Andromeda - Terrain/Landscape editor using Vulkan and Rust

[shadow trench]

sounds pretty fun

[hot yarrow]

liquidgas was its name

hmm maybe i should dig it out from some old backup cds 😄

it looked quite neat

[shadow trench]

would love some screenshots of what young deccer was up to :)

[hot yarrow]

: )

[shadow trench]

Seems about right

You can see the frame lag if I move too quickly though

[hot yarrow]

yeah

you probably need some lod

[shadow trench]

But as long as im standing still its perfectly centered

[shadow trench]

how would that help?

huh thats odd

[hot yarrow]

i mean redude the amound of triangles

[shadow trench]

ah wait nvm I put something there lol

[hot yarrow]

ah

you can also see it in the jiff.gif

[shadow trench]

Oh yeah I can do dynamic lod based on the size of each patch on screen

[hot yarrow]

you clicked

[shadow trench]

eventually

it runs fine for now but yeah ill definitely need lods

okay so the position is fine

then either my komput logic is wrong or computing the uv from the position is wrong

[livid geode]

chunking will work wonderfully to reduce lag once you implement it

in this case probably a quadtree would work best

[young oar]

Yep quadtrees/octrees are beautiful

Without octree

[shadow trench]

ill look into them for sure

[young oar]

With octree

They are op when you manage to fix seams which is super easy for 2d planes

[shadow trench]

ok there now it doesnt do that weird half circle thing anymore

[livid geode]

i have a crate to comvert data octrees into meshes lol

sucks balls tho

[shadow trench]

👀

oh

[livid geode]

but you could look into dual marching cubes

the results are pretty good

much better than marching cubes

[shadow trench]

yeah i have a lot of techniques to still look into

ahhhh

my uv calculation was indeed wrong

the mesh itself has as max uvs 0.98 ish or so because of a bug

but the uv calc from the pos assumed 1 was the max value

[livid geode]

bug accumulation

[shadow trench]

for the longest time that just meant a tiny part of the heightmap was cut off so it was barely noticable

[hot yarrow]

@young oar i like your terrains more 🙂

[shadow trench]

[hot yarrow]

hihi

[young oar]

Heheheha

Tbf it doesn't have editing or anything cool like that

Just procedurally generated

[shadow trench]

next step for editing is holding the mouse down to brush instead of clicking once I guess

[livid geode]

why did it happen? something something x/N vs x/(N-1)?

[shadow trench]

Exactly that

[livid geode]

meheheh

[shadow trench]

huh I think it's also wrong in sasha's samples then

ill open an issue

[shadow trench]

It'll look a lot better once I regenerate normals

I dont think those normals are right

actually maybe it is and the brush just isnt smooth at all

ill have to experiment a bit with the weight function I guess

This is some interesting behaviour

I think I'll have to apply some sort of smoothing after each edit or something

[young oar]

UI looks rlly good tho

[shadow trench]

egui is nice

[shadow trench]

https://cdn.discordapp.com/attachments/714940749707214890/1108324682190573638/2023-05-17_11-25-51.mp4

There we go

return 1; at the start of my weight function was not helping the smoothness

[shadow trench]

i found a really cool crate

https://crates.io/crates/enum_dispatch

#[enum_dispatch]
#[derive(Debug, Copy, Clone)]
pub enum Brush {
    SmoothHeight,
}

#[enum_dispatch(Brush)]
trait ApplyBrush {
    fn apply(&self, bus: &EventBus<DI>, position: Vec3, settings: &BrushSettings) -> Result<()>;
}

struct SmoothHeight {
  // brush-specific settings
}

impl ApplyBrush for SmoothHeight {
  // ...
}

fn use_brush(brush: Brush) {
  // generated code automatically dispatches to the correct impl
  brush.apply(...);
}

[livid geode]

It should be something that is built in

[shadow trench]

Yeah that would be nice

but this is a decent substitute

[livid geode]

but tbh i'd just use dyn

[shadow trench]

meh

[livid geode]

i was like this before, trying to avoid dyn whenever possible

but like who cares the perf cost is minimal

and it is definitely cleaner

[shadow trench]

I mean I use it at times but I dont want to create a new boxed brush object on every stroke

[livid geode]

why the hell you creatin a brush on every stroke

[shadow trench]

the "brush" is only some settings

it doesnt actually hold any real data

[livid geode]

yes i know

but why don't you just store it

or actually you do store it, why do you need to create it on every stroke

[shadow trench]

I keep it around in the gui yeah

[livid geode]

then?

[shadow trench]

idk I feel like with a trait object creating the gui for each type of brush is going to be annoying

bc either I add it to the ApplyBrush trait which is not great

But with this I can just create a new trait, enum_derive it and keep it entire inside the gui crate

[livid geode]

what is gonna be annoying with enum dispatch tho is that every time you want to create, rename, etc a brush, you also need to change Brush

[shadow trench]

Yeah true

But tbh, probably not a huge deal

[shadow trench]

the decal seems a bit big compared to the actual size of the brush

Maybe my distribution falls off too quickly

[livid geode]

can't you also display the distribution on the decal?

[shadow trench]

Yeah I'll try something like that eventually

thats a good idea

[livid geode]

i thought it wasn't a decal but just a vertex shader whose input was the mouse uv pos

[shadow trench]

I was thinking of just using a hardcoded texture but that'd be cool

[shadow trench]

tbh that might work even better

what im doing rn is kind of a hack in the terrain fragment shader

I just check the distance to the mouse

[livid geode]

ye, also supports all kinds of brushes even ones that might be temporally dependant or change with parameters, just need a shader for each

[shadow trench]

Yeah

[livid geode]

ye so do that but in the vertex shader and have the distribution fn in there

[shadow trench]

yeah

[livid geode]

yeah

[shadow trench]

right, the problem with a wider distribution is that you get ugly seams

I can just shrink the decal a bit I suppose

or add the blur pass again

[nimble solar]

hi

[shadow trench]

oi

hello there

[nimble solar]

I am quite intrigued by the terrain generation you have going on there

[shadow trench]

[nimble solar]

Do you plan to extend this generation to infinity and beyond?

[shadow trench]

That's definitely on the list yeah

im not doing any generation rn

Just editing a heightmap

[nimble solar]

Good, I could use your stuff as a reference for my Snow™️ thingy

[shadow trench]

whats that

[nimble solar]

It's basically Steep except it's mine.

[shadow trench]

i see

must be better then

[nimble solar]

How do you do generate the terrain?

