Saurian Sorcery | Graphics Programming | Page 2

main cosmos Nov 24, 2022, 10:37 PM

#

What is that ImGui thing

obsidian wolf Nov 24, 2022, 10:38 PM

#

seems like a layer on top of the imgui input systems for testing

urban lantern Nov 24, 2022, 10:38 PM

#

International multi golden user identifier

obsidian wolf Nov 24, 2022, 10:55 PM

#

ImGuiTest* test = IM_REGISTER_TEST(e, "demo_test", "test1");
test->TestFunc = [](ImGuiTestContext* ctx)
{
    ctx->SetRef("My Window");           // Set a base path so we don't have to specify full path afterwards
    ctx->ItemClick("My Button");        // Click "My Button" inside "My Window"
    ctx->ItemCheck("Node/Checkbox");    // Open "Node", find "Checkbox", ensure it is checked if not checked already.
    ctx->ItemInputValue("Slider", 123); // Find "Slider" and set the value to 123
    IM_CHECK_EQ(app->SliderValue, 123); // Check value on app side
};```

#

what i'd test with this would be like things are damaging as expected, wave spawns are correct,

#

i'm not used to testing but that's because i haven't really had a good way to do it

urban lantern Nov 24, 2022, 10:59 PM

#

usually you test that underneath

#

the spawner logic itself should work correctly already

#

"unit tests"

obsidian wolf Nov 24, 2022, 10:59 PM

#

it makes sense to me to use the same form of input as in normal use

urban lantern Nov 24, 2022, 10:59 PM

#

when the ui is involved you get "end to end" or "ui tests"

#

the input for unit tests is usually described in the tests themselves

#

but this imguitest can be used for that as well

obsidian wolf Nov 24, 2022, 11:22 PM

#

redid this

#

compared to

#

more nightmare fuel

urban lantern Nov 24, 2022, 11:38 PM

#

cute little croco

obsidian wolf Nov 25, 2022, 3:51 AM

#

i now need to figure out what putting IK + rigs into engine looks like

main cosmos Nov 25, 2022, 4:02 AM

#

menacing walk

obsidian wolf Nov 25, 2022, 4:03 AM

#

yeah it would help if i did upper body but i dont think i'm importing IK so no point taking the time

#

jaker you haven't done animation stuff like this, right monkaHmm

main cosmos Nov 25, 2022, 4:08 AM

#

nope

#

I skipped like 40 pages of the animation chapter in the book I'm reading nervous

obsidian wolf Nov 25, 2022, 4:09 AM

#

🙂

#

tbh like

#

there's nothing complex about it

main cosmos Nov 25, 2022, 4:09 AM

#

it went into more detail than I would have been able to absorb

obsidian wolf Nov 25, 2022, 4:09 AM

#

it's just some tedious math

#

i mean maybe? but like?

#

i would think you can get very far with just a rig, weights, IK, and some keyframes

#

far enough for any indie

main cosmos Nov 25, 2022, 4:10 AM

#

and it talked about the minutiae of animation systems in detail like blending animations and compact storage, etc.

#

(game engine architecture 3rd ed.)

obsidian wolf Nov 25, 2022, 4:10 AM

#

i see, yeah, i don't want to store animations

#

i can't imagine it'd be a good idea for me

main cosmos Nov 25, 2022, 4:11 AM

#

I mean storing in memory

obsidian wolf Nov 25, 2022, 4:11 AM

#

huh?

#

yeah like storing the vertex data, no?

main cosmos Nov 25, 2022, 4:11 AM

#

how are you gonna animate stuff?

obsidian wolf Nov 25, 2022, 4:11 AM

#

for baked animations

#

just IK

#

vertex shader takes in a couple matrices, interpolates them based on bone weights

#

then it's just a matter of controlling the IK control points

main cosmos Nov 25, 2022, 4:12 AM

#

what I'm referring to is compressing bones and joints and vertices and stuff

obsidian wolf Nov 25, 2022, 4:13 AM

#

don't some games store actual vertex data with baked animations

main cosmos Nov 25, 2022, 4:13 AM

#

and using bezier curves for animations (I skipped that part so I don't know exactly what it was talking about)

obsidian wolf Nov 25, 2022, 4:13 AM

#

like that one insomniac batman game

main cosmos Nov 25, 2022, 4:13 AM

#

obsidian wolf don't some games store actual vertex data with baked animations

that is one of the 3 major options

obsidian wolf Nov 25, 2022, 4:13 AM

#

that shit i am absolutely not doing

main cosmos Nov 25, 2022, 4:13 AM

#

that way is probably the most powerful, but by far the most memory-intensive

obsidian wolf Nov 25, 2022, 4:14 AM

#

its also a lot more work

main cosmos Nov 25, 2022, 4:14 AM

#

morph targets are probably better

#

then there are bones 🦴

obsidian wolf Nov 25, 2022, 4:14 AM

#

i'd probably set up morph targets with IK lmao

#

boneless morph targets?

#

b💀neless

#

i'd think it's pose interpolation, IK, and baked animations

main cosmos Nov 25, 2022, 4:18 AM

#

If I implemented animations, I'd probably stick to skeletal animation since it seems flexible and there are plenty of tutorials about it

obsidian wolf Nov 25, 2022, 4:19 AM

#

do you mean pose interpolation? idk terminology so i'm just saying what makes sense to me

main cosmos Nov 25, 2022, 4:20 AM

#

It's the one where you animate joints and each vertex uses a weighted sum of its nearest joints as its local orientation

obsidian wolf Nov 25, 2022, 4:22 AM

#

animate how

#

you can use IK or you can interpolate between poses

#

im for sure doing one of those already

main cosmos Nov 25, 2022, 4:35 AM

#

yes you animate one of those ways

obsidian wolf Nov 25, 2022, 4:35 AM

#

think it's worth doing both? monkaHmm

main cosmos Nov 25, 2022, 4:35 AM

#

idk

obsidian wolf Nov 25, 2022, 4:36 AM

#

yeah

main cosmos Nov 25, 2022, 4:36 AM

#

for a tower defense game you probably don't need IK

obsidian wolf Nov 25, 2022, 4:36 AM

#

i'd think IK is easier

#

less factors to worry about, poses feel so artistic to me

main cosmos Nov 25, 2022, 4:38 AM

#

but how do you get the poses in the first place

obsidian wolf Nov 25, 2022, 4:38 AM

#

oh i guess there's always some mix

#

IK for limbs but the rest probably has to be posed?

#

unless i misunderstand what you mean

main cosmos Nov 25, 2022, 4:41 AM

#

I'm asking you

#

don't you need a physics system for IK to work

obsidian wolf Nov 25, 2022, 4:42 AM

#

hmm, i don't think so, but maybe

#

monkaHmm ahh yes, #mathematics fuel

#

https://www.ryanjuckett.com/cyclic-coordinate-descent-in-2d/
i think it's possible to just search for the solution essentially?

RyanJuckett.com

Cyclic Coordinate Descent in 2D

When performing inverse kinematics (IK) on a complicated bone chain, it can become too complex for an analytical solution. Cyclic Coordinate Descent (CCD) is an alternative that is both easy to implement and efficient to process. We will discuss the algorithm works at a high level, followed by a two-dimensional example and sample code.CCD solves...

main cosmos Nov 25, 2022, 4:45 AM

#

but what is the solution?

obsidian wolf Nov 25, 2022, 4:46 AM

#

The goal of Inverse kinematic is simple, given a target point in space T (blue cube) find the best rotation for each joint so that the tip of the last joint reaches the target.
any where the endpoint is the target

#

i guess if you're talking about like

#

energy minimization then you need physics joints

#

i guess that's the hard part of IK

#

with poses you kind of define the joint restraints through the poses that you make

#

and when you test it, if they break it, you change the poses or add another

#

but with IK, each possibly needs to be explicitly defined

main cosmos Nov 25, 2022, 4:51 AM

#

looks like I had no clue what IK was previously

obsidian wolf Nov 25, 2022, 4:51 AM

#

obsidian wolf Nov 25, 2022, 6:16 AM

#

hmmmm

#

main cosmos Nov 25, 2022, 6:16 AM

#

albinism

obsidian wolf Nov 25, 2022, 6:28 AM

#

i was told to put a hat on it

main cosmos Nov 25, 2022, 6:29 AM

#

good suggestion, whoever said that

obsidian wolf Nov 25, 2022, 6:50 AM

#

i just tried to post a picture of it posed

#

alright there was a better pose but it was downvoted by clyde

severe spruce Nov 25, 2022, 7:01 AM

#

explicit froges 😳

obsidian wolf Nov 25, 2022, 7:02 AM

#

:(

urban lantern Nov 25, 2022, 10:05 AM

#

lol the frog is neat too

obsidian wolf Nov 26, 2022, 2:30 AM

#

hmmm

#

my biggest implementation question atm is whether or not its worth having skeletons be optional for vertex formats

#

main downside is i believe id have to write and manage two shaders

urban lantern Nov 26, 2022, 3:08 AM

#

just have two

#

or use some HAS_BONES

obsidian wolf Nov 26, 2022, 3:09 AM

#

but for what monkaHmm

#

i think its bikeshed until there is some tangible reason to fix it tbh, pretty easy to tack on after

urban lantern Nov 26, 2022, 3:12 AM

#

its shrimple

#

what do you mean for what

#

one vertex shader with bone weights, the other without, or just one with #ifdefs

obsidian wolf Nov 26, 2022, 3:13 AM

#

i can just have every draw call pass empty bone info for vertices ;)

urban lantern Nov 26, 2022, 3:15 AM

#

are you worried about "performance" ?

obsidian wolf Nov 26, 2022, 3:18 AM

#

nah, i forgor that i wasn't tho 💀

urban lantern Nov 26, 2022, 3:21 AM

#

: )

obsidian wolf Nov 30, 2022, 6:04 AM

#

in highschool we called joint accessors "drug dealers"

obsidian wolf Dec 2, 2022, 6:35 AM

#

beard dragon, gonna see what posing him does to make him not just stand, or maybe i'll have some standing some not

urban lantern Dec 2, 2022, 11:46 AM

#

cute 🙂

obsidian wolf Dec 3, 2022, 3:06 AM

#

i have been hampered by my lack of practice with modern c++ move semantics

#

i have a ModelGPU, a SkeletonGPU, and a Renderable

#

the relevant parts of each

    string mesh_asset_path;
    string material_asset_path;

