[Archive] The Shader Hole

[snow cradle]

https://lousodrome.net/blog/light/2013/09/15/a-sampling-of-shadow-techniques/
https://www.cg.tuwien.ac.at/research/publications/2013/ADORJAN-2013-ASE/ADORJAN-2013-ASE-thesis.pdf

[fickle ocean]

https://tenor.com/view/dog-reading-dog-reading-read-gif-12347569

[timid pumice]

There's no way to autoreload shaders after compiling right?

[snow cradle]

no. only rename, compile, write to vmt

[timid pumice]

What does this even means
Failed to create shader '49D1F464'

[snow cradle]

_ps30
_vs30
?

49D1F464_ps30

[timid pumice]

yeah, this occurred after supplying a simple _vs30

thanks, found the problem

[timid pumice]

in my custom mat i was supplying twice pixshader, but on the one that override it was using a vs

[hardy acorn]

try to use _ps2x and _vs20 as you can instead of ps30 or vs30, some reports tell that this can causes some troubles for linux user

[wispy birch]

linux uses togl, which is a hellhole either way
the correct way to play gmod on linux is through wine, so best dont complain when shaders dont work as expected

[timid pumice]

https://sharex.imgonzo.dev/f/UYBiLh8CIO.mp4
This looks NEAT, now my problem is that i just cannot force the sector to only apply in that drawable area

Tried SetViewPort, SetScissorRect, CopyRenderTargetToTexture

[timid pumice]

https://sharex.imgonzo.dev/f/lchsiSywol.mp4

https://tenor.com/view/fire-engineer-gif-19422642

    local ow, oh = ScrW(), ScrH()
    render.PushRenderTarget(BLUR_RT)
    surface.SetMaterial(matfb)
    surface.SetDrawColor(color_white)
    surface.DrawTexturedRectUV(0, 0, ow, oh,
        x / ow, y / oh,
        (x + w) / ow, (y + h) / oh)
    render.PopRenderTarget()

    MATERIAL_SetMatrix(ROUNDED_BLUR_MAT, "$viewprojmat", matrix)
    MATERIAL_SetTexture(ROUNDED_BLUR_MAT, "$basetexture", BLUR_RT:GetName())

    surface.SetMaterial(ROUNDED_BLUR_MAT)
    surface.SetDrawColor(color_white)
    surface.DrawTexturedRectUV(x, y, w, h, -0.015, -0.015, 1.015, 1.015)

I hated it

https://sharex.imgonzo.dev/f/ontmO4Vywv.mp4
See you next weekend liquid ice

[stark valve]

Guys, can anyone confirm that the game.MountGMA method to make a shader work on clients work? I've been trying this with no success whatsoever. It seems it will only work if the shader exists in the physical path

[minor raft]

nice ice ^^

https://tenor.com/view/ice-cube-dance-ice-cube-bfdi-bfb-tpot-gif-15057326968531740248

[hybrid trench]

nice balls

[timid pumice]

Im using it, if you ping me in 3 hours i will send you my workflow, i literally copied srlion loader

For some reason, it breaks autorefresh

[stark valve]

I fixed it

Apparently I compiled it wrong or smth

[snow cradle]

4 cascades CSM test using 1 shader

1 Global shadow matrix + float3 offsets + float scale

[snow cradle]

@wispy quartz Hello. can u check it on linux? These arguments are in the source engine. But some arguments are only for D3D9EX. And it is not known how Linux will react to these arguments.

funcs: render.PushFilterMag( number texFilterType ), render.PushFilterMin( number texFilterType )

    TEXFILTER.PYRAMIDALQUAD     = 6
    TEXFILTER.GAUSSIANQUAD         = 7
    TEXFILTER.CONVOLUTIONMONO     = 8             /* D3D9Ex only -- */
    TEXFILTER.FORCE_DWORD         = 0x7fffffff     // force 32-bit size enum

[snow cradle]

Shadow stabilization (ON)

Shadow stabilization (OFF)

[misty surge]

Much better. Although those cascades look a bit off, tbh. Might need to increase the sizes. Games also use a blur to blend that shimmering issue, it looks great so far, keep it up 🙂

Reminds me of my many years ago in vulkan making CSM

Although this sounds dumb, to stop the shimmering some games even use the player as the origin of all the orthos

So your like layering it out just using the player pos

[rain marten]

So... I was experimenting with a way to make a custom shader work on props. I wanted it to look like a emissive VertexLitGeneric in-game, without using any Lua script, and uh...

I just copied the necessary .h files source code from the source SDK, used their functions on my custom shader is the same way I saw them used on VertexLitGeneric, and it just... worked

Here's my old LCD shader, but modified to use the same CalcPixelFogFactor and FinalOutput functions that I saw being used on the source code for VertexLitGeneric:

Oh yeah, on the vertex shader, I also used (mostly) the same logic that VertexLitGeneric used so my custom shader could work on a map with fog & water

Although... "just worked" really only applies on getting the shader compiled, I still played around with every parameter on the .vmt file to see how it affected things. I did find out that $flags2 hack and $softwareskin were unnecessary.

Side note: the game REALLY did not like when I changed $vertexcolor, $vertexnormal or $vertextransform while it was running, it crashed almost every time

[timid pumice]

Amazing

[snow cradle]

130 flags2

for models

max 510-511

do u make output of sv depth?

[rain marten]

nope, just color

here, I wasnt planning on sharing so its a bit messy, but I think you guys can understand it better than I could

I've replaced the texture used for the screen (a RT showing bad apple) with the plastic texture. Also, remember that the shaders uses includes from the Source SDK

Oh and, curiously, the // STATIC/DYNAMIC commented lines at the top actually seem to declare some constants, given that it doesn't compile without them

[snow cradle]

CSM cascade smooth

[snow cradle]

3d drawing quad using cam3d on player's screen

local w,h = ScrW(),ScrH()
local aspect = w/h
local viewSetup = {}

local bias = 0.001

hook.Add("CalcView",  libName, function(ply, origin, angles, fov, znear, zfar)
    local f ,r, u = angles:Forward(), angles:Right(), angles:Up()
    local center = origin + f * znear

    local halfH = math.tan(math.rad(fov * 0.5)) * znear
    local halfW = halfH * aspect

    viewSetup.tl = center - r * halfW + u * halfH
    viewSetup.tr = center + r * halfW + u * halfH
    viewSetup.br = center + r * halfW - u * halfH
    viewSetup.bl = center - r * halfW - u * halfH
end)

function shaderlib.DrawVertexScreenQuad() // Drawing PP shaders with MRT and vertex shader inputs
    cam.Start3D()
        render.SetWriteDepthToDestAlpha( false )
        cam.IgnoreZ( true )
        render.DrawQuad( viewSetup.tl,viewSetup.tr,viewSetup.br,viewSetup.bl )
        cam.IgnoreZ( false )
        render.SetWriteDepthToDestAlpha( true )
    cam.End3D()
end

U can make input to vertex shader on pp shaders using render.SetModelLighting(0, x, y, z). also MRT works with this func

https://github.com/meetric1/gmod_shader_guide/blob/main/lua/autorun/client/shader_examples.lua

[snow cradle]

updated. shaderlib.DrawVertexScreenQuad()
https://steamcommunity.com/sharedfiles/filedetails/?id=3542644649

[snow cradle]

maybe using RenderScene is more stable then CalcView

hook.Add("RenderScene",libName,function(origin, angles, fov, p2)
    local viewSetup = render.GetViewSetup(true)
    local znear = viewSetup.znear

    local f ,r, u = angles:Forward(), angles:Right(), angles:Up()
    znear = znear + bias
    local center = origin + f * znear

    local halfH = math.tan(math.rad(fov * 0.5)) * znear
    local halfW = halfH * aspect

    quadVerts.tl = center - r * halfW + u * halfH
    quadVerts.tr = center + r * halfW + u * halfH
    quadVerts.br = center + r * halfW - u * halfH
    quadVerts.bl = center - r * halfW - u * halfH
end)

[snow cradle]

lol c10 EyePos works

but ambient cube is default color

like blue/dark blue etc

and render.AmbientCube input not works

values is steel blue

idk

why

[snow cradle]

like like like like 🙃
https://github.com/Facepunch/garrysmod-requests/issues/2890

[sonic lynx]

Thanks so much for sharing! Hopefully it's ok with you if I used this in a project?

I can't write shaders to save my life...

