GrubStomper29's OpenGL Sandbox

[plain mantle]

mesh shaders? with OpenGL?

[gilded shell]

that exists, right?

[plain mantle]

Technically yes

but its actively in flux

AMD is adding some kind of extension

Nvidia already has one

[gilded shell]

yeah i use nv hardware

[plain mantle]

you do yeah

[gilded shell]

i thought some amd cards support GL_NV_mesh_shader

[plain mantle]

I'm not sure

[gilded shell]

https://opengl.gpuinfo.org/listreports.php?extension=GL_NV_mesh_shader

theres a few cards on here

wait

im not seeing rtx 3050 on there 😅

oh nvm

its supported

yeah it looks like most modern cards support it

and its not like ill be having a wide audience with this project

i’m hoping ill reach 10ms draw times with that optimization

i dont know if ill be able to get below that without lod, which i refuse to touch

[gilded shell]

screen space derivatives failing? check that red cloth thing I don't know the name of

[plain mantle]

The red thing is called an awning

Not sure about the derivatives

But you could really use some MSAA

[gilded shell]

its deferred, dont know if i could handle that

once i have pbr i wouldnt mind trying out super sampling

@kind sparrow you havent said much in here recently

[plain mantle]

Technically it's possible but theres a reason nobody does it

[gilded shell]

idealy id like to try TAA but it might be scope creep

[junior sparrow]

Yeah I know but in this case it should beat the index buffer generation + rendering

[junior sparrow]

Slap FSR on it and be done

[gilded shell]

i am violently against ai frame generation

[junior sparrow]

You can use it as TAA

[gilded shell]

ill see

todo try min max vectors for radius calc

[plain mantle]

FSR 2&3 are not AI

[gilded shell]

oh ok

[plain mantle]

Dlss & fsr4 are

[kind sparrow]

I lurk

I haven't done CSM before and I'm not that interested in Sponzas etc personally so I don't have much to say

It looks good though

[gilded shell]

fair enough

thanks!

[kind sparrow]

I'll be implementing CSM eventually too, might be a little while though

I'm currently on the audio bikeshed

[gilded shell]

yeah

[night shoal]

ja, sponza == boringue

[gilded shell]

hey, this is bistro

[kind sparrow]

Same thing

[night shoal]

both are also just hacked together, no sense of detail or craftmanship

[kind sparrow]

They're interesting ish but after the Nth time they lose their charm

No doubt they're good test scenes from a utilitarian perspective

[night shoal]

ye, after the nth time you see all the cracks

[gilded shell]

theyre simply the only good stress tests i know of available in gltf

id love to see more

[echo token]

you could ask vblanco for the godot test scene he mentioned

I have it somewhere too I think

[gilded shell]

alright I think I fixed shadow mapping matrix calc

there is some acne on the next cascade

[gilded shell]

i want to try dithering between cascades

[plain mantle]

I use interleaved gradient noise

https://blog.demofox.org/2022/01/01/interleaved-gradient-noise-a-different-kind-of-low-discrepancy-sequence/

[gilded shell]

how does the dithering work?

does the dither value decide what shadow map you sample?

[plain mantle]

I add the dither value to the distance before the cascade calculation

The dither value is between 0-1 so multiply it by the distance you want the transition to happen over

[gilded shell]

i see

[plain mantle]

What does your cascade selection look like in the shader?

[gilded shell]

very primitive lol

[plain mantle]

Ah yeah so it's a flat cutoff

You can get a little more resolution if you use a spherical cutoff

[gilded shell]

yeah but im lazy and its working fine for now

the dithering works good

albiet i am using a random dither function i found on reddit

lets see if yours is better

whats the difference between mod and fmodf

type specificity?

[kind sparrow]

do you mean modf

glsl doesn't have fmodf afaik

[gilded shell]

the article uses std::fmodf

cpp

[kind sparrow]

ah

you can look at cppreference then

[gilded shell]

anyways I just dropped in the code and replaced those calls with mod and it works just fine

[gilded shell]

I did and thats my guess

[gilded shell]

looks about 3% better than the reddit dither

i must say im pretty happy with that

I am still suffering with shadow acne at further cascades, so clearly I need to scale my bias accordingly

[kind sparrow]

modf*(float val, float* intPart) separates the fractional and integer parts of a float,

[gilded shell]

cppreference doesnt mention that afaik

[kind sparrow]

fmod*(float x, float y) performs a floating point division but returns the remainder left over from calculating a whole number quotient

they're two different things

[gilded shell]

I also got rid of the ugly transparent cubes bisecting the scene

[gilded shell]

oh it does

[kind sparrow]

it's confusing but the variants of fmod are fmod, fmodf, fmodl, and the variants of modf are modf, modff, modfl

[gilded shell]

right lol

welp

shadow acnes gone

i guess im done?

if anyone has time I'd love for this to be tested on other machines: https://github.com/GrumStomper29/OpenGL-Sandbox/tree/shadow-mapping

[gilded shell]

I also finally have a thread thumbnail, even if it is very generic 🙂

[night shoal]

im jealous 🙂

[gilded shell]

idk your picture is pretty dope

[night shoal]

i have no shadows going yet 🙂

or hiz meshletisms

[gilded shell]

oh i see

you can see my github and vkguide for hzb stuff

my hzb generation is really brute force (1 compute dispatch per mip level) but its surprisingly fast

i could probably interleave the different hzb dispatches to reduce the impact of stalling for the next mip level write

[gilded shell]

uh oh

[junior sparrow]

daxa is calling

[night shoal]

nsight supports mesh shaders though

you can also ping BoyBayKiller in #opengl about them, if you need more info

[gilded shell]

0(3) : error C0203: extension GL_NV_mesh_shader not supported in profile gp5vp

not sure why this occurs

must I use a compatibility gl profile

[night shoal]

update your gpu drivers

[gilded shell]

i just did that last night

let me see

[night shoal]

what gpu and what os?

[gilded shell]

windows 11, rtx 3050

[night shoal]

weird

you need a core profile

[gilded shell]

hmm

[night shoal]

you cant use GL_NV_mesh_shader as an extension in a fragment shader

mesh and task shaders are completely new stages

[gilded shell]

oh yeah

I probably shouldnt be telling gl that its a vertex shader

[plain mantle]

that would do it

did you read the extension spec?

[night shoal]

https://developer.nvidia.com/blog/introduction-turing-mesh-shaders/ also

[gilded shell]

im using this and another blog

not sure why this happens

figured it out

[echo token]

I'd steal AMD's single pass downsampler

it's what I plan on doing next time I revisit this class ofp roblems

[gilded shell]

preferably but i dont think it works on opengl which is why i wrote my own

[gilded shell]

@night shoal if I were to creddit you for something, should I capitalize the D in your name?

or leave it how you wrote it

[night shoal]

its all lowercase

[gilded shell]

alright, thanks

oh no...