As a mesh, I mean

[shadow trench]

its just loaded in from a heightmap asset i found

ah, the mesh

its a highly subdivided plane, the vertical deform happens on the fly when rendering

using tessellation to generate the subdivided mesh

[nimble solar]

I see, do you plan to use Mesh Shaders instead of Tess?

[shadow trench]

i honestly have not looked into them at all

But maybe

[nimble solar]

They're just a compute shader on steroids

[shadow trench]

Right now this already performs very well, 2ms total frametime with no LODing

[nimble solar]

A compute shader that can write to gl_Position gl_MeshPrimitivesNV

[shadow trench]

im definitely not bound by vertex throughput at least

[nimble solar]

Yeah I don't really care about performance, I just like using the new stuff

[shadow trench]

totally fair lol

[shadow trench]

wow my rendering code is a dumpsterfire

i have to clean that up some day

time to take a massive detour to implement decal rendering

at least itll be reusable for other things later

This whole terrain brush adventure is definitely a good subject for a second blog post

[shrewd lynx]

Relatable as hell, I started cleaning my code lately

Now I have an ubershader

[shadow trench]

not sure if that counts as a win or not

[shrewd lynx]

I mean atleast I only have a single backend for almost everything except depthwrite

But yeah the tradeoff is messy code on That file

[shadow trench]

thats good at least yeah

[hot yarrow]

finally you see it XD

[nimble solar]

Where's your project post btw

My follow list isn't big enough

[shadow trench]

im generally happy with my codebase, just not the rendering backend

[hot yarrow]

im just talking out of my ass

[shadow trench]

[shrewd lynx]

I wish I can but that would just mean advertising a game

[shrewd lynx]

Rendering backends tend to be messy, and I've seen some horrors

So ur fine haha

[hot yarrow]

are you not working on some homerolled engine miyagi?

https://tenor.com/view/miyagi-karatekid3-gif-19396424

[shrewd lynx]

Custom Source engine not unlike bms, respawn

[hot yarrow]

im sure people are interested

[nimble solar]

I am

[hot yarrow]

just saying 🙂

lustri you shouldnt, you got work todo 😄 (jk)

[nimble solar]

I need to understand how other engines work to make my game

Atmospheric Scattering? I don't have the motivation for that 😦

[shadow trench]

but it looks great

[nimble solar]

Yeah I know

I watched Sebastian Lague's video

[shadow trench]

decals ™️

[potent quest]

💀

Wait so is Andromeda just to render terrain? Oh it's edit

[shadow trench]

terrain editing and everything related yes

[potent quest]

it would be cool if you added erosion

[shadow trench]

thats absolutely planned

[potent quest]

oh real

tectonic simulations?

[shadow trench]

maybe

that could be interesting

[potent quest]

https://www.youtube.com/watch?v=GJQVl6Xld0w

Afaik from a very quick read

it looks like they're creating a bunch of points on a sphere and then using a bunch of math equations to describe what happens to them when they overlap with each other

[shadow trench]

Right

[potent quest]

Oh here's the paper: https://hal.archives-ouvertes.fr/hal-02136820/file/2019-Procedural-Tectonic-Planets.pdf if you're interested

[shadow trench]

[hot yarrow]

nice xray

[shadow trench]

Yep

[shadow trench]

hmm that barrier looks okay to me

why is it not happy

SYNC-HAZARD-WRITE-AFTER-READ(ERROR / SPEC): msgNum: 929810911 - Validation Error: [ SYNC-HAZARD-WRITE-AFTER-READ ] Object 0: handle = 0x17e0579e890, type = VK_OBJECT_TYPE_QUEUE; | MessageID = 0x376bc9df | vkQueueSubmit2: Hazard WRITE_AFTER_READ for entry 0, VkCommandBuffer 0x17e2f09a050[], Recorded access info (recorded_usage: SYNC_IMAGE_LAYOUT_TRANSITION, command: vkCmdPipelineBarrier2, seq_no: 3, reset_no: 1). Access info (prior_usage: SYNC_COMPUTE_SHADER_SHADER_SAMPLED_READ, read_barriers: VkPipelineStageFlags2KHR(0), queue: VkQueue 0x17e0579e890[], submit: 476, batch: 0, batch_tag: 9416, command: vkCmdDispatch, seq_no: 14, command_buffer: VkCommandBuffer 0x17e29576430[], reset_no: 1).
    Objects: 1
        [0] 0x17e0579e890, type: 4, name: NUL

ahhhh

I see

the initial start of frame barrier doesnt take the fragment stage too

another day, another bugfix in the rendergraph I guess

[shadow trench]

ffs not this again

no syncval output

only happens when I get too close to the decal

time for a driver update

though i see nothing about an issue like this in the update notes

[shadow trench]

Proper brush decal

[livid geode]

neato

[short surge]

i like this visualisation

how did you do that

[shadow trench]

the DAG crate I use to back my rendergraph can output dot files

[shadow trench]

@livid geode I added the weight shader to the decal as you suggested

float4 main(PS_INPUT input, float4 frag_pos : SV_Position) : SV_TARGET {
    float2 uv = decal_uv(frag_pos);
    float2 centered_uv = uv * 2.0 - 1.0;
    // Discard everything outside the brush area
    float distance = length(centered_uv);
    if (distance >= 1.0) {
        return float4(0.0, 0.0, 0.0, 0.0);
    }

    // We will use our weight function to color the decal
    float weight = weight_function(distance);
    return float4(1.0, 0.0, 0.0, 1.0) * weight;
}

very shrimple to extend if needed

[livid geode]

yoooooooo

i thought that by decal you meant texture

doesn't decal mean that?

like projected texture

[shadow trench]

Yeah generally

But here the texture is just implicitly defined in the shader logic

I can add a texture too

[livid geode]

wait what is this applying to?

this returns a color from black to red right

where are you doing the shading and whatnot

[shadow trench]

i have an include at the top that does all the decal logic

you get the uv with float2 uv = decal_uv(frag_pos);

Then you define what the decal looks like in function of that

Could be a texture, or in this case a circle shape that fades with the brush weight function

[hot yarrow]

looking forward to the "erode" brush

[shadow trench]

I think that would probably be a button/tool, since idk if it makes sense to erode parts of the terrain instead of the whole thing

[hot yarrow]

looks smoof when you watch it in fullscreen

fair

but perhaps you want to draw river beds somehow

[shadow trench]

Yeah thats true

im not sure how ill handle that yet

[hot yarrow]

and some rivers flow faster ie make the bed erode kwiker perhaps

but im no geologist

[shadow trench]

Maybe you first outline the general shape and then apply postprocessing on it when you release the brush? not sure

[hot yarrow]

possible yeah

[livid geode]

so this is not being applied directly to the terrain mesh?

it's something you project on top?

