Foundations - Learning Graphics Programming with the Zig Programming language

[elder kettle]

are the underlying framebuffers the same as here though

or is there some way you can change the attachments cheaply

I thought changing FB configuration stuff was expensive

[late jackal]

yeah afaik you might as well make a new framebuffer and configure that

[elder kettle]

Ok so you sort of cache all the configurations or something?

[late jackal]

in fwog I have a framebuffer cache but I never measured its effectiveness

[elder kettle]

alright

[late jackal]

I would wager that the operation that is actually expensive is changing the render target state

so configuring or reconfiguring a framebuffer is cheap, but the first time you draw with it since it's been bound is probably not

I'm allowed to ponder about perf like this because it's gl

[queen hinge]

I’m pretty sure it’s actually just binding framebuffers in general that’s slow

Regardless of if it’s the first time it’s been bound or not

[delicate delta]

the shadowmaps are flipped, but it doesn't matter during sampling, these look correct for the back light perspective and front light perspective

I have two issues to work through still which is the textures aren't being read correctly for bindless I get warnings that don't correspond to with what I see in Nsight, and the fragment shader code is not applying the shadows

I work through those two remaining issues and I should have multilight shadows

I can actually work back from what is working with my dolphin, which works, to figure both of those issues out

oh there we go

just had the rotation wrong

[delicate delta]

I don't know what I would do without renderdoc

I guess generate textures from shadowmaps manually

oh I should maybe do that anyway

see the shadowmaps in real time

[late jackal]

You can draw textures in imgui with Imgui::Image

You can make a walmart RenderDoc with it

[delicate delta]

nice

[zinc trail]

Belive it or not there is an alternative to it. It's not as easy as overriding a method though. You have to mess about with phantom references n stuff but it's still possible

[ivory verge]

and wiz a little swizzle magic you can make the reddish looking depfthsphf map, look as white as in renderdoc, and a bit of TextureView shmooh

[late jackal]

Yeah, alpha is a bit of a pain with the default backend

Maybe you could modify the backend to support extra stuff in imgui's image handle

Sorry, getting ahead of myself hehe

[zinc trail]

Oh yeah if your drawing your fbo attachments in ingui you have to flip the uvs

Also on the topic ealier too. If you depend on destructor order. Then you probably shouldn't store the object as a global

[late jackal]

Bjorn is using zig so I don't think it's a problem

[zinc trail]

That's one of the few cases where I'm fine with a singleton

[zinc trail]

I've hear good things a out zig. Just havnt given it a look yet.

[late jackal]

It seems comfy

[zinc trail]

I gotta figure out how to do this too

[late jackal]

I used zig once and unfortunately I needed to do a bunch of vector math, so the lack of operator overloading didn't feel nice

Everything else seemed cool

[zinc trail]

aww no operator overloading would probably make me feel like im back in the java days x.x

[ivory verge]

you can just pass those in the ImGui::Image() call no need to flip anything

[late jackal]

I think zig has some built-in vector and matrix types now so if that's some consolation

[zinc trail]

yeah theres params in the call that have the uvs, by flip i meant those two params

[late jackal]

But that's not as good as just letting you overload the dang operators

[wise imp]

that stuff is for simd

[queen hinge]

sounds janky

and like it's probably not worth the effort

[zinc trail]

its java, its all jank built on top of jank

[delicate delta]

I never used operator overloading in the past so I don't have a problem with writing more characters of code to do that
zig is overall more verbose than C or C++
I definitely like zig, but I am fine with relying on a very limited set of C dependencies and updating the compiler every day and dealing with any backwards breaking changes and the overall verbosity of having to explicitly cast and various C++ sugar that lets you avoid writing code explicitly

I am also fine with not having very good editor tooling etc

so if you are ok with not being able to depend on libraries, frequently chance of backwards breaking changes, poor editor tooling/tiny niche ecosystem, no C++ features then you may also like zig

[late jackal]

most of those issues are shared by all new languages

[delicate delta]

yes

it's similar to using rust nightly back before 1.0

[queen hinge]

I honestly kinda actually dislike operator overloading in most scenarios
ESPECIALLY assignment overloading

I sorta rely on keyword patterns to know what’s doing what
Using operator overloads sorta, breaks the keyword patterns I’m used to
And assignment overloads just drive me mad
If I’m gonna use =, I want it to behave exactly how = would
I don’t want to have clamping logic in my = operator or something, and I don’t expect it to write stuff to the disk

vector/matrix/tensor math are like the only things I like operator overloads for

[delicate delta]

I prefer imperative readable code, without hidden control flow

operator overloading is hidden control flow

it's why it's not supported in zig

it's one of the main goals of zig, so it will never be supported

[wild ember]

Well you have the operator right there! it's not hidden . Maybe I'm just to used to C++ but I always read an operator as a function call

I find Vector and Matrix math very hard to read without operator overloading

[elder kettle]

Yes it's terrible without

[wild ember]

We have a very old codebase at work that started in C, but then they added C++ on top of it, so some files have really messy C matrix math and some files are super nice C++ with operator overloading

you just roll the dice

[delicate delta]

I just concatenate my matrices one at a time each line, it doesn't really make a difference after it's compiled

I don't think it's hard to read at all

I could create a function that is concatenate and takes a slice of matrices

I don't really feel the need to

actually think that would compile worse

[late jackal]

I think banning operator overloading was not a good choice but I understand the rationale

Any feature can be abused

[wild ember]

theres no more guarantee that the function add() does what you think it does than the operator + does

[delicate delta]

idk, I don't have any problems multiplying matrices in GLSL with the math operators and the zig way doesn't bother me either, I did like how the stroustrup book overloaded the << or whatever operator for the book's calculator problem that you built as you worked through the book

I can see why it's useful, but I have had zero foot gun. moments with zig in a year now

it's the same with go, it's a very simple imperative language that doesn't offer enough for you to ever get lost with the question "how do I do this in this language??" for hours

which is the opposite experience I had with Scala for example

[delicate delta]

you do have the guarantee that the + never does anything you didn't expect though

have to flip the thought around there to understand the hidden control flow, of course functions can have side effects

operators can't

I don’t think zig is aiming for being a replacement for C++ anyway

More of C replacement

Which also doesn’t have operator overloading

I can’t speak for the core devs though or know their intentions and I don’t want to misrepresent them

idk use the language that works for you.

I sometimes think about how nice it would be to just write C++ with the huge ecosystem, tooling and editor support

If this wasn’t a hobby I would use C++

There’s zero career value for me doing any of this lol

[delicate delta]

I should just use a UBO for all this shadow data

Light and shadow data

[late jackal]

I use ssbos (basically ubos but infinite size and allow writing) for almost everything

[delicate delta]

I made that argument and people argued against it saying ssbos are slower

My little buffer abstraction makes using ubo or ssbo trivial

[late jackal]

ubos go through special memory on nvidia but it's really not that big of a deal

and on amd they are identical to ssbos

[zinc trail]

if your operator = does something you dont expect from an assignment, then you wrote bad code

[delicate delta]

maybe it's not your operator

maybe it's the person on the other team

maybe it's from a library

junior developers are a thing

[zinc trail]

i mean yeah, in most cases you shouldnt need to write your own assignment anyhow

[delicate delta]

and yes, bad code is also a thing

I write bad code

do you not write bad code?