[night shoal]

update nsight

and c++ captures are finnicky

[gilded shell]

looks like what I used before is still there

regardless, nothing is rendering

I got something!

here are the deccer cubes

obviously im suffering indexing errors of some sort

immediately found 1 🙂

looks like were only drawing 3 triangles per meshlet for some reason

YEAH BABY

[night shoal]

et voila

noice

[gilded shell]

bistro works too

for some reason visual studio is hiding my stop debugging button on some submenu

[night shoal]

f10

people who click on buttons to debug are very suspicious 😄

[gilded shell]

lol

i have materials and culling back

i still need blended objects and shadows

its still running real slow because mesh shader thread group size is 1, and im not yet doing anything to deduplicate vertex transforms

each cluster has a max of 64 verts but im transforming vertices per index, so really its closer to 372 vertex transforms per cluster

easy fix though

[night shoal]

still, very cool

[gilded shell]

we still never found an explanation for that threadgroup quirk i found several months ago

where dispatching an 4x4x4 local size compute was much slower than 64x1x1

at least for the cluster batching

i should talk about that in general

[spring kelp]

Don't do that, it's not conservative when it comes to downsampling depth

[echo token]

wdym? also how is it even remotely usable if it doesn't produce the minimum accurate results

and follow up question, can you easily remedy this by modifying it

[spring kelp]

It's usable for other things, just not hiz

The problem is for non power of 2 textures it will generate a hiz where the depth of a coarser mip is not the min/max of a finer mip

It can "miss" pixels

[spring kelp]

Supposedly granite renderer did, I never tested it though. I also don't have a link handy

[echo token]

hm, I might have to dig around granite next time I need a hiz

seems kinda dumb though because if you could tolerate missing pixels, wouldn't you just not bother generating mips to begin with

[spring kelp]

It's only for non power of 2 textures

Hiz isn't spd's intended use case unfortunately

[gilded shell]

testing

looks like i cant send images right now

Neither via mobile or pc

[kind sparrow]

Discord is breaking down it seems

[gilded shell]

oh i thought it couldve been related to the things ive been sending friends for years

well thats fine, im not working on anything visual

by some luck for the past few months I always seem to have just enough padding already in a gl struct to be able to add some more attributes

for example just now i need to add 2 uints to my cluster struct, and by chance there are two ints oh padding

[gilded shell]

mesh shaders are effectively fully implemented and im still getting 30ms frametimes

it seems like the only thing ive accomplished is moving 60% of the load from the meshlet batcher to the mesh shaders

[plain mantle]

What does nsight say is the bottleneck?

[gilded shell]

nsight profiler? not sure itll profile gl mesh shaders but ill give it a try

[plain mantle]

What did it say before the mesh shaders

Gpu trace is what you want

[gilded shell]

yeah it looks like gpu trace isnt going to work for now

here were stats on the heavy meshlet batcher before my mesh shader impl

right now its closer to 11ms meaning the mesh shaders are ~20ms

it probably doesnt help that my local workgroup size is 1

[plain mantle]

Yeah that seems catastrophic tbh

[gilded shell]

welp

ill need to somehow divide the count parameter of the indirect mesh draw buffer by 32

maybe I can use some modulus in the batcher

but that will potentially leave me 31 meshlets less than I need

maybe the first meshlet of each view will add 2 instead of 1 to the buffer

and this will ofc give me a 32x reduction in atomicAdds in the batcher 🙂

[echo token]

you can just use subgroup ops to get that 32x reduction as well

[gilded shell]

lemme think

[echo token]

there's a subgroupAdd-something-or-other that's basically perfect for prefix sum operations

[gilded shell]

ill keep that in mind

i believe that will lock my mesh shaders and batcher to the same threadgroup size

and I still need to track the index for cluster writes

[echo token]

you still get the index, the subgroup op uses magick to make sure that each thread gets its own unique value of the sum, and then you can just elect one thread with the ballot to atomicAdd it

[gilded shell]

man compute really isnt fun

[plain mantle]

I like compute lol

[kind sparrow]

Compute is type 2 fun

Sometimes type 3

But when you get everything working it's great

The main trick is that it's essential that you can contrive test cases that are manually readable

If you have 2,000,000 elements that are being shuffled around it's impossible to debug stuff

You want to have small test cases and you want to have the ability to dump and/or write any intermediate state of the buffers

Because any non determinism can't be debugged with a graphics debugger

[echo token]

what are the types of fun

[kind sparrow]

Type 1 fun is fun as you're doing it, dopamine fun

Type 2 is not very fun in the moment but fun in retrospect, serotonin fun

Type 3 fun isn't fun at all even in retrospect lol

Basically it's mindless fun, tough and challenging but rewarding fun, and not actually fun as a humorous 3rd tier

[echo token]

oh lol

[kind sparrow]

My foliage stuff ended up being borderline type 3 fun lmao because it literally took 6 months of banging my head on horrible nondeterministic issues and when I actually fixed it it was not really that satisfying lmao

Although it laid the groundwork for future work

[gilded shell]

its just been so finicky to me

if I can get each frame under 10ms itll be so worth

[gilded shell]

is this ogl compatible

or am i rotting

[kind sparrow]

I assume he's just talking about doing operations on group local memory

[gilded shell]

im not finding "subgroupAdd" anywhere

[kind sparrow]

You just use a group shared variable and elect one thread to do all the global sync

[fervent lantern]

subgroupInclusiveAdd

or Exclusive

[gilded shell]

this seems like a needless limitation

[gilded shell]

this doesnt seem to appear anywhere except for a khronos blog

im looking for the formal doc

[echo token]

yes you can do it in opengl

[kind sparrow]

Are you just talking about doing shared memory stuff

Or something more exotic

[echo token]

I'm talking about subgroup operations, because it sounds like something vaguely involving a prefix sum

https://www.khronos.org/blog/vulkan-subgroup-tutorial

ignore that it's vulkan, GL has the same ext

particularly subgroupExclusiveAdd

[echo token]

You can find the doc by searching the ext names, or the links here
https://github.com/KhronosGroup/GLSL/blob/master/extensions/khr/GL_KHR_shader_subgroup.txt

but the formal definition is the text files on their github

[gilded shell]

so whenever i do go about making a real game, i’m going to be releasing it under my real name, which i dont really want associated with grubstomper29

to keep yall updated on development id have to make a private server

is that cool with yall?

specifically demon, deccer, lukasino, and jake

[kind sparrow]

sure

[plain mantle]

Yeah sure, why not

[night shoal]

i have no objections either

[kind sparrow]

You can probably make a lot of progress on a prototype before you have anything recognizable enough to warrant hiding though imo

Depending on the project

[night shoal]

and/or develop under another artificial name

StompGrub69 or something

[gilded shell]

eh

i’ll let you guys know when the time comes

[gilded shell]

i think to solve the batching issue ill try and stall thread 0 until all the atomic ops are done, then divide the var by 32

i wonder if there are in-shader barriers?