    SkeletonGPU* skeleton = nullptr;
};```
```struct ModelGPU {
    vector<Renderable*> renderables;
    vector<SkeletonGPU> skeletons;
};```

struct SkeletonGPU {
vector<id_ptr<Bone>> bones;
vuk::Uniquevuk::Buffer buffer = vuk::Uniquevuk::Buffer();
};```

#

the part where i am confused by said move semantics, i want ModelGPU to own SkeletonGPU, so when ModelGPU is destructed, the skeletons will be too, and the buffer needs to be owned by the SkeletonGPU,

#

i think the c ptr on the renderable works with this? (if i make the vector vector<unique_ptr<SkeletonGPU>>) but then if this all makes sense with vectors and such is very monkaHmm

#

i sort of wonder if these move semantics are partially meant to be dealt with by deleting constructors and letting the compiler yell at you when you've done something dumb

main cosmos Dec 3, 2022, 3:29 AM

#

as you currently have it, the SkeletonGPUs will be destroyed when ModelGPU is

#

I'm not sure the relation to move semantics here

obsidian wolf Dec 3, 2022, 3:30 AM

#

ok so i can't copy ModelGPUs

#

because if it contains any skeletons, those need to be moved, since vuk::Unique needs to be moved

#

but i have no clue why stuff like model_comp.model_gpu = instance_model(scene->render_scene, model_comp.model_cpu); doesn't compile (instance_model returns ModelGPU) with
Error C2280 : 'spellbook::ModelGPU &spellbook::ModelGPU::operator =(const spellbook::ModelGPU &)': attempting to reference a deleted function

main cosmos Dec 3, 2022, 3:32 AM

#

which line is that

#

ah nvm

obsidian wolf Dec 3, 2022, 3:32 AM

#

AFAIK shouldn't this use move assignment, not copy?

#

since return should be an rvalue

main cosmos Dec 3, 2022, 3:32 AM

#

show the signature of it

obsidian wolf Dec 3, 2022, 3:33 AM

#

ModelGPU instance_model(RenderScene& render_scene, const ModelCPU& model, bool frame = false);

main cosmos Dec 3, 2022, 3:34 AM

#

I'm guessing throwing a std::move on it does nothing

#

maybe your move assignment operator got deleted somehow

#

try defaulting it

obsidian wolf Dec 3, 2022, 3:34 AM

#

ModelGPU(ModelGPU&& other) { renderables = std::move(other.renderables); skeletons = std::move(other.skeletons); }
i declared it trying to fix it, maybe i need an = move too?...

main cosmos Dec 3, 2022, 3:35 AM

#

just put
ModelGPU& operator=(ModelGPU&&) = default;

#

your linter may ask you to mark it noexcept

obsidian wolf Dec 3, 2022, 3:36 AM

#

fuckk the linter

main cosmos Dec 3, 2022, 3:36 AM

#

frfr

obsidian wolf Dec 3, 2022, 3:36 AM

#

i got so many linter complaints

main cosmos Dec 3, 2022, 3:36 AM

#

sign of a bold programmer

obsidian wolf Dec 3, 2022, 3:36 AM

#

ok i guess this worked?

main cosmos Dec 3, 2022, 3:37 AM

#

I don't think the move assignment operator is implicitly defined

obsidian wolf Dec 3, 2022, 3:37 AM

#

i would've assumed that declaring the move constructor also makes an implicit assignment move 🤔

main cosmos Dec 3, 2022, 3:37 AM

#

#

you trolled yourself by writing a move constructor

obsidian wolf Dec 3, 2022, 3:38 AM

#

🪞

#

now riddle me why tf MSVC doesn't give line numbers with shit liek this >:(

main cosmos Dec 3, 2022, 3:41 AM

#

it does for me

#

is there anything else when you scroll down?

obsidian wolf Dec 3, 2022, 3:42 AM

#

it usually does for me

main cosmos Dec 3, 2022, 3:42 AM

#

I can also double click the message and it will take me to the error

obsidian wolf Dec 3, 2022, 3:42 AM

#

0>C:\spellbook\src\renderer\assets\model.hpp(49): is the only line number i get, and that's just linking to vector<std::unique_ptr<SkeletonGPU>> skeletons;

#

so however i'm adding to that isn't correct but

#

model_gpu.skeletons.push_back(std::make_unique<SkeletonGPU>(game.renderer.upload_skeleton(*skeleton_cpu)));
is incorrect for some reason :)

#

emplace back

#

main cosmos Dec 3, 2022, 3:44 AM

#

does it work

obsidian wolf Dec 3, 2022, 3:45 AM

#

https://tenor.com/view/joker60s-joker-gif-25727518

Tenor

#

yes

#

now i just gotta link the actual bone transform UI up and we can see if it all works!

#

oh wait no i linked it up already

#

PepeWTF

#

actually it's just bad initial values

#

i also badly need a transform widget

main cosmos Dec 3, 2022, 4:01 AM

#

epic

obsidian wolf Dec 3, 2022, 5:14 AM

#

#the-bug-collective message
guessing this has to do with y-up vs z-up, which is incredibly not-epic

#

uhhh...

#

also imma prob gonna be spammin this chat cause these skeleton bugs are funny

main cosmos Dec 3, 2022, 5:26 AM

#

giraffics

obsidian wolf Dec 3, 2022, 5:40 AM

#

giraffes got a fat ass

obsidian wolf Dec 3, 2022, 6:31 AM

#

i feel the need to 🙂 that statement

main cosmos Dec 3, 2022, 6:32 AM

#

🙂

obsidian wolf Dec 3, 2022, 7:03 AM

#

it reminds me of that one thing from that one movie

#

https://tenor.com/view/eyes-hands-creepy-gif-12040142

Tenor

main cosmos Dec 3, 2022, 7:04 AM

#

ah, that one thing

obsidian wolf Dec 3, 2022, 7:09 AM

#

the fuck

#

why is my road kill looking at me like that

main cosmos Dec 3, 2022, 7:14 AM

#

when they hit you with that dollar store roadkill

obsidian wolf Dec 3, 2022, 7:14 AM

#

i think maybe i should go back to the cylinder...

#

[main 5c66d1b] Half-working skinning 47 files changed, 1180 insertions(+), 608 deletions(-)
fairly large commit tho :)

#

skeletons were only like 300 LOC at theend of it

urban lantern Dec 3, 2022, 12:36 PM

#

hmm could be that you use Rider with c++

#

Rider is primarily a c# ide

obsidian wolf Dec 3, 2022, 6:50 PM

#

nah it just pushes out msvc errors

#

it doesn't reformat them or anything

#

well, it does parse them to make hyperlinks and such, but actual message contents should be the same

#

kinda crazy how far reading the actual spec gets you

urban lantern Dec 3, 2022, 6:57 PM

#

i like the dino croco frog dile

obsidian wolf Dec 3, 2022, 11:22 PM

#

i gotta actually figure out how to make this usable monkaHmm

#

i guess i want 3 components

IK
pose interpolation
physics driven
but there's a weird question where.. do i want to interpolate between these?

#

maybe i don't need #3

#

i can imagine tracking at least the first two for all bones, and having some mix between then

#

i have a very large lack of knowledge on how this is handled by other indie games 🙂 i haven't really ever used animation tools in engines

obsidian wolf Dec 4, 2022, 1:00 AM

#

#general chat has updated this to be a relatively complete plan of do keyframes/pose interpolation, then layer the other two on after, and that makes sense to me

obsidian wolf Dec 4, 2022, 1:25 AM

#

#

hmmmm....

#

these are actually super useful 🙂

obsidian wolf Dec 4, 2022, 8:21 PM

#

i have tortured myself by moving from one subject i hardly understand to another :)

#

doing cmake now :)))

obsidian wolf Dec 4, 2022, 10:37 PM

#

man

#

i hate this shit

#

it's so fucking terrible

#

there's no information anywhere that you can actually find

#

every single google result is some unit-A on a forum where there's pages of non information any time there's a question

#

there is so much fucking state in regards to all of this

#

i'm fairly certain that just opening the cmake-gui has invalidated 95% of relevancy of resources online

urban lantern Dec 4, 2022, 10:41 PM

#

https://github.com/deccer/CMake-Glfw-OpenGL-Template

obsidian wolf Dec 4, 2022, 10:41 PM

#

because it's added or removed cache entries

#

doesn't work

#

my current problem is not having the cxx/c compiler paths

#

but of the four ways i've tried to make it work none of them have

urban lantern Dec 4, 2022, 10:42 PM

#

?

#

what doesnt work

obsidian wolf Dec 4, 2022, 10:42 PM

#

where do these entries even come from

urban lantern Dec 4, 2022, 10:42 PM

#

and what does that cxx/c compiler paths thing mean

obsidian wolf Dec 4, 2022, 10:42 PM

#

are they just what's existing in my cache?

urban lantern Dec 4, 2022, 10:42 PM

#

idk, i never used the gui

obsidian wolf Dec 4, 2022, 10:42 PM

#

is there a way i can view what's in my cache?

urban lantern Dec 4, 2022, 10:43 PM

#

clion or vscode+cmake plugin

obsidian wolf Dec 4, 2022, 10:43 PM

#

C:\spellbook\build>cmake .. -- The C compiler identification is unknown -- The CXX compiler identification is unknown CMake Error at CMakeLists.txt:3 (project): No CMAKE_C_COMPILER could be found.

urban lantern Dec 4, 2022, 10:43 PM

#

looks like your environment is fucked

#

what happens if you run vcvarsall and from that prompt run the cmake gui

#

or from the vs developer prompt, run cmake gui

obsidian wolf Dec 4, 2022, 10:45 PM

#

i don't want to use the vs developer prompt

urban lantern Dec 4, 2022, 10:45 PM

#

do it

#

to see if that works

obsidian wolf Dec 4, 2022, 10:45 PM

#

i'd rather use the same terminal for everything

urban lantern Dec 4, 2022, 10:45 PM

#

yes yes

#

try it anyway

obsidian wolf Dec 4, 2022, 10:45 PM

#

i'll try it ofr now

#

this works

urban lantern Dec 4, 2022, 10:48 PM

#

it probably cant find the VS nvironment variables

#

because you did something bogus

obsidian wolf Dec 4, 2022, 10:48 PM

#

yeah i did nothing

urban lantern Dec 4, 2022, 10:48 PM

#

reinstalling vs should work, then you dont need the vs prompt anymore

obsidian wolf Dec 4, 2022, 10:49 PM

#

i fixed it with updating path, but it changing cache like this is really annoying

#

if i run a command and it works

#

i should be able to run the same command later and have it work again unless i changed something

#

explicitly

#

something like opening the gui and clicking generate, or running it trying to generate for ninja, or similar

#

unset something

urban lantern Dec 4, 2022, 10:51 PM

#

vs' cmake integration is still not the best, compared to clion/vscode+cmake