[shadow trench]

To the heightmap

The mesh itself is not modified

[livid geode]

wdym the heightmap

[hot yarrow]

hehe export displacement maps for 4 verticle quads

[shadow trench]

The terrain mesh is a flat plane

[livid geode]

with a really high resolution

yes

[shadow trench]

(after tessellation, yeah)

The brush just prepends commands to the frame's rendering commands to write to the heightmap in a compute shader

[livid geode]

i'm talking about the red dot you put on top bruv

[shadow trench]

ahh

[hot yarrow]

that wasnt really obvious 😄

[shadow trench]

yeah i project it onto the terrain

[livid geode]

so its a high resolution quad you project on top?

[shadow trench]

its a cube

[livid geode]

its a what

[shadow trench]

https://mtnphil.wordpress.com/2014/05/24/decals-deferred-rendering/ this explains it better than i can

its pretty short

(the relevant part at least)

im not doing deferred rendering but for this I dont need a full g buffer, just depth info

[livid geode]

that is pretty cool, thanks

[shadow trench]

[young oar]

Ahhh deferred rendering

I need to implement it too

[shadow trench]

ehh

not really

[young oar]

Sole reason I don't want to is because it makes materials a bit of a pain to write

[young oar]

I want ssr

[shadow trench]

deferred was cool 15 years ago

[young oar]

Is it not anymore??

[livid geode]

what's cool now

forward+?

[shadow trench]

nanite

forward+ is nice yeah

[livid geode]

i have no idea what any of these mean 😎

[young oar]

Still doesn't allow you to do SSR tho right?

[shadow trench]

with forward+ you split your screen into tiles and determine what lights affect what tiles in a culling pass

[shadow trench]

im not sure

possibly not

[young oar]

I always thought only deferred allow you to do screen space stuff like that

(That are based on metallic/smoothness values)

[shadow trench]

that would make sense yeah

problem with deferred is bandwidth

[hot yarrow]

you still have some sort of mini gbuffer with forward, dont you?

[shadow trench]

depth prepass is not uncommon ye

[livid geode]

are these ways to handle lighting? because if so i dont understand how nanite fits in that category

[shadow trench]

nanite is entirely different

[shadow trench]

that may not have been very accurate of me

[livid geode]

that was not very cash money of you

[shadow trench]

I added a reverse mode to decrease height

[young oar]

Awesome!

Those normals look so good

[shadow trench]

thanks

I generate a normal map the same resolution as the heightmap

[young oar]

Oh dang really

I though the normal map was higher res

But eh at least you have a normal map

[shadow trench]

having icons in my source is so cursed

[shrewd lynx]

@shadow trench nice

jwu and u did it already

[shadow trench]

Spent all day on this :P

its 10pm here now

im now fiddling with egui to make a good looking toolbar

[livid geode]

let 🗿 = ctx.sfx.play("vine_boom.ogg")

[shadow trench]

does that look centered

it does right

nvm thats not centered at all

maybe i should just use SelectableLabel but meh

[livid geode]

you can draw the icon yourself

from egui

[shadow trench]

Yeah I ended up copy pasting the selectable_label code and adapting it a bit

[shadow trench]

egui is such a joy to use and look at compared to imgui

[livid geode]

i don't like the default look of egui that much tbh

there's not much contrast

[shadow trench]

That's true

It's easy enough to style though

And way easier to navigate/customize imo

[livid geode]

you think so?

with imgui there were a lot of ready to go customization options

[shadow trench]

Yeah definitely

[livid geode]

i havent seen any for egui

and the style is kinda awkward to navigate thri

[shadow trench]

Creating custom widgets is super simple

[livid geode]

too many substructs

yeah that is a plus

[shadow trench]

idk if I have to look at imgui source code to see how a widget is implemented and how I can adapt it I can dig through a 3000 line cpp file

with egui its not too hard to figure out

[livid geode]

nono for widgets its easy

i mean for changing style settings

[shadow trench]

I havent fiddled with that too much tbh

[livid geode]

i recently fell in love with saleae's logic software

and frankly it doesn't look too off from egui

doesn't look too hard to replicate

[shadow trench]

Is it made using egui or

[livid geode]

nah its electron

[shadow trench]

ah

Ive seen some really cool things in the egui server from people with amazing customization

or on the showcase thread on the repo

[livid geode]

yeah thats true

just need to put in some work

[shadow trench]

yeah

ill definitely do some work to make this not look like default egui eventually

[livid geode]

heres some more inspiration :p

this one is egui

[shadow trench]

that looks really nice

[livid geode]

i wish it was easier to setup animations within egui

[shadow trench]

Is it not?

The custom toggle widget demo has a little animation thingy

But i havent tried much

[livid geode]

does it use the computer api?

value computer/cache

[shadow trench]

// Let's ask for a simple animation from egui.
        // egui keeps track of changes in the boolean associated with the id and
        // returns an animated value in the 0-1 range for how much "on" we are.
        let how_on = ui.ctx().animate_bool(response.id, *on);

i dont think so

[livid geode]

oh it does have animation utils? huh

[shadow trench]

Yeah, maybe not a lot but simple ones I think so

[livid geode]

aha https://docs.rs/egui/latest/egui/struct.Context.html#method.animate_value_with_time

yeah, seems only basics

[shadow trench]

At least you dont need to keep track of the animation progress yourself

[livid geode]

ye, that's p cool

[shadow trench]

Not recommended settings

[shadow trench]

writing a new post about the brush implementation

[livid geode]

lets fucking goooooooooooooooooooooo

[hot yarrow]

can you try this heightmap?

[shadow trench]

sure

[hot yarrow]

found that while looking for some siht

[shadow trench]

its pretty cool

takes ages to load though

I adjusted the scale a bit

I can play a bit more with the settings hang on

i did some brushing too

[hot yarrow]

weee

[shadow trench]

here have some mountains

[hot yarrow]

hehe i expected a little bigger mountain"ridges" somehow

[shadow trench]

I might need more patches

[hot yarrow]

comparfed to the rest of the terrain but it looks neat

if elevation was boosted a little it would probably make a super cool map for some canyon racer thingy

[shadow trench]

its pretty easy to scale elevation

but this was what it looked like on default settings

[hot yarrow]

uh

looks like there are some patches missing

[shadow trench]

this might be reasonable

[hot yarrow]

yeah

[shadow trench]

the shading still looks quite bad imo

[hot yarrow]

it feels like something is missing or cutting out inforation

the heightmap looks continuous

no spikes

[shadow trench]

i might be losing too much precision converting to fp16

[hot yarrow]

but in the last pic you can see some artifacts unless im just blind and moiree is fucking with my eyes 😄