looks like there is a function with no parameter or return, seems easy enough

i wonder if dividing the var will be ub though

i can probably do some atomicExchange magic

if clusterId == 0
memoryBarrier()
fetch index
set index to index/32

seems easy enough

[echo token]

yeah, also how badly do you fear missing out on maybe a couple hundred, couple thousand dollars max? If I ever actually publish anything it'll probably be for free and anonymous

especially since I'm a big fan of copyright infringement among other things

[gilded shell]

itll be free but not anonymous

[echo token]

if you're not making money why pierce the veil

[gilded shell]

because itd be my lifes work and magnum opus so id prefer to take due credit

[echo token]

you can use subgroupElect() to just pick 1 thread in a subgroup to do an operation

if (subgroupElect()) {
  // atomic add or something
}

[gilded shell]

i only need it once per dispatch though

[echo token]

oh then yes, in shader barriers will work

quite literally barrier();

[gilded shell]

but does this make sense

[echo token]

yeah, I've never thought about it that way lol

[gilded shell]

especially if it does become popular, i wouldn’t want someone else claiming credit

[echo token]

I think there's ways around that even still though, like the guy that made yume nikki

[gilded shell]

perhaps

[echo token]

I think only a few businesses and media outlets know who he is

no average person

[plain mantle]

Just a tip, try not to treat any one project like that

ASO isn't gonna be my opus hopefully

[gilded shell]

really if one person plays it and has fun i’ll consider it a job well done

[echo token]

yeah I also try not to think about just how few games you can make in a human lifetime without help

[gilded shell]

i only really want to make one or two published games

i wonder if distributing a game without monetization would affect the price id have to pay for music licenses

ill cross that bridge when i get there

[echo token]

protip: if they can't find you, they can't sue you

[gilded shell]

i’m a lawful citizen

[plain mantle]

You don't need to pay for music if you make your own

I know you have it in you

[gilded shell]

yeah for sure

but there were some cool songs i wanted to cover

[gilded shell]

i think i can do this sometime tomorrow

in the following weeks im going to try to refurbish my Wii, Godwilling

general maintenance + a shelling replacement for the second time

the original shell was yellowed and eroded beyond repair so i bought a 3rd party replacement, which ended up being terrible and didnt even fit itself

so now ive bought a bricked wii to steal the shelling

now that i think about it, im really just migrating all the boards and disk drive from my wii to the bricked one

also replacing thermal pads and the clock battery

if i can repair the other wii’s board/drive or whatever is broken, I might resell it, mod it, or do something with it

[gilded shell]

the Before These Crowded Streets album has the best drumming ive seen in a studio, both in skill and tone

good example here https://youtu.be/O1zYYIXUl8Q?si=YYR7Sse_hmEEzNhH

aux percussion is also great here but is better in Crash by the same band

[gilded shell]

@echo token i believe it was you who recommended i ask vblanco for a test scene. He hasn’t responded. Do you have it?

[echo token]

uhhh let me dig

I definitely have it becaus I generally don't lose/delete this kind of stuff, the question is where

[night shoal]

isnt it on vb's github somewhere, if not some linked google drive

vkguide should know where

[echo token]

if I could just remember what it was called...

unrelated wtf is E1M1

is this a system shock map?

why do I have this

I will tell you that I have like 4 different copies of bistro

[night shoal]

thats doom1

dethraid turned the wad into gltf a year ago or so

[echo token]

ah that's why

I can't find a link on vkguide

but this is it

[fervent lantern]

its in the vkguide repo

https://github.com/vblanco20-1/vulkan-guide/tree/all-chapters-2/assets

structure.glb

[gilded shell]

should be a solid alpha blend showcase too

[gilded shell]

one thing I've been confused about is whether the mesh shader outputs are per invocation or per work group

i think ill assome the former and find out the hard way

from the nv example:

 // the actual amount of primitives the workgroup outputs ( <= max_primitives)
  out uint gl_PrimitiveCountNV;

looks like the output is per workgroup, but that output is still confined by max_primitives

which is pretty low

i dont think a workgroup would be able to output more than 2 meshlets

not sure what to do in this case

I could try to spread a single meshlet accross several threads

each thread transforms ~2 vertices and writes out ~4 indices

[junior sparrow]

Most implementations output just one meshlet

I have seen invocations == max vertices in one meshlet or half of the max vertices. So each thread fetches 2 positions and transforms it. Later writes 6 micro indices

[gilded shell]

i need to write 12 indices per invocation

given thread size of 32 and meshlet triangle count of 124

[junior sparrow]

Why don't you increase the size to 64 so you write only 6?

[gilded shell]

idk if enough hw supports that

ill give it a shot if I can figure that out

[junior sparrow]

Nah it supports it

I would assume it's the same as compute the amount of threads per thread group

I use 64 threads in my engine

[gilded shell]

alright

mabye not

[junior sparrow]

What kind of bullshit is that

[gilded shell]

3050 lol

[junior sparrow]

I think it's some OpenGL limitation than hw one

💀

[gilded shell]

yeah probably

i was gonna say gl_nv_mesh_shader sucks

anyways 32 seems to work well enough; this is so fast man

no lods or hw raster, yet i have bistro csm at 11ms (this is the most expensive camera angle I know of)

which equates to 90fps on my 3050

i can probably get more if I use subgroup atomic ops for the batcher and try interleaving hzb gen

i did not think mesh shaders would give me a 3x speed boost

[plain mantle]

Is this forward rendering?

[gilded shell]

interleaving hzb gen: I use 1 dispatch per mip level because it was real easy to write and is fast enough for my purposes. I generate each hzb sequentially which in theory could waste some time since later hzbs could be waiting on the semaphores of prior hzb dispatches

I could alleviate this by interleaving the dispatches, i.e. gen mip 1 for each hzb then proceed to mip 2 for each, etc.

in practice im not sure the time saved would matter given pcie bandwidth

[gilded shell]

deferred so pbr will be hopefully free or freeish

weird how there's gl_LocalInvocationIndex but not gl_GlobalInvocationIndex

so I have to manually ```uint clusterId =
gl_WorkGroupID.z * gl_NumWorkGroups.x * gl_NumWorkGroups.y +
gl_WorkGroupID.y * gl_NumWorkGroups.x +
gl_WorkGroupID.x;

clusterId = clusterId * 64 + gl_LocalInvocationIndex;```

i STILL need to fix alpha blend stuff though

the vkguide test scene draws pretty fast

cant say im a big fan of the scene; its not very colorful and indoor-ness prevents shadow mapping from doing anything cool

[plain mantle]

You need some other lighting

Do you have spot lights? those are easy to implement

[gilded shell]

nah i’m good

[gilded shell]

also im doing an inverse + transpose in the vertex shader

I def could bake and store that

[echo token]

memory is your bottleneck much more frequently than calculations

I wouldn't be too worried about baking and storing relatively simple stuff like that

[gilded shell]

i see

thats why i dynamically calc tangents

idk why bevy tries to use screenspace derivatives when you can dynamically calc tangents in vertex shader

[night shoal]

you could ask Jasmine 🙂

[spring kelp]

how so?