obsidian wolf Dec 4, 2022, 10:51 PM

#

and the thing it unset is what cmake was using to find the compiler

#

should i be using clion if i want to use ninja?

urban lantern Dec 4, 2022, 10:52 PM

#

all you need on windows is also just mkdir build && cd build && cmake -G "Visual Studio..." ../ if you want to generate the solution manually

main cosmos Dec 4, 2022, 10:53 PM

#

obsidian wolf should i be using clion if i want to use ninja?

You can use ninja and vs

#

It's what I'm doing

obsidian wolf Dec 4, 2022, 10:53 PM

#

how does this work monkaHmm

urban lantern Dec 4, 2022, 10:54 PM

#

-G ninja instead of -G "Visual Studio

obsidian wolf Dec 4, 2022, 10:54 PM

#

are there any steps or does it just work with the -G

main cosmos Dec 4, 2022, 10:54 PM

#

You open the directory in visual studio instead of fiddling with cmake commands

obsidian wolf Dec 4, 2022, 10:54 PM

#

open the directory after cmake -G Ninja .., instead of cmake -build . or whatever

urban lantern Dec 4, 2022, 10:55 PM

#

no

#

just open it with vs, after you cloned it or whatever

#

or clion, or vscode

main cosmos Dec 4, 2022, 10:55 PM

#

You never need to type a cmake command if you open in vs

urban lantern Dec 4, 2022, 10:55 PM

#

same for the other IDEs

obsidian wolf Dec 4, 2022, 10:55 PM

#

where am i asking VS to use ninja

main cosmos Dec 4, 2022, 10:55 PM

#

It uses it by default

#

If you open the directory in vs

obsidian wolf Dec 4, 2022, 10:56 PM

#

after i make the CMakeLists?

urban lantern Dec 4, 2022, 10:56 PM

#

yes

obsidian wolf Dec 4, 2022, 10:57 PM

#

ok 🙂 ty

urban lantern Dec 4, 2022, 10:57 PM

#

you can clone my cmake-glfw repo to test, or jaker's fwog

main cosmos Dec 4, 2022, 10:59 PM

#

Btw opening the directory randomly doesn't work on my work computer, so I have to generate the solution myself

#

So that's a thing

obsidian wolf Dec 4, 2022, 11:00 PM

#

surprisedPikachu

severe spruce Dec 4, 2022, 11:00 PM

#

first the blood sacrifice, then opening the folder smh

obsidian wolf Dec 4, 2022, 11:05 PM

#

this is also epic

main cosmos Dec 4, 2022, 11:06 PM

#

Hey that looks just like what I got

obsidian wolf Dec 4, 2022, 11:07 PM

#

did you solve it 🙂

main cosmos Dec 4, 2022, 11:07 PM

#

nervous

obsidian wolf Dec 4, 2022, 11:07 PM

#

i was expecting like

#

you know the sorts of errors i'd expect out of a poorly configured cmake file

#

include file "<vuk/Types.hpp>" not found. (in source file SkeletonGPU.cpp:7) shit like this

main cosmos Dec 4, 2022, 11:08 PM

#

Instead it fails to compile a shrimple test program

#

It can find cl.exe, so I don't even know what's failing

#

Maybe messed up permissions or something

obsidian wolf Dec 4, 2022, 11:10 PM

#

trade offer:
u get: 10-15 seconds saved per compile, maybe
i get: multiple days worth of brain rot trouble shooting

#

nah fuck this shit

#

it's worse than worthless, this shit siphons actual productivity

#

waste of my fucking time

main cosmos Dec 4, 2022, 11:11 PM

#

Cmake users btfo

urban lantern Dec 4, 2022, 11:13 PM

#

do you use an up2date cmake?

main cosmos Dec 4, 2022, 11:15 PM

#

When I was having the same issue, I reinstalled cmake and VS and it didn't help

obsidian wolf Dec 4, 2022, 11:16 PM

#

i am going to become the game dev joker (jonathan blow)

severe spruce Dec 5, 2022, 7:28 AM

#

its not finding kernel32.lib

#

so maybe a bad path to the WDK

obsidian wolf Dec 5, 2022, 7:34 AM

#

obsidian wolf trade offer: u get: 10-15 seconds saved per compile, maybe i get: multiple days ...

:)

severe spruce Dec 5, 2022, 7:36 AM

#

btw you can run vcvars from any terminal

#

and it will set your env

#

i use a PS startup script so all my terminals can compile

obsidian wolf Dec 6, 2022, 4:17 AM

#

i used imguizmo

#

integrated it quick

#

and was reminded why i don't use it :)

#

NIH time!!!

#

i really shouldn't though

main cosmos Dec 6, 2022, 4:22 AM

#

name a better duo

waste of my fucking time
NIH time!!!

obsidian wolf Dec 6, 2022, 4:22 AM

#

i really dislike this task that takes 10 seconds too long

#

time to spend 3 days

#

but nah there's like 2 inputs to imguizmo and the input doesn't work 🙂

#

i feel like there's too much integration for something like it to work smoothly

#

there's coordinate conventions, there's the input, the rendering (especially depth involvement)

main cosmos Dec 6, 2022, 5:17 AM

#

ye I'm just taking the piss

#

it took me a while to understand cmake enough to not want to commit sudoku, meanwhile NIHing something at least gives you something tangible to work on

severe spruce Dec 6, 2022, 6:54 AM

#

i have a PR open to imguizmo for years now

obsidian wolf Dec 6, 2022, 6:55 AM

#

:)

#

PR my incoming pose widget

#

hehe jk imagine taking the time to make things usable by others

severe spruce Dec 6, 2022, 7:01 AM

#

my PR is something more fundamental, such as making it not broken

#

but thats not worthwhile to merge :3

obsidian wolf Dec 6, 2022, 7:03 AM

#

no i know but surely there will be countless math issues

#

barely related but i actually think a ton of the math for these widgets is super cool

#

like i was just rewriting my method for projecting the mouse to a 3d circle

main cosmos Dec 6, 2022, 7:03 AM

#

severe spruce my PR is something more fundamental, such as making it not broken

tfw your pr is just one loc

severe spruce Dec 6, 2022, 7:04 AM

#

i hope it is fixed otherwise in master

obsidian wolf Dec 6, 2022, 7:05 AM

#

    ray3 os_ray = math::transformed_ray(mvp, uv_mouse_position);

    // We intersect our object space ray with the axis' plane, because the intersect is in object space, we can just atan it for angle
        float t = -os_ray.origin[axis] / os_ray.dir[axis];
        vec3 os_intersect = os_ray.origin + t * os_ray.dir;
        return_info.angle = math::atan2(os_intersect[(axis + 1) % 3], os_intersect[(axis + 2) % 3]);```

#

where transformed ray is relatively simple too

#

basically just

r3 transformed_ray(const m44& transform, v2 viewport_UV) {
    m44 transform_inv = math::inverse(transform);
    v4 origin_hpos = transform_inv * v4(u, v, 0.01f, 1.0f);
    v4 end_hpos    = transform_inv * v4(u, v, 1.0f, 1.0f);```

#

still slightly complicated, but considering how hard finding the closest point on a 3d circle to a mouse conceptually is

obsidian wolf Dec 8, 2022, 12:35 AM

#

#

#

#

this is actually so insane

obsidian wolf Dec 8, 2022, 9:59 PM

#

yeah this aint quite right

urban lantern Dec 8, 2022, 10:05 PM

#

for those moments, the rat-evicts-frog-fromfrogcave-music seems applicable and should be played in the background

obsidian wolf Dec 8, 2022, 10:06 PM

#

https://www.youtube.com/watch?v=ptb5coo5VSo
this? :)

YouTube

Bożydar Grabowski

Rat evicts a frog

Fat bastard

▶ Play video

urban lantern Dec 8, 2022, 10:15 PM

#

yes 😄

obsidian wolf Dec 9, 2022, 3:58 AM

#

also, unironically chatGPT has pushed forward my design documents for the game

#

onenote is weird

#

then there are the like 10 characters i asked it to generate lmao

#

the images aren't actual images i'm using, just refs i'm slowly putting onto them

#

the picture for sauria is the result of asking chat-gpt for a text description, then feeding the result into dall-e lmao

#

also the transform widget is done-enough :)

#

actually jk i still need to pull it into it's own renderpass so that it can render transparently over other objects

#

i added a feature to fade controls when they're basically impossible to use, to keep it visually simpler, i am a big fan of it
there's also more controls, (i.e. rotation around the vector to the camera), i'm just choosing not to expose it :)

main cosmos Dec 9, 2022, 4:16 AM

#

nice widget

obsidian wolf Dec 9, 2022, 7:33 AM

#

still needs a bit of work for the skeletons lmao

#

the inverse bind matrix has got it fucked up

obsidian wolf Dec 9, 2022, 5:30 PM

#

gltf not naturally supporting z-up is annoying as fuck

#

fortunately, blender supports gltf with z-up and therefore i am going to force it for anything with a skeleton

#

cause this is annoying as fuck to debug :)

#

it's better, but no clue why it's rotated like this at the start when the quaternion should be an axis-aligned rotation

#

some of my non-glm math is probably fucked :)

urban lantern Dec 9, 2022, 6:02 PM

#

thats what you get for using z up 🙂

obsidian wolf Dec 9, 2022, 6:02 PM

#

my game is top down

urban lantern Dec 9, 2022, 6:02 PM

#

no excuse

obsidian wolf Dec 9, 2022, 6:02 PM

#

y-up is dogma :)

#

unironically tho it feels weird for a modern model format not to express coordinate conventions

#

it wouldn't really help me here, beyond the fact that blender would probably have those exposed as settings, saving me the work

urban lantern Dec 9, 2022, 6:08 PM

#

can you not just multiply with a matrix to transform from y up to z up

obsidian wolf Dec 9, 2022, 6:08 PM

#

yes but then every component of the model uses y-up

#

it works seamlessly if you never have to touch anything within the model

urban lantern Dec 9, 2022, 6:09 PM

#

i dont get that, i mean you multiply everything once

obsidian wolf Dec 9, 2022, 6:09 PM

#

in the case of skinning, that is most definitely not the case

#

i would do that but it involves every single component (each vertex pos, each normal, skinning transforms, etc.) and when i tried it was a bucket of issues

#

spent a good couple days on it

#

that was a while ago though, i could probably do it a bit easier if i tried again

urban lantern Dec 9, 2022, 6:11 PM

#

ah

obsidian wolf Dec 9, 2022, 6:12 PM

#

it's the sort of thing though where if you know how to fix it, you don't really need to fix it xD

obsidian wolf Dec 10, 2022, 3:57 AM

#

urban lantern Dec 10, 2022, 10:35 AM