[zinc trail]

i mean we all write bad code, anyone who claims their code is perfect are wrong

[delicate delta]

right exactly, and that's kind of explains zig's philosophy

no footguns

[zinc trail]

as far as i see it the only reason to have custom assignment in c++ is if you have custom move or copy

[late jackal]

I only write good code

[delicate delta]

afaik zig doesn't have a move, it's always a copy to do assignment

[late jackal]

yet no borrow checker smh

[delicate delta]

idk what move means I guess

[zinc trail]

in some circumstances it allows you to avoid copying which is good

[delicate delta]

as I understand it any time I do var foo = bar; whatever is in bar was just copied to foo and changing foo will never change bar

if those are storing pointers then you can change what both are pointing to by writing to their fields

[zinc trail]

if you have this in cpp then its usually a copy

auto bar;
auto foo = bar;

[delicate delta]

but the values actually stored by foo and bar are different

[zinc trail]

assuming you dont do weird reference shenanigans

[delicate delta]

what is an example of a move I guess? I don't understand

[zinc trail]

std::string a = "string";
std::string b = a;

so lets say we have these strings

so as expected atm, a gets copied into b

so lets say you know for a fact that a doesnt get used after you copy it to b, this is a situation where you can do a move instead and avoid the copy

[delicate delta]

so a is now not usable?

[zinc trail]

in the example its perfectly usable

std::string a = "string";
std::string b = std::move(a);

[late jackal]

moving semantically "steals" the data from an object. in this case it means a now holds an empty string and b now contains the pointer to chars that a previously had

[zinc trail]

here we can tell b to move a instead of copying

[delicate delta]

why is this useful

[zinc trail]

optimization

[delicate delta]

why not just use a

[zinc trail]

moves are fast, copies are slow

[late jackal]

that's not the main reason

moving lets you express certain semantics that don't exist with only copies

some objects are movable but not copyable, for example, because copying them wouldn't make sense

[zinc trail]

or in the case of many opengl abstractions, where copying the object a lot of the time doesnt make sense

[delicate delta]

how is this different from just using pointers though

[late jackal]

std::unique_ptr enforces single ownership of the pointer it holds by banning copying

so it will only be automatically destroyed in one place

[delicate delta]

you could do b = a; a = undefined; now b has it

and a goes out of scope or whatever, I don't know, set it to something that doesn't cause UB

I kind of get it, sorry

this is like ownership semantics

[late jackal]

yes, exactly

[delicate delta]

rust features ported to C++

[late jackal]

zig doesn't have destructors so it doesn't make as much sense

[delicate delta]

std::cow when

[zinc trail]

that and it lets you keep your value types working as youd expect without heap allocating everything

[late jackal]

it used to be legal to implement std::string with 🐮 semantics

[zinc trail]

like if every object you have behaves like a shared_ptr then yeah moves are almost pointless, ahem... java

[elder kettle]

Move semantics let you describe things that have a unique and uncopyable identity

It's usually not necessary to implement yourself but it's very useful when it is

[zinc trail]

falls into that category of if you implement a destructor you should implement the copies and moves too

[late jackal]

rule of 0 or 5

[zinc trail]

yeah i see way too much newbie code define destuctors that do nothing and just casually makes their code that much slower

[late jackal]

imo a much worse problem is ignoring the rule of 5

[zinc trail]

Heres some cpp guidelines that cover this very well

https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#c20-if-you-can-avoid-defining-default-operations-do

https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#c21-if-you-define-or-delete-any-copy-move-or-destructor-function-define-or-delete-them-all

[late jackal]

noobs make opengl abstractions with completely broken copy and move semantics

[delicate delta]

I kind of never transfer ownership of anything unless I'm doing a job on a thread

[zinc trail]

i legit have this issue come up on the server enough to where if i see opengl code breaking from a mnewbie and i see abstractions, i point em to those rules since thats more often then not whats breaking em

[delicate delta]

at least I haven't needed to

actually I do some initialization in the beginning of my app and then pass that into another thing and that thing becomes the owner, and some zig std library will allocate memory you then have to free, but you pass in the allocator for that thing so if you use an arena it's all handled

and if you don't, you just know to handle it, the general purpose allocator in dev detects leaks, UB is totally possible though if you get it wrong

[zinc trail]

I recently went though a code base where the abstractions were so bad.

Some things were default init in the class, some of it happened on the constuctor, some of it happening inside of it, and there was like a cube class which pushed into a vector of a buffer class with broken semantics and was raising exceptions from the driver, its great

[elder kettle]

Imagine using abstractions

[zinc trail]

the thing that was blowing my mind is that he had some classes that deleted copies, so they knew they could tell it to not allow copies

but deleted copies on a class that was never in a situation where it would ever be copied to begin with

[zinc trail]

write everything in a hex editor then, you dont need all these fancy things like assemblers :>

manually connect the wires many billions of times every second

[elder kettle]

Don't give me that slippery slope bs you know what I mean

[zinc trail]

im just saying it starts with saying you dont need abstractions, meanwhile ur working on top of so many abstractions

i can think of at least 20 abstractions almost every one of you are building your software on

[delicate delta]

everytime I want to do something with my engine any stupid abstraction I wrote before I have to undo

I'm going back to copying and pasting code

[zinc trail]

i usually end up trying stuff and if i hate it i just git stash it lmao

[delicate delta]

I was like ok I have lighting now let me abstract it so every scene can have lighting without me having to do all this all over

and then oh, I want to have multiple lights and I want those to cast shadows

and then it's like jfc what I have done

this is why I am going to make a game with static geometry and small levels

[zinc trail]

yeah its hard to make good abstractions when your needs change a lot

[delicate delta]

every level will just be artisinal shaders and one use everything just for that level

I will do absolutely as little as possible to reuse code across levels

because it just becomes a fucking straight jacket

[zinc trail]

a lot of my higher level abstractions are tied into the ecs system so its super easy to iterate on it

[delicate delta]

ecs is cool, but I'm not going to use one

[zinc trail]

my light struct for example just has a type, a color and attenuation factors

[delicate delta]

yeah I have a light and material struct

that's a data type

that's not an abstraction imo

idk

[zinc trail]

thats fair tbh

i just mention it cause a lot of time abstracting tends to lead to having these massive classes and all

[elder kettle]

ECS is cool and you should use it

[delicate delta]

would you write Doom 1 with an ECS

[zinc trail]

sometimes an ecs is overkill, but if youre making something complex enough its absolutely worth using

[delicate delta]

because that's the kind of game I'm basically going to make

small puzzle escape room levels

[zinc trail]

if you know what an ecs is, then usually if you're good enough you'll know when its time to use one

[delicate delta]

I used one with my voxel game

it was great

[elder kettle]

Yes I would

[zinc trail]

i havnt brought ecs into my voxel stuff yet, but i do plan to bring it in eventually

[delicate delta]

that's fair, I could see using an ECS too

but the main reason that I don't use an ECS is just there's no easy way to port any of the C / C++ ones to zig because they do so much meta programming

[zinc trail]

i technically already have the ecs library in the codebase, i just dont use the ecs features of it

[elder kettle]

You can make one out of static arrays

[delicate delta]

but that's kind of overkill for a small game

[queen hinge]

Abstraction is good in like
1-5% of scenarios in my experience

And I’m a java person

[delicate delta]

I can just not use an ECS and be fine

[zinc trail]

yeah good luck porting the weird c++ template stuff into another lang

[delicate delta]

yeah

or complex C macros that flecs uses

[zinc trail]

entt specifically is pretty heavy on templates, im not sure if any ports of that exist

[delicate delta]

there is a zig port of entt

I don't use zig dependencies

[elder kettle]

ECS can be overkill but that doesn't mean it's bad

[delicate delta]

I said ECS was cool though, I never said it was bad :P

[zinc trail]

nobody here was calling it bad tho?