[shadow trench]

though according to renderdoc its just 8 bit anyway so maybe not

[hot yarrow]

yeh then theh question is

why is the height data loaded as SRGB 🙂

oh you mean for albedo

[shadow trench]

no

idk why renderdoc loads it like that actually

i load as 16-bit uint grayscale i think

[hot yarrow]

its possible that the png has colorprofile information

[shadow trench]

then convert to f16 and normalize

16 bit yeh

fn from_dynamic_image(img: DynamicImage) -> ImageBuffer<Self::Pixel> {
    let img = img.into_luma16();
    let raw = img.into_raw();
    let as_fp = raw
        .into_par_iter()
        .map(|px| f16::from_f32(px as f32))
        .collect::<Vec<_>>();
    ImageBuffer::from_raw(as_fp)
}

hm I cant imagine that im losing precision if I go u16 -> f32 -> f16 tbh

[hot yarrow]

me neither

hmm what happens if you dont convert to f16?

[shadow trench]

so just use f32?

[hot yarrow]

yeah

worf a try

[shadow trench]

pub type HeightmapFormat = Grayscale<f32>; I should only need to swap this out if I wrote this properly :)

[hot yarrow]

and remove the map i guess after iter

[shadow trench]

thats implemented in the Grayscale<T> impl for TextureFormat

[hot yarrow]

https://tenor.com/view/make-it-so-jean-luc-picard-star-trek-the-next-generation-make-it-happen-gif-23455354

[shadow trench]

f32 doesnt seem to help too much no

might be something with how im rendering then

[hot yarrow]

yeah perhaps i found the deccer_heightmap which breaks your terrain engine XD

[shadow trench]

// Normalizes height values in the height map to [-1, 1] based on the most extreme value
fn normalize_height(_width: u32, _height: u32, data: &mut [LumaPixel<f16>]) -> Result<()> {
    trace!("Normalizing heightmap data");
    // Find the largest absolute value in the dataset, and take the absolute value of it.
    let extreme_val = data
        .par_iter()
        .max_by(|lhs, rhs| lhs.to_f32().abs().total_cmp(&rhs.to_f32().abs()))
        .unwrap();
    let extreme_val = f16::from_f32(extreme_val.to_f32().abs());
    let extreme_val_inverse = f16::ONE / extreme_val;
    // Now divide every height value by this extreme value
    data.par_iter_mut().for_each(|value| {
        **value *= extreme_val_inverse;
    });
    Ok(())
}

Maybe I shouldnt do this at all

[hot yarrow]

ye i think everything should be raw

and then you compose it with all your other layers

[hot yarrow]

ahaaaaaaaa

thats more like it

[shadow trench]

Yeh

I wonder why normalizing it breaks the map

[hot yarrow]

rust bug 😛

[shadow trench]

The problem is I dont think the brush works very well with non-normalized heights

Also most f16 precision is in [0, 1]

i will investigate

[livid geode]

precision errors when dividing most likely?

f16 has shit precision, perhaps you should do computation in f32 and then convert

[livid geode]

why are the normals fugged tho

[shadow trench]

yeh maybe thats a good idea

idk

[hot yarrow]

@wispy ore when you are online, give this heightmap a try too

[shadow trench]

i havent heard from him in a while

[shadow trench]

@hot yarrow your heightmap is serving a bigger purpose now

[hot yarrow]

i have no idea where i got this from

[shadow trench]

good height maps are hard to find tbh

[hot yarrow]

yeah

maybe future deccer got it from a program called Andromeda

[shadow trench]

if it can create that then ill be happy

[hot yarrow]

https://tenor.com/view/drums-cat-ba-dum-tss-gif-24587305

[shadow trench]

https://www.notapenguin.blog/posts/terrain-brush/ the cool people who follow my fred have a chance to yell at me with feedback before i post it everywhere

[livid geode]

will do

that is really neat @shadow trench

nice writeup

[shadow trench]

thanks :)

[hot yarrow]

heh

you could mention the normalization fookup with that heightmap

[shadow trench]

Mayhaps

but its not really related to the brush imo

[hot yarrow]

2nd article incoming

[shadow trench]

would be a short one

[hot yarrow]

but consistent

or was that a once off post blog thing only? 🙂

[shadow trench]

hopefully not :)

[hot yarrow]

[shadow trench]

@potent quest

we lost him

[potent quest]

im gone

suffering if you will

[shadow trench]

ive been there

i dont play much anymore luckily

[potent quest]

best decision made in your life

💀

[shadow trench]

💀

my wallet is also glad about it

[potent quest]

let me tell you about genshin impact and honkai star rail

[shadow trench]

oh no

[young oar]

Ohno

[potent quest]

Hour 2 of valorant: I have gone insane

(i got 3 aces :D)

[potent quest]

Hour 3 of valorant: maybe i don't need to understand ray tracing

[shadow trench]

[shadow trench]

i have so many sub crates by now

this is fun to have

publish_error!(bus, "Error loading asset: {error}"); with a shrimple macro to use

[young oar]

Ooo you have hot reloading?? Awesome

(I am going to steal your code)

[shadow trench]

for shaders yeah

its really great

[young oar]

Yea I'm about to implement that for my terrain voxel shader

It'll make prototyping so much faster

[shadow trench]

super useful when debugging

yeah exactly

[young oar]

Good thing is that it wouldn't be hard to implement it for render materials as well

Just rebuild le render pipeline and you're good

[shadow trench]

yep

[young oar]

Now the only thing I'm worried about is that File::open actually locks the file

Which would be not good since you wouldn't be able to write to it, unless you open it and close it multiple times (in code)

[shadow trench]

you only open it once you get a notification that it changed

theres a nice crate for that

[young oar]

Oh awesome

It shouldn't be too hard to implement it myself tho

[shadow trench]

yeah but wrapping around the os-specific apis is ehh

[young oar]

Can't you use something like fs::metadata? I thought that's what the crate you're talking about uses internally, just wraps around it

[shadow trench]

the crate im talking about sends you events when the file gets events

https://crates.io/crates/notify

it has some async watcher thingy that you just await for events

i spin up a thread and let it live there forever

[young oar]

Ohhhh I see I see

Yea that definitely does look like a nice wrapper

[shadow trench]

yeah its great

[shadow trench]

"Why is my blur shader not blurring"

my blur shader:

void main() {
  ...
  output /= accum;
}

what kind of crack was I on

ok i get that its an "equalize" brush but maybe this is a little too flat

heh this could totally be some road up to a castle on the mountain

[shadow trench]

today is a weird day

[shadow trench]

hmm

what do next

i guess its time to fix the rendering

its a bit ugly right now

[hot yarrow]

add texture layers

[shadow trench]

texturing definitely yeah

and some better shading

ill get the rendering code cleaned up a little first because its absolute hell right now

