GrubStomper29's OpenGL Sandbox

[night shoal]

then mdi would be the next logical step ne?

[gilded shell]

Yeah it’s in my notes lol

the only thing I’m binding anymore is my ubos

[night shoal]

thats fine

[gilded shell]

yeah

if i profile and its killing me, i can just make an array

[night shoal]

there are extensions for nvidia from 15 years ago, which allow bindless buffers

[gilded shell]

i checked

im not sure amd supports those

[night shoal]

and you can also track the bind state on your side too, and only bind when it needs binding

[gilded shell]

right

[night shoal]

mdi + bindless textures as a base line

then frustum and hiz culling on gpu

[gilded shell]

yeah

[night shoal]

and then the next step would be nanite, where you split your shit into meshlets and do the same frustum/hiz culling

[gilded shell]

i went bindless now because id rather build off of it than have to change it later

[night shoal]

in my older engines i had a fallback to texture2darrays, so that i can debug in renderdoc, but might not worth the effort, but simply because i dont like nsight much

[gilded shell]

i dont mean for textures, im talking about geometry

[night shoal]

what do you mean

bindless buffers is pretty much out of question for opengl : >

unless you want to support nvidia hardware only

[gilded shell]

technically not bindless

but huge buffers with everything that are always bound

[night shoal]

ah yeah

megavertex/index/material/instance buffers

[gilded shell]

i dont want to write a meshlet generator then immediately refactor it to use megabuffers yeah

[night shoal]

libmeshopt will do the meshlettification : )

or libLVSTRI soon(tm)

[gilded shell]

at runtime?

[night shoal]

at model loading time

which isnt a hotpath

[gilded shell]

alright

[night shoal]

or as a preprocessing step, or even at build time if you want

[gilded shell]

okay good

i dont think writing my own would be out of the question though

[night shoal]

maybe later as an exercise to figure out that meshlettification stuff

[gilded shell]

it won’t be as elegant

[night shoal]

i also believe this is a rabbithole and will grant you a million dollhairs if you actually solve it for every mesh configuration there is

[gilded shell]

lol

[night shoal]

it always sounded like thats an actual maffematical problem or topological or whatever term, when lustri was talking about METIS and whatnot heh

[gilded shell]

my original plan was to tipsify sort the index buffer, then put sequential tris in a meshlet

the results looked good enough for high poly yet simple models like that dragon thing

but im sure it would fall apart with complex or low poly geometry

[night shoal]

i have not put any thought into meshlets myself yet

[kind sparrow]

Bindless textures doesn't improve performance on its own afaik

the point is to do larger batched draws

[night shoal]

yep theoretically they do improve performance, because you dont bind any texture anymore and have to group your meshes by material and whatnot

[kind sparrow]

oh yeah right

I forgot that texture switching itself can be a significant cost

for some reason I've never had that issue

[night shoal]

on the account of not having any textures going in your game ahem

[kind sparrow]

I do though lol

[night shoal]

hihi

those metal plates you can shoot at at the range 😛

just pulling your leg

[kind sparrow]

I make textures that are just solid colors as a pessimization so that later when I have real textures it doesn't cause any perf or memory issues

I have like 1024x1024 textures that are 3 blocks of solid color lol

[night shoal]

: ) absolutely nothing wrong with that

and probably a smarter idea anyway

[kind sparrow]

But yeah probably because I don't have a large enough number of different models yet

actually though I don't even batch by texture binding

so just not enough objects

[night shoal]

you just render what needs rendering neh?

in whatever order it comes in

as in, no batching

[kind sparrow]

I bin by pipeline

[night shoal]

ah

[kind sparrow]

and then just draw the objects naively

batching is implemented bespoke for systems that have a lot of objects

e.g. foliage

[night shoal]

yeah

rendering the 2 quads per grass tile would be insane as individual drawcalls heh

[kind sparrow]

if you think about it, environment stuff is almost always going to outweigh anything the player creates

[night shoal]

yeah

asteroid belts, whole planets, cities full of housing and whatnot, or otherwise terrain rocks and whatnotisms

[kind sparrow]

Yeah

[night shoal]

plus all the fluff which goes with it, decals and foliage

this madrigal guy keeps teasing me

with his little terrain and car fisiks arrangement

(i believe he works in the studio which made alan wake)

explains a lot of stuff 🙂

[kind sparrow]

An average scene in the game is like,
trees/rocks/misc ground clutter: 100,000,000
items/gear/vehicles/etc: 100
buildings: 50
characters: 50

[kind sparrow]

oh yeah that definitely explains it

[night shoal]

: )

i played a bit of dayz earlier again and kept gazing at the trees and grass and sky and the wind playing with all that shit

and pictured some vehicle driving around with wobbly suspension and antennae and whatnot

[kind sparrow]

haha

I'm taking over some structural dynamics sims at work rn so I am learning stuff that will potentially result in lots of wobbly suspension and antennae if I ever make it to SRS DLC #137

[night shoal]

ah sweet

bendy trusses and whatnot 🙂

[kind sparrow]

in this case vibration analysis but yeah

[night shoal]

noice

also 3d visualizationisms involved?

[kind sparrow]

not yet beyond some debug 3D plotting I've done as a basic vis

but maybe I will later if I can justify spending time on it

otherwise the output is just plotting transfer functions and stuff

power spectral density

[night shoal]

your colleagues should hold on to their knickers then

when you drop the srs plugin for that thing : >

where you can walk around in that simulation and inspect values from a virtual camera, or even V R

[kind sparrow]

haha

[night shoal]

like they did when they build CERN

[kind sparrow]

I have actually used SRS code for work stuff lol

in small pieces

[night shoal]

i knew it

hehe

[kind sparrow]

E.g. when I was working on geodesy stuff I transcribed some of the math code from SRS to do my work calculations

same for some aircraft attitude problems

[night shoal]

thats cool when you can apply stuff from your hobby to your work, or other way around

then you did something right 🙂

[kind sparrow]

It's surprising how clunky some of the APIs are when it comes to spatial math stuff, like in matlab or python (even with numpy) I find myself whipping up some glm-like functions that let me do math the way I'm used to doing in GP

[night shoal]

weird, you would think those maffs libs come with those things, neh

since they are so widely used

[kind sparrow]

They kinda do but they're just clunky

like in numpy if you want to take a bunch of vectors and turn them into homogeneous coordinates to multiply them all by a matrix, you have to do a few steps to append the column of 1s and multiply them by the matrix and stuff, so I just made a function that does it in one step

small QOL things

Similar to the debug graphics API I have in my engine

being able to plot a matrix directly as basis vectors in 3D

[night shoal]

in 2005 or so, i rewrote a whole workflow engine from scratch in c# as a demo, see if i can actually do it or not, since iat the time i was working with something java based (had a visual editor in c# which i was able to reuse) but my crude c# bs was magnitudes faster, the core of the engine is basically just xslt transformations, from any input to any output, forks and joins and whatnot

