying to get working; LLama3.1 is somewhat ok at those but it’s noticeably not meant for it)

[delicate delta]

Sounds expensive tokenwise

[queen hinge]

<|begin_of_text|>
<|start_chat_header_id|>User<end_chat_header_id>
Hello! How are you?<|eot_id|>
<|start_header_id|>instruction<end_header_id>
You are a chatbot, designed to respond to the user.<|eot_id|>
<|start_chat_header_id|>assistant<|end_chat_header_id|>
I’m using a prompt format derived from llama’s

[delicate delta]

Neat it’s too close to what I do for work these days

[queen hinge]

Hm?

[delicate delta]

It would just feel like my job to do anything ai related

[queen hinge]

[delicate delta]

I work with azure ai studio and copilot all day now

Writing prompts and state machines for helpdesk support

[queen hinge]

I have plans to try to work around context length; my machine isn’t all that beefy and I’ve been running into the context length limitations of LLama3.1 8b a lot

[delicate delta]

That’s what I was thinking I would do

Summarize context

[queen hinge]

that summary step is what’s running into context length limits

It’s also extremely incoherent with names

Which is
a huge problem for a discord chat bot

[delicate delta]

I have never run a model myself

Well I tried and my fan came on and I exited

[queen hinge]

Basically
My thought is
what if I use up to 4 NNs to encode the context into a fixed size block of data

[delicate delta]

Neat project that’s way beyond my level of understanding. I got enough trying to learn geometry and graphics :P

[queen hinge]

Also
I hate admitting this
But python’s debugger is kinda neat

[delicate delta]

I love jupyter notebooks only time I use python anymore

[queen hinge]

Step over doesn’t murder performance

I’m using HF Transformers for my AI stuff and I’m kinda doing it for a discord chat bot
So my options are either:

• use actual python scripts
• transpile python to java (pretty sure I need to write the transpiler myself)

I am absolutely not writing the discord portion of a discord bot in python, that sounds like a nightmare

[delicate delta]

Yeah I agree

[queen hinge]

I’m currently just using the input/output streams to have java communicate with python and vice versa
which is, not a great solution but it does work

[delicate delta]

Like http2 or bare socket or tcp

[queen hinge]

Those would be the proper solutions, yes

[delicate delta]

Lol

[queen hinge]

I’m using print() and input()

Console IO

[delicate delta]

Ah

Hey if it works

[queen hinge]

I mean so far my program hasn’t crashed in the like
probably already hundreds of messages I’ve had it generate

[queen hinge]

Every time my program starts generating a message
If I’m playing a game when it happens
My game reduces to like 1fps

[delicate delta]

You could spin up a small k8s or docker compose if you wanted something fancy

[queen hinge]

I think I need performance

The game also reduces the program to like 1 message per minute

Which is, another reason I’m trying to make my own model

[delicate delta]

Forgot we’re talking about gpus when I started thinking about services

[queen hinge]

I mean the game I’m referring to is an unreal engine game
And unreal engine is a large GPU hog

[queen hinge]

Also
Worth noting
With LLama, when you go over the context limit, the output is a lower quality and it takes far longer
Which is, not necessarily desired when I’m pretty sure I’m above the limit like 25% of the time

[delicate delta]

Oh interesting hrmm

They do get slower

Yeah even the accuracy thing I noticed too in larger contexts

[queen hinge]

Transformers has a warning when over the length limit
Different models respond differently to the context limit according to said warning
What this means exactly? not sure
So far I have yet to go above… I believe I currently have the limit set as 4096

For some reason I made it so users can upload multiple files to my system at once
users can upload like 5GB of data as a prompt if they have discord nitro

Users can also upload PNGs, which seem to deadlock response generation

[delicate delta]

Do these models have raw embedding extraction options, like just the vectors, for storing contexts

[queen hinge]

Unsure what that means

[delicate delta]

Embedding models let you extract embeddings which are like huge multi dimensional arrays of data

But it would be a more compact format for maybe

Storing conversation context for example

[queen hinge]

I’m not storing conversations, actually

[delicate delta]

I don’t know about these small open source models

[queen hinge]

Based on what I’ve written (with, more than a small amount of assistance from code copilot on chat gpt’s interface)
There’s no way to do that with huggingface models
The embedding layer is part of the forward method, as well as the entire rest of the generation process

[delicate delta]

https://docs.llamaindex.ai/en/stable/module_guides/models/embeddings/

Basically you store shit in binary form, it’s a black box that we don’t really understand

[queen hinge]

That page does not like safari

[delicate delta]

Human brain too small

I loaded it in safari lol

[queen hinge]

It’d be more reasonable for me to store tokenized data instead of embedding data tbh

Which actually I think would be ~40-75% compression ratio
But also I’m using json and I think json would freak out with that

[queen hinge]

iphone safari or desktop safari?

On iPhone it loads the page
Then abruptly reloads
Then abruptly errors

[queen hinge]

I’ve already had a user happen to create a prompt that would get exponentially larger every time I loaded the program
Json is funky with special characters

[delicate delta]

iphone safari

[queen hinge]

first coherent sentence my LLM has ever generated:

I have a problem

[delicate delta]

:S

shut it off it's not a good signal

[queen hinge]

I’ve already shut it off because it was gonna take 5 hours and I kinda don’t want to wait 5 hours

[delicate delta]

you should get back to graphics programming

[queen hinge]

But I wanna innovate and doing that in GP is difficult due to how long GP has been around and been a focus of things-

[delicate delta]

to innovate in GP you have to get really specific

like that fog guy that works at unreal

he's got like a PHD in fog

[queen hinge]

More specifically I also wanna innovate for low to middle end devices
There has been, little to no focus on high end models for low end devices afaik

Oh hey that thing you said about storing the embeddings
that’s a viable idea for one of the things I’m thinking of

So
Considering I want this to be a high end model for low end devices
I kinda, don’t want to actually train everything into it
I’d rather have it deal a lot with what it’s given as inputs, and be able to have it request extra additional inputs for a given topic to substitute the extra training data

[queen hinge]

I think it’s time I mess with activation functions tbh

[delicate delta]

nice

[delicate delta]

this was a tough one

[queen hinge]

my brain can't comprehend the open bottom

anyway, how's that mo-
oh
I see how that's more c-
no I don't-
do I

yeah I don't see how that's more complex than a cylinder

[delicate delta]

it's a smoll brain thing I guess

with a cylinder you rotate all vertices of a quad to assemble the cylinder

with this cone I walked the bottom triangle vertices around a circle

and each triangle's first bottom vertex starts at the previous triangle's bottom vertex

I think cylinder is much less complex

it's just one matrix per quad

[queen hinge]

I think I'm being hecked over by huggingface transformers works

height = 1;
radius = 2;

for (double rad = 0; rad < rad_360; rad += rad_1) {
  double x0 = cos(rad) * radius;
  double z0 = sin(rad) * radius;
  double x1 = cos(rad + rad_1) * radius;
  double z1 = sin(rad + rad_1) * radius;
  
  addTri(
    x0, 0, z0,
    x1, 0, z1,
    0, height, 0
  );
}```

I do, operate on a per triangle basis by default

so I am also biased by that fact, since a cylinder is 2 tris per face

[delicate delta]

you don't think that's more complex

[queen hinge]

than doing the same thing but with 2 triangles instead of 1?