[timid pumice]

https://sharex.imgonzo.dev/f/uOhDzT2pf1.mp4
WHY

struct PS_INPUT {
    // texture coordinates
    float2 uv            : TEXCOORD0;
    // vertex color
    float4 color        : TEXCOORD1;
    // screenspace position
    float2 pos            : VPOS;
};

sampler TexBase : register(s0); // $basetexture RTBuffer
sampler Tex1    : register(s1); // $texture1 Grayscale mask

const float4 Constants0 : register(c0);
float TransitionProgress = 0.5; //Supposed to return c0_x 0-1 CurTime() % 1

float4 main(PS_INPUT i) : COLOR {
    float4 rgb = tex2D(TexBase, i.uv);
    float value = tex2D(Tex1, i.uv).r;

    if (value > TransitionProgress) {
        rgb.a = 1;
    }
    return rgb;
}

[timid pumice]

https://sharex.imgonzo.dev/f/eujBQPOv46.mp4

[timid pumice]

https://sharex.imgonzo.dev/f/8xiXrpVnXX.mp4

https://sharex.imgonzo.dev/f/bblc9Mgzf4.mp4

[graceful lynx]

Looks a lot like https://gmodwiki.com/render.SetWriteDepthToDestAlpha to me

[timid pumice]

It was funny, i was sending the FB into the shader which not contains the viewmodel, still doesnt explains why it becomes less transparent when you get near walls

The solution was just to clear at 0,0,0,0 or the material im just passing

[graceful lynx]

still doesnt explains why it becomes less transparent when you get near walls
Source adds depth to the alpha channel by default. render.SetWriteDepthToDestAlpha( false ) will prevent it from doing that

The example code in that wiki page is bad and should be ignored

[timid pumice]

Shaders are fun
https://sharex.imgonzo.dev/f/cYYCBxluNn.mp4

But boy, they really tax a lot

no shader

[lunar sinew]

i noticed the more fps you have, the more you loose if something is happening

[timid pumice]

https://sharex.imgonzo.dev/f/q02TeUhWiR.mp4

[lunar sinew]

for example, if you normally get 1000+ fps idle but as soon as there is something happening like a battle or something taxxing, it reduces to like 600

but if you normally get like 100 fps, it only reduces by a bit

[timid pumice]

Definitely, losing 500 fps on 1000 fps is nothing like losing 50 on 100

i remember 3kliksphilip did a nice video about that on latency/frame time

[dusty linden]

Interesting

[short dirge]

looks fine from a discord thumbnail, but fullscreen seems a lil slow

like less "on fire" and more "extreme heat/volcano biome"

needs more chaos

1st ver was pretty ok, just lacking flashiness

i think its the speed of the pattern being scrolled being relatively faster there

[snow cradle]

For some users default vertex shader not works

make own default vertex shader for pp

not use $vertextransform 1 instead of $vertexshader

rx 580

[icy plume]

isnt this effect possible with particles? thats how l4d2 does it

still looks cool tho not trying to be mean

[stark valve]

well you could obviously scroll a bunch of flame sprites on the screen but it wouldnt be nearly as cool

[snow cradle]

is that ok velocity buffer?

[fickle ocean]

could look something like a cool echolocation shader

smelly bot, i was saying this is all sorcery to me

[graceful lynx]

DLSS inbound

[snow cradle]

https://github.com/Akabenko/GShader-library/tree/main

[timber narwhal]

would be epic to add some sort of upscaling to gmod

dlss/fsr3 is impossible but im sure you can port the really basic CAS algorithms from AMD and maybe FSR1

run gmod at 128x128 and upscale it to 1080p

peak fps

[snow cradle]

https://github.com/Facepunch/garrysmod-requests/issues/2890

[hollow pollen]

(gmod isnt gpu bound so this wont actually boost performance)

[timid pumice]

So nerd but so right

[heavy blade]

congrats, got added

[timid pumice]

also the shader would have it own impact too

[leaden mesa]

in that srcsdk function you posted, the isfinite() check is missing for the 4th arg v, is this intentional?

[snow cradle]

https://github.com/ValveSoftware/source-sdk-2013/blob/68c8b82fdcb41b8ad5abde9fe1f0654254217b8e/src/public/materialsystem/imesh.h#L2089

I don't know. The function is not mine.

[leaden mesa]

ik

[snow cradle]

Physically Based Bloom release
https://steamcommunity.com/sharedfiles/filedetails/?id=3556046077

[heavy blade]

WHAAA

[misty surge]

Is that even gmod anymore? Obviously not. That’s unreal @snow cradle

[wispy quartz]

How should it look in windows

and how do i even compare them

[snow cradle]

I don't know. It seems it doesn't work on Windows. There are no changes, although these types of filtering are in the source engine code.

[snow cradle]

bumps from framebuffer

#include "common_ps_fxc.h"

sampler FrameBuffer   : register( s0 );
const float strength : register( c0 );
float2 TexBaseSize : register( c4 );

struct PS_IN
{
    float2 P            : VPOS;
    float2 vTexCoord    : TEXCOORD0;
};

float3 CalculateBumpFromDepth(float2 uv)
{
    float3 depthRight   =   tex2D(FrameBuffer, uv + int2( 1, 0) * TexBaseSize ).rgb;
    float3 depthLeft    =   tex2D(FrameBuffer, uv + int2(-1, 0) * TexBaseSize ).rgb;
    float3 depthTop     =   tex2D(FrameBuffer, uv + int2( 0, 1) * TexBaseSize ).rgb;
    float3 depthBottom  =   tex2D(FrameBuffer, uv + int2( 0,-1) * TexBaseSize ).rgb;
    
    float gradientX = Luminance( depthRight - depthLeft ) * 0.5;
    float gradientY = Luminance( depthTop - depthBottom ) * 0.5;
    
    float3 normal = normalize(float3(-gradientX * strength, -gradientY * strength, 1.0)) * 0.5 + 0.5;
    normal.z = sqrt(max(0.0, 1.0 - normal.x * normal.x - normal.y * normal.y));

    return normal;
}

float4 main(PS_IN i ) : COLOR
{   
    float2 center = (i.P+0.5)*TexBaseSize;
    float3 bumps = CalculateBumpFromDepth(center);

    return float4(bumps, 1);
};

same bumps in tangent space (using TBN matrix from GShader lib)

Frenel of tangent space bumps

i need to fix edges

Luminance is comment function from #include "common_ps_fxc.h"

[timid pumice]

Is there any cool usage for screen bump or it’s a cool way to test dynamic bump generation from fb

I think once im done with 2d shaders, i will go with 3d, but doing shaders for me been feeling like taking stuff from other people math and steal it

[jolly talon]

That's like saying putting a wheel on a car is stealing the idea from the guy who invented the wheel

Everything in graphics that Ur gonna be able to do in sm3 has been done by someone else already, just read papers on stuff if you want to do anything

[timid pumice]

The entry field is always knowing about integrals, i barely passed that in university, but at least i can say my brain is more woke than back then in maths wise, but still the entry floor is really high

[stark valve]

https://cdn.discordapp.com/attachments/164779286971154433/1410295893990576268/gmod_2025-08-27_18-11-52.mp4?ex=68b07fe7&is=68af2e67&hm=9b0ec0fa4b3e53d2f7b32f0fd9854b563cd1d0fd863d7944bf22a2121fdbfd54&

simple, but im quite content of the result

[graceful lynx]

Looks great

I bet it would work also as an effect to indicate "This player is cloaked, but you're on their team so you can still partially see them"

[stark valve]

maybbbbeee

[timid pumice]

Looks amazing, is implementation the same than the shader-examples?

[stark valve]

how do you mean

[timid pumice]

In 2D, you have the implementation like a screenspace object with DrawScreenQuad, but not sure how does it works for 3d

[stark valve]

oh yeah its a screenspace pixel shader just like usual

but on a model

i have a very very minimal vertex shader

[dusty linden]

🧐

very nice

I wonder if I could replicate it ingame using the current vertexlitgeneric shaders and mat variables

[timid pumice]

Even with iridiscense is not enough to mimic sub chanel warping

Iridiscense from ministrider tho

Although glasses from portal 2 really blew my mind with the broken reflection

[short dirge]

Isn’t that just a normal map

Or dudv?

P2 had the benefit of higher quality textures and higher poly models, having them able to separate normal islands without anyone seeing texel transition in between

[dusty linden]

hunter's effect is phongwarp

https://developer.valvesoftware.com/wiki/Refract

Refract shader

iridescence could possibly also be envmapped

[snow cradle]

More textures to sceenspace_general. This is code of request for Rubat, P.S. this is code sample. I wanna ask him 8 textures to input total:

code deleted

[hollow pollen]

why tf do you need 8 textures

[fickle ocean]

What if he needs 9 textures

[jolly talon]

Really not uncommon to need 8 textures for a shader

Think of a two plus way blend shader where each material wants diffuse normals etc

My terrain shader is limited to not allowing spec maps since both my channels need colour and heightmaps which already reaches the max of four, and I can luckily reconstruct normals from the heightmap

[hollow pollen]

use a texture atlas

ez

[snow cradle]

PP Snow shader with 2 texture blending:

1. WPDepth (worldpos + depth)
2. NormalsTangents
3. Snow Colormap
4. Snow Bumpmap
5. SnowMap (like shadowmap)
6. Second SnowColorMap
7. Second SnowBumpMap

PBR:

1. ColorMap
2. BumpMap
3. MRAO
4. CubeMap
5. LightMap (theory)
6. Emissive
7. Detailtexture

8 is a power of two. 6 may not be enough. 7 is not a power of two.

So we need 8 textures

We can ask for 6. But wouldn't it be better to ask for 8 right away? And we wouldn't ask for 5. Or 7. Those are odd numbers.

8 is with a reserve, so as not to return again to the question of the number of textures

[snow cradle]

also we can use specural with color texture like .RGB

For 4way blend we will need 9 textures

But I would stop at 8

[snow cradle]

4x colormaps
4x bumpmaps
1x blendmap

[snow cradle]

Atlases are an extreme measure and a loss of optimization. Breaking parallaxes and other defects

For parallax it is not ez

[snow cradle]

Well, what's the result? I request:

• texture4
• texture5
• texture6
• texture7
• c4
• c5

We do this - we close the issue of textures and the issue of constants for input into the pixel shader

[hollow pollen]

u probably could combine all the screenspace stuff into 1 texture

wouldnt hurt perf that much, if at all if you used viewports

[snow cradle]

So what now? Stop at 4 textures?

What are the downsides of having 8 textures?