#

thats cool

#

and i really like the crocodino 🙂

obsidian wolf Dec 10, 2022, 7:47 PM

#

trying to figure out how exactly to integrate it with my input system

#

the main input consumption happens by having a sorted vector of input callbacks which can exit out

void default_scroll_callback(GLFWwindow* window, double x, double y) {
    Input::mouse_wheel = y;
    for (auto ptr = Input::scroll_callback_stack.end() - 1; ptr >= Input::scroll_callback_stack.begin(); ptr--) {
        auto& [fp, name, data] = *ptr;
        bool  esc              = (*fp)(window, x, y, data);
        if (esc)
            return;
    }
}```

#

pushing one of these callbacks for each of the widgets is too annoying to manage, and doesn't really work with immediate mode,

#

so i think i'm going to have some sort of singleton where widgets submit their id and hovered depth every frame, then read back from the singleton to see if they're pressed

#

that way it can have one callback, and can decide which widget is pressed

#

should all work, probably 🙂

obsidian wolf Dec 10, 2022, 10:44 PM

#

i actually also sort of need to optimize lmao

#

:^)

#

debug + validation, but still horrid

#

on release with validation off it's still 600fps, but i want debug+validation to be usable

obsidian wolf Dec 11, 2022, 5:13 AM

#

ok as much as i've enjoyed doing skinning + widget shit

#

i need to take a break from it and do something with gameplay to avoid getting lost in what is ultimately not gameplay

#

so i'm going to reimplement one of the units now that it is not planets and return to fruits dying for the third time

#

actually

#

no i'm not

#

in some of my refactoring since that, i actually wrote in the concept of editor scenes

#

so now i need game scene, then a state transition between the two, which is cool

main cosmos Dec 11, 2022, 5:33 AM

#

are you gonna make a state machine

obsidian wolf Dec 11, 2022, 5:33 AM

#

all of the "gameplay" has been in the map editor xD

#

state machines aren't a real thing

main cosmos Dec 11, 2022, 5:33 AM

#

are you gonna build a real physical state machine

obsidian wolf Dec 11, 2022, 5:33 AM

#

yes

#

i'm going to 3d print it

#

check this out!!!

main cosmos Dec 11, 2022, 5:35 AM

#

my mlg pro state machines are usually an enum + a switch statement

obsidian wolf Dec 11, 2022, 5:36 AM

#

this is only a single line more complicated

#

tbh i actually like jonathan blow's advice for this

#

i'm overanalyzing the joke but

main cosmos Dec 11, 2022, 5:39 AM

#

I'm not joking. I don't know how to make an overengineered state machine

#

or whatever the fancy ones are

obsidian wolf Dec 11, 2022, 5:39 AM

#

i think it's a really good philosophy to implement the solution to your direct problem as quickly as possible to check it off your list, and go onto the next thing

#

where you only build up the engineering on something when you have to keep going back to fix it

#

and even when you go back to fix it for the first or second time, you just retry, and eventually you'll stop coming back to it

#

when you do, it's well engineered

main cosmos Dec 11, 2022, 5:40 AM

#

well-enough engineered

obsidian wolf Dec 11, 2022, 5:40 AM

#

i used to care more about stuff like well engineered state machines, because they provide clean solutions for problems

#

but i don't anymore because of that philosophy

main cosmos Dec 11, 2022, 5:41 AM

#

you got mind Blown

obsidian wolf Dec 11, 2022, 5:42 AM

#

jonathan.

main cosmos Dec 11, 2022, 5:42 AM

#

you got mind jonathaned

obsidian wolf Dec 11, 2022, 5:42 AM

#

https://tenor.com/view/jonathan-name-rocket-blast-off-space-shuttle-gif-15586672

Tenor

#

also, i am accepting ideas for maps :) only restriction is that it has to conceivably be the home or similar of lizards, and grid based

#

ideally in the form of reference images

main cosmos Dec 11, 2022, 5:49 AM

#

for a tower defense game?

obsidian wolf Dec 11, 2022, 5:50 AM

#

for a tower defense game where the defenders are the rpg lizard characters yes

main cosmos Dec 11, 2022, 5:58 AM

#

rpg lizard character

obsidian wolf Dec 11, 2022, 5:58 AM

#

i'm guessing it'll vaguely end up in this realm (i also just found a really good rabbit hole on pinterest)

obsidian wolf Dec 11, 2022, 5:58 AM

#

main cosmos rpg lizard character

this looks AI generated 🧐

main cosmos Dec 11, 2022, 5:59 AM

#

it do be

obsidian wolf Dec 11, 2022, 5:59 AM

#

which?

main cosmos Dec 11, 2022, 5:59 AM

#

stable diffusion

obsidian wolf Dec 11, 2022, 5:59 AM

#

:)

main cosmos Dec 11, 2022, 6:01 AM

#

obsidian wolf Dec 11, 2022, 6:01 AM

#

monkaHmm

#

honestly

#

incredible how many different ways the AI can imagine a corgi wearing a banana

#

scarf, hoodie, ears, neck down, whole thing

main cosmos Dec 11, 2022, 6:04 AM

#

obsidian wolf Dec 11, 2022, 6:04 AM

#

https://tenor.com/view/monkey-mojothemonkey-zzsoobn-monkey-sleep-sleepy-gif-22926696

Tenor

#

we don't need AI for this one

#

https://tenor.com/view/fadsfantasy-monkey-phone-mono-teléfono-george-monkey-gif-26110826

Tenor

#

https://tenor.com/view/kevyn-monkey-kevyntth-gif-24951593

Tenor

#

god i love the human-adjacent nature of monkeys

main cosmos Dec 11, 2022, 6:06 AM

#

I forgot what I was trying to do tbh

#

here's a banana "chicken"

obsidian wolf Dec 11, 2022, 6:07 AM

#

ah yes, horrors beyond my comprehension

obsidian wolf Dec 12, 2022, 4:18 AM

#

just figured out why im on 30 fps with validation off lmao

#

my meshes weren't being added to the mesh cache so they were being reuploaded from disk every frame

#

honestly incredible that it was still at 30-60fps

#

https://tenor.com/view/awaji-monkey-awajishima-monkey-awaji-island-monkey-awajishima-monkey-looking-away-monkey-looking-away-gif-26330925

Tenor

obsidian wolf Dec 12, 2022, 5:05 AM

#

i have no clue why it pauses when i don't have it hovered PepeWTF

#

step on

obsidian wolf Dec 12, 2022, 6:15 AM

#

well this is how it's ending this weekend

#

oh also there's finally a way to edit the z-levels

#

obsidian wolf Dec 13, 2022, 7:19 AM

#

#

shizzzzzzzzzzzzzzzzzzzz

#

basically a damn game

#

whats he doin up there

severe spruce Dec 13, 2022, 7:26 AM

#

he losing it

#

but not the device

obsidian wolf Dec 13, 2022, 7:26 AM

#

so true

main cosmos Dec 13, 2022, 7:27 AM

#

he's busting it down

obsidian wolf Dec 13, 2022, 7:32 AM

#

severe spruce he losing it

he just like me fr fr

obsidian wolf Dec 13, 2022, 11:08 PM

#

z_up_1

obsidian wolf Dec 18, 2022, 12:46 AM

#

void

#

i keep checking my work's slack

#

vacation moment

obsidian wolf Dec 18, 2022, 8:09 PM

#

damn lol i really looked at this, said "why do i have no output" and opened up renderdoc

#

there is output, it's just very dark 🙂

#

was boutta be a 15 minute debugging session if it was sunny outside

obsidian wolf Dec 18, 2022, 8:27 PM

#

looking good

obsidian wolf Dec 18, 2022, 11:00 PM

#

hmmm yes, shadows

obsidian wolf Dec 19, 2022, 4:40 AM

#

#

idk why this guy got so shiny all the sudden

#

but monkaHmm

main cosmos Dec 19, 2022, 4:41 AM

#

he's so happy

obsidian wolf Dec 19, 2022, 4:41 AM

#

i intentionally made him sorta smiling 🙂

#

took the crease vertices up a bit 🙂

main cosmos Dec 19, 2022, 4:42 AM

#

I see working shadows

obsidian wolf Dec 19, 2022, 4:42 AM

#

i gotta figure out what im gonna do to filter these

main cosmos Dec 19, 2022, 4:42 AM

#

the shadows?

obsidian wolf Dec 19, 2022, 4:42 AM

#

yeah

#

idk what works with toon shading

main cosmos Dec 19, 2022, 4:42 AM

#

how do you want it to look?

#

just smooth?

obsidian wolf Dec 19, 2022, 4:43 AM

#

yeah that's what i mean

#

do people ever blur + posterize as a post process for shadows?

#

idk how i'd do that on the renderer side, but that might be a look

main cosmos Dec 19, 2022, 4:43 AM

#

hmm

obsidian wolf Dec 19, 2022, 4:44 AM

#

borderlands only does terrain shadows i believe

main cosmos Dec 19, 2022, 4:44 AM

#

this small piece of code is how I have stochastic disk-sampled shadows (it needs blue noise and maybe a small blur to look good)
https://github.com/JuanDiegoMontoya/Fwog/blob/examples-refactor/example/shaders/ShadeDeferredPbr.frag.glsl#L77

obsidian wolf Dec 19, 2022, 4:44 AM

#

and those are sharp, which i think looks

#

tbh it's probably best to be hard edge

#

just not pixelated, and i'm not entirely sure what people do for that

main cosmos Dec 19, 2022, 4:46 AM

#

someone showed me an algorithm which can be used to upscale hard edges

#

let me find the link

obsidian wolf Dec 19, 2022, 4:46 AM

#

there's that one pixel art upscaling algorithm

#

hx6 or something?

main cosmos Dec 19, 2022, 4:46 AM

#

ok you are thinking of the same thing

obsidian wolf Dec 19, 2022, 4:46 AM

#

hqnx

main cosmos Dec 19, 2022, 4:47 AM

#

ray traced shadow maps might look okay too, but they have some difficult edge cases

obsidian wolf Dec 19, 2022, 4:47 AM

#

not gonna do RT

main cosmos Dec 19, 2022, 4:47 AM

#

but at least you can get almost perfect sharp shadows

obsidian wolf Dec 19, 2022, 4:47 AM

#

not with skinning

main cosmos Dec 19, 2022, 4:47 AM

#

obsidian wolf not gonna do RT

it only uses a single ray-triangle intersection

#

no BVH

obsidian wolf Dec 19, 2022, 4:47 AM

#

where do you pull the triangle from?

main cosmos Dec 19, 2022, 4:47 AM

#

you just store the triangle ID in the shadow map and then ray trace against that triangle when applying shadows

obsidian wolf Dec 19, 2022, 4:48 AM

#

could you store vertex data monkaHmm

#

vertex data is technically 9 floats but

#

then you'd have the advantage of not having to reapply the vertex shader

main cosmos Dec 19, 2022, 4:48 AM

#

your scenes will probably be 🦐le enough that you can get away with some inefficiencies