[elder kettle]

Well you said you wouldn't use it for something like doom 1

[queen hinge]

What is ECS

[delicate delta]

entity component system

[queen hinge]

oh like what unity has?

[delicate delta]

yeah probably

[elder kettle]

No

Unity has a shit version

[queen hinge]

[zinc trail]

conceptually yes

unity does do some weird crap with it tho

[delicate delta]

high level you have entities and they have components and you query components in systems that have side effects/behavior

[elder kettle]

ECS is just splitting your game into pure state and pure logic and creating game objects by composing types of state

[delicate delta]

implementation detail is it's a bunch of arrays of arrays

[elder kettle]

ECS is great for any problem where objects are combinations of smaller things rather than fitting into completely separate categories

[queen hinge]

oh
the performance implications of that offend me

[elder kettle]

No the performance is amazing

[delicate delta]

ECS are fast af

[zinc trail]

thinmatrix has a decent older video about the general idea of it, gives a decent enough visuals

[delicate delta]

small amounts of collocated data queried sequentially goes brr

[elder kettle]

Yea

[zinc trail]

while not really talks about ecs, it goes over why giant class heiarchies are just next to impossible to manage over an ecs

[queen hinge]

I’m the sorta person who will get annoyed about something taking nanoseconds too long

[zinc trail]

implement an ecs correctly, they will easily outperform, dare i say nearly all other methods (of grouping entities and behaviors)

[elder kettle]

ECS is basically cache optimized by default

To make it maximally optimized it takes more work but the naive implementation is like 90% of the way there and faster than most anything else you'd come up with using other architecture paradigms

[queen hinge]

Yea class hierarchies are painful

[zinc trail]

yeah in that video his example was like 7 subclasses deep

and i assume it was based on his real code at the time

[queen hinge]

my class hierarchies usually only go like 2-3 deep

[zinc trail]

i mean it was a java codebase and when its so easy to make a subclass, i cant exactly blame him for using it

[queen hinge]

I mean my stuff is also java, so I have the same conveniences as him for that

[zinc trail]

tbh the only part of java i miss is easily embedding assets

oh and their standard library

[queen hinge]

The standard library is something I wish I could ditch most of the time

[elder kettle]

I don't use classes at all

[zinc trail]

yeah but compare it to c++'s standard library annddd yeah

[elder kettle]

Or almost at all

[zinc trail]

^ c functional purist?

[elder kettle]

No I use C++

I just don't use classes

Yeah mostly procedural code

[zinc trail]

i hope you actually mean classes and not just the class keyword lmao

[elder kettle]

Yes I don't use member functions

[queen hinge]

I struggle to find something that is suitably verbose, with a suitably large collection of libraries made by users of the language, that won’t explode because I looked at it too hard

[zinc trail]

i suppose, i was always able to just throw the files into the classpath and java just sorta knew about them

and majority of the time it will happily load those with inputstreams

[queen hinge]

Java’s dependency management environment and ecosystem is pretty nice
Need something? Someone’s made a library for it
One of the highest compression ratio png encoders I’ve seen is a java library

[zinc trail]

i never got crazy involved in the tooling but what i have used is very nice admittedly

[queen hinge]

If not actually the highest compression ratio png encoder I’ve seen

[queen hinge]

Gradle is a dumpster fire but compared to some other tool systems it’s great

[zinc trail]

the only "proper" tooling ive used for java is gradle (which i hated) and maven

meanwhile theres no c++ tooling out there that i enjoy or know of

[queen hinge]

Those are about the only ones that are still used nowadays as far as I know

[queen hinge]

[zinc trail]

i still maintain some java stuff and it uses maven and i dont have much to complain about there

[queen hinge]

I… don’t like how maven is xml based
I also don’t know if it supports actually scripting your builds

[zinc trail]

eh i could care less if its xml or yaml

just please we gotta get away from using json for everything

[queen hinge]

but json is nice… so long as you don’t need to look at a large file

And yea yaml is a solid choice for human facing data formats

[zinc trail]

json is NOT NICE, who told you that, have you everrrr tried working with it

oh toml is aight too

[queen hinge]

I have written jsons manually
Many times

[zinc trail]

and you enjoyed it?

you're insane i tell you

[delicate delta]

I am getting so good at renderdoc

[queen hinge]

oh no I didn’t say I enjoyed it, but json is not meant to be hand written truthfully

[delicate delta]

it's like boom snap, got my data, confirmed it, move on

like I get a capture and see the output in like 2 seconds

no more of this clicking around

[zinc trail]

love seeing people learning renderdoc, its like ur moving on from grade 1

[queen hinge]

json is meant to be used by the computer
… or at least that’s how I feel about it
I mean json doesn’t even support comments

[zinc trail]

offspring of the devil, javascript

[queen hinge]

gson is also a very convenient thing to use

[zinc trail]

oh that brings me back, i remember using that, but after writting yaml, im never choosing json for my own projects

[queen hinge]

Oh wait

[zinc trail]

anyhow ima clock out of this chat since this is convo is taking away from bjorn's stuff here

[queen hinge]

https://github.com/GiantLuigi4/Smaller-Units/tree/1.19.2-mutliloader/Common/src/main/java/tfc/smallerunits/utils/selection

https://github.com/GiantLuigi4/Smaller-Units/blob/1.19.2-mutliloader/Common/src/main/java/tfc/smallerunits/utils/selection/UnitShape.java
sanity is not a question when talking to me

[zinc trail]

ofc its minecraft related smh

[queen hinge]

I’m the sorta person to casually reverse engineer a system to go implement something more akin to an older version of the system into the newer one for performance reasons

[delicate delta]

ok I have found the most optimal positioning of my lights and objects to test my shadow maps in all 360 degrees

that gets me all four sides and up and down

[queen hinge]

[delicate delta]

I got ubos, I got a good 360 I can fix all the shadows, get shadows working with texture binding and then figure out my bindless problems

this is a plan

I am making progress

someday this will work

just my up and down are fucked

bam

fixed

[queen hinge]

So there’s this optimization you can do
Where you only update shadow maps every N frames
Might also be able to do some reprojection for slow moving light sources

[elder kettle]

I wouldn't bother with such things until your game is bottlenecking

[delicate delta]

guys

I think I should go pro

[queen hinge]

yea

[delicate delta]

I really think so

[queen hinge]

though I know for a fact that my game will bottleneck on lights

[delicate delta]

ok

now to get shadows working with non-bindless, then I can fix bindless

[queen hinge]

[delicate delta]

then, if that is possible, smooth shadows and I'm done with this scene

hrm I just do it every frame right now, I'm not worried about performance at all with these learning scenes

just getting it working

I bind 12 framebuffers every frame

https://tenor.com/bQ6Vf.gif

but I should totally go pro I got what it takes, obviously

back to it

[queen hinge]

Hm actually maybe I can avoid that

[queen hinge]

Makes sense

[delicate delta]

well a lot of shit is fucked, but it kinda works too

[delicate delta]

a ridiculously high glPolygonOffset seems to have made the worst of it go away

[late jackal]

you should add a floor to make shadowing issues more apparent

btw shadow mapping is a constant battle between too little bias (causing shadow acne) and too much bias (causing disconnected shadows/peter panning)

the worst case is when the incoming light is almost perpendicular to the surface normal, as you observed. you need an infinite amount of receiver bias to deal with it

[delicate delta]

yes, I searched this server for glPoygonOffset and found where you had discussed it previously

#graphics-techniques message

I found that diagram really helpful, thank you

although in that case they were using PCF and I am not right now

searching this server for stuff is more helpful than asking google or ai btw

[delicate delta]

what does receiver bias mean?

