Iris - A Journey through OpenGL and beyond to learn Graphics | Graphics Programming | Page 17

wicked notch Feb 22, 2024, 1:10 AM

#

it's still gonna merge some stuff but oftentimes it's good enough™️

#

uh it's -kn btw

#

-km was for materials

delicate rain Feb 22, 2024, 1:11 AM

#

I'll try both

#

we kinda rely on our meshes having few primitives

#

well not few but it does not scale well with increasing number of primitves

wicked notch Feb 22, 2024, 1:12 AM

#

aren't you and potrick meshlet enjoyers

#

it shouldn't matter innit

#

ah wait potrick has the instance culling madness

#

kekkedsadge

delicate rain Feb 22, 2024, 1:13 AM

#

Uh it does matter because of the way we store the data

#

each mesh has an array of primitive indices so that we can allocate GPU buffers and all

#

and the size of this array is set globally

#

(-:

wicked notch Feb 22, 2024, 1:14 AM

#

I don't follow

#

is this for software meshletisms?
(no mesh shader)

delicate rain Feb 22, 2024, 1:14 AM

#

yes no mesh shader

wicked notch Feb 22, 2024, 1:14 AM

#

then I do follow

#

rip

delicate rain Feb 22, 2024, 1:14 AM

#

yeah

#

mesh shader soon as I get new GPU

#

but even if we do mesh shaders, we still would have

struct GPUMesh
{
  PrimitveIndices[MAX_PRIMITIVES_NUM]
  Transform
  BlaBla other data
}

no?

wicked notch Feb 22, 2024, 1:16 AM

#

mm I like using ranges and having it all separate

delicate rain Feb 22, 2024, 1:17 AM

#

so your primitive indices are hiden behind another indirection?

wicked notch Feb 22, 2024, 1:17 AM

#

I set a fixed budget for max primitives and max vertices and allocate gpu buffers with those

wicked notch Feb 22, 2024, 1:17 AM

#

delicate rain so your primitive indices are hiden behind another indirection?

ye

delicate rain Feb 22, 2024, 1:17 AM

#

I see, uhh hmmm it would be like 10th indirection lmao

#

I'll talk to Patrick

wicked notch Feb 22, 2024, 1:18 AM

#

it's fine

#

we're already memory bound anyways

#

kekkedsadge

#

but I haven't profiled in depth so perhaps some investigation™️ is in order

delicate rain Feb 22, 2024, 1:27 AM

#

We'll see how it goes

wicked notch Feb 22, 2024, 10:38 PM

#

everyday I wonder why expression statements are not standard

glass sphinx Feb 26, 2024, 8:36 PM

#

wicked notch is this for software meshletisms? (no mesh shader)

no

#

thats just the meshlet layout from nvidia/ the one meshoptimizer gives us

glass sphinx Feb 26, 2024, 8:36 PM

#

delicate rain but even if we do mesh shaders, we still would have ```cpp struct GPUMesh { Pr...

its the same with and without mesh shaders

wicked notch Feb 26, 2024, 8:38 PM

#

ye, I prefer to keep stuff separate so I can avoid duplication of stuff

#

like two meshes can share a primitive, so I decouple meshes from primitives

glass sphinx Feb 26, 2024, 8:39 PM

#

delicate rain I see, uhh hmmm it would be like 10th indirection lmao

well pointer or offset inside the mesh struct to the arrays doesnt make a difference, its one indirection to these tables in the end

glass sphinx Feb 26, 2024, 8:39 PM

#

wicked notch ye, I prefer to keep stuff separate so I can avoid duplication of stuff

keep what separate?

wicked notch Feb 26, 2024, 8:39 PM

#

mesh info, meshlet info and primitive info

glass sphinx Feb 26, 2024, 8:39 PM

#

you talk in gltf terms

wicked notch Feb 26, 2024, 8:39 PM

#

oh ye primitive in the gltf meaning

glass sphinx Feb 26, 2024, 8:39 PM

#

we call primitives meshes and meshes mesh groups

wicked notch Feb 26, 2024, 8:40 PM

#

ah I see now

glass sphinx Feb 26, 2024, 8:40 PM

#

i thought you mean triangles with primitive

delicate rain Feb 26, 2024, 8:40 PM

#

glass sphinx well pointer or offset inside the mesh struct to the arrays doesnt make a differ...

I was referring to memory locality

wicked notch Feb 26, 2024, 8:40 PM

#

terminology is hard kekkedsadge

#

why does the same word mean 634712763 things

delicate rain Feb 26, 2024, 8:41 PM

#

That's why we have //gltf mesh is our mesh group and //gltf primitive is our mesh all over

glass sphinx Feb 26, 2024, 8:41 PM

#

wicked notch mesh info, meshlet info and primitive info

tido also has meshes and primitives decoupled

delicate rain Feb 26, 2024, 8:41 PM

#

👾

glass sphinx Feb 26, 2024, 8:42 PM

#

we allocate primitives into their own buffers so they are able to be unallocated if we want that

#

simpler to think about imo, one buffer = one primitive

#

doesnt really matter tho, the same in the end

glass sphinx Feb 26, 2024, 8:43 PM

#

delicate rain I was referring to memory locality

?

wicked notch Feb 26, 2024, 8:43 PM

#

ye

delicate rain Feb 26, 2024, 8:43 PM

#

glass sphinx ?

The indices were local to the meshgroup previously

glass sphinx Feb 26, 2024, 8:43 PM

#

ah the cope

#

yea

delicate rain Feb 26, 2024, 8:43 PM

#

Now they are in an array which is some other place in memory

#

Yeh

glass sphinx Feb 26, 2024, 8:44 PM

#

luckily we nearly never pay that price

#

we solve the indirection on cpu

#

when we create the draw lists

#

gpu never needs to do the indirection, it just eats mesh + entity pairs prepared from cpu for each frame

delicate rain Feb 26, 2024, 8:44 PM

#

Beast abstraction

glass sphinx Feb 26, 2024, 8:45 PM

#

now i just need to fix the retard pointer access corrupting tidos buffers

#

im close

delicate rain Feb 26, 2024, 8:45 PM

#

https://tenor.com/view/tnnd-the-noise-next-door-so-close-excited-close-gif-26060628

Tenor

wicked notch Feb 26, 2024, 9:39 PM

#

btw zeux is working on a better shrimplifier that will make my job so much easier froge_love

#

https://github.com/zeux/meshoptimizer/discussions/569

primal shadow Feb 26, 2024, 10:04 PM

#

If you want to see another meshlet impl, here's mine https://github.com/bevyengine/bevy/pull/10164 @delicate rain

GitHub

Meshlet rendering (initial feature) by JMS55 · Pull Request #10164 ...

Objective

Implements a more efficient, GPU-driven (#1342) rendering pipeline based on meshlets.
Meshes are split into small clusters of triangles called meshlets, each of which acts as a mini inde...

#

I sadly don't have mesh shaders either (webgpu)

delicate rain Feb 26, 2024, 10:17 PM

#

Thanks froge_love I'll check it out once I'm on my PC again!

primal shadow Feb 27, 2024, 4:01 AM

#

delicate rain Thanks <:froge_love:1105211408255295624> I'll check it out once I'm on my PC aga...

Happy to answer questions if you have any 🙂

wicked notch Feb 28, 2024, 1:53 AM

#

JetBrains will soon release a software to write plain text and call it "The most advanced writing tool ever"

#

though I have been liking Writerside

#

it's nice to have first class support for both markdown and LaTeX

delicate rain Feb 28, 2024, 5:57 AM

#

Reject modernity, embrace overleaf

primal shadow Feb 28, 2024, 6:01 AM

#

I used this for my thesis https://typst.app/

Typst

Typst: Compose papers faster

Focus on your text and let Typst take care of layout and formatting. Sign up now and speed up your writing process.

primal shadow Feb 28, 2024, 7:07 AM

#

@wicked notch couple questions:

#

Why do you have to lock meshlet borders before simplifying?
What METIS API do I use for partioning the graph?
How do you project whatever this "error" metric meshopt's simplify gives to screen space pixel size?
Do you use meshopt's meshlet builder for the initial clusters, or metis for that too?

#

I don't get why subdividing without locking can cause seams

#

METIS insits on 2 letter variable/API names -_-

#

the heck is ncon

#

https://docs.rs/metis/0.1.2/metis/struct.Mesh.html#method.part_nodal this one I guess?

wicked notch Feb 28, 2024, 10:09 AM

#

I use PartGraphRecursive

wicked notch Feb 28, 2024, 10:10 AM

#

primal shadow I don't get why subdividing without locking can cause seams

different parts of the mesh can be at different LOD levels, if you simplify the borders of meshlets too you will have discontinuities between LOD levels

wicked notch Feb 28, 2024, 10:11 AM

#

primal shadow the heck is ncon

number of shared nodes (vertices) between two elements (triangles) for them to have an edge

#

there is a handy manual for metis, you need to read it at least 45 times before you figure it out though

#

you can project the error in many ways, I just scale it by the size of the sphere in screen space (this is stupid though, do what unreal does)

primal shadow Feb 28, 2024, 5:40 PM

#

wicked notch there is a handy manual for metis, you need to read it at least 45 times before ...

Oh ok, didn't know that

#

Thanks!

primal shadow Feb 29, 2024, 4:59 AM

#

Do I understand the steps correctly?

Split mesh into initial meshlets
While meshlet_count > 1:
PartGraphRecursive, target partitions = meshlet_count / 4
For each group, simplify group
For each group, build meshlets within that group (same as step 1.)
Goto step 2

#

Meshoptimizer and metis already implement the hard parts for simplifying and graph partioning, I just shell out to them and move the triangle data around

#

Seems easy

#

Also mind giving me a list of all the dumb edge cases that break things that I should know about 😅 ?

wicked notch Feb 29, 2024, 11:12 AM

#

primal shadow Meshoptimizer and metis already implement the hard parts for simplifying and gra...

heh

#

hehe

#

he_doesnt_know.meme

#

yes there are so many edge cases that meshoptimizer's mesheltifier just straight up fails, also you forgor a step (building locality info)

#

the most infamous edge case are disconnected meshes

#

then you have the fact that partgraphrecursive doesn't give you a strict upper bound on partition size

#

after that you have to solve vertex welding for close vertices so you have to build locality info between edges of triangles

#

this brings its own set of edge cases like:

what if vertices are close spatially but UVs are completely opposite of each other
same thing applies for normals and tangents, but at least you can recompute those if you accept some more error

#

there are more edge cases but I have an exam

#

https://tenor.com/view/dawn-majora-finalday-gif-9607331

Tenor

faint crane Feb 29, 2024, 12:48 PM

#

Edge cases are the friends we made along the way.

primal shadow Feb 29, 2024, 5:34 PM

#

wicked notch then you have the fact that partgraphrecursive doesn't give you a strict upper b...

Does that matter? Say you get 6 meshlets per partition instead of the target of 4. Your LODs will be more inefficient, but it shouldn't break anything no?

primal shadow Feb 29, 2024, 5:34 PM

#

wicked notch after that you have to solve vertex welding for close vertices so you have to bu...

Can you expand more on this, and the locality info? Whenever you have time

wicked notch Feb 29, 2024, 5:35 PM

#

primal shadow Does that matter? Say you get 6 meshlets per partition instead of the target of ...

you can make a better graph for the GPU if you can make assumptions about partition size

#

also it's strictly required for triangle/vertex limits for meshlets no?

#

partitioning triangles will meshoptimizer will not work

#

it's unusable for any LOD level other than 0

wicked notch Feb 29, 2024, 5:36 PM

#

primal shadow Can you expand more on this, and the locality info? Whenever you have time

meshes with lots of disconnected vertices can't be simplified as well and therefore you must weld close vertices together to keep simplifying

#

locality info just refers to triangle-to-triangle (or meshlet-to-meshlet) distance info

#

building locality info means building edge weights for adjacent triangles

primal shadow Feb 29, 2024, 5:39 PM

#

wicked notch also it's strictly required for triangle/vertex limits for meshlets no?