[hollow pollen]

there arent im just guessing rubat isnt gonna touch it

[jolly talon]

ok sure but adding more samplers would also work and be easier, texture atlas would work tho yh

[misty surge]

Texture atlas would work but they’re annoying to calibrate (since we have to recalculate the UVs, requiring more processing than say storing it in the vram for quick access.) The main issue of using all 8 would be that the shader’s vram may be too bespoke and cause performance issues with potentially calling those 8 textures at once, yet the advantage is that you’re not adding extra math to calculating the correct atlas for that one sampler. Either way, it’ll affect the process.

I’d prefer to have 8, with that option. And allow atlas texturing if you reach 8 as a limit.

Also, if I’m not mistaken we had 4 samplers only because some hardware back in 2013 didn’t support more than 4. Today, it’s more common to have full support of 8 rather than just 4 (default minimal)

And the vram on cards today would have basically negligible draw backs because usually we have more than 2gb vram in most common setups.

[hollow pollen]

uv offset calculations are negligible compared to a texture lookup

again tho not saying ur wrong, 8 would be better im just assuming rubat isnt gonna touch screenspace_general again

[misty surge]

I agree! I’m not always right. Lol but yeah I think ideally 8 is just cleaner

I wanted to push atlases for my planet shaders to allow more than 16 planets at once with one shader using the same concepts. I resorted to just using separate shaders per planet.

[hollow pollen]

ideally you also wouldnt sample 8 times per pixel

memory lookup is slow

especially on older cards

• if you dont optimize for cache locality its even worse

[snow cradle]

8. Flashlight sampler

Forward rendering requires a lot of textures.

[misty surge]

Having 8 would be great, 4 is definitely expensive as it is, but maybe having the choice would be nice, depending on what you’re making.

[jovial tide]

i dont think it makes sense to keep expanding screenspace_general, should make a new one that is tailored for flexibility

name it like CustomGeneric

[hollow pollen]

rubat?? adding something useful??

https://tenor.com/view/lol-gif-7543689799046998153

[jovial tide]

i mean just make the c++ code for it and send it to him, he will add it

i had one nearly done but i didnt finish adding flashlight support and morph weights

it supported everything else like variable mesh formats, skinning, fog, world lights, brush and model support, customizable pixel and vertex constants ($c and $v etc), 8 samplers, depthtesting, alphatesting

[hollow pollen]

oh neat

yeah world lights are definitely missing from screenspace_general

compile time constants might be nice as well

the // STATIC or // DYNAMIC stuff valve added

[jovial tide]

would need to be careful because that immediately opens a security hole

need to check if all the combo code is sanitized (hint: it wasnt, TF2 had some fixes for it recently)

[hollow pollen]

wasnt aware that was a security concern

interesting

[snow cradle]

You wanted to do it yourself.

I have such a goal for me. Screen space general will do. But I would like to add more textures

By the way, people have been waiting for you to make the custom shader you talked about for a long time.

But since the idea has died out, I can ask to improve the screen space general. Move in simple steps

[green forge]

https://steamcommunity.com/sharedfiles/filedetails/?id=3558518483&searchtext=

[snow cradle]

c4,c5 I can't offer it yet, because:

float2 TexBaseSize : register( c4 );
float2 Tex1Size    : register( c5 );

[stark valve]

do stencils fuck up screenspace shaders?

I'm having a damn hard time getting the same result for everyone

some people see white, other transparent, other see it fine

[wispy birch]

they should definitely not

[stark valve]

losing my mind on a shader ive made a while ago, like i said its so fucking weird, at first I thought it was an AMD compat issue, then the way I package gmas and now its happening again

[lunar folio]

if you're compiling as ps3 try ps2. That was the only shader version I was able to get consistent results in with my players. Some people just could not see ps3 shaders at all (fully transparent).

[stark valve]

unfortunately I did just that and I'm not sure it helps

[stark valve]

I am getting that as well

transparent/fully white

[snow cradle]

tangent on models. result is same with$softwareskin 1 and with $softwareskin 0
this not works with $flags2 130. But works with $flags2 2 or $flags2 128

float3 outputNormal;
float4 outputTangent;
_DecompressUByte4NormalTangent( i.vNormalWS,outputNormal,outputTangent );

But without flags2 130 some models may be invisible

With flags 130 the normals become normalized

[snow cradle]

In short, I will ask for +2 textures on the screen space general. So that TexSize c8,c9 does not reach c10 - EyePos and does not disturb the order

I know how to add tangents for brushes. And you can add lightmaps, but it doesn't make much sense without flashlight support.

Adding 2 textures would be rational, but adding other features for direct rendering would not be entirely appropriate.

Just like adding the 7th and 8th texture will break the order of the registers c

[snow cradle]

Hello everyone! Please like the request on GitHub! ❤️❤️❤️ Requested from Rubat +2 additional textures for screenspace_general. This will allow using blending of conditional snow textures in deferred rendering, so that the surface looks more detailed, mixing several snow textures. Or it will help in creating shaders with multiple texture inputs.

This is the most rational request for additional textures, since they do not overlap registers and do not jump over them, preserving their logical sequence.
Thanks @misty dove for help!
https://github.com/Facepunch/garrysmod-requests/issues/2896

[jolly talon]

You can't do wrapped sampling with an atlas, only clamped, plus is requires adding gutters between the textures or making sure you don't sample in a way where the neighbor blends in, it's just overall annoying

[hollow pollen]

not saying ur wrong, 8 would be better. im just assuming rubat isnt gonna touch screenspace_general again

this is a valid, albeit annoying solution to the current problem

also you can still do wrapped sampling

assuming 4 atlas in a texture, like so:
[][]
[][]

might look something like this

// top left
float4 sample_uv00_wrapped(float2 pos) {
    return tex2D(BASETEXTURE, pos % 1.0 / 2.0);
}

// bottom right
float4 sample_uv11_wrapped(float2 pos) {
    return tex2D(BASETEXTURE, pos % 1.0 / 2.0 + 0.5);
}

[jolly talon]

Yea but it won't do blending between the values on opposite sides of the texture like normal

So there will be a seam of sorts

[hollow pollen]

yeah you'd need to do pointsampling or like you said add a "gutter"

[jolly talon]

For point sampling it would be fine but I'm using it for terrain shader so the textures do need to linear sample

And be tiled

[hollow pollen]

if you dont need to edit the texture in realtime you could make an animated vtf with 4 frames

then use a tex3D to sample the texture or "layer" you want

[jolly talon]

True that's an interesting idea

I guess with that you could actually send a crap ton of stuff into a shader

Makes me think how good it would be to get a way to send an array of vectors or something to the shader

Like I wanted to do foot steps affecting the parallax depth but id need a lot of data for lots of steps

[snow cradle]

Treesway have same bug in Source SDK 2013 as in GMOD

[wispy quartz]

No at all. In fact, you can use screenspace shaders as a mask (like aphatest), by simply discarding a pixel

[wispy quartz]

Nah you can still do it via manually making 4 samples of neibhouring pixels and just blending that

[stark valve]

seems like a good trick to do some 2 pass gas / volumetric stuff

[wispy quartz]

nah you can just make volumetric stuff with some hacky way shit

i mean like all those 3d screenspace shit done with th eclouds

[stark valve]

🤔

[green forge]

https://cdn.discordapp.com/attachments/277242276298752000/1412517583718060152/2025-09-02_14-19-48.mp4?ex=68b89503&is=68b74383&hm=520dd07c0e1d2b4b9a4a262081d106125cfb22fa426edfa60fcd9fe974f32142&
Wip gtao with temporal

[surreal parcel]

shader hole mole people when every corner isnt darkened

my favorite shader is fullbright

[green forge]

i mean gtao is indirect shadows SSAO is more of the corner darkening

[timber narwhal]

looks cool but the video is super compressed i can barely see it

[short dirge]

That was compression? I thought that was temporal artifacting.

No, it probably still is.

[timber narwhal]

you can literally see the individual compression artifacts

[graceful lynx]

It's extremely compression artifacting

[surreal parcel]

now THAT is a shader I want

[stark valve]

Brb making corruption shader

[neat terrace]

datamoshing

[green forge]

yeah it was my bitrate

[green forge]

I changed it so it might be better now

[timber narwhal]

still kinda eh. but i can see it much better now, looks sick

[green forge]

yeah i think the main reason is cause of the resolution difference

[green forge]

Did major improvements

[timber narwhal]

looks nice, its so much better as an approximation of indirect occlusion

[green forge]

Yeah much better than SSAO

[green forge]

GTAO vs no gtao

[timber narwhal]

much nicer

[green forge]

This bot i swear

[timber narwhal]

gorgous

would be nice too if you could include normal mapping in the sampling

[short dirge]

Still can’t stand temporal artifacting when you’re moving slightly above walking pace

TSAA has the same issue though so

[green forge]

Your able to change the temporal settings to a point where you don't notice it at all

[green forge]

Also i hope your not talking about bitrate compression that's just obs

[graceful lynx]

Reminder you can change settings in the C menu so the spawn menu isn't in the way of showing the comparison as the setting changes

[green forge]

I acutally didn't know that thanks

[graceful lynx]

Oh great

[stark valve]

do you guys have any clever idea on how I could do a "projected" shader on map geometry?

I want to essentially display the world geometry in an area a particular way and blend its edge with whatever material the current geometry has

[short dirge]

so like, a decal/info_overlay

or are you referrring to a scope/fov vision box?

[stark valve]

a decal of sorts yea

[snow cradle]

screen space decals

using worldpos rendertarget

and model3x3 matrix

maybe u need make manual input of model matrix

[snow cradle]

local mModel = Matrix()
mModel:Identity()
mModel:Rotate(sticker:GetAngles())
mModel:SetTranslation(sticker:GetPos())
mModel:Invert()
mat:SetMatrix("$INVVIEWPROJMAT",mModel)

[jolly talon]

fullscreen pass for every decal 😭

[snow cradle]

why? screen space decal is forward render

this is box-model

[jolly talon]

oh I see clever

like the gta lamps

do you draw inside out box with no depth culling for it?

