DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.tw/irclogs/ Connected with DDNet's IRC channel, Matrix room and GitHub repositories — IRC: #ddnet on Quakenet | Matrix: #ddnet-developer:matrix.org GitHub: https://github.com/ddnet

Between 2021-08-11 00:00:00Z and 2021-08-12 00:00:00Z

cool changes

1

@Deleted User #4003 works but it looks wrong

https://github.com/ddnet/ddnet/pull/4003

Use GL_ARRAY_BUFFER by Jupeyy · Pull Request #4003 · ddnet/ddnet

For older OpenGL contexts @Learath2 can you check this? Checklist Tested the change ingame Provided screenshots if it is a visual change Tested in combination with possibly related configuratio...

Learath2

@Deleted User #4003 works but it looks wrong

can you specify or send screen?

oh actually gl3 and gl2 both look wrong, only gl1 looks correct

gl1

Image attachment

07:10

gl2/3

Image attachment

07:10

the missing quad on the top also makes the animations look very weird

and thats since the patch?

causes these weird corners etc.

Image attachment

Deleted User

and thats since the patch?

I will check

07:13

gl3.3 looks correct, gl3 and gl2 were broken before 4003 anyway. So I guess it's not related to the patch

still weird

07:14

quads use triangles

07:15

directly

07:15

can you see how the HLS color sliders look like in tee settings?

07:15

they also use triangles

let me run my janky debug build in case we have unhandled gl errors still

too bad nobody cares about older GL anymore, makes it just harder to find such errors

sliders seem fine

Image attachment

really weird xD

07:18

maybe the texture binding somehow fails

Learath2

let me run my janky debug build in case we have unhandled gl errors still

okay, no such luck this time

07:20

So unless there are gl calls outside backend_opengl your gl calls succeed

mh yeah, and the map triangles also use the gl 1.x method of rendering them

Maybe some z-ordering issue? my settings screenshot sort of suggests the quad is still there just behind the others

depth buffer shouldnt be active, so first come first rendered

maybe the render order is wrong?

why shouldnt it reproducable in other contexts then

true, so that's not it

wait

07:23

are they even quads

07:23

im not even that sure

I haven't checked

they are tiles

07:25

in fact we only see quads

07:25

normally quad rendering is checked can you try gfx_3d_texture_analysis_done 0 and restart

07:25

maybe your config is in some invalid state

okay, it decided gl1.5 is the best idea

so the fallback works?

07:28

i wonder how it passed the fallback thing, can basically only happen if you copy config or change hardware, or the driver gets a nice downgrade

Now all the versions work, so weird, I guess I had a broken config somehow

yeah it doesnt allow u to use gl 2.x anymore

07:29

it just downgrades the GL version

idk why it decided on 1.5 either, 3.3 works fine on macOS, even better

yeah strange

07:29

gl 2.1+ - 3.0 work with extensions

Anyway, it won't let me have a 2.1 context anymore, it will indeed fall back to 1.5

maybe they are wrongly implemented and gl 3.3 just has it as core feature

Learath2

Anyway, it won't let me have a 2.1 context anymore, it will indeed fall back to 1.5

yeah

Deleted User

i wonder how it passed the fallback thing, can basically only happen if you copy config or change hardware, or the driver gets a nice downgrade

thats why this

07:30

it somehow passed it before, dunno how, almost impossible without a change

And if I request a major 3 context it will properly choose 3.3 now, it used to pick 3.0 iirc

ah yeah bcs minor is set to 5

07:31

bcs it chooses 1.5

Anyway, I was testing around many different versions/commits it is possible the config broke along the way

maybe we should just remove these stupid broken GL versions and default to GL 3.3 and remove the texture unit optimization stuff so it won't break on some systems

07:33

with gl 3.3 we have more debugging too

07:33

renderdoc for example

Eh, quite impossible we have to atleast keep gl1 support because broken drivers

yeah

07:34

gl 1.x and 3.3

07:34

but no inbetween

(or we are doing something wrong because I still think it's insane that we can't render a 2d game in 2021