Well let's say you end up with 6 meshlets in your partition, and after simplifying you have too many triangles for 2 meshlets. Then you just build 3 meshlets instead. Not quite the 4->2 reduction you wanted, but that's ok still. No?

primal shadow Feb 29, 2024, 5:40 PM

#

wicked notch partitioning triangles will meshoptimizer will not work

Oh. You need to use metis for building meshlets for other LODs? Damn...

wicked notch Feb 29, 2024, 5:40 PM

#

even for the leaf LODs

#

meshoptimizer will just not work, forget about it for generating meshlets

primal shadow Feb 29, 2024, 5:40 PM

#

So, meshoptjmizer for simplify only

wicked notch Feb 29, 2024, 5:40 PM

#

it prefers topological locality to spatial locality

primal shadow Feb 29, 2024, 5:40 PM

#

Sadge

wicked notch Feb 29, 2024, 5:40 PM

#

yeah

primal shadow Feb 29, 2024, 5:41 PM

#

wicked notch meshes with lots of disconnected vertices can't be simplified as well and theref...

Ehh edge cases, not important for my first attempt 🙂

#

Well in that case it's more of a problem as metis can't guarantee partition size to ensure <= 64 triangles 😭

#

So manually fixes and such necessary:/

#

That's depressing

primal shadow Mar 1, 2024, 4:16 AM

#

Even with the manual, METIS_PartGraphRecursive is so dumb

primal shadow Mar 1, 2024, 5:04 AM

#

I don't get how to build the initial meshlets

#

each node in the graph is a triangle, ok

#

what the heck are the weights supposed to be?

primal shadow Mar 1, 2024, 6:25 AM

#

How the heck do you turn those into per-triangle pair edge weights

primal shadow Mar 1, 2024, 7:24 AM

#

Atl for now I feel like I should just use meshoptimizer's meshlet builder

wicked notch Mar 1, 2024, 2:40 PM

#

I am

#

free

#

time to eep for 48 hours

primal shadow Mar 2, 2024, 6:41 AM

#

When you have time to answer questions, I'd like to know how you merge the vertex/index buffers for meshlets. I'm curious if you have to group meshlets based on their real original-mesh vertex IDs, or if using the local indices/vertices to the meshlet is fine.

wicked notch Mar 2, 2024, 10:19 AM

#

they remain the same after initial clustering

primal shadow Mar 2, 2024, 6:44 PM

#

wicked notch they remain the same after initial clustering

Wdym?

wicked notch Mar 2, 2024, 7:16 PM

#

The vertex buffer remains the same

#

the index and primitive buffers change per LOD level

wicked notch Mar 2, 2024, 7:17 PM

#

primal shadow When you have time to answer questions, I'd like to know how you merge the verte...

you should use local indices yes

primal shadow Mar 3, 2024, 5:54 AM

#

I think METIS is actually just this easy

// Group meshlets into roughly groups of 4, grouping meshlets with a high number of shared edges
let mut xadj = vec![];
let mut adjncy = vec![];
let mut adjwgt = vec![];
for meshlet_id in simplification_queue.iter() {
    xadj.push(adjncy.len() as i32);
    for (connected_meshlet_id, shared_edge_count) in
        connected_meshlets_per_meshlet[meshlet_id]
    {
        adjncy.push((connected_meshlet_id - simplification_queue[0]) as i32);
        adjwgt.push(shared_edge_count as i32);
    }
}
xadj.push(adjncy.len() as i32);
let mut groups = vec![0; simplification_queue.len()];
Graph::new(1, (simplification_queue.len() / 4) as i32, &xadj, &adjncy)
    .unwrap()
    .set_adjwgt(&adjwgt)
    .part_kway(&mut groups)
    .unwrap();

#

Next up: simplifying each group

frank sail Mar 3, 2024, 5:57 AM

#

you can use vowels in identifiers btw

primal shadow Mar 3, 2024, 5:58 AM

#

frank sail you can use vowels in identifiers btw

Eh I decided just to copy metis for this part, given that adjcny is a meaningless name anyways

frank sail Mar 3, 2024, 5:58 AM

#

uh

primal shadow Mar 3, 2024, 5:58 AM

#

This is what metis calls their parameters

#

I know, it's cursed

frank sail Mar 3, 2024, 5:59 AM

#

I mean, "adjcny" isn't a random jumble of letters

primal shadow Mar 3, 2024, 5:59 AM

#

Yes, but adjaceny isin't any better 😛

#

I'd rather just copy metis's name so it's easy to tell what's what when referencing the manual

frank sail Mar 3, 2024, 6:00 AM

#

ok

wispy spear Mar 3, 2024, 12:14 PM

#

: (

#

yes, please more unreadable spaghetti code

primal shadow Mar 3, 2024, 8:42 PM

#

Why does meshopt_SimplifyLockBorder not work for meshlet LODs?

wicked notch Mar 3, 2024, 9:10 PM

#

it does, I use that

primal shadow Mar 3, 2024, 9:15 PM

#

How does this fit into that? https://github.com/zeux/meshoptimizer/pull/601

#

Also I'm a bit confused on some things. You have:

Original mesh vertices
Meshlet lod 0 "vertices" (points into above)
Meshlet lod 0 "indices" (points into above)

When grouping meshlets, you use edges defined by the meshlet indices
When simplifying meshlets, you do the same?
When you split your groups into new meshlets, what do you do with that info?

#

Also hmm, I somehow need to pass the original mesh vertices to the simplkifier

wicked notch Mar 3, 2024, 9:25 PM

#

primal shadow How does this fit into that? https://github.com/zeux/meshoptimizer/pull/601

yeah honestly I don't understand what they're talking about

#

I fail to understand what they mean by "outer border" of a mesh

primal shadow Mar 3, 2024, 9:27 PM

#

Ok, so ignoring that

#

meshopt_simplify needs the original mesh vertex positions

#

But my index buffer for meshlets points into the meshlet vertices..

#

Am I supposed to remap the index buffer to the original vertices myself before simplifying? Or am I missing something

wicked notch Mar 3, 2024, 9:30 PM

#

I just build another temporary index buffer tbh

primal shadow Mar 3, 2024, 9:34 PM

#

When you split your group, do you just copy the group's error to each new meshlet?

wicked notch Mar 3, 2024, 9:35 PM

#

yes

primal shadow Mar 3, 2024, 9:35 PM

#

Cool, almost done!

#

Ok, testing time!

#

Also curious, did you call any of meshopt's vertex/index buffer optimization functions, before simplifying or building meshlets?

#

The docs imply they help a lot

#

I lost my example code to simplify gltf files 😬

#

Gonna need to write that again 😦

wicked notch Mar 3, 2024, 10:03 PM

#

primal shadow Also curious, did you call any of meshopt's vertex/index buffer optimization fun...

yeah

#

they do help a lot

#

having duplicate vertices/edges is horrible

primal shadow Mar 3, 2024, 10:04 PM

#

wicked notch yeah

Good to know. What/where specifically do you do?

wicked notch Mar 3, 2024, 10:05 PM

#

just call all the optimization functions but the overdraw

hollow pagoda Mar 3, 2024, 10:05 PM

#

does that one even help?

primal shadow Mar 4, 2024, 3:20 AM

#

It crashes 😦

#

somehow I'm indexing out of bounds, somewhere

primal shadow Mar 4, 2024, 3:51 AM

#

LOD 1 bunny. Seems a bit off...

primal shadow Mar 4, 2024, 4:09 AM

#

#

LOD is not supposed to just remove meshlets and keep some others 😅

wicked notch Mar 4, 2024, 8:49 AM

#

welcome to "the problem" I was talking to you about meshopt's generator

primal shadow Mar 4, 2024, 4:37 PM

#

wicked notch welcome to "the problem" I was talking to you about meshopt's generator

If you use metis, what are the edge weights between triangles/meshlets?

wicked notch Mar 4, 2024, 4:38 PM

#

between triangles? spatial distance

#

inverse of the spatial distance actually

primal shadow Mar 4, 2024, 4:40 PM

#

How do you get it to 64 per partition?

#

Oh it's just target partitions = triangle count / 64 ig

wicked notch Mar 4, 2024, 4:46 PM

#

you bisection the mesh (i.e the graph) until it's within your vertex/triangle limits

primal shadow Mar 4, 2024, 4:55 PM

#

You've lost me lol

wicked notch Mar 4, 2024, 4:57 PM

#

you know how graph bisection works right?

#

it's just dividing the graph in two partitions

#

and assigning a good weight to said partitions

#

then, until the number of nodes or edges in the bisectioned graphs are within your limits, you keep recursively bisectioning

primal shadow Mar 4, 2024, 5:03 PM

#

Hmm ok

pale horizon Mar 5, 2024, 9:35 AM

#

I compiled GCC 13.2 and had to fight SPIRV-tools and KTX to get them to compile, but then...

~/work/repos_to_check/Retina/build$ ./RetinaMeshShadingSample 
./RetinaMeshShadingSample: /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by ./RetinaMeshShadingSample)
./RetinaMeshShadingSample: /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.31' not found (required by ./RetinaMeshShadingSample)

bleakekw

#

That's it, I'm updating to Debian Testing and I hope nothing breaks bleakekw

pale horizon Mar 5, 2024, 9:50 AM

#

@wicked notch it can't find Retina/test/Models/compressed/bistro/bistro.glb bleakekw

frank sail Mar 5, 2024, 9:51 AM

#

download it

#

then gltfpack it

#

intermediate blender step may be necessary

pale horizon Mar 5, 2024, 9:51 AM

#

Why doesn't CMake build not do it, 2/10 literally unusable

frank sail Mar 5, 2024, 9:52 AM

#

lvstri doesn't want to upload multi-gb models to github listenyoupieceofshit

pale horizon Mar 5, 2024, 9:52 AM

#

The build script can download it from somewhere gpAkkoShrug

frank sail Mar 5, 2024, 9:52 AM

#

hardcoding my beloved

#

imagine figuring out blender's CLI and shit when you could just convert it yourself

pale horizon Mar 5, 2024, 9:53 AM

#

Oh wait, Bistro is not even in glTF, I guess I have to export it manually via Blender, yeah

#

Most user friendly GP build process be like kekkedsadge

frank sail Mar 5, 2024, 9:57 AM

#

at least I'm kind enough to keep a shrimple test model in my repo

pale horizon Mar 5, 2024, 9:59 AM

#

~/work/repos_to_check/Retina/build$ ./_deps/meshoptimizer-build/gltfpack -i ~/Desktop/Bistro_v5_2/Bistro.gltf -o ~/work/repos_to_check/Retina/test/Models/compressed/bistro/bistro.glb -tc -tq 10 -vpf -kn -km -ke -noq
Error: gltfpack was built without BasisU support, texture compression is not available

bleakekw

frank sail Mar 5, 2024, 9:59 AM

#

oh yeah, you need lvstri's gltfpack fork kekwfroggified

#

I wonder if his renderer has a fallback for non-compressed textures

pale horizon Mar 5, 2024, 10:00 AM

#

Where can I find it though

frank sail Mar 5, 2024, 10:00 AM

#

idk

#

I never tried running it lol

pale horizon Mar 5, 2024, 10:00 AM

#

bleakekw

frank sail Mar 5, 2024, 10:00 AM

#

ratio

#

you can still see his repos btw

#

somewhere

#

no sign of meshopt/gltfpack lol
https://github.com/LVSTRI?tab=repositories

pale horizon Mar 5, 2024, 10:03 AM

#

I did it by reading TFM

git clone -b gltfpack https://github.com/zeux/basis_universal
cmake . -DMESHOPT_BUILD_GLTFPACK=ON -DMESHOPT_BASISU_PATH=basis_universal -DCMAKE_BUILD_TYPE=Release

#

damn son

#

#

This, but only core 23 is looking and doing nothing KEKW

delicate rain Mar 5, 2024, 10:11 AM

#

I think you can also try with the gltfpack from the releases page

#

That one definitely supports basisU

pale horizon Mar 5, 2024, 10:12 AM

#

I compiled it anyways now gpAkkoShrug

#

gltfpack has been doing its work for 9 minutes now bleakekw

#

Hope it's not stuck in some endless loop KEKW

frank sail Mar 5, 2024, 10:14 AM

#

it doesn't normally take that long

#

on my 5950X though

pale horizon Mar 5, 2024, 10:14 AM

#

I have 7950X even bleakekw

frank sail Mar 5, 2024, 10:14 AM

#

lol

#

buy a threadripper NOW

pale horizon Mar 5, 2024, 10:16 AM

#

it's stuck here

#

IT DID IT

frank sail Mar 5, 2024, 10:16 AM

#

high quality bc7 encoding be slow I guess

pale horizon Mar 5, 2024, 10:17 AM

#

And that's it...???