[delicate delta]

I am not so good at math that I can just look at trig and visualize what it does unfortunately

that looks like one triangle per loop but, I have smoll brain

[queen hinge]

I literally wrote this out in discord, lol
sin combined with cos creates a circle

sine on one axis, cosine on the other
doesn't matter which is which, it'll create a circle

[delicate delta]

right

[queen hinge]

that's the only bit of trig you need to know for a cone

[queen hinge]

mhm
with a cylinder it'd be two triangles per loop

[delicate delta]

oh I see what you're saying

[queen hinge]

height = 1;
radius = 2;

for (double rad = 0; rad < rad_360; rad += rad_1) {
  double x0 = cos(rad) * radius;
  double z0 = sin(rad) * radius;
  double x1 = cos(rad + rad_1) * radius;
  double z1 = sin(rad + rad_1) * radius;
  
  addTri(
    x0, 0, z0,
    x1, height, z1,
    x1, height, x1
  );
  addTri(
    x0, 0, z0,
    x1, 0, z1,
    x0, height, x0
  );
}```good chance I have this wrong

[delicate delta]

cool

[queen hinge]

actually kinda curious how scuffed that cylinder algo I just wrote it

but I don't really wanna setup an entire renderer just to test it

[delicate delta]

idk man it was hard for me, but you did it in 2 seconds in discord 👍

noice

obvious in retrospect that is correct

[queen hinge]

What's for lunch?<|eot_id|>
<|start_chat_header_id|>assistant<|end_chat_header_id|>
I'm not? <|eot_id|>

some more training later

What's for lunch?<|eot_id|>
<|start_chat_header_id|>assistant<|end_chat_header_id|>
I'm not have a good. <|eot_id|>

context; these are the best results I've gotten, and it's taken a long time to get to those

I feel like if I make a custom generate algorithm I could probably reach that point in a few dataset entries instead of like 12k dataset entries or whatever

of note; I also do not understand the generate function at all, for I have not yet interacted with it on a lower level than just calling it and debugging some crashes

[delicate delta]

"I'm not have a good" perfectly describes me rn

[queen hinge]

I still love how the first sentence my LLM ever generates was "I have a problem"

also
I don't understand this perfectly yet, obviously
but it seems to me like it's gonna be output count that drags down speed more than input count
so I think I should be able to give it a large input count and small output count for a semi-reasonable speed
or alternatively, I do weird things, which seems like the more fun option

something I could do
is cut my inputs into pieces
cache the embeddings for the inputs into a variable
then when those inputs are next seen, until an unrelated input is provided, I just use those same embeddings
and piece together the embeddings into a single nd array and go from there

I think

[delicate delta]

how many dimensions are the embeddings

[queen hinge]

What's for lunch?<|eot_id|>
<|start_chat_header_id|>assistant<|end_chat_header_id|>
What's a problem? <|eot_id|>
<|start_chat_header_id|>assistant<|end_chat_header_id|>
well that's a new output sentence

[queen hinge]

input_dim * 512

[delicate delta]

what for lunch is too complex of a prompt, ask it something simple, like how to draw a cone

[queen hinge]

that's not in my dataset at all

I don't even know if what's for lunch is anywhere in my dataset
but it seems probably based off the like 2 entries I looked at

but yeah I want this thing to be as close to real time as possible
I'm gonna do my best to accelerate it in any way I can

[delicate delta]

more gpus

[queen hinge]

okay any programmatic way I can

oh I should avoid training the model on full conversations actually

[delicate delta]

well they can run pretty capable models on small devices, it's the training where you need more hardware? idk

[queen hinge]

Well yeah, if you quantize the model it can run at a reasonable speed for short conversations
Problem is, that’s not several simultaneous long conversations

I have
I believe a [high end for home computers] computer, and even that isn’t really good enough for a high enough quality one

Granted I’m also severely limited in the fact that most prompt formats die if you decide to use a username instead of a role, since that’s sorta the best way to prompt-injection-proof naming, and there’s no models trained for that

unsloth 8B models are about the highest I can reasonably run for any kind of performance, and even that’s struggling in both performance and quality

Quantization also tends to slightly reduce output quality or change the output drastically from my messing around with LLama3.1 quants

[granite bolt]

So many text messages and no talk about books

[delicate delta]

I've been in saving my job mode for the last 2 and half weeks, so only have had a tiny bit of time to do any of this, otherwise be boring you with my book reading rn

I usually spend about 4 hours a day on weekdays and nearly all of the weekend working on my game engine when I'm not doing stuff with my family, but it's just a lot of working late nights and weekends right now trying to make unhappy customers happy again

[delicate delta]

I have a pointer

[delicate delta]

I haven't fixed any of the plane rotation around an arbitrary axis and the normals are all wrong also but I am now pointing in the direction of the plane

[delicate delta]

ah I figured it out thanks to renderdoc, I just had to normalize my normals after the inverse transpose

now I just have to fix my rotation

[verbal mason]

yeah, always normalize to be be safe

[delicate delta]

Yup makes sense!

I think I will fix my weird rotation by always rotating around the axes of the plane instead of an invisible arbitrary axis

I will have to convert it to the axis angle though

Well

I have to read the math again it’s just [n|d]

I don’t know why planes are so hard for me

I am going to get it soon

The goal here is to just change the parametric plane values [n|d] and the. draw it based on the values of that four d vector

I am drawing a sub plane of the full plane , the center is the plane’s closest point to origin

Then I will add a sphere and a cube and move them around and find their closest points to the plane and reflect them across the plane

I think I got it

[verbal mason]

cool

[delicate delta]

I fixed a bunch of stuff, I fixed the ridiculous grid scale, I fixed camera, grid and plane positioning, I made pointer look more like pointer, normals are all correct. I'm ready to make progress again

now that the grid scale is actually sane I think maybe the camera sensitivity seems too high again

all of these fixes, fix everything going forward, these are all fixes that impact everything going forward, the camera, the scale of things, the normals of things

these aren't one off fixes for this particular scene, so when I make a game eventually all this will make that game better

[delicate delta]

I added my sphere as an origin marker but it's got some tears in it and issues with normals at the top

the cone normals inverted also

a lot of work to get these normals right 😅

seems correct now though

just the sphere is horrid

that's a little better, now the holes

just bad geometry

[elder kettle]

Kind of a funny looking sphere

But yeah I don't think you want the triangles to align like that

I would generate it basically as if it were quads and then triangulate each one

[elder kettle]

It should be basically the same as an algorithm to generate a grid, except that you take the x and y position of each vertex and use them as lat/long instead

[delicate delta]

yeah

what's funny looking about it?

besides the holes

I made it before I had a projection

so couldn't see all the problems with it then

[elder kettle]

Yeah just the alignment of the triangles, the way the tip of one lines up to the edge of the other

[delicate delta]

yeah

[elder kettle]

that's inevitably going to cause gaps unles the mesh is flat

[delicate delta]

it's just bad geometry code

[elder kettle]

yeah that's all I was referring to

[queen hinge]

Ah spheres
Those are something I don’t dare touch

I favor using a shader to project a sphere out of a quad

Also for spheres
I recommend icospheres, personally

[delicate delta]

this is just a first iteration

I'll probably make a bunch of different ones eventually

[elder kettle]

icospheres are a pain, UV spheres are straightforward

[queen hinge]

I do not like UV spheres

They are straightforward, yes
But I want even point spacing

[elder kettle]

Then you will suffer for it

[delicate delta]

I am going to fix this sphere I guess

[elder kettle]

Might I offer a compelling alternative: a cube

jk the sphere is fine

[delicate delta]

I don't want to spend forever on this right now I'm going to see if I can these holes plugged, I'll make a better sphere eventually

I am going to delete my cube, I have a parallelepied that's a million times better

[elder kettle]

fix it if you like, but cubes can be colored to have the faces represent the XYZ basis vectors too which is nice, it has a recognizable orientation

[delicate delta]

parallelepiped

[elder kettle]

alright

[delicate delta]

oh

I mean it will act as a cube

if I want it to

[delicate delta]

also appreciate your advice thank you!