[late jackal]

one technique to alleviate acne is to simply offset the depth of the receiver (the surface being shaded) in the fragment shader

[delicate delta]

oh yes I tried that

it did nothing

I don't know what values to use for these things

I don't understand the math. I read and I don't really understand it either

maybe because it's late

[late jackal]

glPolygonOffset is different from a receiver bias btw

[delicate delta]

I saw a really nice visualization deccer shared using glPolygonOffset

[late jackal]

but the receiver bias might not have worked for you if you didn't have a factor based on the dot product of the normal and the light

[delicate delta]

oh I see

[late jackal]

at grazing angles the receiver bias approaches infinity

[delicate delta]

I see

[late jackal]

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

[delicate delta]

my math books go into shadows

I haven't gotten to those chapters yet

so if you do a dot product you are getting the projection of the light onto the surface normal

[late jackal]

well we're just using it to get the cosine of the angle

[delicate delta]

right because they're both normalized vectors

oh I can drag the control point in that visualization

I'm going to smooth out the shadow and fix bindless and I think I'll call it on this, I have more to learn about shadows for sure

[late jackal]

nice progress, you're learning really quickly

[delicate delta]

thank you

this is probably the hardest thing I've done so far

[zinc trail]

What exactly is glpolygonoffset for? Or some use cases of it?

Oh nvm. I found a discussion here. Ignore me

[delicate delta]

I may be wrong but it offsets the depth buffer value to be closer to the camera so that on sampling there’s less chance of a precision error where we sample behind the surface and erroneously draw a shadow

[delicate delta]

Are you not using one for your cascading shadowmaps? And that works?

I think maybe my high 4K resolution is part of the problem

[zinc trail]

I'm not using it. Never seen it covered. But as far as ive been reading it's main use cases are shadows and decals

And the function itself can offset polygons based on their normals if I read it right

[delicate delta]

I literally cannot have decent shadows of any kind without it so no idea how you aren’t using it

[zinc trail]

I've just manually calculated biases based on normals

But on the recieving end. So same sorta effect but slower id imagine

[delicate delta]

I think that’s a different issue

[zinc trail]

Wouldn't the bias issues be improved by offsetting the polygons in the shadow pass?

[delicate delta]

I sample inside the mesh, it is similar to z fighting

It’s a precision issue

[zinc trail]

Oh I havnt really had precision issues

Not yet at least

[delicate delta]

I have the normals angle problem too though

Is your code open source?

[zinc trail]

Only certain parts of it. The shadow stuff isn't public rn

[delicate delta]

You’re using textProject with a bias based on cosTheta?

[zinc trail]

No idea what textProject is

[delicate delta]

How are you sampling?

[zinc trail]

But the bias is based on the dot product from the light direction so more or less cosine

[delicate delta]

You are you doing perspective division in your shader code?

Oh your shadows are orthographic

[zinc trail]

I'm rendering stuff to the shadowmap as is. No offsets. And when it comes time to shade objects I transform the fragment into light space. Sample the shadowmap and bias the depth based on the dot project of thr normal and light direction

And if it's in shadow. I don't do thr lighting calculation.

This skips over the technical detail of cascades and pcf filters etc

[delicate delta]

Oh you bias the output of the result of sampling bias cosTheta as in sampleResult * cosTheta? + ?

the dot product of light dir and surface normal is cosTheta

It’s used by your light code too I imagine

[zinc trail]

The bias is basically that yes and it just gets applied to the current depth

[delicate delta]

That doesn’t do anything for me I tried that

[zinc trail]

The lights don't use it at all. It's only used by objects recieving shadows

[delicate delta]

What lightmodel do you use?

[zinc trail]

A modified Blind phong

[delicate delta]

blinn phong and phong use cosTheta

[zinc trail]

I'm not sure if it exactly has a name but it's based on that

[delicate delta]

Are you doing a dot product in your light code

[zinc trail]

Yeah :p

It uses a wider range of the product though instead of only the 0 to 1 range

[delicate delta]

Does it happen to be dot product of the light direction and the surface normal

All my vectors are unit vectors

[zinc trail]

I mean that gets calculated but that's not what's used in thr final calculation

Dot product. Remap the range then clamp

[delicate delta]

Are you using an ortho projection to generate your shadowmaps?

[zinc trail]

Ambient light is handled a bit differently then blind phone too but it isn't significant

It'd a directional light so yes ortho

[delicate delta]

Ambient light doesn’t consider light or view direction

[zinc trail]

But its part of the light model

I'm not seeing how this is relevant to the shadowmap stuff

[delicate delta]

I think that’s the difference I use a perspective projection for both directional and point light shadowmaps

It results it in perspective foreshortening

That makes a huge difference?

[zinc trail]

Yeah in ortho that isn't a big issue

Not an issue at all afaik

The challenge with getting ortho looking nice is biases and very carefully calculating shadow boxes

[delicate delta]

In perspective projection further away objects are smaller and have fewer fp precision

This is a problem with any perspective projection

[zinc trail]

Yeah in ortho you still perspective divide but it doesn't cause foreshortning

[delicate delta]

The range of precision in far plane is bad unless an inverse z matrix is used

[zinc trail]

With point shadows I could see that being a problem

[delicate delta]

Which requires a flipped depth buffer

This all makes sense now

I use perspective with my directional light

Because that’s what the book did

[zinc trail]

That's weird

[delicate delta]

It’s hard

[zinc trail]

Cause directional lights are considered parallel so a frustum doesn't exactly make sense for it

But apparently you can make it work

[delicate delta]

Your ortho near plane is how far from your ground?

[zinc trail]

It moves with your camera frustum so it isn't based on where the ground is at all

[delicate delta]

But where is the near plane? Just the top plane of your view frustum?

[zinc trail]

Basically it calculates a box to fit the frustum. We do some extra calculations to fit it in a certain way and we move the z near plane back some distance to make sure shadows are captures from stuff behind the camera

But if you think about the scene from the pov of the light. You're visualizing a near and far plane.