[hot yarrow]

and better starter heightmaps : >

or perhaps connect to some {Arc}GIS/WGS84/HTM/googlemap/ism (like those addons for blender)

[shadow trench]

maybe

but they usually have low resolutions

like 1 sample/100m

[hot yarrow]

no excuses

i was generally speaking

[shadow trench]

yeah

i had some format implemented but i yeeted it for now

[hot yarrow]

but textures is probably coolest feature next

to get some color in

[shadow trench]

because the parser library took 3 min to build

[hot yarrow]

oof

rewrite it in rust

[shadow trench]

i support textures but i dont have a good one for this mountain

[shadow trench]

im planning to some time when i reintroduce support yeah

[hot yarrow]

then it will take 2.59mins to build 😄

[shadow trench]

[hot yarrow]

hihi

sorry

[shadow trench]

compile times are fine ish, its the linker thats dreadful

[hot yarrow]

someone should rewrite it in rust

ok, ill stop

[shadow trench]

[livid geode]

as much as ppl shit on rust for its compiletimes ime its much faster than c++ compiletimes

obviously c# is a thousand times faster but that's not a fair comparison

[livid geode]

most of the compiletime is moved to the linker, the issue is that since clang/gcc linkers are really slow (gcc specially) you get the impression that rust's compiletimes are really bad

but if you use mold as your linker it's pretty decent even with half a hundred deps

and considering those deps will only need to be compiled once per target config, once you do it once you're good to go for a long while

the real issue is target folder sizes...

[shadow trench]

ubo_struct!(
    atmosphere,
    ifc,
    struct Atmosphere {
        radii_mie_albedo_g: Vec4,
        rayleigh: Vec4,
        mie: Vec4,
        ozone_sun: Vec4,
    }
);
atmosphere.radii_mie_albedo_g = Vec4::new(
    world.atmosphere.planet_radius,
    world.atmosphere.atmosphere_radius,
    world.atmosphere.mie_albedo,
    world.atmosphere.mie_g,
);

cursed macro time

i honestly dont think i should keep this

but it saves a bunch of typing

#[macro_export]
macro_rules! ubo_struct {
    ($var:ident, $ifc:ident, struct $name:ident {$($fname:ident:$ftype:ty$(,)*),*}) => {
        concat_idents::concat_idents!(buffer_name = $var, _, buffer {
            #[repr(C)]
            struct $name {
                $($fname:$ftype,)*
            }

            let mut buffer_name = $ifc.allocate_scratch_ubo(std::mem::size_of::<$name>() as vk::DeviceSize)?;
            let mut $var = buffer_name.mapped_slice::<$name>()?;
            let mut $var = $var.get_mut(0).unwrap();
        });
    };
}

this is the macro in question

maybe its not so bad

[hot yarrow]

yeah i am not seriously shitting on it, i have no say about anything rust related at all, beyond a hello world 5? years ago i have not touche... ah not true, i tried some winit+wgpu bs for like 10mins and gave up

[livid geode]

not now

[shadow trench]

[livid geode]

is the macro necessary?

[hot yarrow]

what does ifc stand for? if condition?

inb4 "ifcourse"

[shadow trench]

InFlightContext

[shadow trench]

no its not necessary, just saves typing out the same exact code to allocate the ubo

[livid geode]

no i mean

is it implementable without a macro

[shadow trench]

should be but a bit less easy to read imo

i was considering ```rs
ubo_struct!(
camera,
ifc,
struct Camera {
projection_view: Mat4 = state.projection_view,
position: Vec4 = state.cam_position,
}
);

but that might be a bit too obscure

[young oar]

Yea I also thought of doing a macro for UBOs and stuff like that

At the end I just sticked with defining structs cause macros scare me

(or at least making them scares me)

[shadow trench]

i think i could get the assign version to work

ill try and see if its not too cursed

you know its not that bad

#[macro_export]
macro_rules! ubo_struct_assign {
    (
        $var:ident,
        $ifc:ident,
        struct $name:ident {
            $(
                $fname:ident:$ftype:ty = $finit:expr,
            )*
        }) => {
        concat_idents::concat_idents!(buffer_name = $var, _, buffer {
            #[repr(C)]
            struct $name {
                $($fname:$ftype,)*
            }

            let mut buffer_name = $ifc.allocate_scratch_ubo(std::mem::size_of::<$name>() as vk::DeviceSize)?;
            let $var = buffer_name.mapped_slice::<$name>()?;
            let mut $var = $var.get_mut(0).unwrap();

            $(
                $var.$fname = $finit;
            )*
        });
    };
}

macros are kinda fun

[hot yarrow]

that is also quite readable

not so much the macro itself, but its what its 🙂

[shadow trench]

are macros readable in any lang

but yeah this is a lot nicer than the alternative

[nimble solar]

What the hell is $(...)*?

Is it regex? It looks like it lol

[shadow trench]

it expands the repeated pattern

uhh

in the pattern for the macro inputs you also have a $(X)*

Which is like a "match zero or more X" in a regex

Then in the expansion you can do a similar thing to say "for each of those matches, expand it to this"

[nimble solar]

Ahh I see

[shadow trench]

so in this case it just takes a list of name: Type = initializer and writes var.name = initializer;

[nimble solar]

So it's the epic thing that C macros don't have

[shadow trench]

Yeah

[nimble solar]

Recursive expansion

[shadow trench]

rust macros are super nice

[nimble solar]

😦

[young oar]

Damn I thought it would much much bigger than this

[shadow trench]

it is quite shrimple

[young oar]

If that's what's all needed to have a fancy macro like that then sign me up lol

[shadow trench]

feel free to copy pasta

you do need to give rustfmt a little hand with it but other than that it works great

[young oar]

Also you were super right about hot reloading

Shit is so nice to use

[shadow trench]

its one of the first things I added lmao

definitely worth it

[shadow trench]

git submodule add ... some/path
git submodule init

fatal: No url found for submodule path 'fsr2-sys/src/vendor/fsr2' in .gitmodules
huh?

am i stupid

what

it did clone properly

it works if i use path/to/foo/ instead of the autocompleted .\path\to\foo

i hate git submodules

ok step 1 of fsr2-rust (generating cmake) is done

next

being able to write rust for build scripts is such a nice feature

[nimble solar]

This isn't fair

You use Vulkan so you get to use FSR2 easily 😦

[shadow trench]

superior API

why

why does it insist on putting the binaries in its own source dir

[young oar]

There's a crate for fsr2?