)

i play with gl 3.3 since 4 years, all bugs we had were fixed

Why did we introduce the gl2 stuff anyway?

mipmaps

Deleted User

i play with gl 3.3 since 4 years, all bugs we had were fixed

Don't we still have that intel issue?

for example yes

07:35

they would then get gl 1.x

07:35

which is slower than 2.0

07:35

annoying af

and iirc our gl1 backend is just flat out inferior, can't zoom out far even, no?

yeah but gl 2.0 isnt much better

07:36

2.0 is basically vanilla 0.7 just a little bit different

07:37

so it uses 3d textures, instead of 2d array textures, thus no mipmaps and iirc it also doesnt use VBOs

Hm, I don't even remember what the initial issue was that required the hack of using 3d textures on vanilla

do you know the border tools?

07:39

they remove some pixels from the tile set images, bcs the UV coordinates aren't correctly calculated with mipmaps

07:39

e.g. 1/1024 is less than 1/512

07:40

so if you get the 512 mipmap the coordinate is wrong

07:40

the problem on vanilla is, it's not worth it

Ah the old borderfix mess

3d textures don't have mipmaps

07:40

well not completly worthless tho tbf

07:41

it still clamps the edges correctly

07:41

but e.g. zoom in the vanilla 0.7 editor on Multimap(the original one)

07:41

and do the same with gl 3.3 on our edtior

07:41

then you see why ddnet's version is supirior

How does the gl33 renderer handle the old borderfixed tilesets?

they arent fixed, the mapres will look 2 pixels streetched if u zoom in very much but vanilla res for example isnt saved in the map

07:43

so its not alot of maps where you will notice it

hm, and the gl1 renderer? I assume that looks correct

07:44

(for borderfixed tilesets)

mhhhhhh

07:45

you could say it looks more correct

07:45

but the math is still wrong

Only if you get the correct mipmap?

gl 1.x doesnt have mipmaps anymore

07:45

we could probably calculate the texel offset for that context too, but i assume only few ppl use 1.x and they probs have a poor resolution anyway to notice

So Q: can't one disable mipmaps? Why did vanilla go with 3d textures if so?

i dont get your first question for the second: bcs 3d textures exists since gl 1.2 and 2d array textures are core since 3.0( and over extension for gl 2.1) (edited)

07:48

vanilla doesnt want modern things xd

Learath2

So Q: can't one disable mipmaps? Why did vanilla go with 3d textures if so?

u can disable it by using gl 2.0 and lower, or maybe in ur driver, but else no

Hm, I thought borderfix was needed to accomodate for mipmaps

it was needed bcs the texel offset math in gl 1.x context is random

07:51

you could say:

07:51

it works for most cases, bcs most ppl have a resolution of something that only chooses a mipmap of 512x512pixels or more

i think sdl 2.0.16 broke wayland backend on wlroots

basically you can imagine it as, just adding an offset of 1/512 so if you have low res screen it looks ok and on higher res it will remove too much from the texture, but the bordertool streetches these pixels, so in the end it looks fine too

07:53

but removing 4 pixels of a 64 pixels tile is quite alot

nori

i think sdl 2.0.16 broke wayland backend on wlroots

rip, have you reported it?

@Deleted User do u use wayland

Deleted User

basically you can imagine it as, just adding an offset of 1/512 so if you have low res screen it looks ok and on higher res it will remove too much from the texture, but the bordertool streetches these pixels, so in the end it looks fine too

but bcs its streetched it looks worse on lower res with boridertools.. thus random xd

Ryozuki

@Deleted User do u use wayland

not rn, bcs i dont want forced vsync

07:55

but i'd use it, once its stable

07:55

so maybe 1-2 years

yeah i hear they are obsessed with real fps or smth

07:55

i dont know the word rn

07:55

xd

the performance is really nice

everyone these days is using meson instead of cmake

i get more fps, KDE's compositor doesnt destroy my fps too

07:55

quite an improvement

Ryozuki

everyone these days is using meson instead of cmake

yeah i don't even get why xd

@Deleted User https://gitlab.freedesktop.org/wayland/wayland-protocols/-/merge_requests/65

Tearing updates protocol (!65) · Merge requests · wayland / wayland...