[elder kettle]

I'm just kidding around too just do whatever you like

I like the vector thingy up there

[delicate delta]

thanks!

[elder kettle]

personally I just use bare lines as debug vectors but that's fancy

[delicate delta]

bare lines are cool, I was also thinking about dashed lines for some debug things

like when I have an AI and it does path finding

I was thinking like a dashed debug line to show where it's going

[elder kettle]

yeah

[delicate delta]

this works

it's got some rough edges, but works for now

I will make something better eventually

yikes

[elder kettle]

There's no escape 🙂

[delicate delta]

this sphere will cut you

hrm lat long huh, I actually am using that

spherical coordinates

[elder kettle]

Yeah

You should just be able to have a double for loop that goes over incrementals of the polar and azimuthal angle

And makes points

And then the index scheme is the same as a grid

[delicate delta]

yeah the holes are gone but this is gross

[elder kettle]

They don't seem to be quite gone in the sense of being an actual manifold mesh they just appear to be gone

[delicate delta]

right

[elder kettle]

Whereas the UV sphere as I'm describing will only have singularities at the poles

And they'll be invisible anyways

It'll only be broken in the sense that the coordinate system doesn't lend itself to calculating a surface normal at the very top point

But actually you can anyways since we can do it analytically

[delicate delta]

I'm getting the normals by just shooting them out from the origin

not using the triangles at all

it looks right though

[elder kettle]

Yeah that's even easier

It should be for a perfect sphere

()

[delicate delta]

how is this not perfect

jk

[elder kettle]

It is, I'm laughing at my own suffering from my non-spherical earth terrain system

[delicate delta]

your terrain is mind blowing, I think it is amazing and if I ever achieve anything half as good I would be pretty happy

[elder kettle]

It's a testament to my stubbornness

And stupidity

[delicate delta]

I am not very fast but I am committed and I do the work. It works for me.

this is not a good sphere but it works for this scene I want to get back to reading so I'm going to do all the plane stuff

I bet one of my many books has a sphere in it

I am going to use a cube actually for the distance measuring

[elder kettle]

Spoiler'd in case you don't want it to be given away but the position part of the UV sphere is just
||```cpp
int stepsAzimuth = 32;
int stepsPolar = 16;
float r = 1.0f;

std::vectorglm::vec3 verts;

for (int i = 0; i < (stepsPolar+1); i++)
{
for (int j = 0; j < (stepsAzimuth+1); j++)
{
float az = 2pifloat(j)/stepsAzimuth;
float polar = pi * (1.0f - float(i)/stepsPolar);
glm::vec3 pos = r * glm::vec3(sin(polar)*cos(az), sin(polar)*sin(az), cos(polar));
verts.push_back(pos);
}
}

I will leave the index generation as a graph paper exercise

[delicate delta]

I was actually writing some code doing quads just now because I couldn't let it go :P

I haven't looked your spoiler yet let me see how far I get

I think my original solution I realized I need more triangles near the equator so I came up with something convoluted one late night

but it doesnt need to be

[elder kettle]

yeah

[delicate delta]

I like the quad idea

[elder kettle]

If you want to be clever you can skip some of the triangles at the top so that the ends come together nicely but personally I just allow a whole row of verts to collapse to a point at the poles and just call it good

at which point it's literally just a grid

[delicate delta]

yeah I'm writing a grid now using spherical coordinates

idk we'll see. I just reduce the grid to very few quads at first it should be somewhat spherical, and simple to debug any issues and then I can increase the grid size and it will look more sphere like

[elder kettle]

good luck

I'm going to sleep but I'll see your results tomorrow

[delicate delta]

gn!

[delicate delta]

already looking better

let me try it with more points

holy fuck

looking great

that was so easy

that took me like 10 minutes lol

I always over think this shit

alright this is my solution, now I will check your spoiler https://github.com/btipling/foundations/blob/2a718120525013bd6661a5183418b2eed03f0876/src/foundations/object/object_sphere/ObjectSphere.zig#L47

ok mine is a bit different, my spherical coordinates do have the same trig though

I'm a bit more explicit in my quad/indices, I can't quite visualize the quad/triangles in your solution, though I can see the points as a sphere

no more sharp edges

it has the same number of triangles though for each band

it's so pretty though

thanks for your help!

I am so glad I stick with that dumb thing I had before

here it is with a orthographic projection

alright time to add a cube

[delicate delta]

these normals are looking great man

I got these things rotating around their centers now too

I wonder if discord is unhappy how much I upload

[ivory verge]

why is your x up?

are you an xup lizzard?

[delicate delta]

so I can make a finger gun when I am thinking about my coordinate system

[ivory verge]

: )

i pictured it immediately hehe

[delicate delta]

I like this scene so much I made it a banner on my githubs

[elder kettle]

Mine just places the points and then you just generate the indices to define the tris

Someone stayed up late hehe

[delicate delta]

Makes sense

I am duplicating vertices I will fix that, it was late

[elder kettle]

Jeez someone's up early too lol

[delicate delta]

Yeah would be better to sleep more

For reflecting the cube across the plane, I am just going to draw triangles, i have only read about the reflection math for points so far. This scene is just about basic plane math

I will just iterate over the points in the cube, reflect three at a time and literally just draw triangles just for the demo. I hope to wrap it up today

Then work on a frustum plane extraction culling scene

Maybe I can get both done today

[elder kettle]

I can't remember off the top of my head what it entails in general

I think just reflecting all the points works

The only potential snag is that it may require reordering the indices to change the winding order

Or just flipping the cull face direction

[delicate delta]

Yes, I will find out. Hehe

[delicate delta]

point in plane closest to origin

AI giving me PR reviews (I hadn't written tests or used this yet so no compile error heh)

[delicate delta]

I got closest point in plane for object working

just need to reflect that cube across the plane and am done with this scene

[delicate delta]

I got the distances working too

it was actually so little code to do this https://github.com/btipling/foundations/pull/13/files :/

[delicate delta]

I want to theme my dear imgui tbh

[ivory verge]

please do 🙂

im curious with what you are going to come up

[delicate delta]

probably won't do it right now I want to get back to reading I'm just tired of looking at it

feels so vanilla, everything else I put so much work into it

[elder kettle]

You could use someone else's

I like the old school steam one

[delicate delta]

yeah that sounds cool even just to look at it to see how theming is done

I'm working on reflecting finally

I know these things aren't games but they're pretty fun to work on

feels great when you see the math work

[elder kettle]

You can theme it just be opening the demo window and tweaking the properties in the theme editor

And then export

[delicate delta]

oh

[delicate delta]

reflection!

