DDraceNetwork - developer

@Koll Potato: is your code already published?

why does the zoom level change when spectating someone?

are They zooming in and out?

does spectating someone clone their zoom?

I just spectated my dummy and it didn't zoom in, so probably ye

wait a minute, I didn't know Quads are used with fragment shaders

You gotta shade your fragments!

09:42

Ohhh, embedding shaders in maps seems like a gloriously bad idea

the new spectate zoom stuff is buggy, if i spectate myself i suddenly switch to a different zoom when i unspectate which makes no sense

risu

Ohhh, embedding shaders in maps seems like a gloriously bad idea

I never considered, that maybe this is why quads are more costly to use than tiles

09:56

aren't there any better ways to transform a texture's 4 end points or rather use polygon shapes to visualize a texture without relying on the GPU to do it? xd

All drawing is done on the gpu

09:59

You can't draw textures without some kind of fragment shader

i didn't think someone would say the most obvious shit ever

1

10:01

what I wanted to point out originally, is that this fragment shader used for all of the Quads are determining the entirety of the transform, which should be done on the CPU anyway, letting the GPU process that shit as well

10:02

meaning, the more quads you flood into the GPU, the worse performance you get: exponentially. meaning the performance decrease will be multiplicative.

10:02

or rather, once the GPU hits a limit on how much it can handle at once, it can drop the game below 60 fps or straight up slow down everything else, only to be able to draw those quads

10:04

I thought the end points of a Quad is calculated on the CPU, so that information can be cached and THEN fed into the GPU's memory when it needs it the most

louis

the new spectate zoom stuff is buggy, if i spectate myself i suddenly switch to a different zoom when i unspectate which makes no sense

I agree, if you manually zoom out while watching someone, it should keep the zoom level between, it's a bit awkward / clunky having 2 different zoom states @eidderFsT

Cellegen

what I wanted to point out originally, is that this fragment shader used for all of the Quads are determining the entirety of the transform, which should be done on the CPU anyway, letting the GPU process that shit as well

which shader are you talking about

And ya, it's a cool new feature, you get to see exactly what people are looking at, as long as they have the latest-ish version

Tater

which shader are you talking about

fragment shader, which is used by every quad

which one

10:07

link it

I literally wrote "quad" inside the repo and found both the fragment shader and vertex shader code, which builds up to the Quads we use in the program.

10:07

I don't think I need you a guide on how to find it, do I?

there is no transformation happening in the "quad" shader https://github.com/ddnet/ddnet/blob/master/data/shader/quad.frag

ddnet/data/shader/quad.frag at master · ddnet/ddnet

DDraceNetwork, a free cooperative platformer game. Contribute to ddnet/ddnet development by creating an account on GitHub.

but since someone is not that familiar with it's own ripo, it's inside data/shaders

10:08

you're welcome

doing a texture fetch at all which is required to draw the texture is likely ~100x more expensive than moving the quad points

10:09

but it does not do that anyway (edited)

even if it does it every single frame?

10:10

even if it has like hundreds of them at the same time?

they all need texture fetches as well

10:10

also where are you talking about

only when it requires it, not whenever a new frame is up

there is no transformation happening in the fragment shader

I just don't understand how this approach is faster, while it's exponentially slower with larger uses

the vertex shader is only run once per vertex which makes it about 10,000x cheaper than the fragment shader (edited)

and I'm talking about the fragment shader, getting transform positions to align colors based on that value, which I think it's not cached

10:13

regardless of this uninteresting fact, Quads are just way worse at maintaining it's performance or even exceeding a simple tile's or entity's performance in a visual test

that's because tiles use a special render mode, you cannot use that for quads

I don't wanna implement a faulty Quad system into my own map editor when I see that it performs worse, than it should be

Skeith

I agree, if you manually zoom out while watching someone, it should keep the zoom level between, it's a bit awkward / clunky having 2 different zoom states @eidderFsT

zoom now sync to server which also makes it harder to decide what zoom is being sent, what zoom is being used, what zoom is being sent to client for other people. they can all be different values. For example, if you spectable, and zoom out. what other people spectating you sees? If someone give me a exhaustive list of zoom behaviour across {LOCAL_ZOOM, SERVER_ZOOM, SERVER_ZOOM_SENT_TO_OTHERS} * {FREE_VIEW, SPECTATING_OTHERS, PLAYING, SPECTATING_SELF} which is 16 behaviours in total. we can then tackle each.