[snow cradle]

I just draw a box. No depth needed for Screen Space Decals

Use WorldPos

[jolly talon]

how to stop waste performance drawing decal through wall

but box mesh culls most waste yh

[snow cradle]

The decal is rendered within the box

[jolly talon]

so where the box is on screen does calculation whether world pos is within box

[snow cradle]

You need 2 WorldPos - from rendertarget and from forward render box from vertex shader

Nothing complicated

[jolly talon]

alright ye, just wanted to know if a method to depth cull the decal box is possible

so if the decal is behind a wall it wont even draw a single pixel

i guess stencil but clearing and writing stencil for every decal is not good

[snow cradle]

How do you draw a decal in a pixel shader if you can't see it and it's behind the wall? It's impossible.

[jolly talon]

if you draw inside-out box for the decal projection it would draw through walls no?

[snow cradle]

You don't have a decal on your monitor. There is no decal on your monitor. The pixel shader doesn't work. The model is behind the wall. You can't see it.

[jolly talon]

you wont see it but the pixel shader will run i mean

[snow cradle]

I was fine and had no problems.

[snow cradle]

How?

A pixel shader can't calculate what it can't see.

[jolly talon]

I think you draw the inverted box with no depth to draw the projected decal to the screen using the depth buffer so if its behind a wall it will still do the world pos calculations?

idk maybe i misunderstand ur idea

[snow cradle]

I don't use depth buffer for screen space decals

i use _rt_WPDepth from GShader library and worldpos from vertex shader from box-model of decal

[jolly talon]

what if your camera enters the box

[snow cradle]

It will be shit. Read the info on the Internet.

[jolly talon]

inverted box no?

[snow cradle]

Do

[jolly talon]

ooh, if camera is outside the box, draw as usual with depth, if its inside draw inverted i guess

When I did screenspace decals I did with purely inverted

Also spherical mapped and cylindrical mapped decals are fun

[stark valve]

very interesting, thanks

[snow cradle]

RESHADE     = file.Exists("ReShade.ini", "EXECUTABLE_PATH")
DGVOODOO     = file.Exists("dgVoodoo.conf", "EXECUTABLE_PATH")
DXVK         = !DGVOODOO and !RESHADE and file.Exists("d3d9.dll", "EXECUTABLE_PATH")

vertex shader fog constanst not works for DXVK

[snow cradle]

Getting videocard name and vendorID if user has installed reshade

RESHADE     = file.Exists("ReShade.ini", "EXECUTABLE_PATH")
local vendorID = 0

VENDORID_NVIDIA = 0x10DE
VENDORID_ATI     = 0x1002
VENDORID_INTEL     = 0x8086

local vendors_id = {
    [VENDORID_NVIDIA]     = "NVIDIA";
    [VENDORID_ATI]         = "AMD";
    [VENDORID_INTEL]     = "INTEL";
}

local driverName = "UNKNOWN"

local function ReadReshadeGPU()
    local i = 0

    while true do
        local file_name = "ReShade.log"
        if i > 0 then file_name = file_name .. i end

        local log_reshade = file.Exists(file_name, "EXECUTABLE_PATH")

        if log_reshade then
            local log_text = file.Read( file_name, "EXECUTABLE_PATH" )
            local log_text_lower = string.lower(log_text)

            for id, vendor_name in pairs(vendors_id) do
                local finded, diver_start = string.find( log_text_lower, string.lower(vendor_name) )

                if finded then
                    vendorID = id
                    local end_test = string.find(log_text_lower, "driver", diver_start)
                    driverName = string.sub( log_text, finded, end_test-2 )
                    print("[GShader library] Found " .. driverName .. " driver")
                    break
                end
            end

            if vendorID != 0 then break end
        else
            if i != 0 then
                break
            end
        end

        i = i + 1
    end
end

if RESHADE then
    ReadReshadeGPU()
end

function system.GetVendorID()
    return vendorID
end

function system.GetVendor()
    return vendors_id[system.GetVendorID()] or "UNKNOWN"
end

function system.GetDriverName()
    return driverName
end

[calm root]

The irony, the reason crash dump can't be read and were blocked (when I reported it) is because they contain this and other private information

[snow cradle]

You can read everything in gmod. Even images - screenshots of the player from reshade by bytes, other saved files. The gmod folder is like a public folder

[snow cradle]

That's why I couldn't read the log from DXVK to read the manufacturer and name of the video card

[stark valve]

poggers

on another note why cant we have system.SysInfo() that returns GPU vendor, etc...

[snow cradle]

Usually, Rubatu requests asked for too much - reading ALL information about the PC hardware. He declined such requests.

But it would be enough to have only the video card manufacturer. And, perhaps, the name of the video card.

• The standard vertex shader screen space general does not work for RX580.
• There are specialized NVIDIA formats such as INTZ

It is also unknown how Intel will behave.

[stark valve]

who cares about intel lol

but yes

[snow cradle]

There is a risk group:

• Linux
• Proton
• DXVK
• DGVOODO
• RX580
• Intel

Proton has almost no chance of being detected, since it is considered gmod like Windows. It has DX95 (although Linux should have 92)

DXVK does not receive fog data from the vertex shader. It is difficult to detect (by exclusion) - d3d9.dll

I have already written about RX580.

If you ignore each group of players, then there will be a solid number of deprived people

[stark valve]

Thats not solid at all, together this isnt even 1% of all players

The only thing worth looking into in this list is proton

[snow cradle]

I think it is wrong to divide people by percentages.

[stark valve]

Garry's Mod on Linux/macos is awfully broken, there are plenty of more pressing issues to address than this if you really want to go that way.

DXVK is very very niche, only developers would know where to get it and how to use it, same with DGVOODOO.

As for the RX580, that one defo deserves attention along with Proton.

And I've yet to see intel GPUs

I'm also not sold on the "divide people with %", thats how you prioritize. You can't sell a feature to rubat if you mention only the 1%

[timber narwhal]

sometimes it feels like nothing in gmod is private 😔

[hexed parrot]

I think when they're talking about dxvk that's in the context of proton? From what I read in their server anyway.

[stark valve]

oh

I didnt know proton uses that

in which case ok

[snow cradle]

How to detect water in frame?

[wispy quartz]

nah that's cringe

dxvk doesn't put d3d9.dll inside gmod's folder

[wispy quartz]

you make ssao before rendering translucent stuff

[snow cradle]

here d3d9.dll in tar gz
https://github.com/doitsujin/dxvk/releases

[wispy quartz]

it still doesn't put it inside gmod's folder

it's contained under System32

[snow cradle]

but it create gmod dxvk logs in gmod folder after putting d3d9.dll inside Garrysmod/bin folder

btw reshade same has d3d9.dll

and it not needed to put it into system32 folder

[vapid reef]

because thats how loading libraries works

order is : Folder > System32

Typically dxvk is used from linux so the system d3d9.dll wine uses will be the one from dxvk

[snow cradle]

hm ok

[snow cradle]

The problem is also much deeper than it seems. Forgetting about players on Linux, we automatically forget about them on the servers that will install shaders. This problem scales as it is examined in detail, shifting to the owners and developers of the servers.

[snow cradle]

PostDrawOpaqueRenderables or PreDrawTranslucentRenderables not helps

[snow cradle]

https://aras-p.info/blog/2016/04/08/solving-dx9-half-pixel-offset/

[snow cradle]

Getting bumps from FrameBuffer. I used the method for removing edge defects as in ACCURATE method reconstruction of Normals.
Due to using DX9 and the pixel shift problem, I had to shift by 1 pixel float depth_c = tex2D(WPDepth, (i.P - 1)*TexBaseSize ).a;.

float4 main(PS_IN i ) : COLOR
{   
    float2 vpos = i.vTexCoord;

    float2 ts1 = float2(1.0, 0.0) * TexBaseSize;
    float2 ts2 = float2(0.0, 2.0) * TexBaseSize;
    float2 ts3 = float2(0.0, 1.0) * TexBaseSize;
    float2 ts4 = float2(2.0, 0.0) * TexBaseSize;

    float3 colorRight   =   tex2D(FrameBuffer, vpos + ts1 ).rgb;
    float3 colorLeft    =   tex2D(FrameBuffer, vpos + float2(-1.0, 0.0) * TexBaseSize ).rgb;
    float3 colorTop     =   tex2D(FrameBuffer, vpos + ts3 ).rgb;
    float3 colorBottom  =   tex2D(FrameBuffer, vpos + float2( 0.0,-1.0) * TexBaseSize ).rgb;

    float depth_c = tex2D(WPDepth, (i.P - 1)*TexBaseSize ).a;

    float4 H = float4(
        tex2D(WPDepth,vpos - ts1).a,
        tex2D(WPDepth,vpos - ts4).a,
        tex2D(WPDepth,vpos + ts1).a,
        tex2D(WPDepth,vpos + ts4).a
    );

    //vertical depths
    float4 V = float4(
        tex2D(WPDepth,vpos - ts3).a,
        tex2D(WPDepth,vpos - ts2).a,
        tex2D(WPDepth,vpos + ts3).a,
        tex2D(WPDepth,vpos + ts2).a
    );

    //find diff of true center and extrapolated one
    float2 he = abs((2.0*H.xz - H.yw) - depth_c);
    float2 ve = abs((2.0*V.xz - V.yw) - depth_c);

    //from which direction to sample
    float gradientX = he.x+bias < he.y ? 0 : Luminance( colorRight - colorLeft );
    float gradientY = ve.x+bias < ve.y ? 0 : Luminance( colorTop - colorBottom );

    float3 normal = float3(gradientX * strength, gradientY * strength, 1.0);
    normal.z = sqrt(max(0.0, 1.0 - normal.x * normal.x - normal.y * normal.y));

    return float4(normal * 0.5 + 0.5, 1);
};

screenspace_general
{
    $pixshader         "pp_bumps_test38_ps30"
    $vertexshader     "pp_vs30"

    $basetexture         "_rt_FullFrameFB"
    $texture1         "_rt_WPDepth"

    $LINEARREAD_BASETEXTURE     1
    $linearwrite             1

    $c0_x 5
    $c1_x 0.000009 // bias

    "<dx90"
    {
        $no_draw 1
    }
}

idk but here i need small bias to remove depth deffect lol

btw i use depth=1/depth/4000 or depth=1/(depth*4000)

[snow cradle]

The essence of the method is to simply zero the bump in the edge zone. Edges can be defined in any way.