obsidian wolf Dec 19, 2022, 4:49 AM

#

im gonna try hqnx first, i think that's literally just like

main cosmos Dec 19, 2022, 4:49 AM

#

it really is like

obsidian wolf Dec 19, 2022, 4:50 AM

#

why has no one done hqnx shadowmapping

main cosmos Dec 19, 2022, 4:51 AM

#

idk

obsidian wolf Dec 19, 2022, 4:51 AM

#

maybe arkham knight?

main cosmos Dec 19, 2022, 4:51 AM

#

@dry monolith suggested it to me at one point

#

seems like a pretty good idea

obsidian wolf Dec 19, 2022, 4:52 AM

#

i know the first step is to sample a 3x3 area

#

i'm not sure how you use the result though

main cosmos Dec 19, 2022, 4:53 AM

#

isn't the result a new 3x3 at the current pixel?

#

I guess that's for the 3x upscaling one

obsidian wolf Dec 19, 2022, 4:53 AM

#

the arrangement of neighbors is looked up in a predefined table which contains the necessary interpolation patterns

main cosmos Dec 19, 2022, 4:54 AM

#

I'm guessing you can just hardcode the tables into the shader if they're small, or put them into an SSBO otherwise

obsidian wolf Dec 19, 2022, 4:54 AM

#

#

tbh this could match the look pretty well

#

u dont understand

#

#

am i a mastermind or is this gonna get it wrong

main cosmos Dec 19, 2022, 4:57 AM

#

it looks incomplete

obsidian wolf Dec 19, 2022, 4:57 AM

#

yeah i think it got it wrong

#

#

that just looks like some gaussian blur

#

actually not even gaussian

main cosmos Dec 19, 2022, 4:57 AM

#

it do

obsidian wolf Dec 19, 2022, 4:57 AM

#

lame

#

chatGPT would be nice if it was just like "yeah this might be fuckin wrong" sometimes

#

https://github.com/grom358/hqx/blob/master/src/hq3x.c

GitHub

hqx/hq3x.c at master · grom358/hqx

hqx is a fast, high-quality magnification filter designed for pixel art. - hqx/hq3x.c at master · grom358/hqx

#

what the fuck

#

https://tenor.com/view/verne-stare-gif-24447130

Tenor

main cosmos Dec 19, 2022, 4:59 AM

#

https://tenor.com/view/no-realization-oh-god-surprised-what-gif-15852396

Tenor

obsidian wolf Dec 19, 2022, 4:59 AM

#

https://github.com/CrossVR/hqx-shader/blob/master/glsl/hq4x.glsl
eh?

GitHub

hqx-shader/hq4x.glsl at master · CrossVR/hqx-shader

GPU shader version of HQx filter. Contribute to CrossVR/hqx-shader development by creating an account on GitHub.

main cosmos Dec 19, 2022, 5:00 AM

#

no results on shadertoy

obsidian wolf Dec 19, 2022, 5:00 AM

#

i think this one is legit

#

just need to fiugre out what the fuck here is trash

#

no way the actual implementation is 100 lines, right

main cosmos Dec 19, 2022, 5:02 AM

#

idk

obsidian wolf Dec 19, 2022, 5:03 AM

#

well

#

we'll see

#

The filter was not designed for photographs, but for images with clear sharp edges, like line graphics or cartoon sprites. It was also designed to be fast enough to process 256x256 images in real-time.

#

i wonder when this was written

main cosmos Dec 19, 2022, 5:25 AM

#

1934

obsidian wolf Dec 19, 2022, 5:26 AM

#

maybe it means cpu lmao

#

even still

obsidian wolf Dec 19, 2022, 7:06 AM

#

https://www.shadertoy.com/view/cls3zN

Shadertoy

#

found a decent starting point, cleaned up their code

main cosmos Dec 19, 2022, 7:13 AM

#

what's the deal with the giant LUT in the OG one?

obsidian wolf Dec 19, 2022, 7:13 AM

#

honestly

#

i have no fucking clue

#

its just some academia shit i think

#

i didn't write this i just found the simplest implementation and made it simpler

main cosmos Dec 19, 2022, 7:14 AM

#

lol it looks quite cursed when you put a noise texture in

#

the rock texture looks like frosted glass or something

obsidian wolf Dec 19, 2022, 7:16 AM

#

honestly

#

kinda sick

#

this one is very cool

#

i wonder if there's any way you could modify this a bit to get rid of the grid artifacts

#

as in like

#

posterize first

main cosmos Dec 19, 2022, 7:19 AM

#

is posterizing the same as quantizing

obsidian wolf Dec 19, 2022, 7:19 AM

#

uhhh

#

just reducing amount of colors by clamping to nearest

main cosmos Dec 19, 2022, 7:20 AM

#

nearest wot

obsidian wolf Dec 19, 2022, 7:20 AM

#

i just mean reducing the amount of colors in the image

main cosmos Dec 19, 2022, 7:20 AM

#

ye that's what quantizing is

obsidian wolf Dec 19, 2022, 7:20 AM

#

then yes

main cosmos Dec 19, 2022, 7:20 AM

#

nice

#

tried quantizing and it looks like ass

#

vec4 quantize(vec4 v, int n)
{
    vec4 noise = texelFetch(iChannel1, ivec2(gl_FragCoord.xy) % textureSize(iChannel1, 0), 0);
    return floor((v * vec4(n)) + 2.0 * noise - 0.5) / vec4(n);
}

//draw diagonal line connecting 2 pixels if within threshold
bool diag(sampler2D tex, inout vec4 sum, vec2 uv, ivec2 p1, ivec2 p2, float thickness) {
    vec4 v1 = texelFetch(tex, ivec2(uv)+ivec2(p1.x,p1.y), 0);
    vec4 v2 = texelFetch(tex, ivec2(uv)+ivec2(p2.x,p2.y), 0);
    v1 = quantize(v1, 2);
    v2 = quantize(v2, 2);
    if (length(v1-v2) < THRESHOLD) {
        vec2 dir = vec2(p2-p1);
        vec2 lp = uv-(floor(uv+vec2(p1))+.5);
        dir = normalize(vec2(dir.y,-dir.x));
        float l = clamp((thickness-dot(lp,dir))*AA_SCALE,0.,1.);
        sum = mix(sum,v1,l);
        return true;
    }
    return false;
}

#

it uses the blue noise texture in iChannel1

obsidian wolf Dec 19, 2022, 7:29 AM

#

tbh i dont even knowhow this algo works still so

#

i aint about to start trying to get funky with it

#

tomorrow i'll actually use it for shadowmaps

#

Bedge now though

main cosmos Dec 19, 2022, 7:40 AM

#

it would probably look better to quantize after

main cosmos Dec 19, 2022, 7:44 AM

#

main cosmos the rock texture looks like frosted glass or something

quantized to 8 values per channel

#

removed the noise because it looks too smooth

obsidian wolf Dec 19, 2022, 7:44 AM

#

btw were u quantizing the original value?

main cosmos Dec 19, 2022, 7:45 AM

#

this one is quantizing at the very end

#

the nyan cat one is quantizing the texture samples

#

but the hqx algorithm makes new colors, so it isn't ideal

obsidian wolf Dec 19, 2022, 7:46 AM

#

all?

#

you only showed the 8

#

but there are 9 samples

main cosmos Dec 19, 2022, 7:47 AM

#

I put it in the diag function, which is called 12 times

obsidian wolf Dec 19, 2022, 7:47 AM

#

this is neat

#

let me try all 13 i guess

#

it doesn't achieve the desired effect anyways

#

#

this one is my favorite

main cosmos Dec 19, 2022, 7:54 AM

#

looks like a brainworm

obsidian wolf Dec 19, 2022, 6:33 PM

#

hmmmm

#

the filtering is working xD

#

okay

#

this isnt' bad at all actually

#

the shadowed edge on the characters isn't great though

dry monolith Dec 19, 2022, 6:42 PM

#

wait what am I doing in this thread again

obsidian wolf Dec 19, 2022, 6:42 PM

#

jaker tagged you

dry monolith Dec 19, 2022, 6:42 PM

#

ah I rember now

#

hqx filtering of shadows

obsidian wolf Dec 19, 2022, 6:43 PM

#

there is some funky-ness, but i think it's mostly just from the fact that i havent' done any slope bias or anything like that

#

overall good idea/10 🙂

dry monolith Dec 19, 2022, 6:44 PM

#

#1019779751600205955 message

#

slope scale bias

obsidian wolf Dec 19, 2022, 6:45 PM

#

hehe

#

when i was in my intro graphics course

#

and our prof told us about bias, i had the idea to try this

#

i could not figure out the math for like 2 days, tried to find it on google, couldn't and gave up 🤠

#

i think i mostly had a problem with the different coordinate spaces back then

#

this server needs a ty emote

dry monolith Dec 19, 2022, 6:50 PM

#

btw since this is cel shaded you can safely do cosine cutoff before you reach slope so big that peter panning becomes too big

obsidian wolf Dec 19, 2022, 6:51 PM

#

monkaHmm

dry monolith Dec 19, 2022, 6:51 PM

#

there is something off about my sentence construction

obsidian wolf Dec 19, 2022, 6:51 PM

#

just clamping?

#

i read it twice and got what you mean, i just don't know what cosine cutoff means here

dry monolith Dec 19, 2022, 6:52 PM

#

whatever, I mean - 0.08 to dot(n, l) like jaker did later on

#

so that it becomes fully shaded before 90 degrees slope

#

this will hide the way too disconnected shadow

#

you'll also need to add some value back to compensate for the cosine never reaching 1 then

obsidian wolf Dec 19, 2022, 6:53 PM

#

ah interesting

dry monolith Dec 19, 2022, 6:54 PM

#

(dot(n,l)-cutoff)/(1-cutoff)

#

https://www.desmos.com/calculator/iu7ijqdqww

Desmos

Desmos | Graphing Calculator

obsidian wolf Dec 19, 2022, 7:00 PM

#

i think i managed to disable everything except self shadowing lmao

#

ah

#

my bias is in the wrong space

obsidian wolf Dec 19, 2022, 7:16 PM

#

#

interesting side effect of hqx

#

it makes shadow acne look way cooler

#

or whatever you want to call this

#

i think it might be a bit big...

dry monolith Dec 19, 2022, 7:23 PM

#

are you doing slope scale?

obsidian wolf Dec 19, 2022, 7:24 PM

#

that's with this

    const float cutoff = 0.1;
    float interior = (dot(normalize(data.normal), normalize(sun_data.xyz)) - cutoff) / (1.0 - cutoff);
    float bias = tan(acos(interior));
    
    float world_position_depth = clip_planes[0] + position_lightspace.z * (clip_planes[1] - clip_planes[0]) + bias;
    float world_read_depth = clip_planes[0] + texture(s_sun_depth, uv).r * (clip_planes[1] - clip_planes[0]);

yeah

dry monolith Dec 19, 2022, 7:24 PM

#

oh my you used the inverse trig version

obsidian wolf Dec 19, 2022, 7:24 PM

#

does it actually matter for testing monkaHmm

dry monolith Dec 19, 2022, 7:25 PM

#

nah but you also used cosine with cutoff which will likely interfere

#

you should tamper with the cosine you use for bias calculation

#

you can do with bias whatever you want after it's calculated