[queen hinge]

#wip message
I do always come back to GP stuff though

[delicate delta]

looks great!

[ivory verge]

thats the green one demon was talking about 🙂

[delicate delta]

back to working on my sphere

I'm going to have a reallly nice sphere

100% worth it

[delicate delta]

it has a single point on top and bottom

it's amazing

[delicate delta]

for my plane extraction I was thinking demoing a little voxel scene with my parallelopiped instances via stb_perlin

the failure to extract planes from my frustum was the reason I stopped working on blockens, this will be a pretty feel good scene once I get it working

[queen hinge]

Oh is this using index buffers?

[delicate delta]

yeah

it's one buffer

vao and index data are in the same buffer

[queen hinge]

huh I have never considered sharing a buffer for the two

[delicate delta]

oh maybe I'm lying

I'm sharing instance data and attribute data

not indices

sorry

[queen hinge]

Also
One of the problems with using a game engine instead of making your own and just hardcoding everything
There’s a chance that your game just decides to corrupt

[delicate delta]

:S

[queen hinge]

Godot crashes ~5 seconds after I open my project, and my fog volumes corrupted

[delicate delta]

I've never used a game engine

nfi how they work

[queen hinge]

So this is actually probably my least favorite sphere meshing algo in terms of sphere aesthetic

[delicate delta]

I love it

although

this is all I've done with it

maybe uv mapping or lighting or other stuff is harder

I have no idea how to do any of that yet

I mean compared to what I had before, which was just broken, I like this for sure :P

I wrote my own trig for this one :O

[elder kettle]

same

[queen hinge]

UV maps are pain with that one and normals and tangents and binormals get a bit wonky at the top and bottom

UV spheres are the easiest spheres to make
They are also the easiest to UV map

Icospheres have an even point distribution (these are my favorite)

The one you did is also a bit too mathy for me, given the fact that I really don’t see any potential advantage to it

[queen hinge]

Lighting is a pretty universal algorithm
What works for one object works for all, until you get to shadow maps

Well
I would say stuff like specular and metalic and stuff but I actually don’t know anything about those

[queen hinge]

I can tell godot also uses this sphere geometry, due to the top of my bubble being a bit wonky; I’ll probably make an icosphere in blender at some point or smth ig

[delicate delta]

looks amazing

icospheres look really cool

I'll probably give that go at some point

[delicate delta]

took me like two minutes to add stb_perlin noise result 0.984375

hrmm

[delicate delta]

ezpz

I like my camera

[ivory verge]

2d equalizer 🙂

[delicate delta]

2 cameras

[delicate delta]

plane extraction from frustum and culling:

[delicate delta]

Back to reading! I got about 5 books that each have a perspective chapter to get through now and then read the 3D math review chapters in the books like GEA and then I am out of the math kiddie pool and on to actual graphics programming topics instead of basic math and geometry

[delicate delta]

I think I might just drill down the opengl book and get through it after I get through these chapters, I'm kind of blocked less on math at this point and more on how to use OpenGL, like I don't know anything about attachments or separate frame buffers or much at all about opengl textures beyond the basics

if I get blocked on math somewhere along the way I'll switch back to the math books, probably on lighting, I don't know anything about lighting or shadow math

if it's just basic vectors stuff though I should be fine

[delicate delta]

I have culling now working with two different methods:

1. getting the planes via the linear combination of the camera matrix columns and the the focal length and aspect ratio
2. by extracting planes via the transpose of the full model to view perspective transform, including the look at and the x up coordinate space transform

they both work

I literally pieced the math together from three different math books

They all had a different piece of the puzzle

One book has a zup coordinate system, one has a yup and the other has a camera y down world y up coordinate system

None of them have an xup

One book is left handed the other are right w

[elder kettle]

Left handed is cursed idk who is using that unironically

[delicate delta]

None of them explain how handedness and up affect the plane extraction math

I am lol

[elder kettle]

oh no

Anyways it just puts a bunch of sign changes in random places

[delicate delta]

It is good for me forcing me to learn more

[elder kettle]

I would really suggest using right handed though that's what like all of math and physics uses

but fair enough

[delicate delta]

Yeah these books just don’t bother to explain how the math is different

The 0 to 1 Z NDC also was a trip

[elder kettle]

I would have to derive the rotation matrices and stuff

maybe it doesn't affect the basic transforms that much

but it would really fuck with my head so I'd prefer not to think about it too much lol

all of my years of math education have been in RH coordinate systems so trying to do it in another is like trying to read a foreign language

I'm sure I could figure it out with some conditioning though

[delicate delta]

I think I kind of am stuck with my left handed x up

Like I am conditioned now by my normals and constantly having to map y up book math to x up

I Actually really like it

Y pointing into the screen is great

x and y are up and forward, z horizontal

[elder kettle]

I have Y into the screen too

I just have X to the right and Z up

Typical ENU (east, north, up) coordinates

[delicate delta]

it's mostly just contained to a single matrix concatenation, but the plane extraction was much harder because of my weird coordinate system

[elder kettle]

what is plane extraction

idk what that's referring to

[delicate delta]

from the clip space

getting the planes that make up the surfaces of the clip space volume

into camera space

[elder kettle]

ah

oh for frustum culling

[delicate delta]

and then into world space

[elder kettle]

I see

[delicate delta]

yeah

so the easy way is to just take your camera matrix and do the linear combination thing

doesn't even involve the perspective matrix

like the left plane from the view frustum I extract via adding the second column of my camera matrix multiplied by the focal length and third columns of my camera matrix multiplied by the aspect ratio

the other way from clip space it's the addition of the inverse of my mvp transform's fourth column by its first

and then inversing the plane distance

I just didnt realize how much harder that would be because of coordinate system decisions

[delicate delta]

it's interesting how the planes are similar but not equal

also my sphere has too much geometry to instance spam it

I need like a sphere LOD

[delicate delta]

The projection chapter in https://www.mathfor3dgameprogramming.com/ is ancient

it uses glFrustum, uses a height/width aspect ratio where the plane of perspective height is the +/- aspect ratio, and mentions that some older GPUs don't do perspective correct interpolation

it also uses the -1 to 1 Z clip space

the math is still correct and interesting though

published in 2011

discusses a reflection perspective matrix also, so not that old I guess

perspective matrices are pretty complex and I understand now why they're so far into these books, people don't want to wait to get to 3D but in a math book you have to work through the math because if you're starting at 0 it's not going to be understood

the whole viewing in 3D is pretty complicated objects get moved to world space, then get moved into camera space via a reverse camera matrix, then projected into a view frustrum onto a projection plane into a clip space which is not 2D, you're still 3D, then you're in shaders, then you're in clipping and rasterization, and now you're working on pixels that get put into framebuffers

[delicate delta]

Not super interesting I'm sure, but I wrote this in my notes when trying to explain how a projection matrix works just from my understanding:

In a perspective matrix the x and y of a point have to be mapped into the x y coordinates of the projection plane, this is done by casting a ray from the camera through the point onto the projection plane. The aspect ratio and the focal length(the distance from the near plane to the projection plane) determine the shape of the projection plane and thus in the perspective matrix the first two entries in the diagonal achieve this transformation of x and y, in the case of older platforms a linear combination required more columns than just the diagonals for anx and y. This is probably because of the shape glFrustum took. In any case the z needs to be transformed into a reciprocal (1/z) by the projection matrix for the purpose of being interpolated into the clip space depth volume to then be transformed into a value in the depth buffer after division by w. The reciprocal is is also used for perspective correct interpolation of vertex attributes. The z value of the perspective requires the depth scale created by the near and far plane and thus the z value ends up being transformed by a linear combination of two components in the third row (in a column major matrix).