[stark valve]

                        pixel_mat:SetString("$pixshader", shader)
                        pixel_mat:Recompute()```

Can you not do that? It doesnt seem to work for me

pixel_mat is an IMaterial

[wispy quartz]

Im sure you missed something on it

[snow cradle]

no

[snow cradle]

I think i can make support of 3d skybox for depthbuffer as pass where in depth == 1 make 3d sky depth render. depth from sky i will write using DepthWrite Shader. so i will save treesway support and alphatest. this pass can be optional. also i will render 3d sky as combined meshes (if mesh has no treesway and no sorted by alphatested texture). but it will requre NikNaks

[snow cradle]

function shaderlib.GetViewProjZ(viewData)
    if !ismatrix(viewData) then viewData = shaderlib.GetViewProjMatrix(viewData) end
    return Vector4(viewData:GetField( 3,1 ),viewData:GetField( 3,2 ),viewData:GetField( 3,3 ),viewData:GetField( 3,4 ) )
end

getting ViewProjZ from VIewProj matrix

fog calculation:

float flProjPosZ = dot( float4(1/tex2D(WPDepthBuffer,i.vTexCoord).xyz,1), ViewProjZ );
float3 fogParams = i.fogParams;
float fog = max( fogParams.z, ( flProjPosZ * fogParams.y + fogParams.x ) );
fog = pow(fog,fog_power);

[blazing mason]

let's say I have an entity that is essentially a prop, I then set the material of that entity to a vmt that uses screenspace_general, with all the proper flags so it shows up on a model correctly
let's say I also use the custom inputs, in this example, just $c0_x
if in the entity's lua, I change $c0_x, does it change it for every instance of the material? or just the specific entity

if it's every instance of the material I think what I'm trying to do is impossible then

looks like its for every instance based on testing with my current implementation....

shared.lua

ENT.Type = "anim"
ENT.Base = "base_anim"
ENT.Category = "Test"
ENT.PrintName = "Test"
ENT.Spawnable = true
ENT.Mat = Material("gmod_shader_guide/test.vmt")
ENT.Index = 0
ENT.AutomaticFrameAdvance = true

function ENT:Think()
  self.Mat:SetFloat("$c0_x", self:GetNWInt("test1"))
end

function ENT:Use( activator )
  self.Index = self.Index+1
  self:SetNWInt("test1", self.Index)
  print(self:GetNWInt("test1"))
end

function ENT:Initialize()
  self:SetModel( "models/test/cube.mdl" )
  self:SetMaterial(self.Mat:GetName())
  self:PhysicsInit(SOLID_VPHYSICS)
  self:SetMoveType(MOVETYPE_VPHYSICS)
  self:SetSolid(SOLID_VPHYSICS)
  if SERVER then
    self:SetUseType(SIMPLE_USE)
  end
  self:Activate()
end

test.vmt

screenspace_general 
{
  $pixshader "test_ps20b"
  $vertexshader "test_vs20"
  $basetexture "gmod_shader_guide/al_child"
  $texture1 "gmod_shader_guide/al_child_palette"
  $POINTSAMPLE_TEXTURE1 1
  $POINTSAMPLE_TEXTURE2 1
  $c0_x 0
  $cull 1
  $depthtest 1
  $model 1
  $softwareskin 1
  $vertexnormal 1
  $linearwrite 1
}

essentially in the shader I am changing the color of the material based on c0_x

spwaning multiple instances of the entity, they all are the same color as the most recently spawned one, ignoring their own index variable, presumably because the custom shader does not instance itself per instance of the material?

[blazing mason]

I had assumed because I'm creating an IMaterial object for each entity by doing ENT.Mat = Material("gmod_shader_guide/test.vmt"), that it was creating multiple instances of each material. Which it's probably doing, but all those material use the same shader, which doesn't instance itself?

[blazing mason]

nevermind, just tried creating a material on the fly, an instance for each and that solution works.
so I think it's a matter of I'm somehow using ENT:SetMaterial() incorrectly with Material()?

currently using CreateMaterial(), but won't work long-term for what I'm trying to do, so its not ideal over having a .vmt that i reference with Material()

[hollow pollen]

if you want multiple entities with the same shader that have different material values, you will need multiple materials

Material() just references an existing one

so your only option in this regard is CreateMaterial

[stark valve]

You can use the same material instance so long as you make sure each of the critical shader parameters are refreshed before the entity is drawn

https://github.com/Metastruct/Arcana/blob/master/lua/entities/arcana_mana_crystal.lua

check the :Draw function

[snow cradle]

perspective matrix:

float4 offset = float4(samplePos, 1.0);
//offset = mul(offset,g_ProjMatrix);
offset.x = samplePos.x*ViewProjX.x+samplePos.y*ViewProjX.y+ViewProjX.w;
offset.y = samplePos.x*ViewProjY.x+samplePos.y*ViewProjY.y+samplePos.z*ViewProjY.z+ViewProjY.w;
float raw_z = samplePos.x*ViewProjZ.x+samplePos.y*ViewProjZ.y+samplePos.z*ViewProjZ.z;
offset.z = raw_z+ViewProjZ.w;
offset.w = -raw_z+ViewProjW.w;

[0.34870,    0.52388,    0.00000,    -2429.19531]
[0.12408,    -0.08259,    -1.10883,    76.26141]
[0.82504,    -0.54915,    0.13322,    -560.02454]
[-0.82504,    0.54915,    -0.13322,    561.02454]

ViewProjZ.xyz == -ViewProjW.xyz only .w is different

getting only projPos.xy:

float2 shadowCoord;
shadowCoord.x = wp_offset.x*ViewProjX.x+wp_offset.y*ViewProjX.y+ViewProjX.w;
shadowCoord.y = wp_offset.x*ViewProjY.x+wp_offset.y*ViewProjY.y+wp_offset.z*ViewProjY.z+ViewProjY.w;

[snow cradle]

p.s. its works

[snow cradle]

Getting ProjPosZ from ResolvedFullFrameDepth https://wiki.facepunch.com/gmod/render.GetResolvedFullFrameDepth:

float projPozW = tex2D(ResolvedFullFrameDepth, uv)*4000;
float raw_z = projPosW-ViewProjW.w;
float projPosZ = -raw_z+ViewProjZ.w;

ResolvedFullFrameDepth got from projPozW https://github.com/ValveSoftware/source-sdk-2013/blob/68c8b82fdcb41b8ad5abde9fe1f0654254217b8e/src/materialsystem/stdshaders/depthwrite_vs20.fxc#L115

But having 2 digits, we can make ProjPosZ out of it

default way to get ProjPosZ:

float flProjPosZ = dot( float4(worldPos,1), ViewProjZ );

This is useful to calculate the fog in the shader and reduce the effect of the shader on the fog.

Making Fog using ResolvedFullFrameDepth:

float4 main(PS_IN i ) : COLOR
{   
    float projPosW = tex2D(ResolvedFullFrameDepth, i.vTexCoord)*4000;
    float raw_z = projPosW-ViewProjW.w;
    float flProjPosZ = -raw_z+ViewProjZ.w;
    float3 fogParams = i.fogParams;
    float fog = max( fogParams.z, ( flProjPosZ * fogParams.y + fogParams.x ) );
    fog = pow(fog,fog_power);
    return float4(fog,fog,fog,1);
};

[snow cradle]

Fog works:

mat_fog:SetFloat("$c3_x", viewProj:GetField(4,4))
mat_fog:SetFloat("$c3_y", viewProj:GetField(3,4))

or u can get ProjPosZ from ProjPosW using precomputed value:

-- lua
mat_fog:SetFloat("$c3_x", viewProj:GetField(4,4) + viewProj:GetField(3,4))

// hlsl
float flProjPosZ = W_X - tex2D(ResolvedFullFrameDepth, i.vTexCoord).a*4000;

but viewProj:GetField(4,4) + viewProj:GetField(3,4) == 1

so ViewProjZ == 1 - ViewProjW

[snow cradle]

offset.z = samplePos.x*ViewProjZ.x+samplePos.y*ViewProjZ.y+samplePos.z*ViewProjZ.z+ViewProjZ.w;
offset.w = 1-offset.z;

Final version:

float4 offset = float4(samplePos, 1.0);
offset.x = samplePos.x*ViewProjX.x+samplePos.y*ViewProjX.y+ViewProjX.w;
offset.y = samplePos.x*ViewProjY.x+samplePos.y*ViewProjY.y+samplePos.z*ViewProjY.z+ViewProjY.w;
offset.z = samplePos.x*ViewProjZ.x+samplePos.y*ViewProjZ.y+samplePos.z*ViewProjZ.z+ViewProjZ.w;
offset.w = 1-offset.z;

Thus, in any layout - both in the ortho projection and in the perspective projection - the last row is not used, so the matrix can be transmitted as float4x3, and not float4x4 - during manual calculation.
That is, the last row of the matrix can be left for any data you want.