[gilded shell]

ahhh i cant find the article

let me paste my code

nvm here it is @spring kelp http://www.thetenthplanet.de/archives/1180

theres probably other ways to do it

[echo token]

afaik you get better interpolation, but also feeding varyings from the vertex to the fragment shader is also a memory bottleneck

it's gotta get there somehow

[gilded shell]

also im doing this in the frag shader for some reason

[spring kelp]

This is basically what I do already, I just use a newer algorithm, no?

[gilded shell]

oh hold on

maybe?

i mean yeah its around 12 years old

whats the new algo look like

[spring kelp]

https://jcgt.org/published/0009/03/04/

[gilded shell]

i wonder what the advantages of this over the previous method are

[junior sparrow]

You can calculate normal matrix much much faster using transform matrix and doing 3 cross products

Also it doesn't introduce errors as inverse transpose

[gilded shell]

can you show code or send a link please

[plain mantle]

{
    return 1.0 / vec3(dot(m[0], m[0]), dot(m[1], m[1]), dot(m[2], m[2]));
}```

you can then multiply the scale back into the matrix or simply apply it separately

so in my case mat3 rotation = mat3(local_transform); vec3 transformed_normal = rotation * (normal * extract_matrix_scale(rotation));

[echo token]

wouldn't you need to sqrt the columns?

I thought the scale was vec3(length(m[0]), length(m[1]), length(m[2]))

[junior sparrow]

https://github.com/lukasino1214/foundation/blob/master/src%2Fgraphics%2Fvirtual_geometry%2Ftasks%2Fresolve_visibility_buffer.slang#L195

The trick is called adjoint/adjogute matrix. It's a special case where inverse transpose isn't needed. I hope I spell the other word correctly

[echo token]

right the cofactor matrix

https://github.com/graphitemaster/normals_revisited

[gilded shell]

splendid

[gilded shell]

before

after?

trying out srgb

i think its fixed

I think it looks a lot nicer

but it could really use some hdr

[gilded shell]

I want good hdr but i dont want to go down a rabbit hole

[night shoal]

opengl doesnt really have colorspaceisms, all you got is tonemappers

vulkan supports the former natively

[gilded shell]

thats fine

[night shoal]

and i always thought you were on vulkan already

[plain mantle]

These two things are not compatible

You can have one or the other

[gilded shell]

i though so

ill probably just throw in the simple logl sample and add controls for that in the gui

[night shoal]

you can probably stay away from the logl sample and throw in a mcmapperface in

https://github.com/h3r2tic/tony-mc-mapface

[gilded shell]

sure

post is, imo, the lamest part of cg by far

[kind sparrow]

Makes a huge difference though

[gilded shell]

yeah

as much as id love retina based solutions thats probably the deepest rabbithole

[kind sparrow]

Retina?

[gilded shell]

eyeball receiver

i’m probably not using correct nerdimology

[kind sparrow]

You mean like a physically based solution

the one true correct™ tonemap

[gilded shell]

empirical would be fine for my case

i just want the image to look somewhat like youre viewing it through an eyeball

which i suppose means absolutely nothing since you already do that with every image

[kind sparrow]

I mean in principle you could have a completely physically based process that is designed to make the pixel the same color as whatever a spectrophotometer would report if you pointed it at that point

The problem is that your overall interpretation of the picture in real life is dependent on more than just one rod or cone

[gilded shell]

right

[kind sparrow]

there is not only automatic gain adjustment via pupil reflex and rhodopsin depletion but also filtering of patterns to do localized contrast leveling

[gilded shell]

right

is there a name for that approach to post processing?

[kind sparrow]

photometry

psychophotometry maybe

[gilded shell]

yeah

i wish more games explored that

i hate to see a first person game still try to look like a film camera

though id love to see a siggraph on this stuff its out of scope for this project

[kind sparrow]

Exhibit A

[echo token]

yeah all the colorimetry stuff is very unfun and annoying

[gilded shell]

piss filter my beloved

https://tenor.com/view/revolver-ocelot-gif-26257771

[gilded shell]

||test||

[west trail]

||Someone discovered spoiler ||

[gilded shell]

had to discuss a show i watched and really like

called neon genesis evangelion

[gilded shell]

i think ill just use Reinhard Extended for tonemapping

via https://64.github.io/tonemapping/

[plain mantle]

So when are you writing a wii game?

[night shoal]

right after releaseing the Bistro DLC for Metal Gear Liquid Solid

[gilded shell]

lol

ive had such a project in mind for over a year now

not a wii game but gc/wii accurate graphics

[gilded shell]

Here's reinhard extended tonemapping

using luma instead of color for calculation 🙂

im also trying to find more accurate sunlight and sky colors but without using real world units, im not sure if its possible

The image above uses https://planetpixelemporium.com/tutorialpages/light.html, but I find it hard to believe the sun emits white light

[night shoal]

you need to show some others side by side

otherwise i have no idea if reinhard is something cool or not 🙂

https://tenor.com/view/reinhardt-overwatch-slam-hammer-gif-16154010

[gilded shell]

its something simple

it fits my needs: I just set a value to the max possible luminance and it maps the scenes colors accordingly

thanks brave

this ones pretty cool https://ancestralis.co/color-of-the-sun-throughout-the-day-in-amsterdam-and-how-it-relates-to-artificial-lighting-choices/

heres something

im not sure how color temperature relates to the strength of a light

[night shoal]

wikipedia probably describes all those terms

luminance, magnitude, flux, etc

and their relationship, candela/lux/lumen

[ornate moss]

https://seblagarde.wordpress.com/wp-content/uploads/2015/07/course_notes_moving_frostbite_to_pbr_v32.pdf
page 21

[gilded shell]

I like this for now. I'll probably use that paper when I impl pbr

[kind sparrow]

It's comparing it to a black body emission spectrum which is parameterized by a single temperature

Otherwise you'd need some other description of arbitrary spectral content

Black body radiation just happens to be a convenient one

Since the sun and a variety of other light sources are approximate black bodies (any thermal light source basically, like an incandescent bulb)

[gilded shell]

i see

[kind sparrow]

It's completely unrelated to strength

[gilded shell]

then i imagine id need some unit of strength for sunlight

[kind sparrow]

Yes

Radiance

[gilded shell]

radiance

[kind sparrow]

This is a major rabbithole though

[gilded shell]

yikes

[kind sparrow]

I imagine you haven't taken vector calculus yet

[gilded shell]

nope

[kind sparrow]

And the units of radiometry are confusing as hell

[gilded shell]

for sure

and i’m not sure how id fit it into my lighting pipeline

[kind sparrow]

You'd use physical units for all your values

[gilded shell]

that doesn’t sound very fun

[kind sparrow]

Such and such watts per square meter, so many watts per steradian per sq meter etc

[gilded shell]

lol

really im looking for a ratio of sunlight radiance to sky radiance, the latter being used as ambient light

i think right now I have it at 4:1

[kind sparrow]

According to wiki it's ~15%

Actually 20% when you remove the outscattered sunlight from the direct lighting

So 5:1 is a good estimate

[gilded shell]

where?