#

KEKW

#

I've been jebaited

#

RayTracing:

[2024-03-05 11:20:13.139] [Device] [critical] Required Extension "VK_KHR_ray_tracing_position_fetch" not supported

AMD sisters...

frank sail Mar 5, 2024, 10:21 AM

#

what driver

#

lol

#

RDNA 2+ supports it though

#

on windows

pale horizon Mar 5, 2024, 10:22 AM

#

    driverID                                             = DRIVER_ID_MESA_RADV
    driverName                                           = radv
    driverInfo                                           = Mesa 23.3.5-1

frank sail Mar 5, 2024, 10:23 AM

#

radv skill issue?

pale horizon Mar 5, 2024, 10:24 AM

#

I guess. pixelduck, plz fix

hallow cedar Mar 5, 2024, 10:24 AM

#

use newer mesa

#

24.0 should have it

frank sail Mar 5, 2024, 10:24 AM

#

there are some radv reports here ye

#

old software-having mf

pale horizon Mar 5, 2024, 10:25 AM

#

It's on Debian unstable, but not on testing bleakekw
Maybe some day it'll reach testing

hallow cedar Mar 5, 2024, 10:25 AM

#

wat distro

pale horizon Mar 5, 2024, 10:26 AM

#

Debian Tesiting aka trixie

frank sail Mar 5, 2024, 10:26 AM

#

why is everything related to that distro out of date

hallow cedar Mar 5, 2024, 10:26 AM

#

oh

pale horizon Mar 5, 2024, 10:26 AM

#

Because stability(tm)

hallow cedar Mar 5, 2024, 10:26 AM

#

"we keep things broken too" stability

frank sail Mar 5, 2024, 10:26 AM

#

use arch (I don't know anything about arch)

pale horizon Mar 5, 2024, 10:27 AM

#

soooo outdated

hallow cedar Mar 5, 2024, 10:27 AM

#

i think the "stable" debian version of mesa has not been updated in literally a full year

pale horizon Mar 5, 2024, 10:27 AM

#

That's what "stable" means, yes KEKW

#

They do stable releases every 2 years

frank sail Mar 5, 2024, 10:28 AM

#

man that's a full gpu generation

pale horizon Mar 5, 2024, 10:28 AM

#

And they only receive security patches

frank sail Mar 5, 2024, 10:28 AM

#

that's the wrong distro for cutting-edge giraffix

hallow cedar Mar 5, 2024, 10:28 AM

#

m'lady distro and arch loonix should have both by now

pale horizon Mar 5, 2024, 10:29 AM

#

Debian Unstable has it, so it will land into testing eventually

#

I'm not brave enough to use unstable, but I guess it will be similar to Arch KEKW

#

m'lady distro
Gentoo?

hallow cedar Mar 5, 2024, 10:29 AM

#

fedora

frank sail Mar 5, 2024, 10:30 AM

#

fedora

pale horizon Mar 5, 2024, 10:30 AM

#

Ah, Gentoo also fits, as it's the target audience KEKW

hallow cedar Mar 5, 2024, 10:30 AM

#

gentoo should probably indeed have it too

pale horizon Mar 5, 2024, 10:32 AM

#

If Debian unstable has it, everyone should have it
But I'll wait for now for it to hit testing anyways gpAkkoShrug

pale horizon Mar 5, 2024, 10:32 AM

#

frank sail that's the wrong distro for cutting-edge giraffix

3 weeks ago my engine was on 3.3 GL KEKW

#

GP will force me to use Arch and buy programming socks mindblown

#

@wicked notch anyways, I had to do this for it to compile. Classic template bruh momento:

📎 patch.diff

#

Also had to update KTX to the latest since they used some funny enum conversions which are deprecared + Werror
Also SPIRV-tools didn't compile for me, but I'll research it and will probably make an issue

wicked notch Mar 5, 2024, 10:42 AM

#

yea nearly all of khronos' tools require some sort of hacks to compile

#

it's actually incredible how annoying it is

#

I should also make public my fork of gltfpack lol

pale horizon Mar 5, 2024, 10:45 AM

#

I actually didn't need to use a fork, just had to pass it path to basisu

pale horizon Mar 5, 2024, 10:45 AM

#

pale horizon And that's it...???

Is it all the MeshShading sample is? KEKW

wicked notch Mar 5, 2024, 10:52 AM

#

yes

#

it's due to the exam speedrunning

#

I am currently resting

pale horizon Mar 5, 2024, 11:02 AM

#

Enjoy your rest 👍
I know how it feels

But also
REMOVE main PLZ AND USE FIXED COMMITS, OK THNX

FetchContent_Declare(
    spirv_cross
    GIT_REPOSITORY https://github.com/KhronosGroup/SPIRV-Cross
    GIT_TAG        main
    GIT_PROGRESS   TRUE
    GIT_SHALLOW    TRUE
)
set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS ON CACHE BOOL "" FORCE)
set(SPIRV_CROSS_ENABLE_TESTS OFF CACHE BOOL "" FORCE)
FetchContent_MakeAvailable(spirv_cross)

FetchContent_Declare(
    spirv-headers
    GIT_REPOSITORY https://github.com/KhronosGroup/SPIRV-Headers
    GIT_TAG        main
    GIT_PROGRESS   TRUE
    GIT_SHALLOW    TRUE
)

#

Because apparently they can have borked commits in main. Cringe

wicked notch Mar 5, 2024, 11:03 AM

#

incredible

#

will do

pale horizon Mar 5, 2024, 11:04 AM

#

https://github.com/KhronosGroup/SPIRV-Tools/pull/5602
I even found the PR which fixes the issue. But seems like their CI is kinda bad if it doesn't get noticed lmao

#

I guess only 2-3 people in the world compile this stuff from main KEKW

wicked notch Mar 5, 2024, 11:05 AM

#

I'm thinking to move to glslang anyways like jaker did

#

shaderc (and all its dependencies so SPIRV-Tools, SPIRV-Headers, etc.) have consistently given me pain

#

or, better yet slang, but I hear potrick is still fighting with the ptr guy KEKW

frank sail Mar 5, 2024, 11:07 AM

#

yeah I'll let them cook for a few years

pale horizon Mar 5, 2024, 11:07 AM

#

Cook FreeSLANG AMD(tm) Edition for us plz

#

glslang is glslValidator, right?

wicked notch Mar 5, 2024, 11:09 AM

#

no this thing: https://github.com/KhronosGroup/glslang

#

it should allow me to ditch many dependencies

pale horizon Mar 5, 2024, 11:09 AM

#

So you're telling me there's the third glsl compiler? KEKW

wicked notch Mar 5, 2024, 11:09 AM

#

I think this is just glslc and glslValidator combined

pale horizon Mar 5, 2024, 11:10 AM

#

https://tenor.com/view/why-huh-but-why-gif-13199396

Tenor

wispy spear Mar 5, 2024, 11:25 AM

#

would be cool if khronosgroup just renames itself to JohnKhronos, it do be more authentic

pale horizon Mar 5, 2024, 11:28 AM

#

Maybe after someone from the GP becomes Khronos chair KEKW

wispy spear Mar 5, 2024, 11:28 AM

#

: )

wicked notch Mar 5, 2024, 11:28 AM

#

JohnKhronos is not a person, it's a title

#

the ultimate title

wispy spear Mar 5, 2024, 11:28 AM

#

John Khronos
John Khronos

#

works for me 😄

#

title adjusted for inflation: Senior Principal John Khronos

pale horizon Mar 5, 2024, 12:30 PM

#

@wicked notch did you compile Retina on Linux before, btw?

wicked notch Mar 5, 2024, 12:31 PM

#

many moons ago yes

#

but I use arch btw

pale horizon Mar 5, 2024, 12:33 PM

#

I’m impressed that it only took me two lines to fix it
Compiling a random Vulkan anime engine now and it’s so full of MSVCisms, it hurts

wicked notch Mar 5, 2024, 12:33 PM

#

btw

#

class NvidiaDlssFeature;
class NvidiaDLSSFeature;
class NVIDIADLSSFeature;
class nvidia_dlss_feature_t;```

#

choose

#

I'll let this go democratically

pale horizon Mar 5, 2024, 12:50 PM

#

Nvidia don't know how to write their name too, I think KEKW

frank sail Mar 5, 2024, 12:53 PM

#

just write it in spongebob-mocking style

#

nViDiA

finite quartz Mar 5, 2024, 1:16 PM

#

NvidiaDeepLearningSuperSamplingFeature; froge_evil

#

or if too long: NvidiaDeLeSuSaFeature; 🎵

wispy spear Mar 5, 2024, 5:24 PM

#

wicked notch ```cpp class NvidiaDlssFeature; class NvidiaDLSSFeature; class NVIDIADLSSFeature...

first tbh... hoever if you stick to your isms_t like in iris, then id go with the last one

#

i also dont know if serious or not bleakekw

wicked notch Mar 5, 2024, 5:25 PM

#

I kinda am

#

I am facing some issues with the C#isms

#

like get this

#

class Graph {
public:
  struct Partition {
    // ...
  };
  Partition Partition();
};```

#

but ultimately it doesn't matter too much

wispy spear Mar 5, 2024, 5:29 PM

#

i decided to prefix mzy Fukisms with Fk

#

and can still go with Pascal style where i want then

#

ala

struct FkPartition { };
FkPartition Partition;

#

mayhaps Retina stuff can be ReThism or ReThatsm

wicked notch Mar 5, 2024, 5:31 PM

#

makes sense

wicked notch Mar 8, 2024, 10:03 PM

#

why the hell does glibc's format not implement 202311L features

#

fucks sake

primal shadow Mar 9, 2024, 6:59 AM

#

Getting LOD building working is so difficult ugh

primal shadow Mar 9, 2024, 7:16 AM

#

@wicked notch If you feel like it, please take a look at my code and tell me if I'm doing something dumb. I know you had better luck using METIS for clustering instead of meshoptimizer, and I'm not weighting clusters by distance between them, but I still feel like I should be getting better results than I am... https://github.com/JMS55/bevy/blob/meshlet-lods/crates/bevy_pbr/src/meshlet/from_mesh.rs#L50

#

Oh hold on, does meshopt simplify take target index count??

#

oh, hell

#

Ok that's a little bit better, but still cursed, and still z-fighting

#

Ok back to sleep since that didn't work

#

Tm I should try visualizing the cluster groupings before I try and simplify

primal shadow Mar 9, 2024, 8:04 AM

#

Hmm I think tuple_windows is not correct to generate edges 😅

#

ok yeah I see a lot wrong fml lol

primal shadow Mar 9, 2024, 8:26 AM

#

Nope still didn't fix things ahhh

wispy spear Mar 9, 2024, 9:50 AM

#

theres just one flaw with your code

#

its written in rust skull_laugh

wicked notch Mar 9, 2024, 11:32 AM

#

primal shadow Ok that's a little bit better, but still cursed, and still z-fighting

are you drawing both LOD 0 and LOD 1 thonk

primal shadow Mar 9, 2024, 5:58 PM

#

wicked notch are you drawing both LOD 0 and LOD 1 <:thonk:728843920393371708>

Shouldn't be...

