DDraceNetwork - developer

There was a discussion about 0.7 lately… why was it even made?

Good catch pro! Yea sure I will :) (@Robyt3)

@ChillerDragon I commented on https://github.com/teeworlds/teeworlds/pull/3194. Can you also port this PR to ddnet after making the change?

def-

[ddnet-web:master] 1 new commit

90d66a3 Frontpage video - def-

https://www.stableattribution.com/

Find the human art behind the A.I. images

10:15

idk how good it is tho

Bro if these images are really from ai it's completely insane

My question would be if these are human guided. AI with a humam filter can actually generate incredible art

After the last ddnet pull I am receiving during cmake build

/server/src/game/server/gamecontext.cpp: In member function ‘OnClientConnected’:
/server/src/game/server/gamecontext.cpp:1544:100: warning: ‘operator delete’ called on unallocated object ‘gs_PoolDataCPlayer’ [-Wfree-nonheap-object]
 1544 |         m_apPlayers[ClientID] = new(ClientID) CPlayer(this, NextUniqueClientID, ClientID, StartTeam);
      |                                                                                                    ^
/server/src/game/server/player.cpp:18:1: note: declared here
   18 | MACRO_ALLOC_POOL_ID_IMPL(CPlayer, MAX_CLIENTS)
      | ^

any idea how to fix? (edited)

Sounds like fails to see the overloaded delete operator. Maybe a false positive

12:11

What is your GCC version?

Jupstar ✪

What is your GCC version?

gcc version 11.3.0

Mh Debian seems to only have 10 or 12

Its compiled on Ubuntu

14:07

Distributor ID: Ubuntu
Description:    Ubuntu 22.04.1 LTS
Release:        22.04
Codename:       jammy

@Jupstar ✪ I want to continue with twgpu (once I stitched together a demo reader with libtw2) and some things are still a mystery to me in particular: how do you work around not having array texture bindings in the web? Array textures just seem so damn comfy. I would like to use them for tilemaps and in the future skins of tees. do you do a draw call for each skin/tee? Currently I hack around using array textures for tilemaps by using those 3D textures, but that will fall apart once I introduce mipmaps, since 3D/cube textures also collapse on the z axis in mip maps

14:16

do I need to create textures atlases? they seem annoying to work with

14:17

-> texture atlases also sound like they will be really annoying to create mip maps for, if that is even a thing, I'll have a look

14:18

yeah okay its possible with spacing but it still sounds weird to work with

Patiga

@Jupstar ✪ I want to continue with twgpu (once I stitched together a demo reader with libtw2) and some things are still a mystery to me in particular: how do you work around not having array texture bindings in the web? Array textures just seem so damn comfy. I would like to use them for tilemaps and in the future skins of tees. do you do a draw call for each skin/tee? Currently I hack around using array textures for tilemaps by using those 3D textures, but that will fall apart once I introduce mipmaps, since 3D/cube textures also collapse on the z axis in mip maps

There are no texture arrays? Use them for tile rendering.. Emscripten is gles3 compatible and they managed to make it work(as e.g. the webclient works) Is that a wgpu limitation or actual webgl limitation? For skins I split them on the cpu BCS gl1.x

14:39

Generally I have to say I am not really a fan of texture atlasses. They don't improve performance, they might make files a bit smaller and reduce io a bit. But they also make customization much harder I don't think u can easily mix them with skins anyway. Body is bigger than feet etc. I'd suggest for skin rendering split them or use integer sampling in shader. Or you need to render many tees at once on the GPU and want to put skins as whole in a texture atlasses? else I don't see a real use case? Also different sized skins make this harder again. Updating the texture can be costly as it is one chunk of memory. For tile maps they are a perfect fit tho^^

ah, would you suggest a draw call for each tee? the transformations of the body parts is probably better calculated on the cpu, right?

Patiga

ah, would you suggest a draw call for each tee? the transformations of the body parts is probably better calculated on the cpu, right?

Well if cpu isn't a hard limit that is probably easier. If™️ we have skins similar to 0.7 u might have to do it anyway. But since u target gles3 u can defs simply use integer sampling

14:43

In glsl

do you mean with integer sampling that I sample for an integer type instead of a float?

Yeah basically u sample the texels itself

isn't float better to get the interpolation?

Yeah it's harder . But floats have a disadvantage as u not know the mipmap and even if u know. U won't get clamped edges

14:49

Then u have texture bleeding

Jupstar ✪

Well if cpu isn't a hard limit that is probably easier. If™️ we have skins similar to 0.7 u might have to do it anyway. But since u target gles3 u can defs simply use integer sampling

hm true one draw call for each tee probably results in easier code. one of my problems is that I still have zero clue which operations are expensive to which degree. I assumed that one draw call for all tees would just be much much more efficient and cleaner. for 0.7 skins it would be possible to create a texture array for each body part I suppose

I thought glsl offers good functions. I can't find them rn tho.

Jupstar ✪

Then u have texture bleeding

okay I guess I'll just not even start with texture atlases as they will probably cause more harm than good

Jupstar ✪

I thought glsl offers good functions. I can't find them rn tho.

so far all my shaders are in wgsl, although wgpu also has optional support for spirv and glsl

Patiga

hm true one draw call for each tee probably results in easier code. one of my problems is that I still have zero clue which operations are expensive to which degree. I assumed that one draw call for all tees would just be much much more efficient and cleaner. for 0.7 skins it would be possible to create a texture array for each body part I suppose

Problem is. U somehow have to tell the GPU which tee uses which skins. And this requires either building up some buffer/CMD every frame. A draw call might carry extra dependencies about which sampler to use. But in the end it's probs not too much difference