I didn't state that because the projection plane's x is twice the size of the aspect ratio it's multiplied by focal length and divided by the aspect ratio, and the y is just multiplied by the focal length, but I know that. The z projection math is harder to explain, but I think I got the high level of it.

probably the hardest to understand matrix so far

[maiden gazelle]

I like Cem Yuksel’s explanation a lot

It is intuitive in that perspective projection is divided into two steps: perspective transformation + orthographic projection

It was easier for me to internalize that way

I also used a combination of sources to tackle this topic

www.gamemath.com (chapter 10 has a section dedicated to viewing transform),
songho’s explanation,
Cem Yuksel’s explanation,
Bremdan Galea’s explanation

Also took some notes if you want to compare

[wise imp]

i feel like directly learning a bit of the core math works well too

khan academy linear algebra stuff

so that you learn about things like vector spaces

[maiden gazelle]

Definitely

Also, I could not keep up with this thread coming in so late as I did

I’m watching you though beware 🦉

Awesome stuff in the works, kudos!

[delicate delta]

understanding linear algebra is definitely a prerequisite for understanding perspective matrices, but it's not enough in my opinion

even understanding perspective geometry isn't enough, because that's not what the perspective matrices we use in our games do, it's not a simple perspective divide

[delicate delta]

the result of a perspective matrix in games and game engines doesn't simply project a 3D point in 2D, it is a homegenous 4x4 matrix that instead performs a transformation from 4D to 3D, the end result being a 3D point in clip space, not a 2D point, because the depth needs to be preserved for hidden surface removal and perspective correct interpolation of vertex attribute data

so you have to also understand graphics apis and the rendering pipeline, and on top of that you have understand how virtual cameras work and coordinate system transformations and NDC because the shape a perspective matrix takes will vary considerably based on these

an infinite reverse matrix takes into account floating point precision near the far plane, so there's even more

or you can just use the glm perspective function and be happy with what you see if that works for you

[delicate delta]

So reading through Essential Math I realized that I don't need to inverse my camera matrix, although that works, there's a more performant option which is to transpose the upper 3x3 of the camera matrix and then multiply the transform the translation from the camera by that and negate it

pub inline fn cameraInverse(m: matrix) matrix {
    var r: matrix = .{
        .columns = .{
            .{
                at(m, 0, 0),
                at(m, 0, 1),
                at(m, 0, 2),
                0,
            },
            .{
                at(m, 1, 0),
                at(m, 1, 1),
                at(m, 1, 2),
                0,
            },
            .{
                at(m, 2, 0),
                at(m, 2, 1),
                at(m, 2, 2),
                0,
            },
            .{ 0, 0, 0, 0 },
        },
    };
    var vPos: vector.vec4 = vector.negate(transformVector(r, m.columns[3]));
    vPos[3] = 1;
    r.columns[3] = vPos;
    return r;
}

works

I think this is better than zmath's lookAt tbh https://github.com/zig-gamedev/zig-gamedev/blob/1c9461eab81a004bf0e3b3064da2487f57a07d77/libs/zmath/src/zmath.zig#L2264-L2273

I think Lengyel didn't even cover the camera matrix

this produces the exact same matrix as inverse did since the upper right 3x3 is orthonormal

[elder kettle]

Yeah

handy trick

orthogonal matrices are convenient for that reason

[delicate delta]

yes, very handy I agree

more evidence that just one math book isn't enough, you have to read a bunch of them

I guess I should have figured this out on my own though, I know that the inverse of orthonormal matrices is the same as the transpose

but I didn't think about how the translation would be affected

[elder kettle]

I just learned that incidentally from group theory I think

[ivory verge]

@delicate delta last night

you were in my dreams

along with your foundation

i dont rember the details, but some maffematical stuff came up and involved vectors or normals of sorts, and then all of a sudden you were there, and all your pictures and clips were parts of a book you browsed through to show/explain to me what normals (or whatever it was exactly i dont rember) were

background was dark blue

and the pages of the book must have been A3 or so

[delicate delta]

oh no

I'm sorry :/

[ivory verge]

😄

[queen hinge]

lol

[delicate delta]

hey yeah you should be having Lyzantra lecture you on maths in your dreams, as Lyzantra is way beyond me

[queen hinge]

I am NOT a math person

[delicate delta]

still better than me

[queen hinge]

I just know how to make the most of the stuff I do know

[delicate delta]

that's great tbh

what version of Java do you use btw

[queen hinge]

Typically J17 ‘cuz that’s what MC used for a while
Though I flip between J8, J17, and J21

[delicate delta]

I knew you were still using java 8

I think like 90% of people writing java are still using 8. I made that up but I think its' true

[queen hinge]

whenever I use J17+, I regret not using J8, and whenever I use J8 I regret not using J17
MC is the one place where I don’t have to deal with this conundrum

[queen hinge]

Most likely correct, yes

[delicate delta]

is java 8 peak java

[queen hinge]

Only reason I ever regret not using J8 is because it’s what everyone has

[delicate delta]

does minecraft ship with its own jre?

who still has java

[queen hinge]

Basically; the java website only has J8
So if you want to use a newer version, you have to download a JDK from elsewhere

[delicate delta]

I think intelij also ships with its own jre yeah?

oh

[queen hinge]

Yes, but that doesn’t install as a default java to run jars with

[queen hinge]

Yes, but that doesn’t install as a default java to run jars with

[delicate delta]

https://www.java.com/download/why-java-8-recommended.html

that page doesn't answer the question

so this is a java licensing issue

[queen hinge]

oh really?

[delicate delta]

looks like because it went GPL?

[queen hinge]

The newer JDKs are faster due to JIT optimizations and easier to write code with due to newer syntax features and methods which exist to reduce boilerplate

[delicate delta]

I am asking the ai

idk seems like it's mostly that java 8 is just the version everyone has I guess

just as you said

[queen hinge]

According to reddit
Until Jdk9, the developers of java didn’t want people to redistribute the JDK, and this mindset changed with J9

[delicate delta]

this all gives me the impression that java is kind of at a deadend, time to switch to something modern like Mono or Perl 6

[queen hinge]

I mean
Code wise Java’s got more potential now than it ever did before

Also got more active development in the JDK than ever before

[delicate delta]

what's a new feature you're excited about in java

[queen hinge]

They’ve been working on redoing native interface is the big thing they’re working on
Project Panama

I have… mixed opinions on it

[queen hinge]

I don’t really keep track of what’s upcoming, nor am I sure if that’s a thing I’m able to do?
But the new syntax features are indeed quite nice

[delicate delta]

new zig nightly doesn't let me do this any more :(

not sure what to name these props now

config2 and scene2 wdyt

config_prop

config_

config_123

[queen hinge]

those are quite the variable names to reserve

[delicate delta]

they're not reserved

it's that I can't use them as props because they're imported

I can rename either one

[queen hinge]

OH

What is this a config for?