[kind sparrow]

haha cool

[night shoal]

i was just a trainee at that time too hehe

[night shoal]

oh, thats what i would have expected to be part of numpy tbh

[kind sparrow]

it is in some limited ways just not as flexible as I'd like

usually it's fine, there are just a few awkward parts that I fixed with bits I have in SRS too basically

[night shoal]

even the VectorGeometry.pas pascal file for all the math stuff had it built in back when i was fiddling with Delphi

from homogeneous to affine and other way around, overloads for everything you need

[kind sparrow]

yeah

numpy is more general, like matlab

[night shoal]

sucks

[kind sparrow]

the helper functions it provides are hit or miss, but 90% of the time it's good

I use it for everything

[night shoal]

if it works most of the time then its probably alright

and then you just add whatever is missing fgor your special corner case

[kind sparrow]

yeah

[gilded shell]

you have been coding since before i was born lol

[gilded shell]

does mdi allow you to change bindings at all?

[kind sparrow]

No

It's the same as regular instancing

[gilded shell]

oh wow

then yeah bindless all the way

[kind sparrow]

Personally I juse use array textures

[night shoal]

yeah, started fiddling with computers in 1990 🙂

[gilded shell]

dang

what about you @kind sparrow

[kind sparrow]

2016

[gilded shell]

oh okay

[kind sparrow]

That's when I started programming

But I was fiddling with computers since maybe 2010 and using them since 2004 or something

Just playing games and drawing on macpaint and stuff

Before that I guess too in limited capacity

I don't think macpaint existed anymore in 2004 so maybe that was earlier

[gilded shell]

yeah

dinosaurs in my chat

[kind sparrow]

Lmao

Most people here are over the age of 25

[gilded shell]

yeah

graduates i assume

[kind sparrow]

Yes

Since most people aren't really equipped to do GP until then

[gilded shell]

yeah

[gilded shell]

~4.5 ms

ignore this im just writing it down

[gilded shell]

I wonder if I should use compute shaders for my mdi

so I can properly choose which SSBOs to read from in vert/frag shaders

and to calculate transforms

[night shoal]

try naive first

then use compute to cull and fill the indirectbuffer with the surviving geometry

[gilded shell]

wdym

[night shoal]

no compute shaders

frustum culling on cpu, populating the indirect/parameters buffer for glMultiDrawIndirectCount

[kind sparrow]

Yes this is good enough for most purposes

Your application may not have enough stuff going on to get much benefit from computer draw generation idk

[gilded shell]

how would that work?

i know you can specify the offsets of each parameter buffer letting you put extra stuff in there ig

but how would you access that extra stuff?

[gilded shell]

it looks like on vulkan you can have a full command buffer for bindless rendering

im jealous now

[kind sparrow]

Which stuff

[night shoal]

you have all your meshes, have all their aabbs, you cull those aabbs against your view frustum, the ones which survive, you add to the indirect buffer, and put the count of all of them in the parameterbuffer

then you bind both buffers and glMultiDrawElementsIndirectCount()

next frame, you do the same, as you might have moved your camera around, frustum changes, you cull all meshes using their aabbs

etc

[kind sparrow]

Maybe worth doing some regular instancing before trying MDI

[night shoal]

yea

and funny enough

if you slowly progress through glDrawArrays, glDrawElements, glDrawElementsInstanced, glDrawElementsInstancedBaseVertex, glDrawElementsInstancedBaseVertexBaseInstance etc you arrive naturally at glDrawIndirect and glMultiDrawIndirect and then the peak of MDIism glMultiDrawElementsIndirectCount hehe

and you can see how all the params of the DrawIndirectElementsCommand came to be

next evolution then is, doing the culling all on the gpu

with compute

then hiz occlusion culling

and then the latest thing, you meshlettify all your geometry and hiz occlusion cull all of them

[gilded shell]

makes sense

[night shoal]

if you have a lot of geometry that is

which overlaps a lot

otherwise a shrimple glDrawElementsInstanced is probably the best

[gilded shell]

I think the structure I will take is

have an array with elements for each primitive/meshlet being rendered containing indices to transforms, materials, etc.

and then have arrays for said transforms and materials

all on GPU side

[night shoal]

use buffers instead of arrays

[gilded shell]

yeah what I mean

buffers of arrays

[night shoal]

buffers of elements : )

sorry for being picky

[gilded shell]

the whole point is to never bind

so i can do mdi

[night shoal]

its not a sin to bind stuff

[gilded shell]

i know but I literally cannot bind

[night shoal]

you might have a depth prepass, a deferred render pass, lights and shadows and whatnot

those require binding of sorts

[gilded shell]

as far as i know mdi doesn't let you bind between draws

[night shoal]

ye its in the nature of the thing

bind all the things you need for the pass to be able to draw its stuff

[gilded shell]

so to go around that I would have to do this

[night shoal]

dont overthink it or overengineer it, you also might need to bind some stuff for debugging purposes

like a line renderer or so to visualize aabbs or frustums

[gilded shell]

right

but then no mdi

[night shoal]

mdi is still very possible

[kind sparrow]

You have a couple thousand state changes per frame before you start getting bottlenecked by the driver

[night shoal]

just in a different renderpass then

yeah also texture binds is what used to be the costly part, but since you go for bindless textures already thats gone pretty much

[gilded shell]

okay

[gilded shell]

how so?