#

I modified the code to throw away lod 0 after calculating lod 1

wicked notch Mar 9, 2024, 5:59 PM

#

does renderdoc have any insight for us

#

do the meshlets appear scuffed or something

primal shadow Mar 9, 2024, 6:09 PM

#

Hmm let me check in a bit, I just woke up lol

wicked notch Mar 9, 2024, 6:10 PM

#

US moment

primal shadow Mar 9, 2024, 8:06 PM

#

oh, wait, hold on

#

I think I was rendering both LODs

#

nope still looks weird though :/

wicked notch Mar 9, 2024, 8:10 PM

#

is this all of LOD 1?

#

it's weird for only a section of the bunny to be scuffed innit

primal shadow Mar 9, 2024, 8:10 PM

#

yeah

#

right??

#

And also even lod 0 looks weird compared to what it used to be

#

It used to be nicely distributed, but now there's some weird patches of small meshlets

#

But the code didn't even change

wicked notch Mar 9, 2024, 8:11 PM

#

I guess you should first focus on fixing that then

#

scuffed leaf meshlets will inevitably lead to scuffed LOD'ed meshlets

primal shadow Mar 9, 2024, 8:12 PM

#

I didn't even change anything though 😭

wicked notch Mar 9, 2024, 8:12 PM

#

debugging is a core part of the process

#

at least you're not hanging your system, like I have many times KEKW

primal shadow Mar 9, 2024, 8:16 PM

#

I'm litterly using the same test model and my previous branch and it's not working as well anymore -_-

#

This is bizzare

wispy spear Mar 9, 2024, 8:18 PM

#

@wicked notch is there any chance that i can make you fix the archived Iris? master is not buildable as is : (

primal shadow Mar 9, 2024, 8:18 PM

#

I'm checking out a much older commit, maybe I broke something at some point

wicked notch Mar 9, 2024, 8:20 PM

#

primal shadow I'm checking out a much older commit, maybe I broke something at some point

ye

#

git checkout saves lives

wicked notch Mar 9, 2024, 8:20 PM

#

wispy spear <@320895822394818561> is there any chance that i can make you fix the archived I...

perchance

wispy spear Mar 9, 2024, 8:20 PM

#

if you are not in the middle of learning for exams : )

primal shadow Mar 9, 2024, 8:23 PM

#

no, older commit is still broken...

wicked notch Mar 9, 2024, 8:23 PM

#

I shall hop on the penguin

wicked notch Mar 9, 2024, 8:26 PM

#

wispy spear <@320895822394818561> is there any chance that i can make you fix the archived I...

the GL one or the VK one

wispy spear Mar 9, 2024, 8:31 PM

#

GL

primal shadow Mar 9, 2024, 8:32 PM

#

Even older commit is not wokring. Maybe this is a bad mesh?

#

time for blender

wicked notch Mar 9, 2024, 8:33 PM

#

wispy spear GL

ouch bleakekw

#

alright I'll dive into the deepest pit of hell

wispy spear Mar 9, 2024, 8:33 PM

#

9110frogcool

#

i think some fixes is using model.h vs mesh.h (which doesnt exist anymore) bt wants mesh_t plus a potential fix for its ctor if i saw that correctly to work in the older examples < 5.2

#

and the warnings from dependencies (looks like ktx is the offender here) need to be silenced

primal shadow Mar 9, 2024, 8:35 PM

#

I mean looks fine

wispy spear Mar 9, 2024, 8:35 PM

#

jasmine, mayhaps try suzanne ~~or the deccer cubes~~?

wicked notch Mar 9, 2024, 8:37 PM

#

we need deccer spheres tbh

#

to meshletize better

distant lodge Mar 9, 2024, 8:37 PM

#

where is the PBR thread

#

those count

primal shadow Mar 9, 2024, 8:38 PM

#

wispy spear jasmine, mayhaps try suzanne ~~or the deccer cubes~~?

Yeah let me try suzanne

#

The thing is the bunny used to work fine

#

The meshlet_mesh I have saved is great

#

But trying to convert it again, even using a very old commit, looks horrible???

#

I feel like I'm missing something 😦

wispy spear Mar 9, 2024, 8:42 PM

#

hmm

#

i thought there was a thread called This other mesh will also crash your engine or something similar

frank sail Mar 9, 2024, 8:44 PM

#

deccer balls
deccer balls

distant lodge Mar 9, 2024, 8:44 PM

#

yeah, I remember it being under a random channel

primal shadow Mar 9, 2024, 8:45 PM

#

I think it looks ok?

distant lodge Mar 9, 2024, 8:45 PM

#

is that the main LOD

wicked notch Mar 9, 2024, 8:45 PM

#

somewhat

distant lodge Mar 9, 2024, 8:45 PM

#

I remember the eyes having more detail

wicked notch Mar 9, 2024, 8:45 PM

#

some of the meshlets look very small

wispy spear Mar 9, 2024, 8:45 PM

#

hmm it looks also fucked somehow, the green center and white bttom one

primal shadow Mar 9, 2024, 8:46 PM

#

distant lodge is that the main LOD

Yes

#

Yeah idk what meshopt is doing all of a sudden

wicked notch Mar 9, 2024, 8:46 PM

#

yeah this definitely does not look like the main LOD

distant lodge Mar 9, 2024, 8:46 PM

#

I mean, I think I recognize some of the main geometry

#

maybe the debug view needs triangle outlines

primal shadow Mar 9, 2024, 8:47 PM

#

I can switch it to per-triangles, 1s

wispy spear Mar 9, 2024, 8:47 PM

#

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

#

i doubt the spheres will produce neat meshlets 😄

primal shadow Mar 9, 2024, 8:49 PM

#

triangles look fine

#

wicked notch Mar 9, 2024, 8:49 PM

#

is this just a very low poly suzanne?

wispy spear Mar 9, 2024, 8:49 PM

#

yeah crank up subdivision

distant lodge Mar 9, 2024, 8:50 PM

#

that's the default one, so I guess you've been using subdivision

primal shadow Mar 9, 2024, 8:50 PM

#

mhm let me figure out how to do that

distant lodge Mar 9, 2024, 8:50 PM

#

subdivision surface modifier

wispy spear Mar 9, 2024, 8:50 PM

#

add modifier on the blue wrench symbol in the bottom right window

primal shadow Mar 9, 2024, 8:51 PM

#

#

triangles

#

Btw, do I even need normals/tangents?? I can just bake detail directly, and use geometric normals derived from the triangle position with this amount of density no?

#

Well lod 0 looks great now

#

Lets try a different LOD and just give up on the bunny for now, idk why

#

Lod 1 looks the same??

#

heck

#

LOD 0: 1968 meshlets
LOD 5: 1968 meshlets
???
each LOD is just a copy of LOD 0
huh

wicked notch Mar 9, 2024, 9:05 PM

#

try building now @wispy spear

wispy spear Mar 9, 2024, 9:09 PM

#

wicked notch try building now <@194502965749350400>

[build] 12 warnings and 12 errors generated.

wicked notch Mar 9, 2024, 9:09 PM

#

epic

#

which errors

wispy spear Mar 9, 2024, 9:10 PM

#

wicked notch Mar 9, 2024, 9:10 PM

#

ah you're trying to build the old schtuff

#

yeah idk how to fix the old ones bleakekw

#

I don't remember anything about them

#

it was also an unwise decision to name those "5.0" and"4.2" kekkedsadge

wispy spear Mar 9, 2024, 9:11 PM

#

let me comment out all the old projects

wicked notch Mar 9, 2024, 9:11 PM

#

try building the "AntiAliasing" target

#

that should work

wispy spear Mar 9, 2024, 9:12 PM

#

#

i wanted the 5.2 thing

#

ah ye

#

AntiAliasing 😄

wicked notch Mar 9, 2024, 9:13 PM

#

I didn't finish implementing FSR btw, it's very aliased bleakekw

wispy spear Mar 9, 2024, 9:13 PM

#

das ist ok

#

im mainly using this to learn me some c++ : )

primal shadow Mar 9, 2024, 9:14 PM

#

I'm suspicious simplification isin't actually doing anything 🤔

wispy spear Mar 9, 2024, 9:14 PM

#

because your code is somewhat readable

wicked notch Mar 9, 2024, 9:14 PM

#

primal shadow I'm suspicious simplification isin't actually doing anything 🤔

this is a good thing to be suspicious of

#

because simplifying can very easily fail

#

if you have lots of duplicated vertices or your error value is too small or your mesh is super disconnected

#

btw look at this guy

#

5.6ms to render bistro with some shadows

#

so slow

primal shadow Mar 9, 2024, 9:15 PM

#

Each group has 4 meshlets, and after simplication and re-meshletifying, there's 4 meshlets... it's litterly doing nothing 😭

wicked notch Mar 9, 2024, 9:16 PM

#

wispy spear im mainly using this to learn me some c++ : )

yeah there's nothing about GL to learn from me

#

bleakekw

wispy spear Mar 9, 2024, 9:16 PM

#

thats not correct

frank sail Mar 9, 2024, 9:16 PM

#

glVertex5f

wicked notch Mar 9, 2024, 9:17 PM

#

wicked notch btw look at this guy

Anti TAA boys look at this and go "it's not aliasing, it's clarity froge_love "

wispy spear Mar 9, 2024, 9:17 PM

#

is that AntiAliasing.exe?

wicked notch Mar 9, 2024, 9:18 PM

#

ye

wispy spear Mar 9, 2024, 9:18 PM

#

ah

primal shadow Mar 9, 2024, 9:24 PM

#

Simplification is litterly doing nothing

#

[crates\bevy_pbr\src\meshlet\from_mesh.rs:140:17] (simplified_group_indices.len(), group_indices.len(), error) = (
    384,
    384,
    0.0,
)

#

same triangle count before and after

wicked notch Mar 9, 2024, 9:24 PM

#

step into meshopt and see why it's failing

#

my guess is bad triangle index data

primal shadow Mar 9, 2024, 9:24 PM

#

yeah me too, lets find out

wispy spear Mar 9, 2024, 9:25 PM

#

did you push the fix already lustri?

wicked notch Mar 9, 2024, 9:25 PM

#

wispy spear did you push the fix already lustri?

one sec

#

I hate ktx

wispy spear Mar 9, 2024, 9:29 PM

#

: )

wicked notch Mar 9, 2024, 9:29 PM

#

why is khronos like this

#

ffs

wispy spear Mar 9, 2024, 9:29 PM

#

iirc if you target_link_libraries(iris SYSTEM ktx) it should be quiet somehow

#

but i kind of forgot how it really worked

wicked notch Mar 9, 2024, 9:29 PM

#

how can they make the greatest graphics API in the entire universe and have the worst possible tooling imaginable

#

ok I push

wispy spear Mar 9, 2024, 9:30 PM

#

because its also just a bunch of weird people and a bnch of smart people like everywhere else : >

#

your .gitignore looks interesting hehe

primal shadow Mar 9, 2024, 9:32 PM

#

@wicked notch can you sanity check me, what are the steps to simplify meshlets? Which index buffers do you use?

wicked notch Mar 9, 2024, 9:33 PM

#

the ones you generate anew from the grouped meshlets

#

for group in groupedMeshlets {
  vertices = [];
  for triangle in group.triangles {
    index = vertexIndices[group.indexOffset + triangle];
    vertex = vertices[group.vertexOffset + index];
    vertices.push(vertex);
  }
  indices = GenerateIndexBufferFrom(vertices);
  (simplifiedIndices, error) = SimplifyHalf(indices);
}```

primal shadow Mar 9, 2024, 9:35 PM

#

What does GenerateIndexBufferFrom do?

wicked notch Mar 9, 2024, 9:36 PM

#

calls meshopt_remapVertexBuffer and stuff

primal shadow Mar 9, 2024, 9:36 PM

#

and vertex is an index into the mesh's original vertex data yes?

wicked notch Mar 9, 2024, 9:36 PM

#

no vertex is the actual vertex

#

like position data

#

meshopt_remapVertexBuffer needs the original vertex data

primal shadow Mar 9, 2024, 9:36 PM

#

hmm

#

I'm doing basically the same thing I think, just not calling any of the meshopt remapping stuff

wicked notch Mar 9, 2024, 9:37 PM

#

you need to do that though thonk

primal shadow Mar 9, 2024, 9:37 PM

#

let mut group_indices = Vec::new();
for meshlet_id in group_meshlets {
    let meshlet = meshlets.get(*meshlet_id);
    for meshlet_index in meshlet.triangles {
        group_indices.push(meshlet.vertices[*meshlet_index as usize]);
    }
}

wicked notch Mar 9, 2024, 9:37 PM

#

in this particular case

primal shadow Mar 9, 2024, 9:37 PM

#

wicked notch you *need* to do that though <:thonk:728843920393371708>

Wdym?

wicked notch Mar 9, 2024, 9:37 PM

#

yes, this is wrong

#

these are not the indices you want

#

you want indices local to the group

primal shadow Mar 9, 2024, 9:38 PM

#

Why not? What indices should I be using, if not the concanated meshlet indices into the original vertex buffer?

wicked notch Mar 9, 2024, 9:38 PM

#

you want to treat this group of meshlets as if it were your only mesh

primal shadow Mar 9, 2024, 9:38 PM

#

I'm confused why that would make a difference tbh?

wicked notch Mar 9, 2024, 9:39 PM

#

because meshlet.vertices is not really an index buffer

#

it does not represent the topology of the mesh

primal shadow Mar 9, 2024, 9:39 PM

#

https://jglrxavpok.github.io/2024/01/19/recreating-nanite-lod-generation.html#simplifying-cluster-groups is doing the same thing I am I think, and it seems to work for them

primal shadow Mar 9, 2024, 9:39 PM

#

wicked notch it does not represent the topology of the mesh

Wdym by this exactly?