[kind sparrow]

First paragraph of https://en.m.wikipedia.org/wiki/Diffuse_sky_radiation

[gilded shell]

and im assuming this depends on time of day? I think I’m shooting for 3:00 pm on this scene

[kind sparrow]

Yeah it does

[gilded shell]

im sure 3 is close enough to noon

[kind sparrow]

There's a formula lower down for calculating it depending on incident angle it seems

[gilded shell]

ah yeah i wouldnt mind simple simulation

i think my shadow angle is probably too extreme for 3:00

[gilded shell]

im actually having a lot of trouble finding a ggx brdf impl

i might just use Cook-Torrance

man google and brave are being so useless right now im getting a headache

[night shoal]

try filament or adobe's materialx

[gilded shell]

can you send a link for the latter

[night shoal]

https://helpx.adobe.com/content/dam/substance-3d/general-knowledge/asm/Adobe Standard Material - Technical Documentation - May2023.pdf

[kind sparrow]

You just have to post the spicy memes in your project thread where the mods aren't looking

[gilded shell]

lol

I really love when code examples use single lettered variables

[plain mantle]

Did I miss a spicy meme?

[gilded shell]

just some screenshots of funny things I see on the roblox shop

im following this https://google.github.io/filament/Filament.html

I wonder why the lambertian is formatted this way? Why not just dot the surfacae and light normal

[echo token]

because it's meant to plug into the render equation

which has cos(w_i) baked into the definition

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

also because the render equation is defined as an integral over the surface area of a unit sphere it leaves you with a 4pi you need to divide off

[gilded shell]

which I just replace with dot(), right

[gilded shell]

can you elaborate the 4pi division

[kind sparrow]

4π steradians cover the whole sphere

Idk what the normalization is for in this specific case though

[gilded shell]

right now Im just using the brdf for direct light

it was my understanding you could just multiply the brdf output by the light-frag cosine and use that

[gilded shell]

Is there any reason to not use a GL_RG16F framebuffer attachment to store deferred metallic roughness data?

[kind sparrow]

Do you need floats to describe that

[gilded shell]

wdym

whats the alternative?

[kind sparrow]

unorm

What's the range of your metallic roughness values

[gilded shell]

0 to 1

so i suppose

why not use unorm for my color framebuffer too

are they ints? reading the spec rn

[gilded shell]

I dont think so. unorm looks like integers? GLTF spec says metal and roughness are floats

...or maybe I can. This is written pretty confusingly:
Unsigned normalized integers store floating-point values on the range [0, 1]

[kind sparrow]

Unorm is what regular texture components are

[kind sparrow]

Yeah that's not a correct description lol

[gilded shell]

so i probably should put my color framebuffers in that

[kind sparrow]

That's what normal framebuffer formats are as well, regular GL_RGBA8 or whatever

Floating point values are for if you want your texture to store values like 3e-4 or 5.1e6 etc

Which is not usually, unless you're doing PBR lighting stuff with high dynamic range values like with physical units

[gilded shell]

not in the deferred buffers

yet I used rgba16f for them

probably way overkill

[echo token]

yes you can see it to the right of the equation I sent

also I don't entirely know why lambertian is defined as C/pi when the total surface area is 4pi, I dunno where the 4 gets lost

[gilded shell]

either im reading materials wrong

or this bistro model doesnt have proper metallicroughness materials

[junior sparrow]

You are getting in quite a Rabbit hole with PBR

[echo token]

https://litter.catbox.moe/tdv5a5.7z
here is what I think is the fixed bistro version pat made a while back, it uses compressed dds textures iirc

[gilded shell]

i forgot to tell yall that nsight is completely broken for this project right now

probably due to mesh shaders

[night shoal]

nsight supports mesh shaders

[gilded shell]

not gl

[night shoal]

of course gl

mesh shader is nv tech

[gilded shell]

nah

whenever i try to capture debug info, theres a message like “unsupported function detected: glDrawMeshTasks”

[gilded shell]

why

I also love that its called "metallicRoughness" and they put roughness before metalness

very nice

[echo token]

it's also in the gb channels

so r and a can be used for... something else

I don't remember what exactly but if you dig around for what's speced to the r and a channels you might see what they expect as an "optimization"

[gilded shell]

I didn't see it

oh well

[gilded shell]

this is more annoying

im just gonna flip it in shaders

looks like all my metallicRoughness vals are 1, 1

i think its an issue with the model

ill try the one yall posted above

7zip my beloved

@echo token mightve asked this already but do you have a project thread I can follow

[gilded shell]

it must be using something because both fastgltf and the khronos validator refuse to load this lol

[night shoal]

you need to enable the extension which is required by that model, most likely some glossiness roughness thing

KHR_materials_pbrSpecularGlossiness perhaps

    constexpr auto parserOptions =
        fastgltf::Extensions::EXT_mesh_gpu_instancing |
        fastgltf::Extensions::KHR_mesh_quantization |
        fastgltf::Extensions::EXT_meshopt_compression |
        fastgltf::Extensions::KHR_lights_punctual |
        fastgltf::Extensions::EXT_texture_webp |
        fastgltf::Extensions::KHR_texture_transform |
        fastgltf::Extensions::KHR_texture_basisu |
        fastgltf::Extensions::MSFT_texture_dds |
        fastgltf::Extensions::KHR_materials_pbrSpecularGlossiness |
        fastgltf::Extensions::KHR_materials_specular |
        fastgltf::Extensions::KHR_materials_ior |
        fastgltf::Extensions::KHR_materials_iridescence |
        fastgltf::Extensions::KHR_materials_volume |
        fastgltf::Extensions::KHR_materials_transmission |
        fastgltf::Extensions::KHR_materials_clearcoat |
        fastgltf::Extensions::KHR_materials_emissive_strength |
        fastgltf::Extensions::KHR_materials_sheen |
        fastgltf::Extensions::KHR_draco_mesh_compression |
        fastgltf::Extensions::KHR_materials_unlit;
    fastgltf::Parser parser(parserOptions);

fastgltf should also tell you why it wasnt able to load

[gilded shell]

yeah something about extensions

domeboy mentioned texture compression

[night shoal]

just look at the error in loadresult

[echo token]

I don't write about my actual personal project anywhere other than discussing aspects of it in other people's threads when it's relevant, primarily because it's not actually that novel or interesting
https://discord.com/channels/318590007881236480/1149482437601153064
but I do have one for my text lib

[gilded shell]

Ah

I love you domeboy but text rendering is so uninteresting to me

[echo token]

though I had a feeling you might be interested since what I do when I feel like working on it is usually copying roblox features

yeah it is boring by most metrics, my engine UI is just basically roblox UI so that lib is basically what I needed to drive feature compliance with all their various things