#1286757925863424063 message

I have this lil gif I make that actually previews the shadowbox

[delicate delta]

Cool thanks for explaining all that

[zinc trail]

The problems I'm currently having is bias issues and some temporal stability issues

And pcf issues but i could care less about that rn

[delicate delta]

I an going to duck these issues with small levels with static geometry lol

[zinc trail]

But to make shadows look better for the sun. You basically divide the frustum up into sections and fit multiple shadowmaps into it

And you do cascades with it. Which makes a huge improvement for thr same resolution shadowmap

[delicate delta]

Yes makes sense

[zinc trail]

If you're light source are static you can get around almost all temporal stability issues

[delicate delta]

Do you use multiple passes and multiple framebuffers?

[zinc trail]

It's one fbo. 4 passes

Basically one depth buffer is split into 4 sections. And using glviewport I draw into the 4 sections

[delicate delta]

With multiple attachments on the fbo or you change the attachment each pass?

Oh

[zinc trail]

Yeah just clever use of the depth buffer

This method was targeting older hardware so there's probably a better way to do it

[delicate delta]

I need to read the spec on how to use multiple attachments

[zinc trail]

But it means the only state that changes between shadow passes is the viewport

[zinc trail]

It's not too hard to do

[delicate delta]

You’re working with others on your project?

[delicate delta]

I am using 12 fbos atm

[zinc trail]

I am but I'm the main one working on code

I have someone who helps with Linux compatability improvements in code sometimes but besides that the code base is my thing

[queen hinge]

Yep

Have not heard of it being used in shadow maps but yeah it offsets the poly in the depth buffer

[queen hinge]

I… think it’s just a flat offset, don’t think it cares about normal vecs

[wild ember]

Not their normals, but their slopes. There are 2 parameters the function, one is constant and 1 is slope dependant

This function is also commonly used in modeling applications, let's say you wanted to draw an overlay of edges on top of a solid model

[delicate delta]

I'm totally unsatisfied with my understanding of glPolygonOffset and light incident bias so instead of rushing this scene to be done I am going to create a scene just to visualize glPolygonOffset and also add some visualizations to my shadows scene to understand the light direction angles on surfaces

I am unsatisfied with pasting some code that fixes problems in a magical way I don't understand, when I can just spend a little bit of time to understand them and learn how they change rendering

reading the spec for glPolygonOffset is just incomprehensible babble because I guess I'm too ignorant

I am confident that the subsequent shadow and lighting related chapters in my math books I haven't yet read cover this

but glPolygonOffset and textProj are like openglisms

[wild ember]

Yes, but there are equivalent isms in d3d and vulkan

[delicate delta]

that makes sense

there's so much I had to learn for this chapter, first time using fbos, first time using more than one rendering pass, first time using depth sampling, first time using attachments

[wild ember]

https://learn.microsoft.com/en-us/windows/win32/direct3d11/d3d10-graphics-programming-guide-output-merger-stage-depth-bias

This is the same concept so you may get more insight here

[delicate delta]

this is actually a really nice explanation

If an application renders a wall first and then a shadow
this is a curious statement though, because the wall's fragments would change the color in the way I'm doing it, I'm not separately rendering shadows

[wild ember]

Yeah don't worry about that, depth bias can be used in multiple ways

Decals for example

[delicate delta]

DepthBias and SlopeScaledDepthBias are such nicer names than gl's "factor" and "units"

[wild ember]

Basically yeah, same thing

[delicate delta]

I could call anything factor or unit, it's like calling things type and item

[wild ember]

This is quite an old feature of GL, im not surprised it has jank naming

[delicate delta]

appreciate the link and context, thank you

I'm just coping

[wild ember]

Even vulkan has better naming https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/vkCmdSetDepthBias.html

[delicate delta]

yes

reading the spec actually makes more sense now too

I'm going to make a scene to visualize what it does

[queen hinge]

opengl is a library built on deprecation

[delicate delta]

we should all be going 4723h54m37s's route and just write assembly renderers

[queen hinge]

Using vulkan as the api

[delicate delta]

assembly to spirv when

[queen hinge]

I might die before I finish a hello triangle if I tried that

[delicate delta]

it would probably be a bit of assembly

[queen hinge]

The amount of C ABI would be lethal to write

[zinc trail]

Opengl is a specification built on depreciation? (Aka compatability)

Yall are being a bit extreme

[delicate delta]

OpenGL is in fact a modern API with a bright future that everyone in the ecosystem is excited to build with and build additional tooling and drive support.

it's the tip of the spear

the best possible choice

[elder kettle]

Expecting APIs to be in active development is web brainrot imo

GPU drivers will support GL 4.6 for longer than GL has existed for sure

[delicate delta]

yes

[queen hinge]

[delicate delta]

glPolygonOffset

it's just a depth buffer manipulation

so it's useful for decals to render something on top of a surface by adjusting its depth buffer so it doesn't end up partly inside the object

I guess this is how you can build a wallhack type deal outline too

so stencil a mesh to produce an outline, then change its glPolygonOffset so that it appears through other objects

it's interesting how I need a 20k range abouts to see it work

that's I guess related to my clip matrix

I'm not sure

that's the maximum depth slope above

[zinc trail]

this is fucking with me

[delicate delta]

I think it's probably more instructive to look at renderdoc than look at the visuals

[zinc trail]

have interactive stuff usually explains stuff to my brain better

[late jackal]

glPolygonOffset is funny when the depth buffer is floating point

and by funny I mean horrifying

[delicate delta]

I don't use floating point for my depth buffer attachment

I saw someone say not to do that

[late jackal]

yeah, just saying

[zinc trail]

i might be missing something but arent depth buffers always float?

[late jackal]

floating point depth is good for your main camera though. it lets you do infinite reverse z

[late jackal]

they are either unorm (fixed point) or float

you should use unorm for shadow maps

[zinc trail]

ohh i didnt realize, DEPTH_COMPONENT24 was a unorm thing

[late jackal]

opengl has bad naming for formats

[zinc trail]

i should have have known this ages ago since i know about the depth float formats

huh good thing i rarely if ever use the floating point depth formats then

[late jackal]

it does have a place as I mentioned

[zinc trail]

ive done infinite z with the unorm format but i probably didnt pay close enough attention to see the problems

[delicate delta]

is the only thing I see in renderdoc that's different

the data is otherwise all the same

[zinc trail]

i mean the gl state yeah makes sense, you should have different stuff written into the depth buffer then no?

[delicate delta]

oh yeah

[zinc trail]

this reminds me i should be able to make my shadow passes faster with the multi viewport stuff

[delicate delta]

multiviewport stuff?

[shy cosmos]

no

this is a common misconception

there is nothing faster about multiviewport

[late jackal]

the fastest way to draw a shadow map is to draw nothing in fact

[shy cosmos]

fr

[late jackal]

(you can cache shadow maps)

[shy cosmos]

yeah the biggest perf givers are cache at n1 and occlusion culling at n2

LOD at n3

[late jackal]

VSM at n0

[zinc trail]

okay assume i have to draw the shadow map, rn i draw everything in the shadow map 4 times, wouldnt i be able to just draw it once instead if i use the multiple viewports?