#

Once you build a list using meshlet.triangles into meshlet.vertices, do you not have a list of triangles?

wicked notch Mar 9, 2024, 9:40 PM

#

meshlet.primitives is the topology meshlet.vertices is just a counter for how many times a certain vertex has been used

wicked notch Mar 9, 2024, 9:41 PM

#

primal shadow Once you build a list using meshlet.triangles into meshlet.vertices, do you not ...

I believe the list of triangles is in meshlet.triangles

#

meshlet.vertices is a counter for vertex reuse

#

if you look at how meshlet.vertices is generated, you will understand

#

nevermind it's the opposite KEKW

#

yeah you're right

#

it should work

#

idk why it don't

primal shadow Mar 9, 2024, 9:42 PM

#

But it dosen't 😭

wicked notch Mar 9, 2024, 9:43 PM

#

did you figure out where meshopt is failing yet

primal shadow Mar 9, 2024, 9:43 PM

#

No, going to go try and figure out how to attach gdb tbh

wicked notch Mar 9, 2024, 9:43 PM

#

inside meshopt_simplify there's a prepass that classifies all the edges

#

look at what the classifier is doing

primal shadow Mar 9, 2024, 9:43 PM

#

I think meshopt-rs is mapping a null ptr to a no-opt instead of telling me it's erroring 😛

wicked notch Mar 9, 2024, 9:44 PM

#

does rust not have its own debugger? thonk

#

https://github.com/zeux/meshoptimizer/blob/master/src/simplifier.cpp#L255-L374 this is the prepass btw

#

classifyVertices

primal shadow Mar 9, 2024, 9:44 PM

#

wicked notch does rust not have its own debugger? <:thonk:728843920393371708>

No? Just gdb/lddb, same as C++/C/etc

wicked notch Mar 9, 2024, 9:45 PM

#

I guess it all works the same independent of lang if you export the symbols ye

#

I figured the rust people would just "rewrite it in rust" 😛

#

@wispy spear it should work now

#

let me know

#

also damn it's almost been a full year since I started this thread

primal shadow Mar 9, 2024, 9:52 PM

#

bash: gdb: command not found

#

ugh

wicked notch Mar 9, 2024, 9:52 PM

#

pacman -S gdb

#

smart

primal shadow Mar 9, 2024, 9:55 PM

#

Running `cargo build --example=meshlet --package=bevy --release --features meshlet_processor --message-format=json`...
error: unexpected argument '--features meshlet_processor' found

  tip: a similar argument exists: '--features'

Usage: cargo.exe build --example [<NAME>] --package [<SPEC>] --release --features <FEATURES>

???

#

ok fixed

wispy spear Mar 9, 2024, 10:01 PM

#

still no worki, lvstri, ill dinglefart from here, no need to be hanging on the penguin for me (for now)

wicked notch Mar 9, 2024, 10:01 PM

#

it's ok I like the penguin :)

wispy spear Mar 9, 2024, 10:01 PM

#

ktx's fmt dependency now complains about warnings treated as error

wicked notch Mar 9, 2024, 10:01 PM

#

btw are you using clang or gcc

wispy spear Mar 9, 2024, 10:02 PM

#

gcc

wicked notch Mar 9, 2024, 10:02 PM

#

13.2 right?

wispy spear Mar 9, 2024, 10:02 PM

#

yes sir

#

let me also yoink build once more

primal shadow Mar 9, 2024, 10:02 PM

#

ok finally got a debugger attached lol

wicked notch Mar 9, 2024, 10:03 PM

#

how am I not getting your same error with the same configuration

#

I'm also using latest gcc

primal shadow Mar 9, 2024, 10:04 PM

#

vscode lldb is frozen wtf

#

I finally got into meshopt and now it's stuck -_-

wispy spear Mar 9, 2024, 10:05 PM

#

heh still this one

#

wicked notch Mar 9, 2024, 10:05 PM

#

can you try cloning the thing again

#

from scratch

wispy spear Mar 9, 2024, 10:05 PM

#

when you updated submodules i dont have to do that, since you modified .gitmodules right

#

kk

wicked notch Mar 9, 2024, 10:05 PM

#

wispy spear when you updated submodules i dont have to do that, since you modified .gitmodul...

idk

#

man I'm glad I switched to fetchcontent lol

wispy spear Mar 9, 2024, 10:06 PM

#

heh

#

oof works now

wicked notch Mar 9, 2024, 10:07 PM

#

pog

wispy spear Mar 9, 2024, 10:07 PM

#

i did have to update the submodule myself

#

wtf

#

sorry for the interruption jasmine, luigi is all yours again

#

thx lustri ❤️

primal shadow Mar 9, 2024, 10:12 PM

#

#

ugh the debugger just freezes all the time

#

I can't step in

wicked notch Mar 9, 2024, 10:13 PM

#

yeah idk vscode

#

I think jetbrains has an IDE for rust now

primal shadow Mar 9, 2024, 10:13 PM

#

nor can I kill my app anymore, it's frozen 😛

wicked notch Mar 9, 2024, 10:14 PM

#

ye jetbrains has rustrover in EAP

#

which means it's free to use

#

maybe give it a try if it's not too much hassle idk

primal shadow Mar 9, 2024, 10:39 PM

#

What exactly should I be looking for? I'm unsure why it's failing tbh

wicked notch Mar 9, 2024, 10:40 PM

#

look inside vertex_kind

#

is everything Locked?

primal shadow Mar 9, 2024, 11:08 PM

#

It's not showing up in my debugger, ahh what

#

wicked notch Mar 9, 2024, 11:12 PM

#

I guess you can use the age old technique

#

std::printf to the rescue

primal shadow Mar 9, 2024, 11:13 PM

#

It'll be annoying to figure out how to bind to a custom build of meshopt, sigh

#

Yeah the debugger doesn't recognize VertexKind, it's just binary

wicked notch Mar 9, 2024, 11:17 PM

#

just printf tbh

#

I want to see LODs working

#

so get a move on 😛

#

(I'm kidding, but I do want to see your LODs working)

wispy spear Mar 9, 2024, 11:21 PM

#

me too : )

primal shadow Mar 10, 2024, 6:47 AM

#

https://github.com/gwihlidal/meshopt-rs/issues/37 Maybe it's this

primal shadow Mar 10, 2024, 7:51 AM

#

Maybe it dosen't like that I'm passing the entire mesh's vertex buffer to simplify, instead of a stripped down one for just the meshlet vertices? idk

#

https://github.com/zeux/meshoptimizer/discussions/569

GitHub

Meshlets Merging and Simplification. · zeux meshoptimizer · Discuss...

Since the meshopt_meshlet is purely indexing structs for outputing meshlet_triangles and vertices. I would like to know how could I merge two meshlets and then simplify their geometry. Could anyone...

wicked notch Mar 10, 2024, 12:02 PM

#

it's always great to read misinfo guy's ramblings KEKW

wicked notch Mar 10, 2024, 12:03 PM

#

primal shadow https://github.com/zeux/meshoptimizer/discussions/569

I read this too, I think I didn't run into the problem because I really create a new vertex buffer for the simplified meshlets

#

this ends up duplicating a lot of stuff so maybe having a deduplication pass later when everything has been shrimplified can be beneficial

#

the remapping will be hell though bleakekw

frank sail Mar 10, 2024, 12:04 PM

#

who is misinfo guy? nervous

wicked notch Mar 10, 2024, 12:04 PM

#

the FuckTAA guy, in #questions

frank sail Mar 10, 2024, 12:05 PM

#

ah, no better way to start the day than with some entertainment from the court jester

wicked notch Mar 10, 2024, 12:05 PM

#

real

wicked notch Mar 10, 2024, 12:55 PM

#

frank sail ah, no better way to start the day than with some entertainment from the court j...

btw what do you do for copy ctor/assign

#

is there a cursed way to avoid duplication here

frank sail Mar 10, 2024, 12:57 PM

#

I delete them

#

Or rule of 0

wicked notch Mar 10, 2024, 12:58 PM

#

what if you actually need a copy ctor bleakekw

frank sail Mar 10, 2024, 12:58 PM

#

You could implement it the same way as my move semantics, where the assignment operator just calls the constructor

wicked notch Mar 10, 2024, 1:00 PM

#

ye but you need swap innit

#

T::operator =(const T& other) {
  if (this == &other) {
    return;
  }
  T(other).swap(*this);
}```

frank sail Mar 10, 2024, 1:00 PM

#

Why is swap needed?

wicked notch Mar 10, 2024, 1:01 PM

#

I guess you could do it with placement new

frank sail Mar 10, 2024, 1:01 PM

#

Yep

#

That's what I meant

wicked notch Mar 10, 2024, 1:01 PM

#

yeah that's cursed enough

#

this placement new thingy is great, too bad it don't work when destructors are protected :(

#

and even if it did, the move constructor would be insane in derived classes

frank sail Mar 10, 2024, 1:03 PM

#

why is it protected

distant lodge Mar 10, 2024, 1:03 PM

#

thing::clone()

#

wow holy shit there wasn't even a hint my messages were behind what the hell is going on

wicked notch Mar 10, 2024, 1:04 PM

#

class Base {
public:
  Base(Base&& other)
    : _state(std::exchange(other._state, {})) {}
private:
  T _state;
};


class Derived : public Base {
public:
  Derived(Derived&& other)
    : Base(std::move(other)),
      _moreState(std::exchange(other._moreState, {})) {}
private:
  T _moreState;
};```

wicked notch Mar 10, 2024, 1:04 PM

#

frank sail why is it protected

ye that's stupid

#

I realized it in the end

distant lodge Mar 10, 2024, 1:05 PM

#

isn't swap only really the elegant thing for move ctors, my understanding was that it was just a convenient way to let the old garbage in the lhs object be swept up because the dtor of the rhs object is about to be called

wicked notch Mar 10, 2024, 1:05 PM

#

having the destructor protected that is

wicked notch Mar 10, 2024, 1:05 PM

#

distant lodge isn't swap only really the elegant thing for move ctors, my understanding was th...

you can achieve a similar effect with the this->~T(); new (this) ...

frank sail Mar 10, 2024, 1:05 PM

#

delete this bleakekw

distant lodge Mar 10, 2024, 1:05 PM

#

something about that seems inherently cursed

wicked notch Mar 10, 2024, 1:06 PM

#

yes, that's why I hide it

#

core::reconstruct(*this, std::move(other))

#

nobody needs to know

frank sail Mar 10, 2024, 1:06 PM

#

I put mine on full display

distant lodge Mar 10, 2024, 1:06 PM

#

I wonder if there's some obscure rule about object lifetimes that makes that ackshually UB

frank sail Mar 10, 2024, 1:06 PM

#

I looked in the spec to make sure it's not UB

wicked notch Mar 10, 2024, 1:06 PM

#

wicked notch ```cpp class Base { public: Base(Base&& other) : _state(std::exchange(othe...

anyways how would you deal with something like this

distant lodge Mar 10, 2024, 1:07 PM

#

are you sure? my dad works at spec and he said you will be banned for this code

wicked notch Mar 10, 2024, 1:07 PM

#

Derived's move ctor is insane and it looks like a use-after-move bleakekw

#

linter complains as well

frank sail Mar 10, 2024, 1:08 PM

#

wicked notch ```cpp class Base { public: Base(Base&& other) : _state(std::exchange(othe...