[gilded shell]

i see

i think i might go back to sponza rendering for now. I’m sure those materials work

better yet, i should probably use a pbr test model for this

[echo token]

there's the gltf one from "this mesh will also crash your engine"

it's a grid of spheres

I don't remember where that thread was though, much less how to necro it

glad so much of my long term memory is reserved for where I acquired GP test models from

[night shoal]

its in #experiments

https://discord.com/channels/318590007881236480/1096136308155498587

[gilded shell]

https://youtu.be/84UoUPcvQCs?si=rX7LqS84s4p0b-1n

i wonder what level of hearing loss this would induce

i sure hope he had ear pro on

especially around 2:39

[kind sparrow]

There are two types of musicians, ones that protect their hearing and ones that are halfway deaf lol

I always wore earpro while playing and I still have mild hearing loss on the right side from the china cymbal

It's not noticeable or anything since my hearing is very good but it's just "merely" normal on that side vs. the left where I have a few dB higher sensitivity than average, and I have mild tinnitus on the right

[gilded shell]

i see lol

[gilded shell]

alright so this is really hard to see but im writing the roughness and metallic as R and G here

and as you can see, it doesn't look right

[night shoal]

that looks familiar 🙂

[gilded shell]

yeah but i wish it worked

i have no clue whats wrong with the metallic roughness textures

[echo token]

if they're uniform, maybe it depends on the values in the material definition in the glTF file?

[gilded shell]

heres the GL side

the whole thing is metal and rough?

[echo token]

it says there you have 1 met, 1 rough and no texture

so yes I guess so

[gilded shell]

that doesnt seem right

thats the only material and its over the entire model

and I may be wrong, but this looks a lot like a metal roughness texture

must be an issue with my loader

[echo token]

yeah

are you using fastgltf

[gilded shell]

yeah

[echo token]

are you checking for the PBR sub-struct in the material

[gilded shell]

I believe so

let me check

if (asset->materials[i].pbrData.metallicRoughnessTexture)
{
    hasMetallicRoughnessTexture = true;

    metallicRoughnessTexture =
        asset->materials[i].pbrData.metallicRoughnessTexture.value().textureIndex;

    metallicRoughnessTexture = mTextures[metallicRoughnessTexture].bindlessHandle;
}```

oh wow

it looks like i just skip those in the final struct init

thanks old me

this is a lot more like it

[night shoal]

this helmet doesnt store AO in the .r channel, so that should literally just be blue and green

there should not be a red

[gilded shell]

im doing some weird deferred junk, dw ab it

[night shoal]

ah

also

since you use bindless textures, you dont need separate hasXXX flags

you can just check if the address is 0 or not

[gilded shell]

is 0 reserved or something

[night shoal]

for what?

[gilded shell]

bindless handles?

[night shoal]

no

0 = no valid handle

[gilded shell]

i see

i wonder if thatll help with cache

ill fix that later

[night shoal]

just removes unnecessary code

[echo token]

I have my 0th bindless texture as my default black texture to be safe

[gilded shell]

this is new

[echo token]

I have handles 0, 1, and 2 reserved for black, white, and normal (0.5, 0.5, 1) respectively

very handy

if it's negative it sounds like an initialization issue

[night shoal]

opengl handles being 0 results in black by default 🙂

[echo token]

also make sure to respect the optionals in fastgltf

[night shoal]

put a break point in your debug callback, when the message type is error

[gilded shell]

already on that

[night shoal]

ah

[gilded shell]

ah yeah i have no alpha blend geometry so the engine tries to make a buffer to store all zero alpha blend indices

this is clearly outline in the README as a feature

anyways this is a bit less nice

[night shoal]

https://tenor.com/view/rorschach-convenient-mask-monster-fitting-gif-5489216

[gilded shell]

yeah idk ab this one

[gilded shell]

ive been using uvec to read/write the unorm g buffer but it seems to be the cause of this

ill just use vec

[night shoal]

vec is float, uvec is uint

[gilded shell]

still a work in progress but heres a look at pbr

some fixes

its still broken btw

[night shoal]

is srgb on?

[gilded shell]

removing the specular denominator seems to get me closer for some reason

the filament paper doesnt seem to have it either

[night shoal]

are you loading albedo/metal/roughness as srgb too?

[gilded shell]

just albedo

[night shoal]

ah pic just loaded in

[gilded shell]

my main issue is fresnel seems to take over when I get too close

[night shoal]

i had metalness and roughness swappen in between somewhere before, make sure you didnt do the same mistake

[gilded shell]

thats not it

i wish the filament paper showed me more of the brdf

[night shoal]

the fresnel part shouldnt be visible when you look straight on to the object

it manifests at grazing angelos

[gilded shell]

im not sure what im missing

[night shoal]

render out the individual things as color

[gilded shell]

heres the brdf ```
vec3 brdf(vec3 v, vec3 l, vec3 n, vec3 f0, vec3 baseColor, float metallic, float perceptualRoughness)
{
vec3 diffuseColor = (1.0f - metallic) * baseColor;

vec3 h = normalize(v + l);

float NoV = abs(dot(n, v)) + 1e-5;
float NoL = clamp(dot(n, l), 0.0, 1.0);
float NoH = clamp(dot(n, h), 0.0, 1.0);
float LoH = clamp(dot(l, h), 0.0, 1.0);

// perceptually linear roughness to roughness (see parameterization)
float roughness = perceptualRoughness * perceptualRoughness;

float D = D_GGX(NoH, roughness);
vec3  F = F_Schlick(LoH, f0);
float V = V_SmithGGXCorrelated(NoV, NoL, roughness);

// specular BRDF
vec3 Fr = ((D * V) * F);
Fr /= (4.0f * max(dot(n, v), 0.0f) * max(dot(n, l), 0.0f) + 0.00001f) ;

// diffuse BRDF
vec3 Fd = diffuseColor * Fd_Lambert();

return Fr + Fd;

}```

[night shoal]

albedo only, metal only, roughness only, diffuse only, specular only etc

[gilded shell]

its def the specular calc

[night shoal]

NoV and max(dot(n, v), 0) looks like a duplicate

[gilded shell]

true

[night shoal]

NoL too

[gilded shell]

yeah

something about that seemed to do it

besides the aliasing it looks perfectly fine to me

now onto bistro

again, idk if the materials are correct but im pretty sure brick walls dont have much fresnel specular

[night shoal]

i had very shiny surfaces too before

because i fucked up metalness and roughness

double check that, just in case

[gilded shell]

yeah I think its an engine issue if even sponza has this problem

[night shoal]

it always is a PEBKAC

[gilded shell]

whats that mean

[night shoal]

: (

problem exists between keyboard and chair

[gilded shell]

ah

yeah

[gilded shell]

has anyone figured out how to render foliage this thick yet

[kind sparrow]

[gilded shell]

thatd be the dream huh

[kind sparrow]