#

maybe clamp like jaker does (and gets acne back at extreme angles), but I think you can use cutoff for cel shading

#

also

#

also also

#

wasn't there another factor for the bias

#

your shadow texel size

obsidian wolf Dec 19, 2022, 7:29 PM

#

yeah but it's just a scale

#

oh wait

#

no that one is real yeah

dry monolith Dec 19, 2022, 7:29 PM

#

the slope scale bias is calculated exactly for the texel size

#

#1019779751600205955 message

#

then if you look in the graph the bias factor is texel size divided by 2

#

https://www.desmos.com/calculator/nce35fulby

Desmos

slope scale bias

dry monolith Dec 19, 2022, 7:33 PM

#

dry monolith you should tamper with the cosine you use for bias calculation

typo, should NOT

obsidian wolf Dec 19, 2022, 7:59 PM

#

    float bias = 1.01 * (1.0 / max(textureSize(s_sun_depth, 0)));
    float slope_bias = length(cross(data.normal, sun_data.xyz)) / dot(data.normal, sun_data.xyz) * bias;
    
    float world_position_depth = clip_planes[0] + position_lightspace.z * (clip_planes[1] - clip_planes[0]) + slope_bias;
    float world_read_depth = clip_planes[0] + texture(s_sun_depth, uv).r * (clip_planes[1] - clip_planes[0]);```
![monkaHmm](https://cdn.discordapp.com/emojis/583705655198744593.webp?size=128 "monkaHmm")

#

the value seems to be too small

#

#

i'm missing the / 2 here but that makes it worse ofc

#

the cutoff ofc also doesn't change anything

#

if it's in clip space (basically just *20.0 for me because orthographic projection), it looks better, still not good though (and it should be in world space AFAIK?)

dry monolith Dec 19, 2022, 8:13 PM

#

should also offset by 1 precision bit value

#

depth map is quantized, I also said that at the time of discussion of slope scale bias in fwog thread

#

and yes I believe all of it should be in world space

#

so you technically need the bias to be the area one texel covers in world space

#

man I really am tired

#

actually let me look up what jaker put in fwog

#

interesting, I don't see any world space scaling in his code

obsidian wolf Dec 19, 2022, 8:19 PM

#

these depth values really shouldn't be floats considering it's orthographic projection, but i've never had issues with depth precision

dry monolith Dec 19, 2022, 8:19 PM

#

https://github.com/JuanDiegoMontoya/Fwog/blob/1d18058dd3d5553446a6042bfc51ab8436c44ce5/example/shaders/ShadeDeferredPbr.frag.glsl#L67

#

texel size in uv space accidentally happens to be texel size in world space or something

#

@main cosmos

#

you can derive how much of world area one texel covers from ortho projection creation code I believe

#

but also can do that by unprojecting two pixels and taking difference

obsidian wolf Dec 19, 2022, 8:24 PM

#

so you technically need the bias to be the area one texel covers in world space
i don't understand this

#

is this the quantize value or separate?

dry monolith Dec 19, 2022, 8:24 PM

#

neither do i

#

not area but max halfextent

obsidian wolf Dec 19, 2022, 8:25 PM

#

the diagonal, yeah?

dry monolith Dec 19, 2022, 8:26 PM

#

yeah I guess diagonal 💀

obsidian wolf Dec 19, 2022, 8:26 PM

#

that should be the width, not max(textureSize())

dry monolith Dec 19, 2022, 8:26 PM

#

that would be length(1/textureSize())

obsidian wolf Dec 19, 2022, 8:26 PM

#

he's also doing it in clip space

#

which is how it works for me

#

    float lightDepth = textureLod(s_rsmDepth, uv + offset, 0).x;

    lightDepth += bias;```

dry monolith Dec 19, 2022, 8:28 PM

#

I am trying to thinl

main cosmos Dec 19, 2022, 8:30 PM

#

dry monolith <@288411937962721305>

dry monolith Dec 19, 2022, 8:31 PM

#

good timing can you tell how your code works

main cosmos Dec 19, 2022, 8:32 PM

#

it applies the bias in ndc space

#

was it supposed to be world space?

dry monolith Dec 19, 2022, 8:32 PM

#

I was trying to remember how shadow mapping works with tex2dproj

main cosmos Dec 19, 2022, 8:32 PM

#

ah, I don't use a shadow sampler anymore

#

but basically with textureProj and a shadow sampler, you pass it a vec4 and it divides xyz by w for you (lol)

dry monolith Dec 19, 2022, 8:33 PM

#

main cosmos was it supposed to be world space?

actually I'm not sure

main cosmos Dec 19, 2022, 8:33 PM

#

then it does the shadow test with the z coord and whatever it samples from the shadow map

dry monolith Dec 19, 2022, 8:33 PM

#

you wrote the code without me handholding you

#

so I don't have all the details 😢

main cosmos Dec 19, 2022, 8:33 PM

#

well I just copied ur formula from desmos

#

and assumed it was supposed to be an ndc space bias

dry monolith Dec 19, 2022, 8:34 PM

#

it was, yeah, I was assuming that too

obsidian wolf Dec 19, 2022, 8:34 PM

#

if i do it in NDC with everything now it's not bad

dry monolith Dec 19, 2022, 8:35 PM

#

but it depends on measure of the space where the comparison is happening, or something, I really can't concentrate fr no cap

#

been working on 1 braincell this whole convo

obsidian wolf Dec 19, 2022, 8:35 PM

#

certainly not perfect 🤠

#

i should figure out if using a shadow sampler would work with hqx

#

i don't know if vulkan is unique to having slope bias but hehe

main cosmos Dec 19, 2022, 8:37 PM

#

opengl has glPolygonOffset, which is identical to vulkan's depth bias stuff in the VkPipelineRasterizationCreateInfo struct

#

but it's kinda lame for shadows compared to computing it in the fs yourself

obsidian wolf Dec 19, 2022, 8:38 PM

#

why monkaHmm

#

NIH?

main cosmos Dec 19, 2022, 8:39 PM

#

the slope bias especially is weird because it depends on the format of the shadow map

#

and it's even weirder with floating-point depth buffers because the bias depends on the maximum exponent of any vertex in the triangle

dry monolith Dec 19, 2022, 8:40 PM

#

you're transforming world by sun viewproj before comparing depth values right

main cosmos Dec 19, 2022, 8:40 PM

#

but using void's formula lets you get the perfect bias everytim

dry monolith Dec 19, 2022, 8:41 PM

#

so that means our measure is entirely in ndc now

#

that means we could use ndc texel size

main cosmos Dec 19, 2022, 8:41 PM

#

obsidian wolf why <:monkaHmm:583705655198744593>

I think the depth bias stuff is useful when you are rasterizing decals and other geometry that you don't want to z-fight

obsidian wolf Dec 19, 2022, 8:42 PM

#

tbh i just don't know why they look like shit for me right now

dry monolith Dec 19, 2022, 8:42 PM

#

I also think that we should use diagonal indeed

severe spruce Dec 19, 2022, 8:42 PM

#

main cosmos and it's even weirder with floating-point depth buffers because the bias depends...

no spoilers

obsidian wolf Dec 19, 2022, 8:42 PM

#

it's 2048x2048 with a 10x10x20 frustrum, feels like i shouldn't be having real bias issues

dry monolith Dec 19, 2022, 8:42 PM

#

so length(0.5/textureSize(shadowmap))

#

guess needs to be tested

#

wherewas my graph

obsidian wolf Dec 19, 2022, 8:43 PM

#

https://www.desmos.com/calculator/nce35fulby

Desmos

slope scale bias

#

fwog code doesn't have any additional bias which doesn't make sense to me either

#

shouldn't you need to add additional bias in order to account for convex objects?

main cosmos Dec 19, 2022, 8:43 PM

#

I'm not sure

#

like a constant bias?

obsidian wolf Dec 19, 2022, 8:44 PM

#

if you're using exact bias on a flat surface, the max(textureSize()) should've had clipped corners

dry monolith Dec 19, 2022, 8:44 PM

#

main cosmos Dec 19, 2022, 8:44 PM

#

I don't think I've rendered a sphere yet lol

#

just flat stuff mostly

obsidian wolf Dec 19, 2022, 8:44 PM

#

obsidian wolf if you're using exact bias on a flat surface, the `max(textureSize())` should've...

even here

#

1.0 / max(textureSize()) < length(1.0 / textureSize())

main cosmos Dec 19, 2022, 8:44 PM

#

I see why a constant bias is needed for convex geo

dry monolith Dec 19, 2022, 8:46 PM

#

fwog has quantization bias additionally

obsidian wolf Dec 19, 2022, 8:46 PM

#

yeah i was thinking that might happen to account for it

dry monolith Dec 19, 2022, 8:46 PM

#

1 bit of precision

main cosmos Dec 19, 2022, 8:46 PM

#

I actually made the quantization bias a little extra large for some reason

#

probably paranoia

obsidian wolf Dec 19, 2022, 8:47 PM

#

use a curvature map to account for curved surfaces and have the perfect bias every time brainworm

dry monolith Dec 19, 2022, 8:47 PM

#

dry monolith

jaker do you understand what I mean here

main cosmos Dec 19, 2022, 8:49 PM

#

that diagram is hard for me to understand

obsidian wolf Dec 19, 2022, 8:50 PM

#

he is just saying the width in 3d is bigger than in 2d

#

because the width in 3d is diagonal, envision the red line at the origin as the cross section of the texel

#

and the box in the bottom left as an alternative perspective of it

main cosmos Dec 19, 2022, 8:50 PM

#

o

obsidian wolf Dec 19, 2022, 8:50 PM

#

obsidian wolf 1.0 / max(textureSize()) < length(1.0 / textureSize())

this

main cosmos Dec 19, 2022, 8:51 PM

#

so the texel size should actually be scaled up by sqrt(2)?

obsidian wolf Dec 19, 2022, 8:51 PM

#

uhhh you can just do length(0.5 / textureSize())

#

for the brainworm of non-uniform shadowmaps

#

and the probably non-existent microoptimization

main cosmos Dec 19, 2022, 8:52 PM

#

like perspective shadow maps?

obsidian wolf Dec 19, 2022, 8:53 PM

#

i don't think i'd ever actually do that

#

non-uniform scale shadowmaps

dry monolith Dec 19, 2022, 8:53 PM

#

obsidian wolf use a curvature map to account for curved surfaces and have the perfect bias eve...

but ur gonna need a smooth surface for curvatyre

obsidian wolf Dec 19, 2022, 8:54 PM

#

yeah not actually feasible i don't think

dry monolith Dec 19, 2022, 8:54 PM

#

I maybe don't understand curvature map