Why is the derived class referring to a private member of the base

wicked notch Mar 10, 2024, 1:08 PM

#

typo

frank sail Mar 10, 2024, 1:08 PM

#

Looks fine to me

wicked notch Mar 10, 2024, 1:09 PM

#

ye but linter complains because use-after-move

#

even though we all know it's actually fine

frank sail Mar 10, 2024, 1:09 PM

#

The linter complains about nonstandard move assignment too

#

Because it dumb

wicked notch Mar 10, 2024, 1:10 PM

#

rip

frank sail Mar 10, 2024, 1:10 PM

#

Just take the linter thing as advice rather than a serious issue

#

And if you know what you're doing you can safely ignore it

wicked notch Mar 10, 2024, 1:13 PM

#

does one ever know what one's doing

frank sail Mar 10, 2024, 1:14 PM

#

C++ lets us try regardless

wispy spear Mar 10, 2024, 1:37 PM

#

can you not add an override for the linter at that spot?

#

with a oneliner comment thing

wicked notch Mar 10, 2024, 1:44 PM

#

perchance

wicked notch Mar 10, 2024, 5:48 PM

#

behold

#

def main():
    includeDirectories = [
        os.path.join(os.environ['VULKAN_SDK'], 'include'),
    ]
    compileDefinitions = [
        'VK_NO_PROTOTYPES'
    ]
    cppStandard = 'c++17'
    cppArgs = []
    for includeDir in includeDirectories:
        cppArgs.append(f'-I{includeDir}')
    for definition in compileDefinitions:
        cppArgs.append(f'-D{definition}')

    cppArgs.append(f'-std={cppStandard}')
    cppArgs.append(f'-x')
    cppArgs.append(f'c++')
    vulkanEnumFile = os.path.join(includeDirectories[0], 'vulkan\\vulkan.h')
    index = Index.create()
    translationUnit = index.parse(
        vulkanEnumFile,
        args=cppArgs,
        options=TranslationUnit.PARSE_INCOMPLETE
    )
    for diagnostic in translationUnit.diagnostics:
        print(diagnostic)

    stack = [translationUnit.cursor]
    while len(stack) > 0:
        current = stack.pop()
        if current.kind == CursorKind.ENUM_DECL:
            print(current.kind, current.location)
        for child in current.get_children():
            stack.append(child)