[shadow trench]

im making one

[young oar]

Ohh yea makes sense

There's none on crates.io

Or anywhere else really lol

Unfortunately there's nothing like that for wgpu

I'm definitely gonna need to swap to vulkan later down the line

[shadow trench]

big progress

now i have to make the actual bindings

im a bit confused how the shader stuff works

ill ask jaker later

[shadow trench]

ok time to start writing bindings

huh?

lmao

[livid geode]

i can imagine that FFX_EOF is used for "ok, completed, encountered end of file" while FFX_ERROR_EOF is for an unexpected end of file

[shadow trench]

probably

looks ugly

anyway, phobos 0.8.0 with fsr2 support soon

[livid geode]

currently pogging

i may actually switch to phobos in rex tbh tbh

[shadow trench]

[livid geode]

library user jumpscare

[shadow trench]

imagine having users

[livid geode]

so true

[shadow trench]

i have a virtual user in danny but hes procrastinating

[livid geode]

lmao

having users takes a lot of your time but its a must to improve the library

[shadow trench]

definitely

[livid geode]

i'm learning lots by maintaining and updating rs-tiled

[shadow trench]

i see it on my github feed sometimes

looks like a lot of work goes into it

[livid geode]

for now its mostly just maintenance and answering user questions

i dont have time to actually update the crate

i was planning to do some work this thursday

[shadow trench]

there are releases with features every now and then no?

[livid geode]

yeah last one was not so long ago and had been planned for a long while

added vfs support

[shadow trench]

good stuff

[shadow trench]

these are some obscure surface formats lol

wtf i just noticed

the fsr2 error codes overflow

nice docs

[livid geode]

fsr2-rs on github: An enumeration of surface dimensions.

[shadow trench]

truly

i have to use fsr2-sys as the crate name since someone reserved fsr2 only to abandon it

oh god no

wchar_t

@livid geode is there a c_wchar type in rust

[livid geode]

isnt that a windows thing

[shadow trench]

no

wchars are 32 bit on lunix

16 on winblows

[livid geode]

https://docs.rs/wchar/latest/wchar/

i found this

[shadow trench]

The standard used to require wchar_t to be large enough to represent any supported character code point. However, such requirement cannot be fulfilled on Windows, and thus it is considered as a defect and removed. 

[livid geode]

wha

where is that from

[shadow trench]

cppreference

consistency

uses size_t for one count, u32 for the other

[hot yarrow]

@livid geode rex as in graphite's rex?

[livid geode]

didn't even know graphite had a project named rex

[hot yarrow]

ah

[livid geode]

rex is a to-be-renamed project of mine which i also posted here

[hot yarrow]

was before your time here

[livid geode]

what was it?

[hot yarrow]

an opengl engine 🙂

actually, not just opengl

[shadow trench]

oh i rember that

[hot yarrow]

dx11 and gl iirc

while i was working on xacor

also a d3d11/gl thing 😄

[short surge]

yeah I saw that, then they put in a stupid message acknowledging it themselves...

[shadow trench]

yep

i would be fine with it if it showed any sign of life or plans to continue but its been empty for 9 months so

im making decent progress on writing bindings i think

its not a whole lot to do manually

then again

that should be all the basic functions

now to do the backend

[shadow trench]

easy enough i guess?

its surprisingly not a lot

[shadow trench]

unsafe everywhere!

also wtf is happening on my pc

[shadow trench]

what

im setting -DCMAKE_BUILD_TYPE=Release why is it not building release

ah

for multi-config (vs) you need to pass the config in cmake --build

ok debug vs release has been fixed

ill figure out how to fix the vk related linker errors tomorrow

good night frogs

[shadow trench]

thank god I spotted this

casting vkGetInstanceProcAddr to vkGetDeviceProcAddr will surely go well

it seems like fsr2 assumes i link to vk to call vkEnumerateDeviceExtensionProperties and other functions not related to the device

i guess ill modify it to pass in those function pointer s instead

kek

of course they use both vkGetPhysicalDeviceProperties and vkGetPhysicalDeviceProperties2

because consistency

not me writing a mini vk loader ish thing so i dont need to depend on ash for the fsr2 bindings

[shadow trench]

surely this will work

[shadow trench]

i despise people who build their examples by default in cmake

or dont even provide a flag to turn it off

at least it compiles

[shadow trench]

so uh, I can create the context fine with validation off

[livid geode]

set(SHITLIB_BUILD_TESTS false)
set(SHITLIB_RUN_TESTS false)
set(SHITLIB_TEST_TESTS false)
set(SHITLIB_BUILD_EXAMPLES false)
set(SHITLIB_USE_STDLIB true)
set(SHITLIB_USE_BOOST false)
set(SHITLIB_LIB_TYPE STATIC)```

[shadow trench]

yes

This is such a cursed workaround lmfao

but it works

[livid geode]

ub

[shadow trench]

idk if its ub or just validation being shit

it does throw some errors about features not being enabled so maybe ill fix those and see if i can remove the thread after that

apparently their shaders use int16 but its not documented that i have to enable that feature

oh yeah that fixed it lol

mfw validation would rather cause a stack overflow than tell me im missing a feature

[shadow trench]

currently stripping down the fsr2 library to its bare essentials to fit in the crates.io upload limit

[shadow trench]

ok i had to completely delete dx12 support because they ship an 18mb binary with dxc

but https://crates.io/crates/fsr2-sys

[shadow trench]

also, phobos is at >400 commits now

[potent quest]

💀💀💀

[shadow trench]

in other news

fsr2 in phobos very soon

you better start learning

[shadow trench]

i think i have the basic infrastructure done so I guess its time to start writing a sample

mostly to see if it works

[livid geode]

kinda incredible that these bindings did not exist already

thank you for your work soldier 🫡

[shadow trench]

o7

yeah i was surprised too tbh

[livid geode]

you should post it in the gamedev rust discord

[shadow trench]

sure, why not yeah

[livid geode]

https://discord.gg/game-development-in-rust-676678179678715904

its bound to get a shitton of stars

post it as a thread on the crates channel

[shadow trench]

will do

[shadow trench]

oh no

some of the fields are getting swapped around

#[derive(Copy, Clone, Debug)]
#[repr(C)]
pub struct FfxResourceDescription {
    pub ty: FfxResourceType,
    pub surface_format: FfxSurfaceFormat,
    pub width: u32,
    pub height: u32,
    pub depth: u32,
    pub mip_count: u32,
    pub flags: FfxResourceFlags,
}