obsidian wolf Dec 19, 2022, 8:54 PM

#

they're more of an artist concept than a graphics programmers

main cosmos Dec 19, 2022, 8:54 PM

#

what is a non-uniform scale shadow map?

#

I just thought you meant perspective

#

parabolic shadow map maybe?

obsidian wolf Dec 19, 2022, 8:55 PM

#

main cosmos what is a non-uniform scale shadow map?

if you used like 2048x1024 res for your shadowmap lamo

main cosmos Dec 19, 2022, 8:55 PM

#

oh

#

void just takes the max of the x and y texel size iirc

#

seems reasonable to me

#

but tbh when would you have a non-square shadow map

dry monolith Dec 19, 2022, 8:55 PM

#

yes but I don't know if corners can still penetrate in 3d

#

so maybe diagonal is reasonabler

#

radius of the circumscribed circle if you will

main cosmos Dec 19, 2022, 8:57 PM

#

a small gnome with a protractor is inside our GPUs froge

dry monolith Dec 19, 2022, 8:58 PM

#

anyways imma take a rest for a bit

obsidian wolf Dec 20, 2022, 2:07 AM

#

moving on from shadows for now

#

these aren't perfect but

#

good enough for why i needed to do shadows

main cosmos Dec 20, 2022, 2:09 AM

#

looks good

#

what res is the shadow map?

obsidian wolf Dec 20, 2022, 2:09 AM

#

2048

#

these are the extents

#

wait

#

#art-discussion message

#

djsp does this

#

i am a follower...

#

peepowait

#

#wip message

obsidian wolf Dec 20, 2022, 11:30 PM

#

hmmm yes

#

need to do the corner pieces still

dry monolith Dec 20, 2022, 11:30 PM

#

did you figure out the shadow bias

obsidian wolf Dec 20, 2022, 11:30 PM

#

obsidian wolf moving on from shadows for now

its okay, this was the main "inspection"

dry monolith Dec 20, 2022, 11:31 PM

#

what's the code is like?

obsidian wolf Dec 20, 2022, 11:31 PM

#

    const float maxBias = 0.005;
    const float quantize = 2.0 / (1 << 23);
    float b = length(1.0 / textureSize(s_sun_depth, 0));
    float bias = quantize + b * length(cross(sun_data.xyz, data.normal)) / clamp(dot(sun_data.xyz, data.normal), 0.0, 1.0);
    bias = min(bias, maxBias);

    float world_position_depth = position_lightspace.z + bias;
    float world_read_depth = texture(s_sun_depth, uv).r;```

#

the quantize is wrong i am just now realizing

#

switched to a 16unorm because orthographic

main cosmos Dec 20, 2022, 11:34 PM

#

obsidian wolf the quantize is wrong i am just now realizing

ye

#

it's either too big or too small, depending on your depth format

dry monolith Dec 20, 2022, 11:34 PM

#

that's basically fwogs code

obsidian wolf Dec 20, 2022, 11:34 PM

#

yeah

#

that's where it started from lmao

#

i had to increase the max bias because of different scene scales though

main cosmos Dec 20, 2022, 11:35 PM

#

1.0 / (1 << 16) should be the largest quantization bias needed

obsidian wolf Dec 20, 2022, 11:35 PM

#

i did 0.5 over that :^)

dry monolith Dec 20, 2022, 11:36 PM

#

jaker did you change max to length

main cosmos Dec 20, 2022, 11:36 PM

#

no

#

t yet

dry monolith Dec 20, 2022, 11:36 PM

#

ok

#

ay

obsidian wolf Dec 20, 2022, 11:36 PM

#

yeahhhhhh

dry monolith Dec 20, 2022, 11:37 PM

#

again instead of clamping bias you could make surface fully shaded by cutting off the cosine early

#

but that's more of an artistic decision

main cosmos Dec 20, 2022, 11:37 PM

#

I prefer the method of just not having grazing shadows

dry monolith Dec 20, 2022, 11:38 PM

#

I prefer the method of just not having rasterized shadows

obsidian wolf Dec 20, 2022, 11:38 PM

#

:^)

main cosmos Dec 20, 2022, 11:38 PM

#

touche

obsidian wolf Dec 20, 2022, 11:38 PM

#

i think i haven't tried the cosine cutoff since i have it working decently

dry monolith Dec 20, 2022, 11:38 PM

#

missed the chance to say minor spelling mistake I win

#

wonder if you can blur shadows so much acne becomes the least of your problems

obsidian wolf Dec 20, 2022, 11:42 PM

#

i used hqx to avoid blur

dry monolith Dec 20, 2022, 11:43 PM

#

tbh

#

you might have wanted stencil shadows

#

those are pixel perfect

obsidian wolf Dec 20, 2022, 11:44 PM

#

never looked into em

main cosmos Dec 20, 2022, 11:46 PM

#

or fake ray traced shadow mapping

#

I think it's called hybrid ray traced shadows or something

obsidian wolf Dec 20, 2022, 11:47 PM

#

is that the triangle id method you were talkin about

main cosmos Dec 20, 2022, 11:47 PM

#

yuh

#

it suffers from holes if your triangles are too small, unless you store a list of triangles IDs per-fragment

obsidian wolf Dec 20, 2022, 11:55 PM

#

first is clamp, second is cutoff

#

assuming i'm doing something wrong with the cutoff?

#

float bias = quantize + b * length(cross(sun_data.xyz, data.normal)) / ((dot(sun_data.xyz, data.normal) - 0.05) / (0.95));

#

not the same sun angle btw, i dont have it exposed as a value anywhere, teehee

dry monolith Dec 21, 2022, 12:00 AM

#

cutoff means not doing something to the bias but to your shading

obsidian wolf Dec 21, 2022, 12:00 AM

#

ah

#

i see what the goal is then

dry monolith Dec 21, 2022, 12:00 AM

#

to hide the panning shadow with early fade to zero

#

sorry I wasn't clear earlier

#

for bias calculation you'll always want accurate cosine

obsidian wolf Dec 21, 2022, 12:02 AM

#

yeah this makes sense

obsidian wolf Dec 21, 2022, 12:20 AM

#

even at 0 it's basically fine

#

i had some wrap before which would force shadowmaps to be the hard edge

#

i kinda wish i could still do wrap without diving a ton into shadows

dry monolith Dec 21, 2022, 12:29 AM

#

so you are testing/showing off stuff on a complex mesh instead of a synthetic scene, and I can't see anything lol

#

there are still weird moments but I can't quite catch the frame

#

ok

#

got it

obsidian wolf Dec 21, 2022, 12:32 AM

#

yeah this is just acne filtered by hqx

#

cutting off the dot at 0.05 hides it completely

#

which... i don't believe is perceivable really? so that works

obsidian wolf Dec 21, 2022, 12:53 AM

#

do you not run into this though jaker?

main cosmos Dec 21, 2022, 12:55 AM

#

No

#

How big is your frustum

obsidian wolf Dec 21, 2022, 12:55 AM

#

the top turning black is from my outline code, i only mean the acne on the sides

#

16x16x20

#

each cube (croc sized) is 1 unit

main cosmos Dec 21, 2022, 12:56 AM

#

I didn't get that side acne, but maybe I didn't test such an extreme case

dry monolith Dec 21, 2022, 12:59 AM

#

nonlinear z issue?

#

cuz perspective

#

what if you use reverse z

obsidian wolf Dec 21, 2022, 12:59 AM

#

i already am using reverse z

#

but the directional shadowmap should already be linear

main cosmos Dec 21, 2022, 1:00 AM

#

Btw what's the frustum of the other camera

#

The player cam

obsidian wolf Dec 21, 2022, 1:00 AM

#

main cosmos Btw what's the frustum of the other camera

infinite, it's reverse depth

dry monolith Dec 21, 2022, 1:01 AM

#

I mean if you go from screen-space to world space by unprojecting there may be an issue with reconstruction caused by depth precision

obsidian wolf Dec 21, 2022, 1:01 AM

#

yeah that'd make sense

#

imma raise the near plane and see if that changes anything

dry monolith Dec 21, 2022, 1:02 AM

#

also reverse z not really linearizes it

#

just distributes it more evenly than it usually is

#

and I think it only works with D32F

main cosmos Dec 21, 2022, 1:03 AM

#

reverse z + floating point depth makes it even more linearer

obsidian wolf Dec 21, 2022, 1:03 AM

#

i am realizing that my depth outline works in clip space so it changes with near plane lmao

main cosmos Dec 21, 2022, 1:04 AM

#

Kej

obsidian wolf Dec 21, 2022, 1:04 AM

#

that's unrelated though, player cam is D32F with reverse depth, shadowmap is D16U with orthographic projection

#

it does slightly better so i guess that's it

#

#questions message
does this mean there is a reason to store position monkaHmm

#

i dont think i actually would

main cosmos Dec 21, 2022, 1:06 AM

#

It means there is no reason to store position

obsidian wolf Dec 21, 2022, 1:06 AM

#

i'd rather fix it with adding some bias factor for reconstructed position inaccuracies

#

no that's what he said, not storing position has the downside of losing precision

main cosmos Dec 21, 2022, 1:08 AM

#

yeah, but if your depth is high quality then it shouldn't matter much

#

and I suspect it may actually be better in some cases because depth doesn't lose precision as you move away from the origin

obsidian wolf Dec 21, 2022, 1:51 AM

#

you'd need to work in some local space to actually make those gains

#

and at that point, the g-buffer would probably have positions in that same space

main cosmos Dec 21, 2022, 1:59 AM

#

you could store view space position for slightly increased precision, but the extra 96 bits in the g buffer is a bruh moment

obsidian wolf Dec 21, 2022, 1:59 AM

#

yeah im not gonna do it

#

its not a problem i need to fix

main cosmos Dec 21, 2022, 3:32 AM

#

main cosmos it suffers from holes if your triangles are too small, unless you store a list o...

the other thing I forgor to mention is that you have to also use conservative rasterization to capture all the small triangles

obsidian wolf Dec 21, 2022, 3:46 AM

#

i probably have to in other places monkaHmm

#

didn't know that existed

obsidian wolf Dec 22, 2022, 6:12 AM

#

the way that i generate icospheres makes a pretty interesting pattern if culling is enabled

obsidian wolf Dec 29, 2022, 6:22 AM

#

this left my chat list ;(

#

christmas vacation moment

#

been working on cleaning up all of my assets to have the sort of "asset editor" working, kind of boring and tedious

#

it is annoying to get data set up properly to work for managing an animation library

#

ex; i have a Model component which has a ModelCPU (contains the hierarchy) and ModelGPU (contains the renderer items), the Model component is for moving data from the CPU side to the GPU side