This protocol is meant to be used by drivers to be able to hint to the compositor if and when it should employ asynchronous page flips for presentation....

yeah

Image attachment

07:57

they want it "frame perfecT"

yeah xD

he probs doesnt play games

rule number one if a gamer wants smth, they want it xD

games are for kids tho

Image attachment

@Deleted User found the pr that cause it, i'll report

I sort of agree that tearing doesn't help much, but you leet gamers are very passionate about it and I've promised myself I won't get mad this month

it makes insane difference xD

greenthing

08:00

ye

for ddmax maps it might not matter but as soon as you have more action in it, rip (edited)

I've been playing games almost as long as you and I've almost always played with vsync, there is a difference but calling it insane is a bit overreaching

i get if you'd play on my setup you would never want to go back to another xD

08:02

bet*

anyway, wayland shouldn't be the one enforcing this, I agree on that

my brother even uses 144hz just for desktop xD scrolling etc just smoother

08:03

mobile phones also get 120hz screen nowadays

I mean I played on many setups over the years, starting from a janky old crt tv all the way to a 4k 144hz monitor

maybe vsync will work with 1000hz screens xd

60hz to 144hz is indeed a large change, but go back to 60hz and you'll forget about the jerkiness in a matter of days

60hz is nice, it made me not playing teeworlds in my holidays anymore

(unless you obsess about it, then it will stay there forever, the brain is good at making the motion continuous, but if you keep looking for the gap it can't do it's job)

i am obsess with such stuff

Anyway, at 144hz vsync is even better for me

if there would be an OS with lto and Ofast i'd use it for the 5-10% performance uplift xD

https://www.nytimes.com/2021/08/10/world/europe/germany-covid-test.html

By Christopher F. Schuetze, Nicholas Casey, Eileen Sullivan, Davey Alba, Jason Gutierrez, Constant Méheut, Aanya Wipulasena and Saif Hasnat

Germany will stop paying for virus tests for people choosing to rem...

People who choose not to get vaccinated will no longer get free antigen tests, starting in October.

08:06

monkalaugh

Deleted User

if there would be an OS with lto and Ofast i'd use it for the 5-10% performance uplift xD

You can compile most everything with Ofast and LTO on gentoo, I used such a system for a long time

ah yeah i dont want to compile myself

Sort of gave up on it tho since LTO was very immature back when I was using it

takes too much time

08:07

but yeah generally i like the idea

Deleted User

takes too much time

Doesn't, unless you use a toaster. The initial setup is longer but after that modern compilers and linkers are quite fast

i want to switch to gentoo so badly but im too lazy to redo my setup

08:08

and i have accumulated lot of files

08:08

i dont know which are important

I even had distcc setup so I could compile packages on my desktop for my laptop

well i sometimes compile KDE, takes ages

08:08

maybe if i'd use a home server, so it doesnt block my main setup

lucky me i use a wm compiling that is probs way faster

well next debian finally enables lto, thats already a nice step forward

why have a desktop when rofi/dmenu is all u need

08:09

greenthing

arch probably already has quite alot lto packages i assume?

https://github.com/davatorium/rofi

GitHub - davatorium/rofi: Rofi: A window switcher, application laun...