[shy cosmos]

VSM is too powerful

[delicate delta]

what are n1,n2 and what is VSM

[shy cosmos]

sure, are you actually saving anything? no

[late jackal]

the amount of work on the gpu is the same

you might save cpu overhead though

[shy cosmos]

even then it's just 4 indirect calls vs 1

you only get a measurable CPU speedup if you're doing basic glDrawArrays

[late jackal]

n = number
VSM = virtual shadow maps (or variance shadow maps if you're a pleb)

[delicate delta]

I am a pleb

[late jackal]

I'll allow it

[zinc trail]

was gonna say this is basically what im doing atm, i didnt get into stuff with indirect draws

[late jackal]

variance sm is not too difficult to implement

[delicate delta]

n1 means number of shadow maps what are we counting?

[shy cosmos]

you still don't want to do multiview because eventually you're going to go indirect

[late jackal]

it's a ranking

[shy cosmos]

multiview can only save you so much

[delicate delta]

oh tier list

[zinc trail]

so basically just do the 4 indirect draw calls and change my viewport between them

[shy cosmos]

yes

[zinc trail]

fair enough i suppose

[shy cosmos]

or even just have a layered shadow map if you're not doing fancy stuff

[delicate delta]

I thought this was a graphics jargon thing and it's just a reading comprehension thing

[zinc trail]

i would use the layered rendering stuff but i dont think theres a way to set the layer in the vertex shader thats in the core spec

and i really dont want to have a geometry shader for that

[shy cosmos]

yeah gl is sad like that

have you heard of our lord and savior

[zinc trail]

eh swapping to vulkan for me would slow me down in terms of development time, plus i just simply dont need the performance gains vulkan provides

i really should just a better method to do shadowmaps though

[shy cosmos]

the better method is glMultiDrawElementsIndirect

[late jackal]

erm how do you render to multiple shadow maps

[shy cosmos]

for shadowmap in shadowmaps glMultiDrawElementsIndirect();

[late jackal]

wahoo

[zinc trail]

its basically just fitting multiple shadow boxes to the furstum and its just a couple small things to help looks

its nothing fancy and im not at all vested into using it forever

[elder kettle]

You could make your transformation matrix and indirect buffers such that it renders the whole shadow map atlas as a single scene

[delicate delta]

yeah I'm just drawing in a loop to do my multiple shadowmaps atm

[zinc trail]

it was just my introduction to getting shadows working and i crave better shadows lmao

[delicate delta]

not sure if there was a better way

[shy cosmos]

have you heard of our lord and savior VSM

[zinc trail]

vsm and pcss are two things ive been wanting to get that workin

[shy cosmos]

it's only 15 bababillion steps to get up and running

or 2 if you're cool like us

[zinc trail]

oh and shadow samplers, i keep seeing them in the spec but never tried them

[shy cosmos]

they're just free 4 tap pcf

not too useful

[zinc trail]

oh i aint missing out then

would the gpu gems 3 about summed area vsm be a good place to start?

[shy cosmos]

I don't like the bad vsm

light leaking and self shadowing errors are nasty with bad vsm

[delicate delta]

I use that

that's what I mentioned before

[shy cosmos]

yea in shrimple scenes it's fine

[zinc trail]

i mean good luck avoiding artifacts with any shadow map method

[shy cosmos]

but with high depth complexity you just die

[shy cosmos]

uhm

have you heard of our lord and savior VSM (the good one)

[delicate delta]

sampler2dshadow

[zinc trail]

i have but nobody suggested a good resource for me to learn about it

[delicate delta]

it's like a one line change

[shy cosmos]

eventually it'll come out

[delicate delta]

textProj

[shy cosmos]

but for the time being #1168692074447642664

the VSM team is working on it

[zinc trail]

and what about completed stuff that i can follow, discord threads are horrid to navigate

[shy cosmos]

go read this https://therealmjp.github.io/posts/shadow-maps/

[late jackal]

virtual shadow maps not found, useless article smh

[zinc trail]

ill take actual articles about stuff over stuff that isnt documented well

[shy cosmos]

we made VSM with our blood and tears

remember when we were struggling to understand how to go from virtual UVs to physical UVs Jaker?

hold up I gotta find that conversation

[late jackal]

there were many challenges

[delicate delta]

sounds like niche stuff that is unecessary

[late jackal]

I originally proposed rendering each page as a separate bindless texture

[shy cosmos]

#1090390868449558618 message

it's almost been one year since VSM

how time flies

[delicate delta]

oh so the VSM is making one gigantic texture and then using lookups to index into it

this avoids additional passes?

hrm

[late jackal]

it's about applying the concept of virtual memory to textures (so you have virtual textures) and then using that to minimize memory consumption

and it will consume basically a constant amount of memory no matter how many shadow cascades you have

[shy cosmos]

https://zhuanlan.zhihu.com/p/488930416

the random chinese articles about VSM

remember those?

[late jackal]

xiao hong zhu

[shy cosmos]

https://tenor.com/view/chili-sauce-very-good-chili-sauce-lao-gan-ma-john-cena-gif-23277520

[late jackal]

lao gan ma

[shy cosmos]

need some lao gan ma on my vsm

[delicate delta]

memory consumption or memory allocation? how does virtual memory reduce consumption, reducing memory consumption would involve compression? sorry if this is a dense question

[shy cosmos]

it achieves better resolution with less memory overhead

so for example a classic VSM setup is to use 16 shadow maps, all of which are 4k in resolution

[late jackal]

in the context of shadow maps, you can analyze the visible scene to see which pages from which shadow maps need to be allocated

[shy cosmos]

if those are D32_SFLOAT shadow maps they would consume 16 * 4096^2 * 4 bytes of memory (1 GiB basically)

[late jackal]

being able to allocate individual pages (not whole shadow maps) gives you the granularity to minimize memory usage

[delicate delta]

yeah I allocate a lot of memory with my 12 4k shadowmaps

[shy cosmos]

yup, in the classic case you allocate only 64MiB to achieve 16 4k shadowmaps

[delicate delta]

that's incredible

hrm

I don't understand it though

[late jackal]

it's complicated

[shy cosmos]

https://tenor.com/bUDAr.gif

[late jackal]

here's a cool presentation about shadows
https://research.activision.com/publications/2021/10/shadows-of-cold-war--a-scalable-approach-to-shadowing

consider these things a curiosity rather than something you need to implement immediately

[delicate delta]

gotta be cool to have an entire team of graphics programming experts building just the shadow maps for your game's rendering engine

[shy cosmos]

they could've called us smh

[delicate delta]

they couldn't afford it

so what's cool

in renderdoc is you can hover over the pixels in the texture map

and get the actual values thank god

and see the math

the output from the math

but maybe it's deceiving because it's a float

hrm

I'm trying to understand what I should learn from this glPolygonOffset visualization

the constant I understand

the slope is harder to understand

by changing the factor parameter, the angle of the surface changes its position in the depth buffer

the factor parameter changes the position of a surface in the depth buffer given the surface's slope

so if I increase the factor the angle of of a surface results in the surface being given a greater distance value in the depth buffer

so factor * angle + r * unit = offset

m is the slope of the angle

I get it!

hrm

and this is useful for reducing shadowacne because I can use it move the surface further into the depth buffer so that when I sample the surface as part of the render pass it will appear closer to the light than what's in the depth buffer

and this can cause peter panning because the actual visible surface is now at a non-zero distance from what's in the depth buffer

[delicate delta]

hrm

so the reason I saw artifacts

was because it was self shadowing

and I have to offset it just enough so that when I sample it the surface that actually is visible to the light doesn't test as being a more distant value than what's in the depth buffer

[delicate delta]

I'm going to add these sliders to my shadows scene

that seems good

except when there's movement

hrm

ok

that shadowing while movement happens kind of makes sense though

that's also self shawdowing

the geometry is out of sync with the shadowmap for a frame or something

I don't know how to fix that

let me just skip a frame?

that was a terrible idea

so

I fixed it

I now generate all 12 shadow maps twice in a frame

before a draw and after a draw

but only if the light and objects moved

setting the polygon offset and the full depth pass twice a frame when the lights or objects have moved

jfc that looks nice and smooth

[late jackal]

Factor: increase depth offset as the angle between the camera and surface normal increases
Constant: constant depth offset

[delicate delta]

I clean this up and done?