[vague jetty]

Did you ever solve this? I'm having issues with the same thing. My shader rendering code that worked a long time ago no longer works, presumably from an update. I'm able to make it work with this code:

    local shader = Material("effects/shaders/testshader")

    hook.Add("RenderScreenspaceEffects", "MyScreenOverlay", function()
        cam.Start2D()
            surface.SetMaterial(shader)
            surface.SetDrawColor(255, 255, 255, 255)
            surface.DrawTexturedRectUV( -1, -1, 2, 2, 0, 1, 1, 0 )
        cam.End2D()
    end)```
And it works with r_screenoverlay, but my old code does not work and instead draws incorrectly on the top right quad.
```lua
    local shader = Material("effects/shaders/testshader")

    hook.Add("PostDrawTranslucentRenderables", "TranslucentTextDisplay", function()
        render.UpdateScreenEffectTexture()
        render.SetMaterial(shader)
        render.DrawScreenQuad()
    end)```

I'm trying to render it in PostDrawTranslucentRenderables so I can render certain UI elements properly.

[snow cradle]

Sponza map ft. СЛон for shader dev and tests

[snow cradle]

https://steamcommunity.com/sharedfiles/filedetails/?id=3574744621

[snow cradle]

@wispy quartz IA88 is works good for u? (for linux)

linux user screen:

he have colors in IA88 lol

R32F looks like red for linux

@wispy quartz
Is your R32F not working? As far as I remember.

wtf

lua say: dx90
mat_dxlevel say: dx92

[wispy birch]

mate you know u can just test it all under wsl right

[wispy birch]

mat_dxlevel might just be the target, that might not be available at all times

[snow cradle]

how?

[wispy birch]

windows subsystem for linux
officially supported linux under windows (if you run win11 you have graphics by default)

google the installation process if you want it

[snow cradle]

In general, I'd also like to hear @wispy quartz's opinion on this whole IA88 situation.

Because, for example, I don’t understand how the colors were preserved in IA88

And I wonder what could be connected with this.

[stark valve]

Running steam on that has been a pain

I do remember spending quite a long time trying to setup gmod for it

And it didn’t even work

[blazing mason]

... is there any way to get flex/morph data using screenspace_general? or am I shit outta luck

[misty surge]

Not really, unless you make the bone tree yourself somehow, and include bone weights, at that point you might as well use VertexLitGeneric because this is supported, but screen space is limited to screenspace mainly. There’s only reconstruction available because 3D geometry isn’t processed the way a 2D shader is. Screenspace uses your screen to make shaders, when the ideal way for 3D is to include skins in the render call which refers to the bone tree.

A mesh you produce outside the screen space meaning you can inject your mesh data, it’s possible to make somewhat of a replacement, but it won’t work without properly distributing weights and bone IDs, which is called in a normal draw model, but not in screen space.

[blazing mason]

yeah I ended up having to go with a workaround

[snow cradle]

tangents, binormals for models:

no $flags2, no $softwareskin, #define COMPRESSED_VERTS 1

[timid pumice]

That's fucking huge

[stark valve]

is the "carved in" part your shader?

not sure what tangents/binormals mean in this context

[snow cradle]

Tangents and binormals are needed for parallax, specular highlights, etc. Many couldn't get them from a vertex shader. But it's possible.

I don't have a ready-made parallax shader, but I know a working way to get tangents and binormals, and I've uploaded it.

[stark valve]

its very nice

[snow cradle]

btw i try to add flashlight support. now i add fog support, 6face lighting. and now i calc matrix of flash on lua (3st steen my wip test)

we can add flashlight support using only 1 texture and some data

with no NoiseSampler (shadows disable)

at final screespace general can get light, fog, depth, flashlight support but only for dynamic ents (to get info from lua)

we can make 2-3 lighting support for bumps and spec

now i have textures:

1. basecolor.rgb
2. bump.xy, height.z
3. cubemap
4. flashlight tex

we need 5th, 6th texture
https://github.com/Facepunch/garrysmod-requests/issues/2896

[misty surge]

I love it. It’s definitely looking fantastic, I think I had a working version on EGSM but with screen space shaders? That’s epic.

[snow cradle]

I8 and IA88 not works on proton. use it to detect proton
https://github.com/solsticegamestudios/GModPatchTool/blob/master/examples/detection_example.lua

[snow cradle]

what is that?
https://wiki.facepunch.com/gmod/render.GetFullScreenDepthTexture

[graceful lynx]

I believe this is just the depth map in the red color channel

[nocturne patio]

Does anyone know how to pass the intensity of lights around the mesh to the shader?
I'm trying to draw a model with basic lighting to make it look just like VertexLitGeneric.

My current setup is as follows:
I tried to use render.GetLightColor but it brightens up all the surface uniformly like the screenshot below.

// Pixel shader
sampler Albedo : register(s0); // $basetexture
const float4 c0 : register(c0);
struct PS_INPUT {
    float2 pos : VPOS;
    float2 uv  : TEXCOORD0;
};
float4 main(const PS_INPUT i) : COLOR0 {
    float4 albedo = tex2D(Albedo, i.uv);
    return c0 * albedo;
}

// Vertex shader
const float4x4 cModelViewProj : register(c4);
struct VS_INPUT {
    float4 pos    : POSITION0;
    float2 uv     : TEXCOORD0;
    float3 normal : NORMAL0;
};
struct VS_OUTPUT {
    float4 pos : POSITION;
    float2 uv  : TEXCOORD0;
};
VS_OUTPUT main(const VS_INPUT v) {
    VS_OUTPUT output;
    output.pos = mul(v.pos, cModelViewProj);
    output.uv = v.uv;
    return output;
}

"Screenspace_General"
{
    "$pixshader"    "pixelshader"
    "$vertexshader" "vertexshader"
    "$basetexture"  "uvtest"

    "$cull"         "1"
    "$depthtest"    "1"
    "$model"        "1"
    "$softwareskin" "1"
    "$vertexnormal" "1"
}

local mdl = ClientsideModel "models/props_junk/CinderBlock01a.mdl"
mdl.RenderOverride = function(self)
    render.OverrideDepthEnable(true, true)
    self:DrawModel()
    render.OverrideDepthEnable(false)
end
local mat = Material "myshader_vmt"
local light = render.GetLightColor(mdl:GetPos())
mat:SetFloat("$c0_x", light.x)
mat:SetFloat("$c0_y", light.y)
mat:SetFloat("$c0_z", light.z)
mat:SetFloat("$c0_w", 1)

[wispy birch]

you should have this from the engine from the start

[wispy birch]

just look at what LitmappedGeneric uses inside of itself, and youll be good to go

[nocturne patio]

thanks, I'll try

[hollow pollen]

render.setlocalmodellights, then you can use the cLightInfo array in the vertex shader

ideally source would set the lights up for you but I’m unsure if thats the case

[nocturne patio]

Interesting!

but I set the first light's dir to Vector(1, 1, 1) but it seems I'm getting (1, 0, 0) in the pixel shader.

// Vertex shader
#include "common_vs_fxc.h"
struct VS_INPUT {
    float4 pos    : POSITION0;
    float2 uv     : TEXCOORD0;
    float3 normal : NORMAL0;
};
struct VS_OUTPUT {
    float4 pos   : POSITION;
    float2 uv    : TEXCOORD0;
    float4 light : COLOR0;
};
VS_OUTPUT main(const VS_INPUT v) {
    VS_OUTPUT output;
    output.pos = mul(v.pos, cModelViewProj);
    output.uv = v.uv;
    output.light = float4(cLightInfo[0].dir.xyz, 1.0);
    return output;
}

// Pixel shader
float4 main(const PS_INPUT i) : COLOR0 {
    return tex2D(Albedo, i.uv); * float4(i.light.rgb, 1.0);
}

local function RenderOverride(self)
    render.SetLocalModelLights({
        {
            pos = self:GetPos() - Vector(50, 50, 50),
            dir = Vector(1, 1, 1),
            color = Vector(1, 1, 0),
            type = MATERIAL_LIGHT_DIRECTIONAL,
        }
    })
    render.OverrideDepthEnable(true, true)
    self:DrawModel()
    render.OverrideDepthEnable(false)
end

[hollow pollen]

you will need to use render.supressenginelighting to prevent source from internally overriding those values

[nocturne patio]

Thank you!! it seems working now 🤗

I wish I could use render.SetModelLighting and then cAmbientCubeX/Y/Z 🤔
Sometimes it seems working, though

[nocturne patio]

Drawing VertexLitGeneric once fixed the problem with cAmbientCubeX/Y/Z 🤗

local function RenderOverride(self)
    render.SuppressEngineLighting(true)
    render.OverrideDepthEnable(true, true)
    local rainbow = HSVToColor(CurTime() * 120 % 360, 1, 1):ToVector()
    local dummy = Material "metal2"
    render.SetModelLighting(BOX_FRONT, rainbow:Unpack())
    render.SetModelLighting(BOX_LEFT, Color(255, 0, 0):ToVector():Unpack())
    render.SetModelLighting(BOX_TOP, Color(0, 255, 255):ToVector():Unpack())
    render.MaterialOverride(dummy)
    self:DrawModel()
    render.MaterialOverride()
    self:DrawModel()
    render.OverrideDepthEnable(false)
    render.SuppressEngineLighting(false)
end

// Vertex shader
#include "common_vs_fxc.h"
struct VS_INPUT;
struct VS_OUTPUT; // same as above
VS_OUTPUT main(const VS_INPUT v) {
    VS_OUTPUT output;
    output.pos = mul(v.pos, cModelViewProj);
    output.uv = v.uv;
    output.light = AmbientLight(v.normal);
    return output;
}

[wispy birch]

i guess the engine sets it in the VertexLitGeneric's prelude

[nocturne patio]

Honestly I don't want to draw a model twice but I guess this is the best bet for now.

[nocturne patio]