[night shoal]

like i just mentioned

[kind sparrow]

MDI is just another type of draw call

[night shoal]

you might have a depth prepass, then some deferred geometry pass, then a pass lights and shadows pass, then a pass to resolve the two

[kind sparrow]

You don't have to put every single thing into one MDI call, that's not really desirable

[night shoal]

and in the depth prepass and deferred you ahve the mega vbo/ibo bound for mdi

yeah

[gilded shell]

I see

change of plans then

[night shoal]

all the post effects if you ever add some will do their own thing

[gilded shell]

i'll stick to singular draw indirect which I assume is about a nanosecond faster lol

and upgrade to mdi in the future if needed

[night shoal]

drawindirect and mdi is virtually the same

[gilded shell]

correct

[night shoal]

but starting with non indirect draw is a good start too

it does have implications on the "renderer" "architecture" a bit but thats just 1 vs many (2 per mesh)buffers really

[gilded shell]

I think I'm just bikeshedding the wrong stuff

[night shoal]

: )

thats what we like to do too hehe

[kind sparrow]

Well it depends on your goals I guess, deccer and I are fairly practical about this so our advice is always going to be "you don't need to bikeshed this until your application demands it"

But if you don't want to make an application and you just want to make a meshlet bikeshedding application then that's up to you

But you can't really dive into it without some decent experience using the API the normal way imo

[gilded shell]

I plan to implement gi, so I do want to optimize this well but I really doubt I'm cpu bound anyways

I do wonder how people are drawing sponza at 0.1ms though

probably just culling

honestly before I do anything else I want some OIT

and maybe alpha to coverage

[kind sparrow]

OIT supplants A2C as far as I know

The point of A2C is to avoid blending altogether

[gilded shell]

ohh

I thought it was a technique to reduce aliasing

[night shoal]

: )

[gilded shell]

I read a paper on something called "fourier opacity mapped OIT"

it sounds like a neat place to start

[night shoal]

for some reason i have old game magazines in front of my third eye right now explaining MSAA with some telefone cable poles in games

[kind sparrow]

Lol sounds ambitious

[gilded shell]

perhaps

[kind sparrow]

I guess if you're just taking stuff straight off the paper maybe it's doable

[gilded shell]

if it's too much I can always fall back to whatever method logl has up

[kind sparrow]

I wish there were more test scenes that were better than sponza and bistro

[gilded shell]

weighed blended

[night shoal]

i have this neat antenna thing from mjp's shadow demo thingy

[kind sparrow]

I've never rendered them but afaik they are basically just one giant mesh right? Nothing to really instance

[gilded shell]

uhhhh

[night shoal]

yeah more or less

lustri is also complaining about it

[gilded shell]

I think bistro has some identical bush meshes

but repeated less than 10 times so most likely not worth any instancing

[kind sparrow]

Yeah that's not that interesting though, what we need is a test scene that's like an actual game and not just a blender scene

[night shoal]

thats why i always wanted to make some "bricks" in various default sizes 1x1x2, 1x1x3, and so forth per axis, and "rebuild" sponza out of those, would be suitable for some fisiks testing too hehe

[gilded shell]

lol

actually

doesnt sponza have that 1 thousand candles add on

[night shoal]

yes

and ivy

and curtains

too bad the fucking knight model only comes in obscure file formats

[gilded shell]

fr

[night shoal]

<rant> i have no fucking clue whats so hard to properly name the model, its parts and whatever else is in there, and then export them into all the fileformats they need individually </rant>

[kind sparrow]

I'm going to be the first one here whose engine includes a glTF exporter and then others can render SRS scenes in their engines lol

[gilded shell]

absolute W

[night shoal]

i screenshotted that

i enjoy being distracted grumpy

otherwise i would need to attend my 4k lines of mess.cpp

[gilded shell]

yeah

doing controlled sixteenth notes on a single kick is hard

but i refuse to buy a double pedal

[kind sparrow]

Rip

Double pedals are great

[gilded shell]

well

people can do 32nds on their doubles

[kind sparrow]

Although I exclusively play 16ths on the kick lol

[gilded shell]

so naturally, 16ths on singles are possible

[kind sparrow]

I thought you meant sustained

[gilded shell]

wdym sustained

as in doing it for a while?

[kind sparrow]

Yeah

[gilded shell]

yeah that's what I mean

[kind sparrow]

I can play 8ths sustained on one pedal up to a certain bpm but not 16ths

Idk how you'd do that unless at very low bpm

[gilded shell]

https://youtu.be/edLhkzAynY4?si=sOLAVGIzCkvofHON&t=283

[gilded shell]

to my understanding its much like 32nds with your hands: much less technique and mostly just brute force practicing it

[kind sparrow]

Where does he play 16ths on the kick

[gilded shell]

he does 32nds

while doing those cymbal rolls

[kind sparrow]

Oh I see it

He's got a double pedal though

[gilded shell]

still thats an equivelant of 16ths on one foot

unless hes doing some sort of double stroke

[kind sparrow]

Timestamp? I'm at work I can't watch the whole thing

[gilded shell]

4:43

but the whole video is certainly worth a watch 😁

[kind sparrow]

Those sound like 16ths to me

[gilded shell]

he speeds up to 32nds

this isnt metal, the song is like 120

[kind sparrow]

Ah I'm counting it at cut time I guess

[gilded shell]

yeah

im learning this song

its hard lol

[kind sparrow]

I can sustain 8ths at 200 I think

Or could rather

Haven't played in years

[gilded shell]

16ths at 100 sound more impressive

theres plenty of songs at 100

[kind sparrow]

Well that's 8ths doing a blast beat on a single foot so I can do 16ths on a double pedal at that BPM

[gilded shell]

yeah

[kind sparrow]

That's pushing the range of what I could sustain for more than 30 seconds or so though

Ok I found maybe the only recording of myself playing that I can find online, it's two songs I tracked for my friend's meme grindcore project lol

I will DM you a link so as to not dox myself

[gilded shell]

lol

[gilded shell]

darn, none of my projects are cool enough to go in that comp

[night shoal]

dont say that 🙂

or take it as an inspiration

like these tiktokers who ask random 1 man burgerflippers if they can make them a 60s intro video 😛

[gilded shell]

put the deadline a year ahead and there might finally be some cool stuff in here lol

[kind sparrow]

Is this compilation thing going to be publicly associated with this server, I don't want it to attract more windowlickers

[gilded shell]

i was a windowlicker once

[night shoal]

#1252708855176106086

[kind sparrow]

Advertising it as a place to come hang out with the cool kids or see cool projects without contributing

[kind sparrow]

No you weren't

[night shoal]

i like aphex twin's window licker

[night shoal]

not sure, perhaps there could be various formats, jake is the brains behind this idea

time to create an alt account, demon 😛

where nobody will totally recognize all the DLCs which went into srs when showing srs reel

[gilded shell]

maybe we just show it to our wives/gfs and no one else

[night shoal]

that also works

[kind sparrow]

I would include SRS if there was no way a viewer could follow it back to this server

[night shoal]

i will also refrain from fan commentary under that video then

[gilded shell]

i would love to see a demo of what you have so far

all ive seen are some models and a terrain and fooliage system

[kind sparrow]

There's not a whole lot to see yet tbh I just test it in small pieces and right now it's all taken apart to rewrite most of the whole game

The most cohesive things to see is just at the top of my thread basically

[gilded shell]

lol

i need to escape this habit of doing some tiny task then being done for the day

[kind sparrow]

I need to stop anxiously avoiding my projects because of the crushing weight of their difficulty