#[derive(Copy, Clone, Debug)]
#[repr(C)]
pub struct FfxResource {
    pub resource: *mut c_void,
    pub name: [wchar_t; 64],
    pub description: FfxResourceDescription,
    pub state: FfxResourceState,
    pub is_depth: bool,
    pub descriptor_data: u64,
}

/// A structure describing a resource.
typedef struct FfxResourceDescription {

    FfxResourceType                 type;                                   ///< The type of the resource.
    FfxSurfaceFormat                format;                                 ///< The surface format.
    uint32_t                        width;                                  ///< The width of the resource.
    uint32_t                        height;                                 ///< The height of the resource.
    uint32_t                        depth;                                  ///< The depth of the resource.
    uint32_t                        mipCount;                               ///< Number of mips (or 0 for full mipchain).
    FfxResourceFlags                flags;                                  ///< A set of <c><i>FfxResourceFlags</i></c> flags.
} FfxResourceDescription;

/// An outward facing structure containing a resource
typedef struct FfxResource {
    void*                           resource;                               ///< pointer to the resource.
    wchar_t                         name[64];
    FfxResourceDescription          description;
    FfxResourceStates               state;
    bool                            isDepth;
    uint64_t                        descriptorData;
} FfxResource;

compare time

[shadow trench]

i got trolled by the formatting lmao

[wise viper]

that is very evil formating

[shadow trench]

i blame amd programmers

[wise viper]

smh jaker not having his company in order

[shadow trench]

unbelievable 🙄

[nimble solar]

Clearly Jaker is to blame for all AMD problems as he is the CEO, CTO, Chief Programmer, Chief Designer, Chief Marketing Manager, Chief Relationships Manager, and whatever other title there is

[shadow trench]

exactly

[polar otter]

Project looking really good Penguin

any future plans for water / grass rendering, or purely focusing on the terrain part for a while?

[shadow trench]

absolutely! pretty soon probably :)

[shadow trench]

real fsr2

no validation errors though

[hot yarrow]

perfectly upscaled black

[shadow trench]

zero artifacts

[shadow trench]

ok i think all i need to do to get it working now is jitter and motion vectors

test scene is ready

[shadow trench]

that looks antialiased to me :)

[livid geode]

Yayyyy

In phobos too?

[shadow trench]

this is all in phobos yea

what do you mean I can use it below but not clone it and pass it along

ah i get it

meh

There we go

[hot yarrow]

hard to tell

you need an actual noisy scene

[shadow trench]

yeah true

but this is good to test if its not crashing completely

and it seems to at least not be aliased

[nimble solar]

You seem to be having the same motion problem, hmm 🤔

[shadow trench]

i calculated motion vectors by doing the shrimplest thing i could find

FragMotion = PrevClipPos.xy / PrevClipPos.w - ClipPos.xy / ClipPos.w;

[nimble solar]

How do you do them

[shadow trench]

and then set motion vector scale to render w/h

[nimble solar]

I read frogmotion

[shadow trench]

[nimble solar]

The scale should be set to w / 2, h / 2

[shadow trench]

wtf

[nimble solar]

Because math @distant radish did

Oh shit, sorry

[shadow trench]

why does the readme say otherwise lmao

[nimble solar]

I have no idea

[shadow trench]

i copied this

worf a try

[nimble solar]

#1019779751600205955 message

This be the math

[shadow trench]

that makes sense

That seems alright

might be better yeah

[livid geode]

this mf really think we can see the difference from a gif

[shadow trench]

gifs do be trash

but im too lazy to open obs every time

[shadow trench]

im so stupid i figured out why its stack overflowing

#[derive(Copy, Clone, Debug)]
#[repr(C)]
pub struct FfxFsr2Context {
    pub data: [u32; FFX_FSR2_CONTEXT_SIZE],
}

i tried to keep this on the stack

i made it upscale with the Quality preset now and its definitely not perfect

maybe its better with infinite reverse depth as they said

wtf i forgot to enable depth write lmao

much better

[shadow trench]

@potent quest https://github.com/NotAPenguin0/phobos-rs/releases/tag/v0.8.0

FSR2 has been added

[nimble solar]

How does reverseZ make such a big difference wtf

[shadow trench]

Well a big part of that difference you see is because in the first screenshot I had depth writes off entirely

[coarse yoke]

what are you recording with

sharex can do 60fps mp4s :)

question i actually came here for, do you store a copy of the terrain on the CPU? (i'm guessing not since you'd have to manage, but i guess i'm curious if you've considered whether you'll need to)

other questions, if i understand correctly, you store a normal map of the terrain? why?

obviously normals can add detail, but it doesn't feel realistic to me for a terrain editor to actually utilize it unless you're just exporting hi-res from houdini, at which point it's a bit

or maybe blending from brushes? but that too also feels messy

you also said you use deferred decal rendering to render the brush

why implement that over virtual texturing for the terrain? i.e. why not do it as a component of the terrain rendering, rather than in compositing? i would expect this to be a lot more powerful, given that with other stuff in the future, you can modify heights with it too

these may seem like suslik/nano/jasper/etc.. .. . . style questions where i'm challenging decisions, they are not, i am not knowledgable enough for these to not be curiosities

[wispy ore]

VTM would be a better fit for non-dynamic decals. Ofc. it'd make sense for static decals and textures.

Any decal that is movong just increases bandwidth used there proportional to decal count

Say you're doing stochastic texture mapping etc.
You need to redo everything for the area then reapply the decal at the new pos

[coarse yoke]

so do you cache the sample in chunked textures with vtm then? i haven't implemented it

[wispy ore]

You cache it there yeah

[coarse yoke]

gotcha, i understood vtm as just taking the expensive sample every time when rendering the terrain

[shadow trench]

Sounds interesting, I can’t say I know enough to answer all those questions properly but it’s definitely worth looking at

[nimble solar]

I was about to destory everything just to implement reverse Z

[shadow trench]

The normals are mostly for adding detail while rendering, without it looks really quite flat which just is kinda ugly

[shadow trench]

Its pretty easy tho

[nimble solar]

I don't understand the math behind extracting planes from an infinite reverse projection 😦

[shadow trench]

I briefly thought about that for the mouse picking but figured it’s not worth the extra hassle just to eliminate the one frame delay

[shadow trench]

anyway, time to integrate fsr2 in andromeda now

[shadow trench]

uhhhh

that's not right

its not just me right its shaking

why is it shaking

[shadow trench]

wat

how is it all zero now

it wasnt 10 minutes ago

[potent quest]

LETS GOOO
Wouldn’t this made the crate size really big since you’re including all of FSR in it?