I’d name it after what it configures, personally

[delicate delta]

oh game config like right now just screen settings

well they're both the same thing one is an instance of the other

so it's like having a function parameter called x setting an x prop on a struct inside the function

hrm

no that's allowed

src\foundations\app.zig:1:1: error: duplicate struct member name 'scenes'
scenes: *scenes,

const scenes = @import("scenes/scenes.zig");

I import this at the bottom

[delicate delta]

Chapter 10 of game math is like 120 some pages

it's gonna take a while :<

it's basically a summary of an entire book on rendering

it's a good intro because that's what's next

I just have to power through it

[delicate delta]

I've never read so much in my life as I have in the last few months tbh

[delicate delta]

you can tell when I started getting into graphics programming and zig

[delicate delta]

half way through chapter 10, with notes, covered the phong and other basic lighting models

[delicate delta]

I have finished Chapter 10!

Of game math (gamemath.com)

it is a massive chapter

I actually really enjoyed it

I learned so much, and because of how old this book is, I got a little bit nostalgia and I felt like I was really learning about how to build a game

like I didn't know anything about lighting at all before this chapter, never once read about phong or blinn or doom style volumetric lighting

the amount of detail it went into the render pipeline and graphics pipeline, even though it is a bit dated, was really cool and I learend a lot

a 150 pages that covered like all of rendering at a very high level and ended with some HSLS examples of how to do bitmapping and skinning and lighting

I also really liked that it kind of broke down how games are built, at least how the rendering is built

like what is done where and when

I am actually done with math books for a while. I am going to read three chapters of Game Engine Architecture and then just plow through my OpenGL book, Computer Graphics Programming in OpenGL with C++
3rd Edition, https://athena.ecs.csus.edu/~gordonvs/textC3E.html

because I'm actually like kinda more limited by lack of OpenGL knowledge than anything else before I get into advanced math I can't use because I don't know how to add attachments and whatever

I emailed the publisher for the files for that book, I will actually build all of the examples in my engine

[elder kettle]

Nice

Time to whip up some gameplay

It will require you to expand your GL knowledge pretty quickly

[delicate delta]

nice, that would be fun

yeah everything I read from here on out should lead to some cool rendering I hope, out of the kiddie pool now

[maiden gazelle]

Nice progress!

[maiden gazelle]

Most of the lighting stuff is decades old knowledge anyways, so it might be less outdated than you think

[maiden gazelle]

I have finished and implemented in-engine the first 8 chapters

And the viewing transform part of Chapter 10

I think the book is solid from an education point of view

[maiden gazelle]

Our curriculum's are similar 🙂 I'm also reading Game Engine Architecture on the side. Though for OpenGL I use LOGL. Once my current term ends in this curriculum, I'll pick up Real Time Rendering I think.

[delicate delta]

Nice, it’ll take a bit before I get to real time rendering

I want to finish all the math books first

I am about halfway through most of them now

So probably at this rate maybe next year

idk

I want to read the books I bought they aren’t there to just look pretty

[queen hinge]

To this day
I still know nothing about phong and blinn

I’ve yet to really have a reason to read about those tbh

[delicate delta]

It’s trivial

It’s super basic

It’s cool how basic it is and how well it works

[maiden gazelle]

Solid mentality

You'll be a force of nature next year

Though a little overpowered math-wise

Me personally

[delicate delta]

Nah

[maiden gazelle]

I tend to forget math topics if I don't apply them frequently

[delicate delta]

Everyone is like that mostly I think

[delicate delta]

You don’t need lighting or because you use godot?

Your godot project looked really cool

[queen hinge]

I’ve always just sorta made my own shading model when I wanted to do lighting stuff outside godot

Physical accuracy isn’t really something I desire, I like my stuff being stylized

[delicate delta]

Makes sense

[queen hinge]

In my opinion
Realistic rendering is best for horror games, because it makes it a lot easier to get immersed in it
Beyond that, I don’t generally care much for realism in games, and generally see it being more performance intensive or more file size than it’s worth

[delicate delta]

I like cool lighting, like a cyberpunk scene in which all the neon and digital lights refract and reflect all over and every other thing is itself emissive

[queen hinge]

This is part of why I like to minimax my understanding
A lot can be done with math using a minimal understanding

[delicate delta]

I just write notes

[queen hinge]

There’s not all that many games that actually do that though

also I’ve never actually seen cyberpunk

Oh shoot
I just discovered something neat about discord mobile

[delicate delta]

The Ascent is a small indie cyberpunk game that has cool lighting

[queen hinge]

You can tap on the top of the screen to jump to the top of the forums posts

So I can just immediately jump to #1248864886772072591 message

That’s neat

[queen hinge]

Yea it’s also cool if you design your game around it I guess

[delicate delta]

It is a short game though

It is very fun

[queen hinge]

Actually
A neat mechanic I’ve only seen once or twice

Having stuff react to light

Like
In gameplay
You shine a light on an object and it acts as a trigger
Not just a visual thing

[delicate delta]

That is cool

[queen hinge]

There’s a game with a combat system essentially built around that

[delicate delta]

Metro games do that

[queen hinge]

I have never heard of metro

[delicate delta]

Some of the creatures hide from light and avoid coming near you

[queen hinge]

One I was thinking of is alan wake

[delicate delta]

https://en.m.wikipedia.org/wiki/Metro_(franchise)

I only played 2035 it was great

[queen hinge]

For alan wake
Enemies can have these, essentially shields
Which much be removed by shining a flash light on them

[delicate delta]

The one from 2021

[queen hinge]

Iirc the flashlight can also act as a weapon on its own

[ivory verge]

this looks cozy, settings wise

i dont know the game

[queen hinge]

The Ascent

[ivory verge]

seems to be some sort of transistor clone but without pixelart bs?

not transistor, the other game with H

[maiden gazelle]

Hades?