I waste a lot of time

Also I feel bad working on that one because there's another I'm supposed to be working on

[gilded shell]

huh

[kind sparrow]

I am beholden to another volunteer project that I can't disclose but it's a huge PITA to work on so I've been avoiding it

[gilded shell]

i see

[night shoal]

i could temporarily ban you from if that helps : >

[gilded shell]

i think my ubo alignment is all messed up

ah

I think my booleans were being optimized to 1 byte or something

anyways it took embarrasingly long to get this demo up

[gilded shell]

True order independent transparency

(weighed blended)

[gilded shell]

I have dozens of gl objects that i dont delete when the program closes

"ill handle it later" lol

[gilded shell]

Forgive me

but I might make a Scratch project

[kind sparrow]

Oh no lol

Graphics not masochistic enough is it

[gilded shell]

yeah

[gilded shell]

now would be an appropriate time

https://tenor.com/view/voices-in-my-head-the-voices-meme-meme-asylum-gif-25250651

[gilded shell]

I have finally gotten my model class written

theres a couple more things to do to finalize it but other than that were good

and I am proud to report that I did not write a Scratch project

[gilded shell]

And—I might switch to vulkan later down the road for rt support

we’ll have to see

[gilded shell]

Tbh

I might just

step through the hierarchy a second time for transparent primitives

I doubt it's much slower than making some sort of queue system

and if it's a perf problem I can probably just store two seperate hierarchies in the future

[night shoal]

the material should be able to tell you whether its transparent or not

and from that you can divide your primitives in Opaque and Transparent-isms

i suppose

[gilded shell]

There is an issue

blender didn't actually correctly write the materials upon export

[night shoal]

spooky

there is a correct bistro model (as glb) out somewhere, check the searchbox

[gilded shell]

perhaps this? #1237853896471220314 message

[night shoal]

idk thats a foreign discord

it was mentioned in #ray-tracing somewhere, perhaps ping jaker or BayBoyKiller

i believe those 2 were referring to the good one some time agio

[gilded shell]

ill try jaker

[gilded shell]

The gltf file is worse than I imagined

it's storing duplicate texture-sampler pairs

ig I gotta make a hash table now 😐

[gilded shell]

the fix was simpler than imagined yet took some time to write

if (!set.contains(mTextures[i].bindlessHandle))
{
    glMakeTextureHandleResidentARB(mTextures[i].bindlessHandle);
    set.insert(mTextures[i].bindlessHandle);
}

[night shoal]

use the variant which takes samplers into account already

i believe i mentioned that some time ago, that when you go bindless to also look at sampler objects

otherwise you have no way of using different samplers per texture

[gilded shell]

thats what I do

[night shoal]

ah right the sampler and texture goes into the GetSampledTextureHandleThingARB not this function

i

[gilded shell]

yeah lol

i am smart

[gilded shell]

shader hot reloading

[gilded shell]

@night shoal cool finding

those depth issues with the sponza decals don't occur with OIT for some reason

[night shoal]

interesting

[gilded shell]

todo: default sampler

[gilded shell]

5 on the ap csa exam

[gilded shell]

finally correctly rendering sponza

[west trail]

That's bistro mate

[gilded shell]

oops

[gilded shell]

normal mapping

[oak moat]

doesn't look very normal to me

[gilded shell]

its just lighting

everything is white, and the light color is red

the code is all updated on the github if anyone wants to see

https://github.com/GrumStomper29/OpenGL-Sandbox

[night shoal]

i dont like the highfov

[gilded shell]

do you know its high from looking at screenshots, the code, or by running the code?

[night shoal]

looking at the screenshots

[gilded shell]

i changed it to 75

[night shoal]

dont get me wrong, just beacuse i dont like it doesnt mean you have to change it 🙂

[gilded shell]

i mean i didnt really like it either lol

would die for some code review though

[night shoal]

i scrolled through a little and i think its perfectly fine

i personally dont use glGetXXXLocation to map strings to locations, i just use locations directly because they will never change really

there is a slight problem with creating textures, regarding color space/format, diffuse or albedo and emissive textures need to (or should) be in GL_SRGB8_ALPHA8, the rest in whatever... when you add compressed textures later then that matters for normals and single channel stuff too (the BC format, not srgb)

you can also save some bandwidth, by encoding normals into an uint using orthogonal encoding and glm::packSnorm2x16 (plus its counterpart in glsl to unpack and decode orthogonal into vec3)

looks like you create materialbuffer etc per model you load

that most likely will work when you render each model individually (bistro + deccer cubes + sponza + whatever)

hehe reading c++ which doesnt use trailing type thingy syntax is weird, now that i kind of like it and use it everywhere

[gilded shell]

I see

[gilded shell]

lol

i personally don’t like it

if you noticed i also dont store vertex tangents at all

A lot of models don’t include it anyways so I just calculate them on the fly with no major perf penalty

[night shoal]

gltf will provide tangents

[gilded shell]

optionally

[night shoal]

tick "tangents" when exporting gltf

and they will be vec4s

[gilded shell]

It’s just easier to calculate them lol

[night shoal]

sure fallback to have makes sense

[gilded shell]

anyways, what do you mean compress them later?

[kind sparrow]

I don't like the trailing type thingy either

Declaring everything as auto only to separately declare its return type is just pointless imo

But to each their own

The thing about C++ is that there are so many ways to do things that you just need to learn to be flexible with different code styles with it

[night shoal]

yeah

[gilded shell]

yeah

I just dislike it because it makes the actual function name harder to find

[night shoal]

i do use fuzzy finders in whatever tool when present, but i understand what you are saying

[gilded shell]

deferred shading

[night shoal]

im jealous 🙂

[gilded shell]

Why?