there is also SkeletonCPU which has bones and animation data now, and SkeletonGPU which has the bones buffer pretty much, the ModelCPU has a SkeletonCPU, and then the ModelGPU has ptrs to the SkeletonGPU which also has a ptr to the SkeletonCPU for updating

the point of me explaining it is that it is overly-complicated, and that's after a bit of cleanup (future cleanup will probably be using the model component to move data from SkeletonCPU to SkeletonGPU in the same way as with models, rather than keeping the ptr on SkeletonGPU)

#

that cleanup point probably isn't actually true, it'll probably stay as is, Model can move data pretty nicely because ModelGPU is a vector of renderer items that can have their transforms set (i.e. it's pre-upload data), but SkeletonGPU is just a buffer, so Model or whatever equivalent would have to upload data and that is renderer side

obsidian wolf Dec 30, 2022, 7:42 AM

#

this is not bad anymore

#

actually feels very usable

dry monolith Dec 30, 2022, 7:53 AM

#

what happened to fps

obsidian wolf Dec 30, 2022, 7:57 AM

#

its debug + validation and the way im doing the skeleton is not friendly with that

#

still gonna optimize it a bit, but its gonna be like 70 max

#

uploading the skeleton debug mesh every frame, same with the pose widget mesh

#

actually, more accurately, i'm not uploading "the" skeleton debug mesh, i'm actually uploading a mesh for each bone ;)

#

thats the "optimize it a bit" part tho

main cosmos Dec 31, 2022, 8:26 PM

#

Leaked image of your game

obsidian wolf Dec 31, 2022, 8:26 PM

#

monkaHmm is that cereal

main cosmos Dec 31, 2022, 8:26 PM

#

I'm being cereal right now

#

Super cereal

obsidian wolf Dec 31, 2022, 8:37 PM

#

its so small

#

reptile moment i suppose

#

🙂

#

obsidian wolf Dec 31, 2022, 9:40 PM

#

ok, this is actually intended now

#

the animations suck but they are working!

obsidian wolf Jan 1, 2023, 2:05 AM

#

obsidian wolf Jan 1, 2023, 8:40 PM

#

hmmm, funny problem

#

if i want changing elevations, i need to figure out how the dragging (above vid) works across heights

#

the dragging works by plane intersection with the ray the mouse makes, but if the height changes the plane intersection changes

#

so i think i could either do some ray voxel intersection, which probably is bad, or i could use the same query that i use for starting it

#

difference is that'd be every frame and i still don't know the proper way of doing this

#

actually

#

i think i need to render a heightmap of the scene for a couple reasons

obsidian wolf Jan 2, 2023, 3:20 AM

#

obsidian wolf Jan 4, 2023, 12:06 AM

#

need to do animations for this, but i started on the code paths for basic abilities

#

my plan for abilities is to only allow targeting of squares at most precise, it massively simplifies a lot of gameplay code and gives an interesting restriction to work around

obsidian wolf Jan 4, 2023, 7:13 AM

#

meehehehe

#

is a LIL buggy with the dragging though

obsidian wolf Jan 4, 2023, 9:42 PM

#

flail on the old model

obsidian wolf Jan 6, 2023, 2:36 AM

#

she be castin

main cosmos Jan 6, 2023, 2:39 AM

#

awesome

#

your hard work do be paying off frfr

obsidian wolf Jan 6, 2023, 3:05 AM

#

🙂

#

the asset editor is specifically nice and thats not really demonstratable through the vids unfortunately

#

thats the part thats been the coolest, being able to open a model and make another pose quick or edit the particle effects feels fairly nice actually

#

i wanna do water soon

#

tired of black background

obsidian wolf Jan 7, 2023, 7:26 AM

#

i've got some ugly code

#

that i really want to make not ugly

#

but i have spent the time to make code like this not ugly before and it wasn't rewarding 🙂

#

we will see how long i can last

#

https://github.com/cajallen/spellbook/blob/7b4ae996af4e970e6b8fdf8dd7dde0b2d02be871/src/game/lizard.cpp#L65

GitHub

spellbook/lizard.cpp at 7b4ae996af4e970e6b8fdf8dd7dde0b2d02be871 · ...

C++ tower defense game . Contribute to cajallen/spellbook development by creating an account on GitHub.

#

this 150 line long switch statement of lambdas for abilities

#

currently it's only 2.5 abilities, so at ~60 lines per ability, that quickly balloons, obviously this shouldn't be in a switch statement with hard coded values like this

#

but the intentional thing is just to work in this for a bit more until i actually have a handful of abilities that actually give me a real direction to take it

#

rather than the hypothetical one that i want to replace it with now and have done in the past

#

it's funny though because it's such an intellectualization of it, like i have to remind myself that i haven't actually dealt with any friction at this point

#

it's not like i'm gonna write less code with some abstraction, really the only benefit i get is better system interaction and integration, which if i haven't done any game system interaction then i have no fucking clue of knowing what i need for that

#

tl;dr this is what peak code looks like 🙂

obsidian wolf Jan 9, 2023, 5:56 AM

#

animations still suck ass!

main cosmos Jan 9, 2023, 6:02 AM

#

those animations are froggers fam

obsidian wolf Jan 9, 2023, 6:02 AM

#

the effects help, but the actual skinning animation sucks

#

i just need to actually practice with it and try stuff that is overdone, i've spent hundreds of hours modeling but only a couple posing/animating so practice is needed

#

also i dropped the witch hat on this one because i think another will be more deserving and i don't want to have like 50% witch/wizard hats xD

dry monolith Jan 9, 2023, 12:09 PM

#

health bars literal blocks

obsidian wolf Jan 9, 2023, 5:22 PM

#

they used to have another mesh to give them a border but since the outline shader they look relatively fine at gameplay distances lmao

obsidian wolf Jan 12, 2023, 3:46 AM

#

https://twitter.com/OskSta/status/1590669875869286400

Oskar Stålberg (@OskSta)

An early look at grass billboards. I really like how they add a completely new kind of shape and detail frequency.

Likes

1967

Retweets

119

▶ Play video

#

i might look at this for grass

main cosmos Jan 12, 2023, 3:57 AM

#

interesting

obsidian wolf Jan 12, 2023, 3:57 AM

#

i was having a very hard time getting grass that looked at all decent

main cosmos Jan 12, 2023, 3:59 AM

#

I'm also interested in grass rendering suddenly due to my failed ld52 project

obsidian wolf Jan 12, 2023, 3:59 AM

#

assuming you mean actual grass, not stylized grass :^)

main cosmos Jan 12, 2023, 3:59 AM

#

ye

obsidian wolf Jan 12, 2023, 4:00 AM

#

tbh the main trick with grass is just aligning the normal to the terrain

#

once you do that it looks good

#

some fudging with SSAO and other shading can be important but like... relatively... it's just the normals

main cosmos Jan 12, 2023, 4:02 AM

#

the grass in valheim looks really flat if you don't have SSAO enabled

#

but when it is enabled, the grass becomes overblurred in shadowed areas (where SSAO is affecting the only lighting source)

obsidian wolf Jan 12, 2023, 4:03 AM

#

what about SSAO + no DOF?

main cosmos Jan 12, 2023, 4:03 AM

#

I always have DoF disabled

obsidian wolf Jan 12, 2023, 4:03 AM

#

gothca

#

#

on the left here?

main cosmos Jan 12, 2023, 4:08 AM

#

ye

obsidian wolf Jan 12, 2023, 4:08 AM

#

tbh it looks like some GI mess

#

its probably whatever the out of box unity solution is for what its worth

main cosmos Jan 12, 2023, 4:08 AM

#

SSAO disabled

obsidian wolf Jan 12, 2023, 4:09 AM

#

i doubt they have anyone doing rendering beyond unity toggling

main cosmos Jan 12, 2023, 4:09 AM

#

this game definitely doesn't have very complex GI

#

at most, it does something interesting for water reflections

#

they aren't screenspace, and you can see the reflections update occasionally

#

looks pretty ugly without SSAO lmao

obsidian wolf Jan 12, 2023, 4:11 AM

#

yeah but the first screenshot does not look like SSAO to me

#

well

#

i guess it could be

main cosmos Jan 12, 2023, 4:12 AM

#

it's hard because it seems like there is a lot of video compression

obsidian wolf Jan 12, 2023, 4:13 AM

#

trees are way harder than grass FWIW

#

giving trees any decent shading is really hard

#

though

#

if you're doing trees you also need to be an artist for a bit

#

and i'm not sure if you want to do that x)

#

i guess the way you'd actually say it is trees are kind of more tech art-y

main cosmos Jan 12, 2023, 5:00 AM

#

ye my trees are just cylinders with cones on them

obsidian wolf Jan 12, 2023, 5:01 AM

#

trees have: leaves which are a good bit transparent + there's a shit ton of bounce lighting + they're moving

#

honestly GP's worst nightmare even aside rfom modeling them

#

and ignoring modeling when talking about trees is ignoring a lot 😬

#

but do trees!!! catblush

#

leaves are basically just grass but on branches instead of dirt smart

obsidian wolf Jan 12, 2023, 5:07 AM

#

obsidian wolf trees have: leaves which are a good bit transparent + there's a shit ton of bou...

actually, i'm kind of ranting, but speaking of that comparison, i'm pretty sure a big part of why aligning the grass normal to the terrain works decently well is because it approximates the first two

obsidian wolf Jan 13, 2023, 3:13 AM

#

https://twitter.com/OskSta/status/1448248658865049605
more from him that's useful :^)

Oskar Stålberg (@OskSta)

More talk prep. Gonna have another go at persuading people to cut their tiles along the dual grid instead of the main grid. I genuinely don't understand if this is rare because people don't know about it or if there is some drawback I'm not seeing.

Likes

8081

Retweets

1037