    return

#

worst python code ever written by humans

#

AI had no hand in the creation of this abomination

#

this has been painfully handcrafted by yours truly

pale horizon Mar 10, 2024, 5:50 PM

#

It’s fine. What years of bikeshed to a man KEKW

wicked notch Mar 10, 2024, 5:51 PM

#

I now have a list of all vulkan enums

#

this is so good

pale horizon Mar 10, 2024, 5:58 PM

#

Now write a tool that does vk validation in compile time KEKW

wicked notch Mar 10, 2024, 8:05 PM

#

duplicateKeys = []
for (k, v) in enums.items():
    if HasVendorSuffix(k):
        removedVendor = RemoveVendorSuffix(k)
        if removedVendor in enums:
            duplicateKeys.append(k)
``` mmm efficient python code

wispy spear Mar 10, 2024, 8:16 PM

#

what are you building exactly? : )

wicked notch Mar 10, 2024, 8:20 PM

#

thing to convert vk enums to retina enums

#

I got tired of writing them by hand

wispy spear Mar 10, 2024, 8:25 PM

#

ah

wicked notch Mar 10, 2024, 8:26 PM

#

python is almost not cancerous to write with type annotations btw

#

who knew that types were useful

wispy spear Mar 10, 2024, 8:27 PM

#

the only real gripe i have with it is tab and the stupid indentation

wicked notch Mar 10, 2024, 8:27 PM

#

if only

wispy spear Mar 10, 2024, 8:27 PM

#

and 100 ways of doing 1 thing

pale horizon Mar 10, 2024, 8:43 PM

#

wicked notch ```py duplicateKeys = [] for (k, v) in enums.items(): if HasVendorSuffix(k):...

Cringe, use list comprehension NOW

wicked notch Mar 10, 2024, 8:47 PM

#

no :gigachad:

pale horizon Mar 10, 2024, 8:49 PM

#

Append and loops in Python are forbidden frog_whip

wispy spear Mar 10, 2024, 8:54 PM

#

its weird why they call it like that

#

rather than just map reduce blabla

#

like any other lang....... wait.... c++ also calls it different, c# too 😄

wicked notch Mar 10, 2024, 8:55 PM

#

alright the conversions have been executed

wispy spear Mar 10, 2024, 8:55 PM

#

now rename vulkan_core.hpp to retina_core.hpp

wicked notch Mar 10, 2024, 8:55 PM

#

def as_retina_enum_values(name: str, values: {str: int}) -> {str: int}:
    output: {str: int} = {}
    common = find_common_enum_string(name)

    for (entity, value) in values.items():
        entity = entity.replace('VK_', 'E_', 1)
        entity = entity.replace('_BIT', '', 1)
        entity = entity.replace(f'_{common}', '', 1)
        entity = remove_vendor_suffix(entity)
        if entity.endswith('MAX_ENUM'):
            continue
        output[entity] = value
    return output``` peak python code

wispy spear Mar 10, 2024, 8:56 PM

#

readable

wicked notch Mar 10, 2024, 8:57 PM

#

I could make it more readable

#

there it is

#

now I only have to figure out mako templates and the codegen is officially complete

#

it only took a few hours

#

@pale horizon mistrusted me

#

smh

pale horizon Mar 10, 2024, 9:02 PM

#

It’s only the beginning… you think you’re almost done(tm)

wicked notch Mar 10, 2024, 9:03 PM

#

the good news is that I don't have to write any more python

#

thank god

wicked notch Mar 10, 2024, 9:38 PM

#

lads

#

it's done

#

📎 Enums.hpp

#

269KiB of pure enum goodness

wicked notch Mar 11, 2024, 12:25 AM

#

it got worse btw

📎 Enums.hpp

#

aight now I have all the enums it's possible to wish for

#

time to continue the engine™️

wicked notch Mar 11, 2024, 9:03 PM

#

9000 untested lines of code of allah froge_love

wispy spear Mar 11, 2024, 9:08 PM

#

you little nutjob

#

Regtina better be a baller engine then

wicked notch Mar 11, 2024, 9:09 PM

#

it is currently all infrastructure kekkedsadge

#

I don't even have a triangle yet

wispy spear Mar 11, 2024, 9:09 PM

#

: >

#

https://tenor.com/view/dredd-i-knew-youd-say-that-stallone-gif-23229845

Tenor

wicked notch Mar 11, 2024, 9:09 PM

#

it'll be baller though

#

I can promise that

#

actually you know what

wispy spear Mar 11, 2024, 9:09 PM

#

noice

wicked notch Mar 11, 2024, 9:09 PM

#

code review time

wispy spear Mar 11, 2024, 9:11 PM

#

https://tenor.com/view/running-late-check-watch-late-hurry-up-surprised-gif-16482455

Tenor

wicked notch Mar 11, 2024, 9:11 PM

#

@frank sail are you in as well

#

for the code review

#

I want the two professional™️ engineers of the server to see this

#

(because everyone else is a fake engineer frfr)

frank sail Mar 11, 2024, 9:13 PM

#

I'll look later

#

!remindme 169.420 minutes review da code

vivid boughBOT Mar 11, 2024, 9:13 PM

#

Alright gpgpu, I'll remind you about review da code in 2 hours, 49 minutes and 25 seconds. ID: 66379570

delicate rain Mar 11, 2024, 9:32 PM

#

wicked notch I want the two professional™️ engineers of the server to see this

When you ask so nicely

#

I guess I can spare a minute

wicked notch Mar 11, 2024, 10:56 PM

#

https://github.com/LVSTRI/Retina code review time

#

small caveat, you may or may not have to update your Vulkan SDK

#

enums were generated from 1.3.275

glass sphinx Mar 11, 2024, 11:06 PM

#

🚨 https://github.com/LVSTRI/Retina/blob/a3e3c55471658ca2e439d94e91dfa85f561f048b/src/Retina/Graphics/CommandBuffer.cpp#L298

GitHub

Retina/src/Retina/Graphics/CommandBuffer.cpp at a3e3c55471658ca2e43...

Contribute to LVSTRI/Retina development by creating an account on GitHub.

#

inserting barriers non batched like this will have performance penalties on at least nvidia

wicked notch Mar 11, 2024, 11:07 PM

#

interesting

#

something to solve when I make the rendergraph

glass sphinx Mar 11, 2024, 11:07 PM

#

they arent auto merged

wicked notch Mar 11, 2024, 11:07 PM

#

ye right now I can't merge them, unless I do it manually I guess

#

rendergraph is probably next on the list

glass sphinx Mar 11, 2024, 11:08 PM

#

good luck

#

daxa has a cursed auto merge in command recorders

wicked notch Mar 12, 2024, 12:03 AM

#

man smooth resize is stupid easy lol

#

why did nobody coerce me into doing it

wicked notch Mar 12, 2024, 12:24 AM

#

(there were bugs, of course there were)

#

@frank sail I await your code review

wispy spear Mar 12, 2024, 12:42 AM

#

its readable

#

i like that

#

i also like that not everything is public and static like in any other engine

#

CommandBuffer's interface is dangerous : )

#

the uninitiated could just call .End().Begin()

#

which you might block with some asserts inside, hehe you dont

#

im not a fan of the screaming constants E_ISMS but ill let it slide, this time

wicked notch Mar 12, 2024, 12:45 AM

#

wispy spear CommandBuffer's interface is dangerous : )

it do be

wispy spear Mar 12, 2024, 12:46 AM

#

to solve this puzzle, temporal coupling could solve it, but might explode the class count

wicked notch Mar 12, 2024, 12:46 AM

#

I've been lazy because it maps to vulkan 1:1, so validation layers will automagically catch stuff like that (End() before Begin())

wispy spear Mar 12, 2024, 12:46 AM

#

fair

wicked notch Mar 12, 2024, 12:46 AM

#

wispy spear to solve this puzzle, temporal coupling could solve it, but might explode the cl...

what ist temporal coupling?

wispy spear Mar 12, 2024, 12:47 AM

#

you would not be able to call on .CommandBuffer at all, but .Begin() would return a new object, which allows all the things you usually record in order when recording a record

#

and only then you can call .End() on it

#

which returns CommandBuffer& again

wicked notch Mar 12, 2024, 12:48 AM

#

ah I see

#

smart

wispy spear Mar 12, 2024, 12:49 AM

#

recording commands probably follows the same patterns for various things

#

with that in mind one could cook up a few of those "temporal couplings" probably

#

anywho was just saying

wicked notch Mar 12, 2024, 12:50 AM

#

ye it's definitely something I will use when rendergraph times come about

#

because command buffers in that thing will be more clamplicated

wispy spear Mar 12, 2024, 12:50 AM

#

heh i can imagine

#

the class prefix C is also weird

#

but its tolerable

#

because all the other stuff is so neat and readable 😄

wicked notch Mar 12, 2024, 12:52 AM

#

ye I hate it

#

But I decided to stick with it because it made everything else nicer

wispy spear Mar 12, 2024, 12:52 AM

#

you could name all your classes and things RtXXX or RetXXX or even RetinaXXX

#

ah, oki

wicked notch Mar 12, 2024, 12:53 AM

#

I'm still thinking about that

wispy spear Mar 12, 2024, 12:53 AM

#

including enums and schtructs

wicked notch Mar 12, 2024, 12:53 AM

#

It doesn't feel "right" because you're already in a namespace

#

but who knows

wispy spear Mar 12, 2024, 12:53 AM

#

yeah

#

i know what you mean

#

im in the same dilemma rn 🙂

frank sail Mar 12, 2024, 12:56 AM

#

wicked notch what ist temporal coupling?

deccer be teaching the dark arts to yet another apprentice

wispy spear Mar 12, 2024, 12:59 AM

#

https://tenor.com/view/greggifs-koidl-stefan-koidl-demon-demons-gif-14263793

Tenor

frank sail Mar 12, 2024, 2:12 AM

#

wicked notch <https://github.com/LVSTRI/Retina> code review time

these barriers are annoyingly explicit, but obviously they map well to vulkan so eugh
https://github.com/LVSTRI/Retina/blob/master/src/Retina/Entry/Application.cpp#L158-L212

#

so far it looks clean though

frank sail Mar 12, 2024, 2:45 AM

#

weird
https://github.com/LVSTRI/Retina/blob/master/src/Retina/Graphics/CommandBuffer.cpp#L159-L160

#

also I'm not sure how useful it is to have these things that return a vector of objects

#

oh hell naw
https://github.com/LVSTRI/Retina/blob/master/src/Retina/Graphics/Device.cpp#L260C7-L260C72

wicked notch Mar 12, 2024, 6:15 AM

#

frank sail oh hell naw <https://github.com/LVSTRI/Retina/blob/master/src/Retina/Graphics/De...

mesh shaders require geometry shader kekkedsadge

wicked notch Mar 12, 2024, 6:18 AM

#

frank sail weird <https://github.com/LVSTRI/Retina/blob/master/src/Retina/Graphics/CommandB...

good catch

frank sail Mar 12, 2024, 6:19 AM

#

your code is very clean btw

wicked notch Mar 12, 2024, 6:19 AM

#

frank sail your code is very clean btw

here's hoping it stays that way

frank sail Mar 12, 2024, 6:30 AM

#

sometimes I feel like like a princess when I sprinkle auto, const, noexcept, and [[nodiscard]] around my code, but then I see your code and it looks like a c++ genie visited

wispy spear Mar 12, 2024, 11:11 AM

#

there is one more thing which i dont like that much

#

MakeXXX instead of CreateXXX

#

Create shows intent better than make

#

i blame std::make_xxx<T>

frank sail Mar 12, 2024, 11:13 AM

#

std::make_lifetime_at

wispy spear Mar 12, 2024, 11:13 AM

#

😛

#

c++ people dont know how to name shit properly imho

#

but its also a hard problem

#

id also get rid of the Entry folder and have everything in Regtina's root

wicked notch Mar 12, 2024, 11:15 AM

#

wispy spear c++ people dont know how to name shit properly imho

real

wispy spear Mar 12, 2024, 11:16 AM

#

i should hand in "N6969: Name your shit properly ffs"

#

or was it P

glass sphinx Mar 12, 2024, 11:29 AM

#

i will steal the way you do features

#

long overdue refactor in daxa

wispy spear Mar 12, 2024, 11:58 AM

#

lustri is the Dr. Frankenstein of

wheat haven Mar 13, 2024, 5:37 AM

#

ah good another engine to ~~steal~~ inspire my own design choices

wheat haven Mar 13, 2024, 6:04 AM

#

What is the benefit of postfix return types anyway?

delicate rain Mar 13, 2024, 6:04 AM

#

All your functions are nicely padded

pale horizon Mar 13, 2024, 8:06 AM

#

When will this thread be renamed? KEKW

frank sail Mar 13, 2024, 8:11 AM

#

we "beyond" now

wicked notch Mar 13, 2024, 11:13 AM

#

pale horizon When will this thread be renamed? <:KEKW:666849321462792234>

I left it as is for nostalgia

delicate rain Mar 13, 2024, 11:20 AM

#

Everyone hyped for Retina while I'm out here waiting for the "optic nerve" (version 3.0)

pale horizon Mar 13, 2024, 11:20 AM

#

LVSTRI gets Neuralink and starts hacking it KEKW

#

Finally, ray tracing IN YOUR BRAIN mindblown

frank sail Mar 13, 2024, 11:21 AM

#

ray tracing at 3am (gone sexual)

wispy spear Mar 13, 2024, 12:05 PM

#

need a postfix ctor and dtor too

#

auto Foo -> ()

#

auto Foo -> ~()

#

😄

frank sail Mar 13, 2024, 12:08 PM

#

did you know that you can put [[nodiscard]] on constructors

wicked notch Mar 13, 2024, 1:16 PM

#

how do I stop clangd from complaining about hidden non virtual functions ffs

delicate rain Mar 13, 2024, 1:21 PM

#

You can add comment decorations to turn off warnings

wheat haven Mar 13, 2024, 1:47 PM

#

hidden non-virtual functions?

wicked notch Mar 13, 2024, 2:01 PM

#

ye, when you do

struct Base {
  int Get();
};

struct Derived : Base {
  int Get();
};```

wheat haven Mar 13, 2024, 3:31 PM

#

ah, that does feel a bit weird

pale horizon Mar 13, 2024, 3:48 PM

#

You should put override here though KEKW

wheat haven Mar 13, 2024, 3:49 PM

#

is that allowed on a non-virtual?

pale horizon Mar 13, 2024, 3:50 PM

#

No misinfo

#

clangd is being goofy ah then

wicked notch Mar 13, 2024, 3:52 PM

#

wheat haven ah, that does feel a bit weird

it's perfectly reasonable to want this though

#

I don't need dynamic dispatch, I'm 100% sure I'll always know the type statically

wheat haven Mar 13, 2024, 3:53 PM

#

I would just name them differently then

pale horizon Mar 13, 2024, 3:54 PM

#

Just don't inherit from stuff. Ever smart

cold sky Mar 13, 2024, 4:00 PM

#

wicked notch I don't need dynamic dispatch, I'm 100% sure I'll always know the type staticall...

use CRTP then

wicked notch Mar 13, 2024, 5:45 PM

#

ye perhaps that's a solution

#

RetinaDeclarePushConstant() {
  uint u_ViewBufferIndex;
};

RetinaDeclareBuffer(restrict readonly, SViewInfoBuffer) {
  mat4 Projection;
  mat4 View;
  mat4 ProjView;
};
RetinaDeclareBufferPointer(SViewInfoBuffer, g_ViewInfoBuffer, u_ViewBufferIndex);``` this feels so good to use

#

With this the base Retina is ported over

#

raytracing is sadly missing but I didn't like the original API anyways

wheat haven Mar 13, 2024, 6:03 PM

#

what do you do if the buffer has no qualifiers?

wicked notch Mar 13, 2024, 6:03 PM

#

you just omit the first parameter

wheat haven Mar 13, 2024, 6:04 PM

#

I thought macros didn't resolve like that, huh