[night shoal]

my renderrer is fucked atm : >

[gilded shell]

Language

but mine too slightly

those lamps hanging from the rope should be colored

[night shoal]

indeed

[gilded shell]

I also need to debug this

obviously it's working but I'm still scared that it's somehow not

It might be something about emissiveness

[night shoal]

i think they have emissive properties yeah, but also some diffuse

[gilded shell]

Well im not rendering emissive properties

nor do I plan to

I went to a gltf viewer and turned emissiveness off, and it looks like the colors do dissapear

so it's fine

[night shoal]

[gilded shell]

time to choose a meshlet generator

any suggestions?

also i went and saw dmb last night

their drummer is insane

[gilded shell]

this is not real

[gilded shell]

maybe you meant this https://github.com/zeux/meshoptimizer

I've downloaded it

the documentation seems very lacking

ohhhhh nvm

[night shoal]

: )

[gilded shell]

I am a little confused by what I'm left with after using meshopt to generate meshlets

there's meshletVertices which are meant to be indices to the original vertex buffer

so I suppose the new meshlet vertices are sorted

and then there's meshletTriangles

which ig is the actual index buffer, but it's std::vector<unsigned char>

[gilded shell]

maybe theyre just 8 bit uints?

yeah that would seem so

[gilded shell]

why not just use uint8_t though

[gilded shell]

Taking a week off as I plan to go on vacation

[night shoal]

https://tenor.com/view/just-relax-have-a-good-time-enjoy-it-have-fun-chill-gif-17170907

[gilded shell]

I know its far off

but when I’m done with college, I want to make a really good 2D game and distribute it for free

using my own engine and music of course

[night shoal]

im looking forward to it

[gilded shell]

The thing is I kinda want to release it anonymously

I think that could be cool

or at the very least be silent about it during development

[night shoal]

you could release it as StrubGromper69

[gilded shell]

I’d probably use my real name

which is a great reason to not dox myself here

maybe I will personally DM you and demon about it lol

[kind sparrow]

Yea the separation of identities is pretty tricky

It's hard to keep them perfectly isolated

I have a more professional handle that I don't mind being associated with myself but obviously I do all my main discording and stuff with this one so if I were to release under another name then everyone who's seen me here will know the connection between the identities

[gilded shell]

Interesting

well if a side scroller is released in 2030 with dave matthews styled music youll all know who made it

[kind sparrow]

I'll keep my eyes peeled

[gilded shell]

goodness gracious water is so beautiful

[kind sparrow]

Midwesterner moment hehe

[gilded shell]

Actually I’m from Maryland

but md water sucks so true

[kind sparrow]

Oh I thought you were from Ohio for some reason

[kind sparrow]

Seeing the curvature of the earth always reinvigorates my desire to stick with my round earth terrain

[oak moat]

just assume earth is a 10 dimensional object with a perfectly flat projection in 3D

[gilded shell]

waterposting gets this thread more activity than actually taking about the project

well I love water so I’m not too bothered by that

[kind sparrow]

Gotta do more bikeshedding

[gilded shell]

there’s no code to bikeshed

i impress myself with how small my programs tend to be

[gilded shell]

https://youtu.be/gsZiJeaMO48?si=CKvEquNACDeq4E8f

I found this video to be a good resource

[gilded shell]

i am home now

[night shoal]

wb

[gilded shell]

I hate these bugs:

    throw;

std::size_t maxMeshlets{ meshopt_buildMeshletsBound(tmpIndices.size(), 64, 124) };```
It doesn't throw, but meshopt_build fails because 
`assert(index_count % 3 == 0);`

alright we're getting somewhere

Okay index bufs are per meshlet

[gilded shell]

a lot closer

even closer

the meshlets seem to be cut short

[gilded shell]

meshlet gen works, but that frametime 😓

[gilded shell]

7x slower

I need to use smaller indices, and maybe mdi for the meshlets

ahhh I see why its slower

I updated my little stats thing and before meshlets, there are 1538 drawcalls, and now with meshlets there are 50213

[junior sparrow]

How slow we are talking about? Also what gpu do you hve?

[night shoal]

the trick is now to cull all the non visible meshlets

[gilded shell]

35ms for a frame on rtx 3050

[gilded shell]

true but theres still 50k drawcalls lol

mdi is much needed here

[junior sparrow]

jesus on my 1050 its 7ms without culling

[gilded shell]

how many drawcalls lol

[junior sparrow]

58k

but indirect

[gilded shell]

exactly

all 50k calls are over the pcie

[junior sparrow]

the opengl driver has to have insane overhead

for just executing drawcalls

[gilded shell]

yeah

try direct calls

[junior sparrow]

not really possible cpu almost doesnt know what is loaded

[gilded shell]

uh

you can do singular draw indirects

thats what im doing

bind the indirect buffer then call drawindirect

[junior sparrow]

Its not simple for me to just switch it

[gilded shell]

unless you’re using vulkan, im not sure what the equivalent will be

God willing, I’ll start with MDI and see if that resolves it

[junior sparrow]

i am using vulkan but switching things around would be too much effort

only thing missing is lods and sw raster

[gilded shell]

i see

[junior sparrow]

wait for hiz its going to be big pain in the ass

I still havent nailed it. There are some edge cases that are hard to nail down

[gilded shell]

Time to bikeshed drawcall batching

Ideally I want to use one MDI per mesh so that all meshlet draw calls are indirect

however I still need to sneak culling into that process

afaik you can't have MDI skip an instance

I have heard of mesh and meshlet shaders but iirc theyre vender specific so I want to avoid those

The vulkan switch may come sooner than we thought

[kind sparrow]

What are you skipping instances based on

Is this for culling

[gilded shell]

yeah

[kind sparrow]

Normally you just build up the list of draws from scratch and you skip in that process

[gilded shell]

so

via compute i generate the indirect draw arrays each frame?

[kind sparrow]

Or just on the host, whenever you build up the draw data

Oh wait this is for meshlets is it

[gilded shell]

yeah

i do have meshlet data on host

[kind sparrow]

Ok I guess it's still similar, but you have sort of a fixed "whole scene" buffer that's static and then fill the MDI buffer and transform buffer out of that I suppose

I'm not really sure

The difference for me is that normally your draw data is changing every frame

[gilded shell]

right

[kind sparrow]

But if you're just selecting meshlets to be drawn out of a fixed array it's a bit different

You'd either have to copy the scene data into a compact instance buffer, sort the scene data, or add another layer of indirection so that your instances can randomly access the meshlets they're drawing

[gilded shell]

but the takeaway is im rebuilding indirect arrays each frame, right?

[kind sparrow]

Most likely

How do the meshlets work, is each one an indirect draw?

[gilded shell]

yeah

[kind sparrow]

I don't see how they could be instanced

Ok

Yeah just rebuild the draw list

[gilded shell]

so i could get the scene represented on gpu

then use compute to gen the arrays

[kind sparrow]

You can do this on the host too which is easier to start with but yeah you can do it in compute as well

[gilded shell]

okay

this would be a lot easier if i could make a drawcall from a compute shader

[kind sparrow]

Wdym isn't that what we're talking about doing

[gilded shell]

i thought we were using compute to make the draw arrays then make the few drawcalls from the cpu?

[kind sparrow]

The CPU just does kicks off the indirect work

[gilded shell]

so why do we need to write the arrays if the compute shaders can just directly make the drawcalls?