10:15

for now if you just manually zoom out while spectating to the same level when you are playing. it will stay like that

I'm not nerd enough for that, I can only give you feedback happy

eidderFsT

zoom now sync to server which also makes it harder to decide what zoom is being sent, what zoom is being used, what zoom is being sent to client for other people. they can all be different values. For example, if you spectable, and zoom out. what other people spectating you sees? If someone give me a exhaustive list of zoom behaviour across {LOCAL_ZOOM, SERVER_ZOOM, SERVER_ZOOM_SENT_TO_OTHERS} * {FREE_VIEW, SPECTATING_OTHERS, PLAYING, SPECTATING_SELF} which is 16 behaviours in total. we can then tackle each.

I don't care about the zoom level of the tee I'm spectating it should just be the client's zoom

Also, as long as at least 500 quads using HD textures can achieve a limit of 1/60th of a second performance, I'm more than happy to use any approach that is below this requirement

@Tater gm I slept like ass

gm

the auto spec cam is to make sure the cursor don't just fly out and matches dyncam. we can remove that and just allow cursors to be out of screen for zoomed players

10:16

you can turn it off just by zooming

Tater

I don't care about the zoom level of the tee I'm spectating it should just be the client's zoom

I personally disagree, it's good information, you can tell if someone is doing something wrong easier based on the things they can see

@Tater did you fix your blend mode?

no I gave up

In my own map editor, the only way I can achieve this is if I don't transform each individual end point of a mesh, but rather use a 2d rectangle shape to define a Quad and then manipulating the UV only, which in my case can be achieved without making a fragment shader

the blend mode is not the hard part

Skeith

I personally disagree, it's good information, you can tell if someone is doing something wrong easier based on the things they can see

ᶜˡᶦᵉⁿᵗ ˢᵉᵗᵗᶦⁿᵍ ᵗᵒᵍᵍˡᵉˀ

10:17

i'd prefer not to see people's terrible default zooms

I just don't like how you have to manage 2 different zoom levels in the process, rather than just having an auto feature when you reset

I can make spec zoom use client zoom automatically if auto cam is not available

I don't want the zoom level changes unless I press the zoom binds

10:18

it should just never change automatically imo (edited)

which is mostly the cases since it is not released

10:18

actually, i'm in the middle of something else, if someone can make that change it would be great

Skeith

I just don't like how you have to manage 2 different zoom levels in the process, rather than just having an auto feature when you reset

i think you need to remove auto spec cam. or make it a manual turn on. but we don't have cursors and idk what bind to use. client settings maybe but then it would be more confusing to turn it off and on.

10:19

i think to solve this, we should just remove auto spec cam

10:20

for now

why did the zoom need to change at all

it's the auto spec cam thing which is turned on and off by zooming or reset zoom. it tries to match what other players sees

10:21

free view uses client zoom still

10:21

if it doesn't then it is a bug

10:22

although currently there is no auto spec cam so it default to a internal spec zoom variable which is indeed adjusted separately

10:22

i can also just remove spec zoom and if you turn off auto spec cam you change the player zoom too.

It uses default default zoom level if they're using a version below 18090 I think, which is definitely hard to look at

true

10:23

remove the spec zoom value then

10:23

and just changes client zoom when auto spec cam is turned off, but it'd be harder to tell if player want it on tho

Would it harm the auto feature by doing this

i can also just make client zoom matches auto spec cam zoom when exiting spectate, it's like copying other people's zoom settings, and we keep the auto spec cam state too. as long as you turn off auto by zooming out and never press reset zoom when spectating other players it'll match the old behaviour

10:25

but still, removing auto spec cam might be easier (edited)

10:26

OR, we can also just make spec cursor default to off, and auto spec cam would just stop make sense, so we don't use it. if people like having two separate zoom, they can turn spec cursor on.

10:27

spec cursor is yet to have a menu settings

10:28

i mean the original goals is to not change your playing zoom when you spectating, so you always go back to what you have when you resume

I think it benefits new players to have auto work by default, since it allows them to see how good players see the game (changing their default zoom level, the ability to zoom in and out)

and auto spec cam will change it so i made it separate

I think it's a good feature

still, sounds like removing auto spec cam is easier

Easier, but I don't think it'd be better for the health of the game

1

how about just use client zoom if you self spectate