[ivory verge]

yes!

i want something like that for my scifi shit too, instead of the dude, have my ship hover in ze hangar while you look at things 🙂

[maiden gazelle]

Sounds sick

When

[delicate delta]

The Ascent is real fun

[queen hinge]

Game’s name: Hades
Deccer’s memory: transistor

[delicate delta]

Can play it fully in half a day I think

They added content

Since I played it

It’s like original the Fallout games

Very story based and character driven

[queen hinge]

Honestly
I should probably actually do a project where I go through and make a bunch of test scenes for sake of generalizing my engine

[ivory verge]

https://www.youtube.com/watch?v=LyjnQDYee4o its the same shit

same studio too iirc

[queen hinge]

Also
I’m trying to do a software raytracer in MC
… need to figure out how to build a BVH for anywhere from 0 to… I think it was 500k triangles entirely on the GPU

every single reference I’ve found of BVH construction is on a light theme website, for C++, on a single thread, and does not assume a fixed memory layout like GLSL has
I also cannot preallocate large blocks of data, as that makes switching to the shader take a while (iirc, ~15ms or smth; more than I have leeway for)

literally writing the number 4 can cripple performance with this due to that

[delicate delta]

I saw your ray tracing recently looks impressive

[queen hinge]

That was hardware raytracing with vulkanite

that could not have been much more simple to write

If I can get this shader to perform well I will probably put it on github as a minimal software rt shader type of thing

[queen hinge]

#ifdef FALSE
void main()
#endif

{
    int att = 0;
    uint oldBits = DATA;
    float oldFloat = uintBitsToFloat(oldBits);