[shadow trench]

fsr2-sys (the bindings lib) is a separate crate

but i made a fork of FSR2 thats stripped down to just the source and stuff needed for the vk backend

Because yes initially it didnt fit

now fsr2-sys is like 5 mb or so

[potent quest]

OHHH

[shadow trench]

I also changed some stuff in the fork but its quite minimal

the biggest thing is I changed the vk backend api to take in function pointers stuff like vkEnumerateDeviceExtensionProperties

because it was calling those directly

[potent quest]

Never understood function pointers but sounds good

[shadow trench]

basically what it means is the bindings dont depend on vulkan being linked at all so it can use any loader potentially

[potent quest]

Ohhh so you can do stuff like test cases

[shadow trench]

possibly i suppose

but also i dont want to link vulkan directly

that would break with using the dynamic loader in ash for example

[potent quest]

Ohh

Wait there are multiple loaders for Vulkan?

[shadow trench]

even in c++ you can link statically or use volk

[livid geode]

give the flying cameraman a break

[shadow trench]

right, of course

sorry sir

[potent quest]

hire a new one

clearly they arent fit enough for the job

[shadow trench]

when are you learning rust

[hot yarrow]

🇸🇴 🔛 🇹🇲

[potent quest]

wish i had time 💀 im coping so hard for summer. only up to crates rn in the rust book

[shadow trench]

[nimble solar]

Give me incentive

[shadow trench]

you can use vulkan easily with phobos ™️

also

cool lang

[nimble solar]

What are you talking about

This is NIH land isn't it

[shadow trench]

naturally

[nimble solar]

I make my own bread

[shadow trench]

its like fwog but for vk

and in rust

it does a little more but thats the gist of it

i have this overview thingy of features

[nimble solar]

Looks cool, I'll use it as reference for my own thingy

Virtual resources are particularly attractive

[shadow trench]

admittedly it doesnt do any aliasing yet

but one day

some initial ideas are heavily inspired by vuk (martty's thing for c++)

but its nothing like current vuk

[shadow trench]

no vuk does a lot more with fancy rendergraph combination thingies

[potent quest]

i am proud to present bad news

i got deferred from University of Toronto CS to their math department

[shrewd lynx]

Sadge

[potent quest]

maybe living in a cardboard box isn't that bad

[shadow trench]

i dont know what that means

[potent quest]

the cardboard box?

or university

[shadow trench]

the university thing

[potent quest]

UoFT (university of toronto) is one of the most prestigous universities in Canada

if not the

but I got deferred from computer science (rejected) and instead got an offer to their maths

[shadow trench]

ooh I see

Yeah piece of advice dont go study maths

Just dont

[potent quest]

LMAOOO

is it that bad

i'm coping so hard rn

[shadow trench]

i take a minor in maths and its horrible

i mean

i quite like math but i didnt sign up for not getting my degree ever

[potent quest]

💀
The second option is a math offer from another uni

[shadow trench]

damn you really dont get to choose there

[potent quest]

it's math but i pay 20K or math, but no one knows about my university

[shadow trench]

💀

[potent quest]

the other option is CS, but no one knows about this uni and half of my high school is going there 💀

[shadow trench]

yeah that sound even worse

[potent quest]

(i do not want to see my high school grade again)

yeah but from what i heard no one there tries so its easy to get 4.0 gpas 💀

[hot yarrow]

i dont understand what that means that nobody knows this uni?

[potent quest]

it's known within my province well

actually country, but if i were to fly out, it wouldn't really be known but tbf

trying to get a job outside of my home country is

[hot yarrow]

is that a thing over there?

that you are only getting a job if you go to a place ppl know?

rather than what you can/have researched/have mastered in

[potent quest]

from what i heard from my techers

you use the co-op programs to help you get jobs?

[shadow trench]

that’s not usually how we eu frogs do it

But you can also generally attend (almost) any uni you like so it’s not as important

[potent quest]

huh

yeah co-op is usually the route most people go down

people can/do apply using degrees right out of grad

but co-op has become really big at least for se

as it's more or less how you connect to and from potential employers

[shadow trench]

Hmmm

[hot yarrow]

never heard of that before

you either studied an interesting topic or not

and then people hire you according to what you can do/sold yourself best with

not which university you went to

[shadow trench]

Yeah

[hot yarrow]

but that could be a cultural thing... ive seen that in south korea/india as well

us of freedom perhaps too to some degree

but anyway, im not judging, just wondering 🙂

[potent quest]

tbf i am being completely over dramatic. both unis are perfectly fine and in truth, i'm aiming way too high for my caliber trying to get co-op at the top unis

💀

but i know for sure that co-op has been the primary route that my teachers at least

have been touting as the "this will land you a job"

however i am not being over dramatc in saying half of my grade is going there 💀 I don't wana see them again

[hot yarrow]

just make sure you learn/study/research on something interesting

and when you get the chance to travel/see the world no matter how short it is... 3 weeks, 6 months, 2 days... do it

exchange programs or whatever

[potent quest]

I really wanna do an exchange program since I haven't travelled since I was a child

Wanna visit Europe one day fr

[hot yarrow]

if you can other cultures too, not just the cheese eaters, wine drinkers and beer brewers 🙂

[potent quest]

lmaooooo ill try

[shadow trench]

im making some drastic changes to how in-frame resources are allocated and passed around

going to make it more streamlined with out of frame stuff

should also lead to major api simplifications

[shadow trench]

i was fully prepared for this to come crashing down burning but after fixing all the compiler errors it just works

epic

fences and scratch allocators are now globally pooled

as an added bonus you now no longer need to manually keep track of pipeline/descriptor caches or pass them to cmd buffers

[shadow trench]

frame.new_frame(..., || {
  // global_pool is a ResourcePool, you only really have one across the entire program
  let pool = LocalPool::new(global_pool)?;

  pool.allocate_scratch_ubo(...) // memory for this frame only
});

The nice thing about this is that it doesnt matter if this is inside a frame or not

[nimble solar]

How do you delete it?

[shadow trench]

i was just figuring that out

[nimble solar]

Rather, do you keep track of the GPU using the resource or do you just kill it after some fixed time has passed?

[shadow trench]

im thinking the cleanest way is to make SubmitBatch::submit_for_present() take ownership of the local pool

[shadow trench]

kill it after the fence completes

well

release back to the pool

[nimble solar]

Hmm

So for each queue submission, you keep track of the resources it uses and assign a fence to each?

[shadow trench]

not really

each queue submission has a single fence associated with it

But fences are a bit beefed up

They store a chain of cleanup functions called after wait() completes

I can move the local pool into such a cleanup function so it is kept alive until wait() is done