I have the makings of what could be basically nanite specialized for plants

but it's like multiple man-years away from being that mature

I would need to literally hire a professional team

the other issue is that a lot of the complexity you see comes from the infinite variability of the plants, in a game you need to reuse assets or the plants in this picture would be like gigs of data

but you could probably get close with the right pipeline

[gilded shell]

id love to know how it works

[kind sparrow]

It's basically just a graph representation of the plants that could be used for dynamically generated LODs while allowing arbitrarily complicated plants

but the actual infrastructure required to scale it to the level of generality and performance necessary to render a scene like yours is immense

also it would require my plant simulator to get way more advanced to be able to generate that scene in the first place

[gilded shell]

yeah

you should write a paper

[kind sparrow]

If someone gave me half a million dollars to quit my job and hire a few collaborators for a year or two I probably could

[gilded shell]

lol

[plain mantle]

What if I gave you no money to add the plant tech for ASO2

[kind sparrow]

if only

[gilded shell]

i think the foliage would just be so hard to cull

too many “holes” in a plant to offer good occlusion of plants behind it

hzb culling would probably do something but not enough

i also imagine a need for imposters, but like demon said, keeping imposters for each unique plant sounds like a huge memory hog

then theres lod popping

problems exist to be solved 💯

[plain mantle]

Yeah foliage pretty much guarantees overdraw

[gilded shell]

putting dense foliage on procedural terrain sounds fun

i’ll add that to the long list of potential projects

[gilded shell]

as a plus, so much light is scattered from the canopy and fog of the Appalachians that shadows dont seem to be much of a thing

[gilded shell]

plant generation doesnt sound too fun though

@kind sparrow can you confirm or deny this

[kind sparrow]

it's !fun!

[gilded shell]

welp, i’m sure theres a lib for it somewhere

i can rip yours

[kind sparrow]

There isn't

there are probably some people's various experiments but idk if there's anything up to the task of real general-purpose plant simulation

Part of the problem is that modeling plant growth habits is real hard

it's very difficult to match observation to a model

for me anyways

I spent 6 months straight studying algorithmic botany and didn't get to the point where I could look at a plant and figure out its growth patterns

[gilded shell]

yikes

ferns tend to look uniform

100 billion ferns

take it or leave it

[kind sparrow]

I started messing with fern models but part of the problem is that kind of plant's morphology is heavily governed by physics

with trees you can sort of get away with treating them as perfectly rigid

[gilded shell]

im not terribly worried with trees atm

[kind sparrow]

but a fern needs to bend under the weight of its leaves

trees are the easy part though

everything else is hard because it's flexible

[gilded shell]

yeah

[kind sparrow]

and its growth patterns are in some ways even more context-sensitive

[gilded shell]

i’m looking at the various ivies and leaves

[kind sparrow]

so you end up needing structural dynamics models and sensory processes and approximations of the behaviors plants use to decide which direction to grow

[gilded shell]

i wonder if cutting corners to make a rudimentary approximation is practical, hoping the density of the scene would mask the cracks

or would it be much like using broken bricks to build a castle

[kind sparrow]

that's sort of what I'm hoping

I burnt out on it though

[gilded shell]

yeah

i can see why 💀

[kind sparrow]

I spent 6 months studying books and doing field observation and then it took another 6 months just to get my basic GPU culling/instancing stuff working so I just shelved it as-is as soon as it worked and haven't gone back

it's probably well poised to make more improvements quickly but I don't have the interest atm

maybe later

[gilded shell]

for sure

[gilded shell]

this would be my second forest project though

so maybe i do it for the heritage

[ornate moss]

https://developer.unigine.com/en/docs/future/content/tutorials/vegetation/clutter_vegetation.gif
source:
https://developer.unigine.com/docs/future/content/tutorials/vegetation/

[night shoal]

if oyu hire me, ill make sure no bat will attack you during that time

[echo token]

unfortunately in order to afford the salary we had to diversify into guano collection

[plain mantle]

This is a fantastic link. Thank you for sharing

[ornate moss]

also this is quite nice for clutter plants:
https://www.fab.com/sellers/Quixel?listing_types=3d-model&categories=nature-plants--plants

[gilded shell]

@kind sparrow i was watching a movie and a character in it held his pistol weird, with his non dominant elbow above the slide like so

is this hollywood shenanigans or something esoteric

[kind sparrow]

Not sure

Possibly some defense against someone grabbing the gun

[echo token]

it seems like a weird hollywood interpretation of when you usually see guys stabilize their gun with their elbow below (also usually that's when they're also holding a flashlight, so it naturally faces forward)

[gilded shell]

that would indeed fit into the context of the scene

the gunman in question was apprehending someone

that being said its Michael Bay so you never know

[echo token]

usually your best defense against someone grabbing your gun while you're aiming it is to pull the trigger

[gilded shell]

theres more missing context

the movie is The Rock and i enjoyed it

[night shoal]

https://tenor.com/view/nicolas-cage-nic-cage-ball-roll-almost-gif-7519943

[gilded shell]

idk if i told yall but a few years ago i showed my friends a color interpolated square i made in vulkan

like the hello triangle but a rectangle

and I told them it took me two weeks to make so its been referred to as the 2 Week Box since

one of my friends made it his wallpaper too lol

[night shoal]

best friend!

[gilded shell]

back to my program, you can see strong specular reflection on rough surfaces at grazing angles

Filament article says

so maybe its correct

[junior sparrow]

https://x.com/BenSimsTech/status/1933128205457830213

Ben Sims (@BenSimsTech)

Came up with a nice way to handle backfacing normals for PBR. When NdotV < 0, rotate N to V so that NdotV == 0. This prevents issues with negative NdoV (Fireflies) and preserves relation between N and V. Alternative reflect method produces dark edges and loses edge detail:

X · Jun 12, 2025, 11:44 AM

try this

@gilded shell

[gilded shell]

I already clamp such vals

[gilded shell]

maybe it looks wrong because I dont have ibl or gi?

well this certainly isnt a grazing angle so its probably not correct lol

[junior sparrow]

💀

[gilded shell]

odd how filament uses vec3 as fresnel return

[junior sparrow]

its because of the metals

[gilded shell]

yeah my mistake

I wish they included the specular denominator

anyways I dropped in the opengl functions as quick replacements for the filament ones

now it looks a little lame lol

it makes me think filament is doing something different for the specular denominator

which is a shame since there's no code example for it. It might be buried in the explanations somewhere.

[gilded shell]

@night shoal would you know anything about this

[night shoal]

https://github.com/google/filament/blob/main/samples/sample_full_pbr.cpp

[gilded shell]

Why is float NoV = abs(dot(n, v)) + 1e-5; when the other dot calcs clamp and omit the addition?