wicked notch Mar 13, 2024, 6:04 PM

#

I didn't test this but iirc they to

#

lemme try

#

ok yeah you're right

#

shit

#

fucking glsl piece of trash

wheat haven Mar 13, 2024, 6:06 PM

#

it's a C thing too 😄

wicked notch Mar 13, 2024, 6:07 PM

#

yeah

#

but I'll still take it out on glsl

distant lodge Mar 13, 2024, 6:07 PM

#

I think you only had good variadic macros in C++ 17 or 20

#

that can automatically tolerate 0 args to the variadic

wheat haven Mar 13, 2024, 6:08 PM

#

RetinaDeclareBufferEx

#

or keep with vulkan style and go RetinaDeclareBuffer2 😄

wicked notch Mar 13, 2024, 6:09 PM

#

one could do RetinaDeclareQualifiedBuffer(, SViewInfoBuffer)

#

but it's garbage

wispy spear Mar 13, 2024, 6:09 PM

#

wicked notch ye, when you do ```cpp struct Base { int Get(); }; struct Derived : Base { ...

no reintroducing methods should be a bannable offense

wicked notch Mar 13, 2024, 6:10 PM

#

why :(

wheat haven Mar 13, 2024, 6:10 PM

#

wicked notch one could do `RetinaDeclareQualifiedBuffer(, SViewInfoBuffer)`

yeah macros are pretty primitive still unfortunately

wispy spear Mar 13, 2024, 6:10 PM

#

because you hide Base::Get

wicked notch Mar 13, 2024, 6:11 PM

#

that's intended tho

#

I don't want you to call Base::Get

wispy spear Mar 13, 2024, 6:11 PM

#

then inheritance isnt the right tool for the job here

wicked notch Mar 13, 2024, 6:11 PM

#

how would you do Buffer and TypedBuffer<T> then

wispy spear Mar 13, 2024, 6:11 PM

#

same way jaker and i did itin the past

wicked notch Mar 13, 2024, 6:12 PM

#

class Buffer {
  GetSize(); // size in bytes
}

class TypedBuffer<T> {
  GetSize(); // element count
}```

wicked notch Mar 13, 2024, 6:12 PM

#

wispy spear same way jaker and i did itin the past

I don't remember bleakekw

wispy spear Mar 13, 2024, 6:12 PM

#

have GetElementCount

#

and GetSize returns the size in bytes in both classes

wicked notch Mar 13, 2024, 6:13 PM

#

ok that's fair

#

but I still need buffer and typedbuffer to be related

#

that's a hard requirement

wispy spear Mar 13, 2024, 6:13 PM

#

thats ok

#

https://github.com/JuanDiegoMontoya/Fwog/blob/main/include/Fwog/Buffer.h

#

or crtp it as devsh suggested

wheat haven Mar 13, 2024, 6:17 PM

#

also random question, what exactly is ArcPtr? it seems to be a smart pointer of some kind, but it's...odd to me. Reset will delete the underlying pointer which implies the class owns the pointer, but the class is copyable so it's not unique either

wicked notch Mar 13, 2024, 6:17 PM

#

wheat haven also random question, what exactly is `ArcPtr`? it seems to be a smart pointer o...

it's an intrusive shared_ptr

wheat haven Mar 13, 2024, 6:18 PM

#

ahh, I missed the Grab and Drop calls; I thought you'd have a requirement on the template

wicked notch Mar 13, 2024, 6:19 PM

#

ye perhaps I should add that requirement

wheat haven Mar 13, 2024, 6:19 PM

#

where'd the name come from? 'cause I have the same thing but I just called it IntrusivePtr (which is more of a mouthful)

wicked notch Mar 13, 2024, 6:20 PM

#

Rust

#

but rust's Arc is actually not intrusive

#

so it's a bit of a misnomer

wheat haven Mar 13, 2024, 6:21 PM

#

I might end up renaming mine to just Ref or something

distant lodge Mar 13, 2024, 6:22 PM

#

I renamed my IntrusivePtr to RefPtr some time ago after a heavy dose of apple webkit code

frank sail Mar 13, 2024, 6:22 PM

#

wheat haven I might end up renaming mine to just `Ref` or something

call yours &

distant lodge Mar 13, 2024, 6:22 PM

#

AmpersandOf<T>

frank sail Mar 13, 2024, 6:23 PM

#

berfect

wispy spear Mar 13, 2024, 6:25 PM

#

KaufmannsUnd<T>

wicked notch Mar 13, 2024, 6:26 PM

#

I forgor to check whether my thing builds on the penguin

wispy spear Mar 13, 2024, 6:26 PM

#

i can try

wicked notch Mar 13, 2024, 6:26 PM

#

I can also probably integrate CI as well because I don't clone shaderc anymore kekkedsadge

wispy spear Mar 13, 2024, 6:26 PM

#

i am on a penguin laptop atm

frank sail Mar 13, 2024, 6:26 PM

#

wispy spear KaufmannsUnd<T>

KaufmannsUndies

wispy spear Mar 13, 2024, 6:28 PM

#

btw thank you for not using vcpkg

#

ah build fails

wicked notch Mar 13, 2024, 6:30 PM

#

sadness

wispy spear Mar 13, 2024, 6:30 PM

#

xD

#

[build] /home/deccer/Code/External/Retina/include/Retina/Core/STL/ArcPtr.hpp:143:18: internal compiler error: in build_comparison_op, at cp/method.cc:1461

#

gcc 13.2.1

wicked notch Mar 13, 2024, 6:30 PM

#

what in tarnation

wispy spear Mar 13, 2024, 6:31 PM

#

📎 message.txt

wicked notch Mar 13, 2024, 6:32 PM

#

actually incredible

wispy spear Mar 13, 2024, 6:32 PM

#

and with clang it stops at including windows.h 😛

#

[build] /home/deccer/Code/External/Retina/src/Retina/WSI/Window.cpp:5:10: fatal error: 'Windows.h' file not found
[build]     5 | #include <Windows.h>
[build]       |          ^~~~~~~~~~~
[build] 1 error generated.

wicked notch Mar 13, 2024, 6:33 PM

#

oh shit I forgor to remove it yesterday night KEKW

#

does it work without windows.h

frank sail Mar 13, 2024, 6:41 PM

#

it makes sense that window.cpp would depend on windows.h

wicked notch Mar 13, 2024, 6:57 PM

#

time to submit the Nth compiler bug

#

C++ is pain

wispy spear Mar 13, 2024, 7:02 PM

#

you could always switch to c#

#

and leave c++ in the sewers where it belongs

wicked notch Mar 13, 2024, 11:28 PM

#

figure out why

📎 message.txt

wicked notch Mar 13, 2024, 11:44 PM

#

figured it out yet?

#

I forgor to delete the copy constructor

#

clang decided to tell me in the most unhinged way possible

delicate rain Mar 13, 2024, 11:46 PM

#

I tried to read the message on my phone, first I got a pop up to download it, then to read it in Google docs then another pop up that I first need to upload it to Google drive after which a popup of "give me permission to view your files" followed

#

After which I gave up

wicked notch Mar 13, 2024, 11:46 PM

#

Bruh

delicate rain Mar 13, 2024, 11:46 PM

#

agonyfrog

wispy spear Mar 14, 2024, 11:22 AM

#

@wicked notch we need to link with tbb as well

#

on lunix

wicked notch Mar 14, 2024, 11:22 AM

#

huh wat

#

I don't remember using any parallel algorithms ever

wispy spear Mar 14, 2024, 11:25 AM

#

[build] /usr/include/oneapi/tbb/detail/_task.h:169:(.text+0x3d9): undefined reference to `tbb::detail::r1::execution_slot(tbb::detail::d1::execution_data const*)'
[build] /usr/bin/ld: /tmp/cco0F0AY.lto.o: in function `tbb::detail::d1::current_thread_index()':
[build] /usr/include/oneapi/tbb/task_arena.h:451:(.text+0x3ed): undefined reference to `tbb::detail::r1::execution_slot(tbb::detail::d1::execution_data const*)'

wicked notch Mar 14, 2024, 11:26 AM

#

does this tell you where the reference was named?

wispy spear Mar 14, 2024, 11:27 AM

#

gcc complains only

#

clang has no problem and builds

wicked notch Mar 14, 2024, 11:27 AM

#

agonyfrog

wispy spear Mar 14, 2024, 11:29 AM

#

thats weird, i can only find tbbisms in tracy but it also seems to be disabled there

wicked notch Mar 14, 2024, 11:30 AM

#

it's probably cause of LTO innit

wispy spear Mar 14, 2024, 11:30 AM

#

i wouldnt know, i only know what lto is spelled out

#

but not how it conducts its bisnes

#

i did the classic delete build, lets see if that fixes it : )

#

ah, itnt

#

📎 message.txt

wicked notch Mar 14, 2024, 11:33 AM

#

ye it's definitely because of tracy

#

well that's my fault

wispy spear Mar 14, 2024, 11:34 AM

#

NO_TBB = 1

#

i read that as "dont touch or use tbb"

wicked notch Mar 14, 2024, 11:35 AM

#

you can pull now

#

I fix

hallow cedar Mar 14, 2024, 12:03 PM

#

wicked notch I don't remember using any parallel algorithms ever

do you use std execution policies

wicked notch Mar 14, 2024, 12:07 PM

#

I do not

#

it's tracy's fault, though idk why it links fine with clang but not with gcc

wispy spear Mar 14, 2024, 12:13 PM

#

nope

wicked notch Mar 14, 2024, 12:15 PM

#

rip

wispy spear Mar 14, 2024, 12:15 PM

#

did rm build/ and reconfictured

#

idk mayhaps pull tbb (if it has sources available soewhere) if gcc

#

build and link with it for the time being

#

or require lunix people to instal libtbb and you just link with it if gcc

pale horizon Mar 14, 2024, 3:00 PM

#

wicked notch it's tracy's fault, though idk why it links fine with clang but not with gcc

idk, I'm using GCC and there's no link to TBB

#

~/work/edbr/build$ ldd src/game 
    linux-vdso.so.1 (0x00007ffe2bfd9000)
    libglm.so => /home/eliasdaler/work/edbr/build/third_party/glm/glm/libglm.so (0x00007f1b3db17000)
    libfreetyped.so.6 => /home/eliasdaler/work/edbr/build/third_party/freetype/libfreetyped.so.6 (0x00007f1b3da1c000)
    libSDL2-2.0d.so.0 => /home/eliasdaler/work/edbr/build/third_party/SDL/libSDL2-2.0d.so.0 (0x00007f1b3d600000)
    libTracyClient.so.0.10.0 => /home/eliasdaler/work/edbr/build/third_party/tracy/libTracyClient.so.0.10.0 (0x00007f1b3d200000)
    libvulkan.so.1 => /home/eliasdaler/work/vulkan_sdk/1.3.275.0/x86_64/lib/libvulkan.so.1 (0x00007f1b3d9a2000)
    libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f1b3ce00000)
    libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f1b3d521000)
    libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f1b3d951000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f1b3cc1e000)
    /lib64/ld-linux-x86-64.so.2 (0x00007f1b3dbd1000)
    libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f1b3d94a000)
    libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f1b3d945000)
~/work/edbr/build$ ldd /home/eliasdaler/work/edbr/build/third_party/tracy/libTracyClient.so.0.10.0
    linux-vdso.so.1 (0x00007ffe0fef8000)
    libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f1abf400000)
    libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f1abf721000)
    libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f1abfa45000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f1abf21e000)
    /lib64/ld-linux-x86-64.so.2 (0x00007f1abfa98000)

wicked notch Mar 14, 2024, 3:01 PM

#

ye it be weird

#

does it compile for you?

pale horizon Mar 14, 2024, 3:05 PM

#

fatal: reference is not a tree: 923e865b7da81bfcab5c54f260c0f034953d406e
CMake Error at spirv-tools-subbuild/spirv-tools-populate-prefix/tmp/spirv-tools-populate-gitclone.cmake:49 (message):
  Failed to checkout tag: '923e865b7da81bfcab5c54f260c0f034953d406e'

idk KEKW
Don't want to debug for now

wicked notch Mar 14, 2024, 3:22 PM

#

incredible

pale horizon Mar 14, 2024, 3:24 PM

#

FetchContent issue
Git submodule chads win once again

wicked notch Mar 14, 2024, 3:35 PM

#

I don't even clone spirv tools

wispy spear Mar 14, 2024, 4:16 PM

#

i think i saw it the conf part too

#

spirv_cross-src is in build/Deps 😄

#

also elias is on some outdated debian 😛

pale horizon Mar 14, 2024, 4:34 PM

#

I’m on testing! KEKW

wheat haven Mar 14, 2024, 4:39 PM

#

wispy spear spirv_cross-src is in build/Deps 😄

spirv-cross isn't spirv-tools; I double-checked the spirv-cross repo and they don't fetchcontent it in either

pale horizon Mar 14, 2024, 4:52 PM

#

Okay, it builds, I had some merge conflict KEKW

#

Did git pull again:

CMake Error at Dependencies/CMakeLists.txt:174 (target_sources):
  Cannot find source file:

    /home/eliasdaler/work/repos_to_check/Retina/Dependencies/cgltf/Implementation.c


CMake Error at Dependencies/CMakeLists.txt:173 (add_library):
  No SOURCES given to target: cgltf

FetchContent is a meme

#

(Yes, I removed build dir and created it again)

#

Dependencies/CGLF
bro...

wheat haven Mar 14, 2024, 4:57 PM

#

common windows L

#

case sensitivity

pale horizon Mar 14, 2024, 4:57 PM

#

More like case insensitivity here

#

diff --git a/Dependencies/CMakeLists.txt b/Dependencies/CMakeLists.txt
index 4ee1997..ec72b9c 100644
--- a/Dependencies/CMakeLists.txt
+++ b/Dependencies/CMakeLists.txt
@@ -172,7 +172,7 @@ if (NOT cgltf_POPULATED)
   FetchContent_Populate(cgltf)
   add_library(cgltf STATIC)
   target_sources(cgltf PRIVATE
-    ${CMAKE_CURRENT_SOURCE_DIR}/cgltf/Implementation.c
+    ${CMAKE_CURRENT_SOURCE_DIR}/CGLTF/Implementation.c
   )
   target_include_directories(cgltf PUBLIC
     ${cgltf_SOURCE_DIR}

xoxo

wheat haven Mar 14, 2024, 4:57 PM

#

wicked notch Mar 14, 2024, 4:59 PM

#

pale horizon ```diff diff --git a/Dependencies/CMakeLists.txt b/Dependencies/CMakeLists.txt i...

incredible

#

ye that's on me KEKW

pale horizon Mar 14, 2024, 4:59 PM

#

Also what do I launch now KEKW

wicked notch Mar 14, 2024, 5:00 PM

#

are you on latest

pale horizon Mar 14, 2024, 5:00 PM

#

Yes

wicked notch Mar 14, 2024, 5:00 PM

#

Retina.Entry

pale horizon Mar 14, 2024, 5:00 PM

#

Beautiful, 11/10 raytracing

#

~/work/repos_to_check/Retina/build$ ldd ./src/Retina/Entry/Retina.Entry
    linux-vdso.so.1 (0x00007ffce9340000)
    libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fcd50600000)
    libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fcd50521000)
    libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fcd5088c000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fcd5033f000)
    /lib64/ld-linux-x86-64.so.2 (0x00007fcd508df000)

hmm, no Tracy and no TBB

frank sail Mar 14, 2024, 6:02 PM

#

wispy spear i read that as "dont touch or use tbb"

It's actually "no, tbb"

wheat haven Mar 14, 2024, 6:20 PM

#

pale horizon ```sh ~/work/repos_to_check/Retina/build$ ldd ./src/Retina/Entry/Retina.Entry ...

option(RETINA_ENABLE_LOGGER "Enable logging" OFF)
option(RETINA_ENABLE_PROFILER "Enable CPU profiling" OFF)

#

also they're all static so they wouldn't be in ldd anyway

pale horizon Mar 14, 2024, 6:41 PM

#

Build shared libs is ON by default I think. Unless LVSTRI overrides it

wheat haven Mar 14, 2024, 6:44 PM

#

true, but all of the parts Retina.Core, Retina.Graphics, etc are set as static, which...I think forces all of their dependencies to be baked in too?

pale horizon Mar 14, 2024, 6:44 PM

#

Ah, I see

wicked notch Mar 14, 2024, 8:31 PM

#

we're back

#

now, time to make that VSM sample I promised to jaker

#

and then rendergraph

frank sail Mar 14, 2024, 8:31 PM

#

VSM shrimple

#Iris - A Journey through OpenGL and beyond to learn Graphics