It turned out that I can use DoLighting if I draw VertexLitGeneric once

local function RenderOverride(self)
    render.MaterialOverride(vertexlitgeneric)
    render.SetBlend(0) -- Do not actually render
    self:DrawModel()
    render.SetBlend(1)
    render.MaterialOverride()
    render.OverrideDepthEnable(true, true)
    self:DrawModel()
    render.OverrideDepthEnable(false)
end

// Vertex shader
#include "common_vs_fxc.h"
struct VS_INPUT;
struct VS_OUTPUT; // same as above
VS_OUTPUT main(const VS_INPUT v) {
    VS_OUTPUT output;
    output.pos = mul(v.pos, cModelViewProj);
    output.uv = v.uv;
    output.light = DoLighting(v.pos, v.normal,
        float4(0.0, 0.0, 0.0, 0.0), true, true, true);
    return output;
}

[snow cradle]

@nocturne patio

shaderlib.BuildPerspectiveWorldToFlashlightMatrix is afloat4x4 g_FlashlightWorldToTexture:

function shaderlib.BuildWorldToShadowMatrix(translation, ang) // View Flashlight
    local vForward = ang:Forward()
    local vLeft = ang:Right()
    local vUp = ang:Up()
    
    local matBasis = Matrix()
    matBasis:SetForward(vLeft)
    matBasis:SetRight(vUp)
    matBasis:SetUp(vForward)
    local matWorldToShadow = matBasis:GetTransposed()

    translation = matWorldToShadow * translation
    translation = -translation

    matWorldToShadow:SetTranslation(translation)

    matWorldToShadow:SetField(4,1,0)
    matWorldToShadow:SetField(4,2,0)
    matWorldToShadow:SetField(4,3,0)
    matWorldToShadow:SetField(4,4,1)

    return matWorldToShadow
end


function math.cot(x) // cotangent
    return 1 / math.tan(x)
end

function shaderlib.BuildPerspectiveWorldToFlashlightMatrix(viewSetup) // ViewProj for Flashlight g_FlashlightWorldToTexture
    local pos, ang = viewSetup.origin, viewSetup.angles
    local mFlashlightView = shaderlib.BuildWorldToShadowMatrix(pos, ang)

    local fov = viewSetup.fov
    fov = math.cot( math.rad(fov * 0.5)  )
    local f = viewSetup.zfar
    local n = viewSetup.znear
    local aspect = viewSetup.aspect or 1

    local mProj = Matrix({
        {    fov,  0,            -0.5,              0,            },
        {    0,  fov*aspect,        -0.5,              0,            },
        {    0,     0,                1,                 1,            },
        {    0,  0,                -1,             0            }
    })

    mProj:Mul(mFlashlightView)

    return mProj
end

sampler FlashlightSampler           : register(s3);

const float4x4 g_FlashlightWorldToTexture : register( c15 );
#define FLASHLIGHT bool(Constants0.w > 0)
#define g_FlashlightPos EyePosition
#define m_fConstantAtten 0
#define m_fLinearAtten 100
#define m_fQuadraticAtten 0
#define m_FarZ 750
#define g_FlashlightAttenuationFactors float4(m_fConstantAtten, m_fLinearAtten, m_fQuadraticAtten, m_FarZ)

#define bDoShadows          false   // need RandRotSampler
#define bAllowHighQuality   false
#define bClip               false
#define nShadowLevel        int(0)

// code
if (FLASHLIGHT) {
    float4 flashlightSpacePosition = mul( float4( worldPos, 1.0f ), g_FlashlightWorldToTexture );

    diffuseLighting += DoFlashlight( g_FlashlightPos, worldPos, flashlightSpacePosition, worldNormal,
    g_FlashlightAttenuationFactors.xyz,
        g_FlashlightAttenuationFactors.w, FlashlightSampler, FlashlightSampler, // 
        FlashlightSampler, // RandRotSampler
        nShadowLevel, bDoShadows, bAllowHighQuality, i.pPos, bClip);

    pixLight += diffuseLighting;
}

vmt:

$texture3             "effects\flashlight001"  // FlashlightSampler

default view matrix +- same as flashlight view matrix

they have very minor differences

I mean in terms of the result

local viewSetup = render.GetViewSetup()
viewSetup.aspect = 1
viewSetup.fov = GetConVar("r_flashlightfov"):GetFloat()
flashlightMatrix = shaderlib.BuildPerspectiveWorldToFlashlightMatrix(viewSetup)

this is g_FlashlightAttenuationFactors. but i hardcode it. i have no free constants

local atten = Vector4(
    GetConVar("r_flashlightconstant"):GetFloat(),        // Set the flashlight attenuation factors
    GetConVar("r_flashlightlinear"):GetFloat(),
    GetConVar("r_flashlightquadratic"):GetFloat(),
    GetConVar("r_flashlightfar"):GetFloat()
)

This is in case you want to make a game-ready shader.

This will be useful to you

[wispy quartz]

its an alias of main rendertarget

well, a directx alias

[wispy quartz]

Don't use it then, that's all

There's no any particular reason to actually do any of that since vram size have changed from 256-512mb up to 16gb

even low end video cards have minimum 4gb of video memory, that's really huge for gmod

[snow cradle]

rave

No matter how long I've been here, I constantly hear destructive comments:

1. "Linux? Who the hell uses Linux? Just cut off their support, that's all. They only play on Windows."
2. "IA88? What kind of format is that? Increase the video memory consumption, that's all."

We need to think about all users—Linux users, weaker PCs. Everything. We can't ignore it. Nothing. Your idea is negligent.

With your thoughtless actions, you are weeding out players due to your recklessness.

[snow cradle]

https://wiki.facepunch.com/gmod/Shaders/DepthWrite
https://github.com/Facepunch/garrysmod-requests/issues/2939

Completing the request will help avoid similar depth buffer bugs.

[wispy quartz]

Your shader stuff is already unoptimized and laggy stuff that will work only on gpus that are way more powerfull than they were back in 2004

You trying to squeze extra 2 megabytes in vram is dumb considering the amount of vram you consumed already via your extensive usage of deffered rendering technique, not talking about the sttuff that you used multiple render passes to do stuff that could be done in forward rendering

that was a silly attempt to gaslight me dude

[snow cradle]

Deferred rendering render targets consume a significant amount of video memory.

worldpos + depth = 32323232F bits, despite the fact that a separate R32F depth buffer already exists.

World normals can be stored in either RGBA16F or RGBA32F. They could also, in theory, be stored as RGBA8 (alpha channel - sign of tangents).
Also for Worldnormals IMAGE_FORMAT_RGBA1010102 would be nice, but it is not available to us.

But I don't use multiple render passes, that is, I don't call any render.RenderView, but reconstruct the data from the depth buffer.
Btw, if it is possible to use IA88 and I8 for other purposes, then there is no point in abandoning them.

You don't determine where the necessary video memory consumption is and where it's negligence.

[wispy quartz]

So you consume 64 megabytes already, then whats the difference in those 2 megabytes anyway?

[snow cradle]

The difference is in the cumulative effect.

[wispy quartz]

Yes i do determine it, since system specs of gmod did already increase

There's no reason to overoptimize some unesufficient stuff since the way you do your unreasoned shading methods are already unoptimized

"wow my stuff needs 8mb of vram!!! But it requires rtx 5090 to actually compute my screenspace shadows!"

that's how you sound

like

[wispy quartz]

Original hl2 requires only 128MB of vram, that's 10 times increase of vram usage just so the modders could use it for 4k anime textures

and that's minimal system specs required for gmod

[snow cradle]

1080p:

WorldPos + depth.a rgba32f: 32,4m
resolved full frame buffer 32f: 8,1m
WorldNormals+Tangent - Saving method: ||8 bit 8,1m||/16 bit 16.2m/32 bit 32,4m

It would be nice to make WorldPos as IMAGE_FORMAT_RGB323232F = 28 (3 layers), without adding depth buffer to alpha channel, to save video memory, but this image format is not available to us.
Also i haven't made 8-bit normals yet, but I have plans to implement it. Because it is important to save video memory.

So, ideally (from the point of view of video memory), I would do it like this:

WorldPos: IMAGE_FORMAT_RGB323232F
Depth: IMAGE_FORMAT_RGB32F
NormalsTangents: IMAGE_FORMAT_RGBA1010102

But the only render targets that allow me to save any video memory are IA88 and I8 in specific cases. And they should be used if possible.

[wispy quartz]

8mb is not a real deal in computing costs, therefore you can just use that rt and store something elsewhere in it, so yeah, you do it basicalaly for no reason

and your worldpos/depth stuff is actually 256mb

[wispy quartz]

you do realise that 32 bit * 4 * 1920 * 1080 is not 32mb

same stuff for your normals, its 128mb not 32mb too

[snow cradle]

The weight has already been calculated here. One channel weighs 8.1 megabytes (32 bits).
Four channels (32 bits) weigh 8.1 x 4 = 32.4. All numbers correspond to the screenshots.

I can also mention the importance of saving video memory in an article by MJP, Lead Rendering Programmer at Sony Santa Monica Studio, where he discusses the attitude of video game developers and video memory consumption:

https://mynameismjp.wordpress.com/2015/02/18/shadow-sample-update/
The paper primarily suggests two variants of the technique: one that directly stores the 1st, 2nd, 3rd, and 4th moments in an RGBA32_FLOAT texture, and another that uses an optimized quantization scheme (which essentially boils down to a 4×4 matrix transform) in order to use an RGBA16_UNORM texture instead. The first option most likely isn’t immediately interesting for people working on games, since 128 bits per texel requires quite a bit of memory storage and bandwidth.

[wispy quartz]