[kind sparrow]

Wait is there no transformation data for the meshlets

[gilded shell]

there’s material/transform data per mesh but not meshlet

[kind sparrow]

Ah ok

Then I guess you don't need the arrays

But either way you're just filling the indirect draw buffer every time

[gilded shell]

excellent

this seems a little bit more trivial

[night shoal]

meshlets are nothing else but ordinary mesh primitives

you treat them the same way as your cube or sphere or motorcycle or deccer cube

they all go in your whole scene geometry/andotherwise indirection buffers

then on the gpu you ask the compute shader to cull invisible ones

and the survivors are appended into the indirect(and parameter buffer if you use mdic) buffer for the next mdi call

[junior sparrow]

First what I would do move the data needed for rendering to gpu. So materials, transforms, mesh groups and meshes are on gpu in their own buffers and are reachable from shaders.
My simple render loop looked like:

1. compute dispatch which reads a scene data and writes a indirect compute dispatch with ammount of mesh groups
2. indirect compute shader goes over all meshgroups and their meshes to write all meshlets into one giant buffer with a counter.
3. compute shader that read a counter and writes a single indirect draw call command with the instance count which equals to meshlet count in the scene
4. indirect draw call which reads a previously written buffer with meshlet data and renders them.

This is my vertex shader that renders meshlets

I abuse instancing to render meshlets

[kind sparrow]

That's true if the meshlets are all the same size you can just use instancing and pull the vertices huh

[junior sparrow]

I set vertex count to a max of vertices that meshlet can have and if meshlets has less then max I clip out the vertex

[night shoal]

instancing for meshlets?

[junior sparrow]

As I said I abuse instancing

[night shoal]

is their vertex data always the same? hehe

ah

[kind sparrow]

Yeah if each meshlet is e.g. 64 vertices then you can instance a 64-vert mesh and use the instanceID to fetch the vertices from an SSBO

[junior sparrow]

if you mean if they have different amount of vertices you can simply handle it same way I do

[junior sparrow]

you can instance them even tho they have different sizes

its not optimal but vertex shaders are fast enough

[kind sparrow]

How do you discard the unused invocations, just emit NaN?

[junior sparrow]

yes ☝️

I linked the shader how I do it

[night shoal]

indeed, i was thinking wrong

with vertex fetch you can do whatever you like

[junior sparrow]

https://tenor.com/view/mando-way-this-is-the-way-mandalorian-star-wars-gif-25407126

[night shoal]

i also have yet to descend into the meshlet realm

[junior sparrow]

If you can use mesh shaders its much better

otherwise you have to cope with compute and vertex shaders

[night shoal]

yeah so i heard

fortunately i have some 40xx grade hardware

but a fallback would also be nice, for the learning purposisms

[kind sparrow]

My entire player character is a meshlet

[night shoal]

sounds like an insult if you say it like that

[junior sparrow]

learning purposes for sure but realistic wise you can just assume you will have mesh shaders

[kind sparrow]

Well I am in the 11.2% and am not sure I agree with this message

I want to be able to play Deccer's game

Although I'm going to put a 30xx in my other computer eventually so all will be forgiven if he decides to go the mesh shader route

[junior sparrow]

This is only discrete gpus on steam

[kind sparrow]

Actually weren't there whisperings of cross-platform mesh shaders coming in GL soon

Although idk if that would work on old cards still

[junior sparrow]

I assume most players who will want to play my game will have one

[kind sparrow]

Are mesh shaders a hardware thing

[junior sparrow]

sadly they are

turing+

[kind sparrow]

rip

[junior sparrow]

so gtx 1660 up in nvidia terms

[gilded shell]

is there anything non-vendor specific?

[kind sparrow]

Vulkan

But it still requires a newer card

[gilded shell]

oh brother

i am kinda tired of dealing with ogl's crap

[kind sparrow]

Like what

[gilded shell]

only having vendor specific mesh shaders

[kind sparrow]

Lol

[gilded shell]

lack of hw rt support

and overall inconsistencies

[junior sparrow]

this is non vendor specific but vulkan

[gilded shell]

I don’t want raw vulkan though; i think an abstraction would be better

[junior sparrow]

I can recommend daxa

[gilded shell]

is that locally made here in this server?

[junior sparrow]

made by lpotrick

and gabe rundlet with saky

[gilded shell]

hmm

[junior sparrow]

very easy to use

[gilded shell]

a quick glance tells me it might be too high level

[junior sparrow]

it exposes everything you want to use

but no raw vulkan

[gilded shell]

i see

ill keep this in mind, thanks

[junior sparrow]

also to mention it has rendergraph

[gilded shell]

uh oh

what’s that?

[junior sparrow]

it handles synchronization for you and easier to code stuff

[gilded shell]

i see

[junior sparrow]

also it requires to share code between C++ and glsl/slang

[night shoal]

OpenGL also supports mesh shaders

on nwidea hardware

[junior sparrow]

only vendor specific extensions

mesh shaders are also coming to amd

[night shoal]

yeah, turns out i cannot read, as usual

@gilded shell you can find the daxa thread in #1019722539116802068

[junior sparrow]

kinda dead tho

[night shoal]

https://tenor.com/view/cat-revives-gif-24153050

[gilded shell]

lol

[junior sparrow]

daxa has its own server but that one has occasional activity

[gilded shell]

i changed my mind

raw bulkan sounds funner

[junior sparrow]

https://tenor.com/view/copium-gif-766857345458198993

[night shoal]

good thing is you can still pester potti or gabe or any other frog who took the daxa pill

[junior sparrow]

daxa pill is very nice

[kind sparrow]

If you want to make a specific game then raw Vulkan is fine

If you want to make a flexible engine or sandbox tool for graphics prototyping you should use something like vuk or daxa

[gilded shell]

i see

vuk vs. daxa?

[kind sparrow]

Otherwise you're going to just spend huge numbers of hours writing infrastructure code

[night shoal]

we need a srs contender

aptly named "srsly"

[gilded shell]

i suppose calling this a sandbox is wrong because i do have end goals

[night shoal]

famous last words

[gilded shell]

but i do want this to be sandbox-like

[junior sparrow]

thats up to you vuk is also nice but I like daxa for better ergonomics and ability to annoy lpotrick

[kind sparrow]

End goals like what though

End goals as in a specific game or end goals as in a certain set of graphics features

[gilded shell]

features

which is why for now im going with daxa or vuk 👍

maybe vuk because it has a better name

[kind sparrow]

It's also made by a funnier person

[night shoal]

vuk has a dedicated server too, its in #related-servers martty is on travelz atm might not get immediate feedback atm

but Hek is there to help as well

[gilded shell]

i see

[junior sparrow]

with daxa you have to either infer from the code or timerdoodle or one of my projects

Also you can ask me

[gilded shell]

daxa does have a dedicated tutorial too

[junior sparrow]

I dont think its up to date but its really close

there were breaking changes and lpotrick is lazy to write it

[gilded shell]

darn

[junior sparrow]

I would rather consult one of my projects or timberdoodle

I checked the tutorial and it should work fine
https://tutorial.learndaxa.com/installing-dependencies/

[oak moat]

why is none of you suggesting the good ways of doing meshlets without mesh shaders

@gilded shell #1262676828322271293 message

[gilded shell]

legend

thank you

[gilded shell]

@oak moat if I’m “decoding” a vertex index in the vert shader, doesn’t that mean i cant use conventional vertex buffer layouts where the shader fetches the vertex automatically?