    do {
        uint newBits = floatBitsToUint(OP(oldFloat, VALUE));
        uint prevBits = atomicCompSwap(minX, oldBits, newBits);
        if (prevBits == oldBits) {
            // Swap was successful
            DATA = newBits;
            break;
        } else {
            // Swap failed due to concurrent modification, try again
            oldBits = prevBits;
            oldFloat = uintBitsToFloat(oldBits);
        }
        att++;
    // 500 for safety reasons
    // should never hit this point, but maybe
    } while (att < 500);
}
```cursed atomics

[delicate delta]

The function returns the original content of mem regardless of the outcome of the comparison.

I'm not sure if (prevBits == oldBits) is valid

isn't that always true

or that means the previous swap, but not this one worked

idk

this feels terrible

If the content of mem is equal to compare, then the content of data is written into mem, otherwise the content of mem is unmodifed. The function returns the original content of mem regardless of the outcome of the comparison.

I see

prevBits == oldBits means it wrote newBits

well

[queen hinge]

basically
I'm trying to make sure it's written to and also wasn't overwritten by another thread

now the thing is
I feel that should be checking prevBits == newBits

[delicate delta]

you can't know that

at the time of comparison it may get overwritten

idk

don't you have to get the context each thread

[queen hinge]

well
it sure seems to be working

[delicate delta]

ok it is an atomic compare

[queen hinge]

this is GLSL

[queen hinge]

yes

[delicate delta]

oh I am dumb

       if (prevBits == oldBits) {
            // Swap was successful
            DATA = newBits;

this though

this can be wrong without a lock or something?

because now you're comparing without atomic

[queen hinge]

yes

seems to reliably converge after 2 iterations, so 500 should be more than good as a safety

[delicate delta]

why 500?

[queen hinge]

arbitrary value

[delicate delta]

ok

this is for ray tracing?

or a new project

[queen hinge]

computing the bounds of a box that encloses the entire scene

[delicate delta]

why do you have to compute the bounds of your scene? is it generated?

[queen hinge]

well
it's a minecraft world

almost nothing is consistent in minecraft

[delicate delta]

you're doing this for light sources from outside the view frustum?

[queen hinge]

I'm doing this because I'm going to get approximately -1000 fps if I don't have a bvh, and I need to know the size of the root node in order to be able to start subdividing the BVH

[delicate delta]

idk what I'm saying tbh

can't you just create an arbitrary box and see how it behaves

[queen hinge]

I could
or I could use 16 threads to essentially instantly compute the bounds that perfectly enclose all of the triangles in the scene; no more no less

[delicate delta]

I don't know anything about bvh or ray tracing

[queen hinge]

I know minimal, lol

[delicate delta]

are you using stb_include

[queen hinge]

whatever iris does

[delicate delta]

have you considered a nested loop from 0 to 10

er

0 to 15

[queen hinge]

glsl hates that

I can see that my code is doing stuff based upon the values
can't tell if it's actually right or not but

also why is it flickering sometimes

maybe my triangle bin is full?
can't tell

ok setting it to 1 thread tells me that either these cursed atomics do not work, or my loop is wrong

[delicate delta]

:P

After this current chapter in GAE I'm going to read a chapter about parallelism and concurrency, maybe after I can say things that are helpful, although this is all GLSL and not application side

[maiden gazelle]

That chapter is fire

[delicate delta]

This my current stack of books, doesn’t include game math obviously since it’s digital

[maiden gazelle]

That's a fortune in my country

I have Game Programming Patterns & Mathmetaic for 3D Game ... in my library as well

[delicate delta]

I did also just order The Modern Vulkan Cookbook also

I had some of these books, just sitting around for a couple of years

[maiden gazelle]

Lovely collection, drooling just looking at it 😅

[delicate delta]

well I have to read them

I mean

to me it's a bunch of pending work

or work in progress

[maiden gazelle]

Is it fun?

[delicate delta]

sometimes

I wish osmosis worked

[maiden gazelle]

I was gonna reply with "then all the more reason to do it" to cheer you up, provided you responded with "yes" lol

"Sometimes" works as well though

It's kind of like a roller coaster

[delicate delta]

it's like exercising, I just do it because it helps me reach my goals

[maiden gazelle]

Motivation does not grow on trees

[delicate delta]

I don't have motivation, I have goals

[maiden gazelle]

Damn

I can't achieve the latter without the former

[delicate delta]

motivation is temporary, a routine is daily

[maiden gazelle]

💯 % true

That's why I'm still not where I wanted to be perhaps

But I'm getting there

[delicate delta]

where are you from?

if you don't mind me asking, just curious since you brought it up

[maiden gazelle]

Turkey

[delicate delta]

oh cool

I want to vist Ankara some day

[maiden gazelle]

I personally don't like Ankara that much tbh, worst drivers on the planet for some unknown reason, and the weather is too dry for my crumbling retina (I have severe eye dryness)

[queen hinge]

I have no control over application side for what I'm doing
everything has to be done in GLSL

[delicate delta]

time to let go of minecraft?

[queen hinge]

no

[delicate delta]

you need an intervention

[queen hinge]

I'm using a premade mod to implement this into

[delicate delta]

how are the trains?

[queen hinge]

https://modrinth.com/mod/iris

Because I’m sorta trying to challenge myself

[maiden gazelle]

I just visited relatives from time to time so I'm not sure

The one going from/to İstanbul is good though, the "speed train"

[queen hinge]

People have already made raytraced shaders with this

[maiden gazelle]

My humbler collection

[delicate delta]

The Lengyel book is so great, it's basically all the math

that you need

it's just not really explained conceptually

but you would know when you need it

what it is for

it's like the ideal math reference I think

what's the math for this problem? it will have an answer

[queen hinge]

float ov = x
if ov < v break
float nv = min(ov, v)

pv = atomicCompSwap(x, ov, nv)
if x<=nv break```
I think this is probably what I want

[maiden gazelle]

Funny story: that Lengyel math book is printed reverse. It is perfectly usable, provided you flip it 180 degrees on your x axis (direction to your right).
I asked Amazon to take this back and send me a new copy, they countered with "keep the book & we'll refund your money". So it is free 😅

[delicate delta]

your real time rendering is digital?

[queen hinge]

How did that occur

[maiden gazelle]

Yeah it is thorough I think, I read a small portion of it yeeears ago. But I remember it being good.
gamemath.com is more intuitive in my opinion. More beginner friendly.

[maiden gazelle]

No idea 😅 Additional photos incoming wait

[delicate delta]

I don't like digital books

[maiden gazelle]

It is surreal

[delicate delta]

but I have to with gamemath

[maiden gazelle]

As in I don't have it yet

[delicate delta]

oh I see

[queen hinge]

The WORST part is that there IS a glsl extension that does exactly what I need but I can’t use it

[delicate delta]

only good about digital is you can search them

[maiden gazelle]

I was like that too but I don't know reading from a tablet, laying down on a couch is good too

[delicate delta]

true

[maiden gazelle]

For example I'm reading a digital copy of GAE on the tablet

[delicate delta]

also lighter to carry around

[maiden gazelle]

even though I have a physical copy

[delicate delta]

I just don't like reading on a screen

[queen hinge]

decided to abandon the cursed atomics and just do two shaders
Not 100% convinced it’s working though, turning down the invocation count causes it to break

[delicate delta]

what is it

[queen hinge]

These colors represent the bounds I need for my BVH’s root node
… not sure if they’re actually correct, gonna have to validate that tomorrow

[fleet dew]

hey

[fleet dew]

I kinda assumed it from this, lol

also from turkey 😅

[maiden gazelle]

Hey 👋

As bayrakları

[ivory verge]

https://tenor.com/view/turkey-turkey-looking-at-friend-turkey-looking-at-oven-gif-7839521338030702394

[ivory verge]

@delicate delta andrew left discord and the zig server behind? do you know why?

[delicate delta]

#785499283368706060 message

he's still active until end of september

I also dislike the quests

but like

I'm not going to start using IRC

I haven't used IRC in like 10 years

all the clients are unmaintained and it is not great for sharing media and such

idk I don't really talk to him much anyway

I don't go into the dev channels, they are serious in there

like you say the wrong thing while they're trying to get work done it's a ban

[ivory verge]

its quite funny

i am thinking of leaving discord too

and find a new place where you just hang with your friends

no kids homework solving bs like we have in #opengl or #questions all day

just bikeshedding your frog's projects

[elder kettle]

We need the elite gp inner circle private element chat or something

where it's mostly just project threads

general discussion can still take place here

[ivory verge]

yeah

but basically a new server

where those peasant channels dont exist which keep teasing you to take a look

[elder kettle]

yeah I imagined matrix/element or something

away from AI training bs too

[ivory verge]

yep something selfhosted perhapsly

[elder kettle]

I just wish there were some better solution for user uploaded data

[ivory verge]

yea

[elder kettle]

There needs to be an "upload manager" system like old forums used to have, where when you upload files it goes into a little personal cloud storage drive with a fixed size

Then it's up to the user to manage their data

If they are running out of space they can delete old files or reduce the resolution of old images

then it puts the data management in the hands of each user and makes the storage scaling on the server vastly more predictable

Also gives an easy way to look back at your pictures so I don't have to search from:the_Demongod has:file every damn time

and click through their slow pages

[ivory verge]

a background job could replace all the images by low quality ones for posts which are older than 90d or so

and if you click on it to enlarge and the user still has it in their store you get the original pic or something

small thumbnails could be stored globally somewhere, so that you dont lose pics from old messages

[elder kettle]

Yeah

[delicate delta]

there's matrix.org

revolt

oh you mentioned matrix

[fleet dew]

yeah. I don't get purpose of it. imho, it just creates burden for anyone that wants to occasionally chat/follow up

like, discord search for example, I use it extensively to figure lots of random stuff. and sometimes I stumbled upon zig std channels, read some random interesting stuff

[delicate delta]

yeah I learn a lot from being on the servers

[delicate delta]

have been sitting on the zig irc channel on libre via the web client, it's super dead

[ivory verge]

yeah kids these days dont want irc anymore, i dont blame them

[delicate delta]

idk if andrew can really stay off of discord

[queen hinge]

I have never used an IRC client

[delicate delta]

ohh

he did leave discord today

he didn't wait a month

and he's not on irc he's on zulip

[ivory verge]

yeah

that platform which uses "leadership" in every other word on their website 😄

[delicate delta]

ok I'll try zulip I guess

oh just work channels

doesn't seem like a community chat

[ivory verge]

it is more thread centered

something what demon and i would want probably

[delicate delta]

oh is very thread based

I like it

[fleet dew]

I know rust compiler and stuff related uses it successfully

wish it was possible to move existing zig discord conversations to that without hassle, that would be awesome

[delicate delta]

Having lurked on Zulip for a day, I think it’s pretty great actually

[ivory verge]

i sink i have to ackchually check this out too

[delicate delta]

it's not a very pretty UI

but I like it how it works

[ivory verge]

yeah

[delicate delta]

GAE's third chapter has the best explainer for alignment I've ever read

it's super, it's like two pages and just fully explains it in clear and simple terms whereas everything else seems to have struggled

[ivory verge]

i find GEA quite bad in general

its how one would write engines 15 or 20 years ago

not sure why so many praise that book so much

[delicate delta]

can't speak to that but the parts I've read so far have been very good, I'm still in the first section of the book

[ivory verge]

perhaps im biased somehow

because its heavily OGRE influenced, and ogre was shit back then

[delicate delta]

honestly it kind of comes across as a how to pass a technical interview at a game studio more kinda book

[ivory verge]

yeah

you wouldnt waste your time reading it

it helps to understand concepts in general and thats more important than knowing what assembly you have to use to implement SSAO for a specific gaming platform or soemthing like that

[maiden gazelle]

This shows yeah

I treat GAE as a general-knowledge kind of book

[delicate delta]

I think maybe the more recent edition might have reduced ogre, I see some of it

[maiden gazelle]

Showing the big picture that is a game engine

Lots of better books on specific topics of course but as a general surface-level book I think GAE is not bad

[delicate delta]

I think it's more like a great intro for junior engineers at game studios or something more than a how to build your own

[maiden gazelle]

Yeah

[delicate delta]

or as I said for how to prepare for an interview

[ivory verge]

yeah

its GEA my frogs 😛

[delicate delta]

right GEA :P

[ivory verge]

everyone on the server keeps saying GAE 😛

[delicate delta]

idk why I did that

it's GEA

[maiden gazelle]

I'm only saying it because Bjorn said it first

[ivory verge]

: >

vb also says GEA hehe

[delicate delta]

lies

but actually truth

[wise imp]

why do you think gea can be bad?

its only a high level overview but its a pretty decent book to get started with

i dont remember any sort of ogre influence on it 🤔

OGRE was truly awful even at its time

and i think first edition of GEA is older than ogre

[delicate delta]

I think maybe older versions of the book had more about OGRE

it has appeared a little bit so far, but I'm only chapter 3

[maiden gazelle]

It appears every now and then

I’m on chapter 5 I think

[delicate delta]

that's the math chapter yeah

[maiden gazelle]

Ok then I’m on chapter 4 😅

Resources and whatnot

[delicate delta]

that's the parallelism chapter in my issue

[maiden gazelle]

Hmmm maybe I skipped math

And on chapter 6

Let me check

[delicate delta]

we have different editions

it goes overview of game engines, dev tools, how to write c++, parallelism, game math