[gilded shell]

and yeah for some reason I can't find the denominator anywhere in the code

I wonder if theyre just omitting it

or they've replaced it with "pixel.energyCompensation"

idk what gives but dropping it seems to give correct results

i have no clue why the aliasing is like that one the edges of the model, thats really weird

[night shoal]

many solutions to the same problem

[gilded shell]

now im really curious about this

the filament cook-torrence impl doesnt have the specular denominator

[gilded shell]

I might implement RSM

i still probably need ibl for specular light though

[gilded shell]

im still really confused why the outside edges are way more pixely than other geometric edges

i dont have any aa

[night shoal]

if you are referring to this one, it just looks like you are rendering into a low resolution render target perchance

[gilded shell]

wdym

[night shoal]

wdym wdym

it looks like this is rendered into a 256x256 or something texture

rather than a 1080p one

[gilded shell]

I'm still getting used to android

I just switched from ios

idk what to put on my homescreen so i just have nothing

(the wallpaper is a picture i took)

if you guys have tips for smoother android usage or helpful settings id love to hear it

[kind sparrow]

Idk you don't have to do much

[plain mantle]

Keep it in your pocket, it can do less mental damage to you that way

[gilded shell]

Well now that i have android i can play cave story on my phone

[gilded shell]

I replaced most of the google apps with othet stuff

but theres no replacing google maps

and for some reason my investing app needs google play installed and enabled

[junior sparrow]

That's for security reasons

Most apps use google services for security reasons

Graphene OS had some beef with companies about this bs

[junior sparrow]

This is how mine looks quite simple

[gilded shell]

transcribed part of a song by ear

it sounds a bit off, probably beacuse im not trying hard enough to stay on key

@plain mantle

its crazy how much better I am at the drum composition than anything else lol

[plain mantle]

Cool! what song is it btw?

[gilded shell]

https://www.youtube.com/watch?v=klzB1Q1zqLo

at 0:40

[plain mantle]

I shoulda guessed it was DMB

[gilded shell]

lol

I probably didnt do enough with the bass

[plain mantle]

I dunno the bass part sounds pretty close to the recording

hes not going wild

[gilded shell]

but hes not playing the same note for 16ths

he dances around a bit

regardless, its too late; trial FL wont let me open the file again so this is what we have lol

but yeah I wonder if doing little transcriptions like these will be beneficial

[plain mantle]

Yeah its good to train your ear

[gilded shell]

what about creativity for composition

[plain mantle]

Best way to get better at composing is to compose

[gilded shell]

i see

I think these are solid baby steps

[plain mantle]

You will inevitably make some stuff that doesn't sound great, but eventually you will have something that sounds good. And you can choose to only show us the good stuff 😉

[gilded shell]

yeah

anything but programming

ive been putting off ibl

[gilded shell]

im proud of this lol

[gilded shell]

I forgot to tell you all; I got a 5 (best possible score) on the ap calculus exam

[plain mantle]

Congratulations!

[gilded shell]

i think i wanna hold off on vk forest scene 2

and instead make a basic game first, mostly to learn better engine structure

frog platformer or smth like that

[plain mantle]

Yuss! Do frogge platformer!

Is it gonna be gamecube styled?

[gilded shell]

you know me so well

[plain mantle]

Like mario sunshine type stuff?

[gilded shell]

perhaps

do you all have scenegraph impls?

[kind sparrow]

No

Depending on what you mean

[plain mantle]

Same

I just loop through everything with a mesh component and draw it

Highly recommend EnTT btw

[kind sparrow]

Yeah and I just have a specific component and system to handle parent/child spatial relationships

[plain mantle]

I have like 12 different components for that

Like I've said before ASO is a DAG

[gilded shell]

I see

[gilded shell]

Why this as opposed to my own?

[kind sparrow]

It just makes it easy

Your own would basically do the same thing

If your game is going to be simple you can roll your own easily entt is just convenient

[echo token]

yeah its not worth rolling your own tbh

I did and eventually just switched back to entt because maintaining a lesser featureset just wasn't worth it

[night shoal]

demons nih rolled one is also sufficient enough, but i also stick with entt

[kind sparrow]

If your game has a known finite scope you can just use flat static arrays of size MAX_ENTS which is easy

That's my NIH one

But if your requirements are basically any bit more complex it's worth just using entt

[night shoal]

yeah

perhaps if you have hundreds or thousands of entities then entt's group-thing might be a better choice

i also dont want macro-isms 😛

[kind sparrow]

Mine only uses macros to define the types

After that it's macro-free

[night shoal]

yeah

hence the 😛

[kind sparrow]

Deccer you need to learn to become based and macropilled

Macros are great when used judiciously

[plain mantle]

Who are you PQCraft?

[kind sparrow]

Don't slander me

He's referring to my minimal_game ECS impl

[night shoal]

https://tenor.com/view/keeping-a-secret-star-trek-tng-hidden-meaning-coy-gif-17292531

[gilded shell]

ive never macrod

my code is perfect and doesnt need changed on the fly

[kind sparrow]

macros are for when you have a pattern that would result in copy/pasted code that's either not worth abstracting or not possible to (e.g. in the case of declaring named static arrays)

It's rare to need them but when they are the right tool they are super useful as long as you don't abuse them

[gilded shell]

yeah

i definitely havent rewritten the same functionality multiple times across different glsl files

[kind sparrow]

stb_include is your friend on that front

[gilded shell]

does anyone know good resources on shoreline water rendering

as shown in this fine chap's video

https://youtube.com/shorts/YiyKpwQf_zw?si=w3JmiLt2muEoG8hy

[echo token]

there's an article from assassins creed which is decent

the one before black flag

https://www.fxguide.com/fxfeatured/assassins-creed-iii-the-tech-behind-or-beneath-the-action/

[gilded shell]

ive been had

I'm seeing instagram ads based on some research i did earlier today

likely because some of it was on youtube

[plain mantle]

Yeah trackers

Not using adblock I see

[gilded shell]

i havent found a good one for android

are you referring to dns, vpn, or something else

[plain mantle]

No just the browser plugin

Ads that are embedded into sites usually contain trackers

So that when you go to another site that also serves ads.

It remembers you

Of course this won't stop the instagram app from tracking you

[gilded shell]

ofc but im not quite willing to give it up just yet

[plain mantle]

I hate instagram. But unfortunately I need it

[echo token]

do you get a significant follower base from instagram

[gilded shell]

Not at all

[plain mantle]

Nah I need it for music stuff. All musicians use insta so when you meet them after a show or whatever they always ask for your insta

[gilded shell]

i just use it for acceletating my dementia and communicating with friends

[plain mantle]

Well you are gen Z so the brainrot is unfortunately terminal

[echo token]

it has booted into gui mode for me

[gilded shell]

well i dont outsource thinking

nor do i post anything anywhere but github

[plain mantle]

Very good

[kind sparrow]

You are basically gen Z too

We just got called millennials growing up because nobody knew what gen Z was yet

[plain mantle]

I very importantly came in just under the cutoff

[kind sparrow]

Haha

[echo token]

yeah I am very glad to not be a millenial

[plain mantle]

I am very glad to be one

[night shoal]

you were born <81?

[kind sparrow]

No lol

[night shoal]

then DR is a millenial too 😛

[kind sparrow]

We are all gen Z

You're thinking of gen alpha

We're just old gen Z