I can be happy with this?

[late jackal]

Yes

Spheres are the worst case for shadow artifacts hehe

Try a bigger, more diverse scene

[delicate delta]

the final boss

[delicate delta]

yes eventually want to do sponza and all those and the deccer cubes

[late jackal]

Is the dolphin a triangle mesh?

[delicate delta]

yes an obj

I wrote my own very simple obj parser

idk what I'm going to do for gltf yet

my options are cgltf, write a C++ lib to use in my engine that uses fastglft or write my own gltf parser

they're all kind of bad options

[late jackal]

I'm telling sean

oh wait fastgltf is c++

yeah cgltf is probably going to be the easiest to use then

[delicate delta]

yeah, cgltf is kind of bad though

[elder kettle]

continue using obj

[delicate delta]

https://github.com/jkuhlmann/cgltf/blob/master/cgltf.h#L690

cgltf_node** nodes;

[late jackal]

two-star programmer

[delicate delta]

imagine using ** for a format that heavily relies on indexing

cgltf_size cgltf_mesh_index(const cgltf_data* data, const cgltf_mesh* object)
{
    assert(object && (cgltf_size)(object - data->meshes) < data->meshes_count);
    return (cgltf_size)(object - data->meshes);
}

[elder kettle]

Consider that if you don't pick one of those 3 options gltf then your next best option will likely be to write another parser for some other format

[delicate delta]

is how you get the index

I am not sure that works with zig

[delicate delta]

that's true

I should test if those functions work in zig

and if they do just go with cgltf

[elder kettle]

I used IQM which is basically like binary .obj but supports animation

It does have an example you could copy but it's still probably more work than just writing some abstraction over cgltf

Just standardize on an output format that your engine will store and just have a function that takes a gltf and returns that format

[delicate delta]

ok well gouraud is not going to get any shadows :/

hrm

maybe I need to remove ambient from the lights in the shadows too

in gouraud

that doesn't make sense

[delicate delta]

idk, good enough for gouraud

[late jackal]

Looks good. I wouldn't worry too much about fixing the tiny artifacts. If you look closely, most games have them too

[delicate delta]

yeah that's true

[delicate delta]

I am spamming these I know

[late jackal]

hmm the latency between the cursor and imgui is sus, but maybe it's a recording artifact

[delicate delta]

oh you mean how the cursor is always ahead of the drag handle?

it's really like that

this is all through renderdoc

nah it's like that without renderdoc too

is that not normal?

[late jackal]

hmm let me see

[delicate delta]

literally rendering only 8 vertices in this scene with no lighting or anything

the cursor is still ahead of the handle

even when I go slow

[late jackal]

mine has latency too but not as much

[delicate delta]

I think it's my laptop

and a 4k screen

it's just a bad laptop

[late jackal]

I recorded that on my crappy 60hz monitor

with double buffering I think

anyway it's not a big deal or even a problem

[delicate delta]

I don't have evidence, but I think the gpu can't power the pixels

[merry surge]

L

should’ve used C++

[delicate delta]

do you search this server for usages of gltf?

[merry surge]

I was searching for something else and found that by accident

[delicate delta]

what was that something else

[merry surge]

looking for users with my name

[delicate delta]

maybe I will get tired of zig and rewrite in C++

I don't think so though

I would definitely use fastgltf if I was writing C++

[late jackal]

me too

[delicate delta]

idk I might use it with this project too idk yet

[queen hinge]

Reduce your resolution to 1080p through windows settings, see if that helps

[delicate delta]

nah

it doesn't work

that's what deccer says he does with his, he has a smiliar laptop

it doesn't scale well

[queen hinge]

idk why th people decided 4k monitors should be more common on laptops than on desktop computers

the only differences with a 4k monitor on a laptop to me are

• the device runs slower
• the device uses more battery
• text is basically unreadable unless I reduce to resolution to 1080p in windows settings

[delicate delta]

^^ all of this is true

and I regret buying this thing

[queen hinge]

Which… I’d imagine a lot of people would have those same complaints with 4k monitors for laptops, lol

[delicate delta]

it's punishment for just looking blindly looking at a spec with a higher number

and pressing the buy button

[queen hinge]

[delicate delta]

nah macs have a lot of pixels

and I wanted something mac like

and dell didn't deliver

[queen hinge]

My laptop’s a gaming laptop and it only has a 2k monitor with a weird aspect ratio

[delicate delta]

mac like in terms of screen quality

and that was a foolish thing to want

[queen hinge]

The weird aspect ratio is eh, but the 2k monitor is relatively nice

I use it as ~1080p because I want to be able to read text
but it performs comparable to my desktop computer

A lot of programs I use don’t acknowledge dpi, and thus don’t scale for 2k or 4k monitors

[late jackal]

The text should be more readable on 4k

If it's too small, increase the scaling

[delicate delta]

the default scaling is readable on my laptop, it's just a performance issue, readability is an issue with imgui apps that haven't applied scaling themselves

the resolution is a performance issue

anyway I suspect it is

I don't actually have any evidence other than a blank glfw window that just calls clear color maxes at 500 fps

when using the full window resolution

and the smaller the resolution the more the fps go up

a max 500 fps for a blank window seems to indicate poor hardware performance yes?

at 4k

it's all spent in swapbuffer

[late jackal]

2ms to blit a single framebuffer is not a good sign

[delicate delta]

right

the laptop is otherwise really good

well there are touch screen issues

I think given I have an nvidia gpu pc my next laptop will just be amd linux

framework 16 or something

my goal is to get through the books I have and then if I make it maybe that will justify the purchase idk they're not cheap

I can't work on a PC and be glued to a desk I am more productive on a laptop

[queen hinge]

too small

[queen hinge]

app doesn't support that

[late jackal]

bad app

[queen hinge]

also fonts just look
off to me at 2k res for some reason

[late jackal]

Higher resolution text should always look clearer in an app that correctly respects content scaling

[delicate delta]

I should really go pro

don't look at my shader code though

it has a lot of if statements

        float bias = 0.0;
        float not_in_shadow = 1.0;
        float normal_offset = 0.5;
        if (i == 0) {
            if (f_can_get_light_from_z_pos_0 < normal_offset) {
                not_in_shadow = textureProj(f_shadow_texture0, fo_light_1_z_pos_0, bias);
            }
...

super professional

unreal recruiters breaking down my doors rn

hey it works

[queen hinge]

you have caused me to start going through LearnOpenGL myself

[delicate delta]

yes I saw you were active in your thread

I gotta lurk all the unread community threads

been super busy all day

I had a bug where I was shadowing the wrong direction

so that's why I added the normal check

it's not perfect but like

it's just a learning scene

I'm good, I did it, we did it, we have shadows

[queen hinge]

what does that even mean?