You threw me an article describing the games released in 2015 for ps4/xboxone era having problems with 2gb vram barrier
Your're really thinking that's your case?

your level is literally ps3 graphics, same for effects

the issue they had with so huge vram usage is that they used pbr + higher res textures for almost everything

(Still not talking that gmod's recommended vram is 4gb, 2 times higher than that)

[hollow pollen]

why are you complaining about optimization

AAA gamedev over here

[wispy quartz]

In fact i'm not, it's him doing this all over his urge of using I8 if it's not working on linux

[wispy quartz]

my original statement here

[hollow pollen]

fair enough wtf is that format

[wispy quartz]

I8? Used for black/white stuff, like digit clocks, ui elements, w/e
Same thing for IA8, which can be used in ui too

IA8 also holds alpha channel, but i don't think squezing anything of that cuz of 1 rt is like a huge need to actually hack around that to support linux / ditch linux cuz of that (including proton/dxvk)

[hollow pollen]

tbf this kind of thing happens all the time, he can just have different pipelines for different systems

if (windows) {I8} else {RGBA8888}

[wispy quartz]

since there's already more huge bottlenecks in akabenko's shaders, such as tflops/fillrate limitations for gpus
So it's not really supporting so budget gpus (like 560ti/etc) with low vram available since they couldn't really handle shadowmaps/volumetric lighting with descent fps

It's just decreasing vram usage for almost no reason

[wispy quartz]

Proton also does that shit too, so yeah....

well, as well as windows users with dxvk applied

[hollow pollen]

memory consumption is the bottleneck in most applications, cache optimization is more important than youd think

but youre probably right for his cases

[wispy quartz]

In the way i meant is that computational powers of GPU is his bottleneck right now

[hollow pollen]

is it? if hes doing CSM and stuff thats multiple texture lookups per pixel

[wispy quartz]

not VRAM usage.
The fps with like 100mb of vram usage will be around the same as with 10gb vram usage basically (if gpu has 10gb of memory, of course)

[wispy quartz]

Well he also computes a lot of stuff with reconstructing that from depth buffer, which also tanks texture lookups as well as fps

[hollow pollen]

oh yeah if hes using that F32 depth reconstruction that’d kill any old gpu

[wispy quartz]

There's actually a way to somehow fix it, like using vertex shader to precompute offsets and use it in pixel shader, but it gives like 10% of perf boost

[wispy quartz]

I saw that technique in someone's paper with dual kawase's blur

[hollow pollen]

could maybe precompute the direction in the vertex shader but you’d still need that texture lookup

[wispy quartz]

that's what i said

[hollow pollen]

ya

[wispy quartz]

it will have so tricky optimizations by gpu, but still

[wispy quartz]

And well

system specs of gmod is huge

a lot of space for modders

and shaders

to not care about additional 8mb of vram memory

[hollow pollen]

its mostly about optimizing those texture lookups

a GPU cache is like 128kb or something

big performance difference between sampling a higher res texture and lower res

I presume same goes for RGBA8888 and I8

~3x less “compute”

[wispy quartz]

Yet still it'll be better to just do that stuff in forward rendering way
using info from vertex shader

(just use new assets/maps with your own shaders, that's all)\

waiting for CustomGeneric from ficool2 / Material:SetShader return

[hollow pollen]

i wouldnt hold ur breath

lol

[wispy quartz]

nah i'd still do my stuff with screenspace_generic stuff, i don't really think doing that all work with post processing stuff would really make such a difference vs doing that shit with remade assets + replacing default shaders to custom ones

Like ditching all source lighting and baking it into textures myself, like that guy did with blender
https://www.youtube.com/watch?v=so1TWKkfhXQ

That will really result in good looking graphics in gmod, i think

[hollow pollen]

holy crap lol

clean as hell

[wispy quartz]

That's nothing new at all!

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

I mean, it's just people didn't use this custom shit since it requires baking that into models, etc

[hollow pollen]

right

[wispy quartz]

and therefore you have to ditch brushes (only use them for pvs/collision, etc)

and use models

well, basically it's possible to bake all that shit into brushes

just will need custom compiler toolkit

custom shaders for brushes

etc

[wispy quartz]

Btw that's what cs go actually did, they ditched brushes like in 2016 or so, their last maps were mainly made by models, like ancient

but that's made mainly for other reasons

[hollow pollen]

¯_(ツ)_/¯

[wispy quartz]

well my rant is over

[wispy quartz]

i'd stick to this way to enchancing graphics in source 1
instead of relying to hacky workarounds about custom shadowmaps/volumetric lighting

and w/e akabenko is doing rn

[hollow pollen]

source is old as hell

I think the bad graphics give it charm

if you wanted “realism” you had reshade for that

[wispy quartz]

yup, that's why ditching a lot of the stuff as lighting/default shaders will have a big diff

[hollow pollen]

what akabenko is doing is cool though

im not gonna stop him from doing a personal project

[wispy quartz]

Well, no one does this rn so he's the only one to do it rn

[placid stream]

I'm not even sure if most GPUs natively support I8 or IA8, they might just end up getting converted to something like BGR888 or BGRA8888 when they reach the GPU

[hollow pollen]

be the change you want to see in this world

https://tenor.com/view/chill-guy-chill-guy-just-a-chill-guy-relax-gif-1579445140409713118

[wispy quartz]

I want to kill gmod

how can i do that lol

🤔

[hollow pollen]

by not playing it

[wispy quartz]

i still do it but gmod is not dead yet

what to do then

[hollow pollen]

make a better sandbox game

[wispy quartz]

i'll call it roblox

still people play gmod for w/e reason

🙁

[hollow pollen]

gmod isnt run by pedos

[timber narwhal]

dont worry garry'll do it eventually

[wispy quartz]

still better than nazi-pedos

GMOD 1942 RP

[hollow pollen]

lol

[wispy quartz]

with minors... you know what minors i mean

[wispy quartz]

i recall even garry himself went into that stuff and banned anything of it

like made a statement

no nazi

no nsfw

etc

[wispy quartz]

After that i'll wish to kill S&Box

[snow cradle]

extremism

[wispy quartz]

That shit literally runs worse than TLOU2 (high/ultra graphic settings) on my pc

idk how but that's funny

[timber narwhal]

oh it'll take care of itself with how its going 😂

[snow cradle]

how to use D16 SHADOWS D32 INTZ D24X8 image formats?

[snow cradle]

and why RGBA1010102 has shader api error?

[calm root]

I am dealing with those server adverts from time to time, yea

[graceful lynx]

I like your silly images fyi

[misty surge]

If we are the shader hole, where is water?

[calm root]

in his bio (/s)

[snow cradle]

https://steamcommunity.com/sharedfiles/filedetails/?id=3586554495

[jovial tide]

good news: ive been researching the depth buffer situation, because im making something and this has become an obstacle

it should be possible to fetch the hardware depth buffer, without ever having to draw a depth pass

if i get it working (and test that it works on amd + nvidia + intel), i'll forward the details to rubat

[wispy quartz]

You mean fetching RT's depth buffer? That's cool, but what about stencils? Afaik they are stored with depth buffer together. Is it possible to obtain them?

[jovial tide]

stencil should also be fetchable

i wont test this though as i dont need it

[jovial tide]

stencil will not work if MSAA is enabled

[timber narwhal]

with INTZ?

[jovial tide]

no

[timber narwhal]

cool

[jovial tide]

INTZ doesnt work with MSAA

this is RESZ

[timber narwhal]

aahhh ok

[jovial tide]

with an asterisk, because nvidia needs a hack

[timber narwhal]

last time i tried intz it was really buggy on msaa like you said

sweet

[blazing mason]

#1324464901401477191 message this?

[jovial tide]

no

[wispy quartz]

well that actually sucks, we would have to force people to disable it and use other antialiasing options, such as taa, etc

thanks for clarification though

[timber narwhal]

taa ❤️

[wispy quartz]

yeah the problem is that iirc Rubat doesn't like when lua creators focrfully change mat_antialias and etc

at least without notice

[jovial tide]

this is speculation based on what im reading

it may end up being better (or worse) in practice

[wispy quartz]

then we'll hopefully wait for your glorious work then

[graceful lynx]

[Archve] The Shader Hole

[heavy blade]

R.I.P

[rain marten]

[fickle ocean]

why

[vapid reef]

@graceful lynx Archve

[fickle ocean]

oh the new channel was hidden okay yea

[wispy birch]

archve 🗣️

[graceful lynx]

I do not have the ability to archive the thread

[vapid reef]

sir i do believe you left out the i of archive

[timid pumice]

you cannot name it archive as it dissapears due Peacekeeper, you guys should do your homework

[lunar sinew]

archyve

[graceful lynx]

[Archive] The Shader Hole

[timber narwhal]

now we can say goodbye to this thread

[graceful lynx]

🫡

[surreal parcel]

I feel like being the last poster here 🦹🏻‍♂️

[stark valve]

too bad