Rofi: A window switcher, application launcher and dmenu replacement - GitHub - davatorium/rofi: Rofi: A window switcher, application launcher and dmenu replacement

i remember we had an compiler bug once, that destroyed the arch package, i could reprod with lto

Deleted User

maybe if i'd use a home server, so it doesnt block my main setup

Have you tried the muqss patchset? I compile in the background all the time and my system stays fairly responsive

Deleted User

arch probably already has quite alot lto packages i assume?

you only compile packages from AUR

ah yeah AUR is the thing

i wonder how much more speed native arch compilation gives u

08:10

-march=native

Learath2

Have you tried the muqss patchset? I compile in the background all the time and my system stays fairly responsive

never heard of it

@Deleted User what distro u use

I think ck-sources includes the patch

Ryozuki

i wonder how much more speed native arch compilation gives u

ah yeah thats the next step, an OS for AMD CPUs xD

Ryozuki

@Deleted User what distro u use

debian unstable

08:11

mostly bcs i hate when KDE is buggy xD

08:11

and its basically always buggy

@Deleted User im surprised AAA games dont give out binaries with amd/intel cpu simd stuff

Ryozuki

@Deleted User im surprised AAA games dont give out binaries with amd/intel cpu simd stuff

ikr

08:12

especially on steam it shouldnt be so hard

Image attachment

08:12

stockfish does it

08:12

they are years ahead

maybe steam allows it, dunno, i dont have a dev account there

Ryozuki

i wonder how much more speed native arch compilation gives u

Probably not much at all

if i made a game i would add it as a custom version

08:12

beta

08:12

steam probs allows it that way

https://openbenchmarking.org/embed.php?i=2106164-IB-CLANGOPTI18&sha=461cf373e986&p=2

08:14

on clang 12

o3 lto is slower

08:15

than o3

except for openmp clang is also often faster than gcc, thats also a nice battle

xd

yeah really weird

08:16

i think phoronix also has benchmarks showing that O3 didnt change anything over the last 10 years xD

08:16

so sad

08:17

performance wise

The linux kernel also fares better on gcc still iirc.

yeah possibly

08:18

still i wonder why gcc almost didnt improve, are the tricks over, or is it bcs gcc fears to break stuff or smth

So what do you want to do about #4003 @Deleted User?

https://github.com/ddnet/ddnet/pull/4003

Use GL_ARRAY_BUFFER by Jupeyy · Pull Request #4003 · ddnet/ddnet

For older OpenGL contexts @Learath2 can you check this? Checklist Tested the change ingame Provided screenshots if it is a visual change Tested in combination with possibly related configuratio...

just merge i guess, it won't break anything, still better than having it wrong

08:22

GL_COPY_WRITE_BUFFER So made for copy and writing but we only bind one buffer at a time in <gl 3.0 context

08:22

so should be fine

@Learath2 https://github.com/Jupeyy/ddnet/commit/3b5eddbe4607f342c8931ff8601eb33220bf6434 This checks glError after every opengl call, and also checks the current attached shader for correct runtime(e.g. missing texture etc.) If you want you could try to use it with GL 2.1(u gotta enable gfx_3d_texture_analysis_done 1 for that again) Maybe it shows why it cant render tile sets

CallGL debug macro · Jupeyy/ddnet@3b5eddb

09:20

oh and its only in debug mode

c

@Deleted User I had a very similar patch, I checked the error log after each command and didn't find anything

k

free node is where (edited)

can someone remove this from being counted as worse time on ddnet

16:26

Image attachment

16:26

its on Puzzle Partners, it has a 15min time penalty u can spam

u

https://github.dev/ddnet/ddnet

17:31

u dont need ur editor anymore

17:32

just press "." on any github repo

17:32

and it opens this editor

is that new?

17:42

if yes its suprisingly late considering how long ago ms bought github and gitlab already had a more solid IDE right? I mean wasnt vscode developed as web app and then later "ported" to desktop

17:42

why is is white themed :D

i dont see it white themed

ddnet client white theme when

@Ryozuki: do u use dark themed github?=

y

like everybody except chiller xD

I used light github while working outside the last week

20:37

I also use a light themed ide when out, it's actually fairly hard to see when the sun is out, we sort of forget how insanely overwhelming the sun can be compared to our puny indoor lighting

Embedded image

Exported 242 message(s)