[junior sparrow]

you fetch from buffers

[junior sparrow]

I am using bda/pointers☝️

[oak moat]

you can't do the classic vertex buffer stuff, yes

[gilded shell]

i see

[gilded shell]

"dispatch a compute shader to write an index buffer of size meshletCount * maxPrimitives * 3 that encodes primitiveID and meshletID, then in the vertex shader you can decode the vertex index and proceed as usual"

Not sure why it needs to write the index buffer

[night shoal]

you bind it as an indexbuffer

and make use of indexed drawing as usual

[gilded shell]

right but why? I already have index buffers; rewriting them with compute seems wasteful

I can understand the primitive and meshlet id part

[junior sparrow]

its not index buffer in traditional sense

its buffer that contains indices for meshlet

[oak moat]

I mean you can do this offline for twice the memory usage

[gilded shell]

why not do it offline and drop the original indices?

[junior sparrow]

here you have highlighted code to understand
https://github.com/bevyengine/bevy/blob/a6f8f62748677a8a04b86231870e8e6de5505bc9/crates/bevy_pbr/src/meshlet/write_index_buffer.wgsl#L41
https://github.com/bevyengine/bevy/blob/a6f8f62748677a8a04b86231870e8e6de5505bc9/crates/bevy_pbr/src/meshlet/visibility_buffer_raster.wgsl#L41-L44

[oak moat]

you will pay twice the memory usage

[gilded shell]

I do not understand why

[junior sparrow]

because you write same info 3 times instead of just once

same packed index for every vertex in triangle or you could do it once per triangle

[gilded shell]

Sorry but I'm incredibly confused lol

[oak moat]

I will simplify

[oak moat]

suppose you do this offline

you have a buffer of size N

now suppose you want culling

[gilded shell]

just to clarify offline means on the CPU at loadtime?

[oak moat]

it doesn't matter

point is that it's not done in compute

[oak moat]

you need somewhere to store the culling result

worst case for culling is that every meshlet and every triangle is visible

so the worst case size for the new buffer that contains the culling result is N again

therefore the total size is 2N

[gilded shell]

right

so how does doing this in compute fix this?

[oak moat]

you don't do it offline

and the culling shader is the same as your "make buffer" shader

[gilded shell]

okay

[oak moat]

I think it's better if you actually implement it

then it'll become clear

[gilded shell]

perhaps

in conclusion,

all the geometry is loaded into storage buffers and when its time to render, a compute shader runs through, does the culling, and writes out the survivors to new buffers?

or am I back at square one

[oak moat]

yes

[gilded shell]

I see

Thank you for giving me widsom

wisdom

after that, I suppose the CPU issues an MDI command that draws all those survivors

[oak moat]

no

[gilded shell]

No?

[junior sparrow]

you write the command in compute shader and cpu does indirect with the buffer

[oak moat]

actually yes I thought you were doing the other strat

yes it's a single command with indexCount set as survivingMeshlets * triangleCount * 3

[gilded shell]

of course

one more thing

is this solution for per-triangle culling as well? That would explain the index buffer rewrite

[junior sparrow]

yes

[gilded shell]

I see

I wonder how per-triangle culling can be faster than just letting them through

it depends on the method I suppose

[junior sparrow]

you reduce vertex invocations

[gilded shell]

I assume per-triangle culling would use a frustum intersection method, rather than the projection method used in vkguide 2

[oak moat]

no

you do small triangle culling

then backface culling

and then you profile if occlusion/frustum culling is worth it

typically it's not

[junior sparrow]

also from lpotrick its not worth if you dont use mesh shader

[gilded shell]

i can't really do backface culling before vertex transform though

[junior sparrow]

you can

[oak moat]

yes you can

[gilded shell]

🤔

via dot products?

[junior sparrow]

no by using determinant

[oak moat]

v0 = projView * worldTransform * vertices[indices[i0]];
v1 = projView * worldTransform * vertices[indices[i1]];
v2 = projView * worldTransform * vertices[indices[i2]];
cull = determinant(mat3(v0.xyw, v1.xyw, v2.xyw)) > 0;```

[gilded shell]

i see

you both are very smort

[oak moat]

nah zeux is

[junior sparrow]

yeah

[gilded shell]

well

my plans get closer to nanite daily

[junior sparrow]

Also I need to figure out why my hiz is not 100% correct

it just sometimes breaks

[gilded shell]

now all i need is a lod tree and sw raster lol

[oak moat]

all I need is lod tree

u-huh

let me know when you're done

[gilded shell]

just a couple business days

[junior sparrow]

I tried sw raster but it was hilariously broken like whole sponza was splitted weirdly in half and both parts were going in different direction

[gilded shell]

lol

you have to make a scratch project first

[junior sparrow]

god no

I am already suffering with my jobs

angular + php will make you hate programming

[gilded shell]

right

i actually did write a clusterizer and dynamic lod engine on Scratch and it was hell

[junior sparrow]

I saw that

I would question your sanity

[gilded shell]

yeah

[junior sparrow]

Also I do wish upon worst enemy to write javaish php code or deal with css/js spec

[gilded shell]

which is less than a pixel, no?