[delicate delta]

hrm I lost my capture afte nvidia driver reboot

well actually I only fixed for one lighting

I can show you

that's broken

that's fixed

[queen hinge]

huh

[delicate delta]

just had to use the surface normal vs the shadowmap direction

dot product

[queen hinge]

next thing for me is textures
(I'm like 90% done, just need to install it into my shader)

good job brain
install is not the right word there
conveys the point but like

[delicate delta]

nice

I am going to use this picture for this chapter's learnings on my github readme, it doesn't look like much but it was the hardest thing to get right

that's applying the shadow across three axes onto a sphere

[late jackal]

what does it mean to apply the shadow across three axes

is it a point light

[delicate delta]

it means there's three shadow maps

involved

yes

[late jackal]

sun + two point lights?

[delicate delta]

no sun here

it's six shadow maps per light

like a cube encasing the light

maybe that was dumb idk

[late jackal]

no that's normal

idk where the three shadow maps are coming from is all

[delicate delta]

oh so

in this picture it's

y neg, x neg, z neg

idk

[late jackal]

oh like three faces of the shadow cube

[delicate delta]

a part of that sphere is spread across each of those three maps

is what I meant

[late jackal]

it's on the diagonal of the shadow cube?

where three faces meet

[delicate delta]

I think so

I didn't verify in renderdoc when I took the screenshot

it was a cube there instead of a sphere

[late jackal]

hmm you could color each shadow differently to check

[delicate delta]

and the shadow was on three surfaces

of the cube

idk

[late jackal]

so basically it's seamless

[delicate delta]

it's not actually, there's definitely a problem there at parts of the diagonal

but it's good enough :P

[late jackal]

seamless enough

[delicate delta]

if I wanted to spend more time on it I could make it better

my frag shader is awful

[late jackal]

next time you write it, it'll be better

[delicate delta]

yes

[late jackal]

I still think you should load a bigger model

[delicate delta]

there's still artifacts, but which for a game I will want to get just right

[late jackal]

There are some good objs on casual-effects.com

[delicate delta]

I will, I have tons of progress to make still, I'm not done with shadows yet

it's just learn a little bit about this, move on to the next, come back around learn a little bit more

after my math books I have four books on just rendering, and then a bunch of vulkan cookbooks

[late jackal]

wow

[delicate delta]

I love taking pictures of my books :P

I need help

[late jackal]

I have a pile of books too, but they're collecting dust 😦

[delicate delta]

you already read them?

[late jackal]

I read a little of all of them

[delicate delta]

nice

[late jackal]

I need to read them fully

[delicate delta]

you know a lot already

[late jackal]

Math is my weakness

[delicate delta]

you work on graphics at AMD?

[late jackal]

I was a gpu devtech

Then layoff nation attacked

[delicate delta]

oh no I'm sorry

[wild ember]

das alot of books

[late jackal]

It's all good, I was looking for an out anyway

[delicate delta]

I wish you luck

[late jackal]

I'm having fun working on my renderer

I should've used this free time to read those books lol

I should go through math textbooks too but I don't think I have the discipline to teach myself college level math beyond what I already know

[delicate delta]

my degree is social science, I am anthro major and have a library science masters so not a lot math, I like learning the graphics and physics math though

I'm definitely not interested in math for math's sake

just to get stuff done in a game and that's it

I do like math a lot

[late jackal]

Same, but sometimes I want to understand papers better

[delicate delta]

I just need it to be useful

[late jackal]

Yeah having a use for it really makes you keen to pick it up

I didn't like math classes, but then I discovered graphics programming and now I wish I paid better attention

[delicate delta]

I was able to solve a shadow problem today because I understood what a dot product does, that felt nice

[delicate delta]

same!

[late jackal]

When I was making a software renderer for a college class, I had a eureka moment when I understood how the dot product fit into lighting

[delicate delta]

that's a great feeling

gaussian elimination and the determinant matrix construction is basically just a black box to me still, I did think about it while reading through it, and it made sense then, but I couldn't explain it to anyone

[late jackal]

for me, it takes multiple reads to understand things

[delicate delta]

same

[late jackal]

and from multiple sources

[delicate delta]

yes, reading through multiple math books was very so helpful for understanding the look at/view matrix and the perspective matrix

they all had a piece of the puzzle

[late jackal]

brain works best when you feed it lots of sources

[delicate delta]

that is very true

[delicate delta]

I think Blinn Phong looks better than Phong, at least with what I've done so far

it just looks better, I like the specular better

[late jackal]

it does

[delicate delta]

I was looking at the latest FIFA game because it was being advertised on Steam and you know I think all the shadows in that game are simple projective shadows

the characters are all on a flat surface

the shadows are all highly PCF'ed. I bet those shadows are cheap af

they're all directly beneath the players, there are lights from every direction in the stadium

I bet that game uses 0 shadowmaps

I bet 90% of that game's costs is just the FIFA license

it probably is a single shadowmap

actually idk

[ivory verge]

2 times mastering thing with vulkan book one on each pile

[delicate delta]

You’re the first person to notice that

I accidentally ordered it twice

[ivory verge]

: >

[delicate delta]

No idea what to do with the extra copy

[ivory verge]

find one who is also interested in giraffics in your area 🙂 or donate it to the library?

[delicate delta]

Libraries don’t like donations fwiw

[late jackal]

burn it for heat during a particularly cold sf summer day

[delicate delta]

Librarians curate the books at a library

[late jackal]

yeah it's like how museums curate their collections and can't just put everything in an exhibit

[ivory verge]

so you need to plant the thought into their brains that they should cure their curations by acureately adding a new bok to their curation

[wise imp]

well that depends on the library

you can just ask them

most libraries do take books in donations

[delicate delta]

It’s not pauper house for orphaned books to find parents

[delicate delta]

They do, but, they don’t like it

[wise imp]

i mean that depends if you library has stuff like graphics programming books

[delicate delta]

It’s extra work and they again curate the books

[wise imp]

ive donated mine to my school library

[delicate delta]

Idk what they do with donated books tbh

They’re usually low quality

And old and very used

I think Americans libraries are different from European libraries

Being a librarian requires a lot of education

I forget it’s been awhile

I think saying vulkan in a community thread summons vblanco like saying gltf summons lethargic 😛

[wise imp]

nah ive been on this thread for a while

im only on a few

[delicate delta]

My thread is probably pretty basic for you but always appreciate when you pop in

Probably basic for most people

In a year I will have a game I am working on again

And hopefully it will go better this time

[maiden gazelle]

Wait what

You can use operators in glsl to multiply matrices tho

[ivory verge]

yes

mat4 a = b * c; // b and c are mat4s

its all in the glsl spec iirc

[maiden gazelle]

Okay I misread

I read the opposite of what Björn said

Nvm

This channel and Björn is too active

I should read it more often

[ivory verge]

😄

i keep mixing the 3 channels all the time

[maiden gazelle]

That’s an insult to Björn :p

[ivory verge]

nah 🙂

i like to hang out in all of them

[maiden gazelle]

Just joking

I like having you on my channel 🫶

And Björn of course