[oak moat]

sure

more rigorously: projected area of triangle is less than threshold (with subpixel scaling taken into account)

[gilded shell]

right

but i dont own any models with that level of fidelity lol

maybe it makes more sense for far away triangles, not sure if bistro is expansive enough for that to apply

[oak moat]

if you go far enough away every triangle is subpixel

but it is pretty marginal

it may even give you a little overhead

[gilded shell]

i see

ill probably go with the backface cull

ahhh i need to make my materials bindless too

[kind sparrow]

The thread has been pulled

[gilded shell]

what

[kind sparrow]

You're pulling the thread and now the whole thing is unravelling lol

If you're going to go down this route you may as well switch to Vulkan

[oak moat]

fake you can implement nanite without vulkan

you need vulkan only if you want to torture your GPU with CAS loops and devicecoherent memory

for that beautiful cross queue async visbuffer writing and bvh trasversal

[gilded shell]

i dont want tooooo

this meshlet frustum junk + two pass occlusion culling is all i have planned

the vulkan switch might be inevitable depending on how performance critical ray trace operations will be

[oak moat]

what kind of tracing

[gilded shell]

not sure what you mean

raytracing against a bvh for lumen styled radiance caching

[oak moat]

unless by bvh you mean distance field then you're gonna get "interactive framerates" at best

and even then it'll be painful

[gilded shell]

so

i need hw rt?

meaning i need bulkan?

im going to kill myself

[gilded shell]

when you said this i thought you were taking about how this discord thread has gotten more activity than ever

its at about 1800 messages and yesterday i think it was 1500 lol

meaning ~17% of this threads messages occurred today just because i wanted meshlets to draw a bit faster

[kind sparrow]

No I meant the sweater is unravelling

[gilded shell]

well i can either switch to vulkan now or later

[oak moat]

I would suggest avoiding anything RT

[gilded shell]

why?

[oak moat]

slow

[gilded shell]

darn

but i dont like ibl

[oak moat]

raymarching and conetracing still exist you know

also IBL is useful for sky lighting

[gilded shell]

how are those any faster?

[oak moat]

it depends

generally if you want HWRT there is only one way of doing things

with everything specified by the API

instead of bringing your own representations which may or may not work well with your scene

anyway it depends, just do whatever you want

[gilded shell]

what about shader ray queries?

[oak moat]

what about them

it's the same thing except without RT pipelines

[gilded shell]

that sounds like it nullifies this if its my own pipelines though

unless you’re talking about the scene and bvh setup?

[oak moat]

legitimately it's really just traceRayEXT but in compute

you're still using your HW's BVH format and HW's tracing algorithm and all that

it's fast*

*assuming you provide a scene that is a good fit

[gilded shell]

i see

i will take this into consideration

thank you for taking the time to help

[oak moat]

you should read more

what I said is just the tip of the berg

generally one can only say "it depends"

[gilded shell]

I know, thats the plan

[gilded shell]

Actually

I think I'll start with screenspace

since I want to use screenspace for specular reflections anyways

[gilded shell]

closer

the code in question

oh my shader version wasn't high enough

alright

all the scene's materials are now in one big ssbo

[gilded shell]

geometry now mostly bindless

this is all so that I can do aforementioned culling and batching methods

the only thing left being bound are material indices and and model transforms, which all can probably be built into another large ssbo with per-meshlet elements

after that I can use compute to build the index and mdi buffers

[night shoal]

what do you mean by geometry bindless?

[gilded shell]

only one vbo, ibo, and vao

[night shoal]

ye thats still not bindless 🙂

you mean indirect drawing perhaps

[gilded shell]

how isn’t that bindless?

[night shoal]

bindless means you dont bind anything

hence the name

like bindless textures, you just grab a handle from the resource, and pass the handle to the shader

that exists for buffers too on nvidia, not just textures

[gilded shell]

oh

well then bind-once i suppose

[night shoal]

most people refer to that as mdi or mdic

[gilded shell]

im not using multidraw indirect yet

just huge buffers

[night shoal]

ah

and glDrawElementsInstancedBaseVertexBaseInstance?

[gilded shell]

just singular draw indirect

since it lets me offset indices

[junior sparrow]

That isn't true. You need to still bind something. But you bind once per pipeline

Correct name would be bind less

[night shoal]

no bindless means you dont bind

its not "bind less textures" 🙂

[night shoal]

ah so you do use mdi pretty much, just not the multipart, but indirect draw

[junior sparrow]

in vulkan bindless is descriptor indexing and bda. you dont need to bind bda because its pointer but for descriptor indexing you need to bing big descriptor set to index into in shaders

you still have to bind but only once per pipeline

[night shoal]

ye but we are in opengl space here 🙂

no vulkan debauchery yet

[junior sparrow]

I am aware that in opengl you dont need to bind textures since nvidia has some extension but theres nothing for buffers

[night shoal]

that nvidia extension i was mentioning earlier is also BDA for bindless buffers

[junior sparrow]

hmm

[night shoal]

but for indirect draw you still need to bind the indirect (and parameter buffer)

[junior sparrow]

still stinky

vulkan is much better

[night shoal]

yeah

i agree

gl4.6 is not so bad

[junior sparrow]

Its little better but still have global state and you get surprises like nvidia doing mip mapping of 3d textures on cpu

[gilded shell]

Whenever I say bindless I mean the technique rather than the APIisms

[west trail]

You store animated and non-animated objects in single buffer? Or you dont have vertex skinning yet?

[gilded shell]

no skinning yet, why?

[kind sparrow]

Because skinning adds a ton of attributes that you'd have a second vao for

You wouldn't want to have those attributes on every mesh if they're unused

[west trail]

Still struggling to understand how to do it right. Creating 2 types of vertices, one with skinning attributes,one without and then loading them based on user choice seems wrong

[night shoal]

why user choice

you render static objects with vao 1, and all your animated ones with vao 2, or you dont use vaos at all (besides the default one for gl core)

[west trail]

loadAsStatic, loadAsAnimated. Maybe user doesnt want to import anims from gltf

• creating 2 types of vertices seems wrong to me

[night shoal]

then what?

you usually have more than 1 vertex type anyway

[west trail]

Didnt know,thx

[west trail]

Saving a bit of memory on stuff like animated gates/doors/etc. You may not want to import animations and just use them for background

[night shoal]

your artist will provide the animations used in the game

not loading animations to save memory doesnt matter

[gilded shell]

@junior sparrow @west trail Do you have project threads I can follow?

[west trail]

not me, still figuring out stuff

mind updating your github repo btw? I want to stole check out your code

[junior sparrow]

I keep switching projects
I should create a thread about what I am working on currently ofc GP related

[gilded shell]

Uh

Probably not today

its at a pretty incomplete state lol

what code are you looking for?

[west trail]

just checking out

[night shoal]

nothing wrong with creating new threads for another project : >

[gilded shell]

alright

then maybe today

but its gonna be super messy

[kind sparrow]

Gates and doors wouldn't be skinned they would be animated programmatically

We're talking about skeletal animation

[west trail]

okay, no user choice then

[west trail]

hell yeah

[gilded shell]

Watch your language in my thread please

We do not speak obscenities in this chat

[gilded shell]

@west trail its up

[west trail]

Its night time in my timezone man, let me sleep😔

[gilded shell]

i think when i finish this project id like to dabble with 2D

or not