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 2022-01-18 00:00:00Z and 2022-01-19 00:00:00Z

[ddnet/ddnet] Issue closed: #4610 Exit cleanly on kill signals

00:02

[ddnet:master] 6 new commits

eecd134 handle SIGINT in server to properly shutdown everything - Robyt3 a69852b force quit on second Ctrl+C - Robyt3 259a449 use compliant sig_atomic_t for signal handler, use _Exit - Robyt3 f16e1f0 Also handle SIGTERM (fixes #4610) - def- 8a6cb2b Simplify signal handling by unregistering - def- 054e752 Merge #4612 - bors[bot]

00:02

[ddnet/ddnet] Pull request closed: #4612 Handle SIGINT and SIGTERM (fixes #4610)

im trying to implement some format, and they use "shall be"

09:34

this mean "must be" right?

09:34

its so confusing

https://datatracker.ietf.org/doc/html/rfc2119 yes

Key words for use in RFCs to Indicate Requirement Levels (RFC )

ok good

09:56

(im writing a pdf parser in rust)

09:57

https://www.adobe.com/content/dam/acom/en/devnet/pdf/pdfs/PDF32000_2008.pdf

09:58

oh here is a newer edition here https://www.adobe.com/content/dam/acom/en/devnet/pdf/pdfs/pdf_reference_archives/PDFReference.pdf

I was gonna ask wtf is this

Image attachment

10:01

but it's not in the new version I think

10:01

... headlines are relevant

Image attachment

i love how the ISO website sells this pdf when u can get it for free

10:03

such is humanity stupidity

You have a good website to get other up to date ISO documents? Send PM with link. @Ryozuki

no i dont

11:24

i just googled

Standards costing money is a giant scam, similar to research papers costing money

12:02

And what you can (legally) get for free is usually the last draft, which might miss a few correction compared to the final version. So people who are serious about implementing the standard are forced to buy them

How is standards being paywalled supposed to help with standardization, it just sounds like an oxymoron to me

that could help with further standards being developed

12:24

but I don't buy it tbh

This comment make you laugh but also make you think https://news.ycombinator.com/item?id=26393117

pronik

It's way beyond comical right now. PDF standard document has been available from evil monopolist's Adobe's website for free up until version 1.7. Version 2.0 has been "freed" from Adobe and is now an "open" ISO standard and costs about €180. Guess what I'm reading when I need to work with PDFs...

https://news.xbox.com/en-us/2022/01/18/welcoming-activision-blizzard-to-microsoft-gaming/

Welcoming the Incredible Teams and Legendary Franchises of Activisi...

Creators of Call of Duty, Warcraft, Candy Crush, Tony Hawk, Diablo, Overwatch, Spyro, Hearthstone, Guitar Hero, Crash Bandicoot, StarCraft and more join Team Xbox As a team, we are on a mission to extend the joy and community of gaming to everyone on the planet. We all know that gaming is the most vibrant and […]

13:36

microsoft bought activision Blizzard

13:36

lmaooo

13:36

this means wow, warcraft, diablo starcraft overwatch, etc

13:37

well blizzard was on the verge of death

13:37

https://twitter.com/Nibellion/status/1483431169526050818

Nibel (@Nibellion)

IT'S OFFICIAL: Microsoft has acquired Activision Blizzard https://t.co/ddzJ9J8xfD

13:38

70 billion dollars

13:38

insane

thats actually crazy

13:57

i hope they will fix the broken cod franchise

Jupstar ✪

yeah rosetta something like an emulator so thats expected i guessthe openbenchmark ddnet site says macos is supported and last time i checked the test suite didnt really check what client version you use, I could simply replace ~/.phoronix-test-suite/installed-tests/pts/ddnet-1.1.0/build/ with any build (@deen)

I didn't get pts to work on macos. so what do you want to get benchmarked?

c++
for (int i = 0; i < (int)m_vVector.size(); i++)
{
    if (condition)
    {
        m_vVector.erase(m_vVector.begin() + i);
        i--;
    }
}

Is this considered safe? (edited)

Invalidates iterators and references at or after the point of the erase, including the end() iterator.

14:12

probs not

14:13

hmm

14:14

a maybe it is safe

14:14

xd

14:14

cppreference does a similar thing i guess

Image attachment

14:15

u probs should use the iterators this way

14:15

since erase returns a iterator

deen

I didn't get pts to work on macos. so what do you want to get benchmarked?

At 1920x1080, not as many fps as I saw on other maps, but I guess we benched some harder demos

multeasymap.log

292.56 KB

841.6 KB

fokkonaut

c++
for (int i = 0; i < (int)m_vVector.size(); i++)
{
    if (condition)
    {
        m_vVector.erase(m_vVector.begin() + i);
        i--;
    }
}

Is this considered safe? (edited)

If you're not sure about the vector index, try with std::vector::at (http://www.cplusplus.com/reference/vector/vector/at/), throws an exception (http://www.cplusplus.com/reference/stdexcept/out_of_range/) if not exists

The iterator returned from a.erase(q) points to the element immediately following q prior to the element being erased. If no such element exists, a.end() is returned.

14:31

that could be used then in a loop

14:31

yea, like u said @Ryozuki

14:32

I think then i-- should do the same thing, the size updates and therefore also the loop, right?

ye

thanks, yeah the demos were a bit harder indeed Was this normal zoom level? (@deen)

At 1920x1080, not as many fps as I saw on other maps, but I guess we benched some harder demos (@deen)

Learath2

I ask for directions in plain english, the guy gives them in french...

Lmao

15:21

What a dumb guy

I wrote that in 2020, wtf stalker

justatest

15:42

I just scrolled too much

justatest

Jupstar ✪

thanks, yeah the demos were a bit harder indeed Was this normal zoom level? (@deen)

ooh, I forgot that I always play with more zoom. Running again with normal zoom

multeasymap.log

288.33 KB

711.23 KB

ok thanks alot :D (@deen)

ooh, I forgot that I always play with more zoom. Running again with normal zoom

would some moderator be so kind and ban BulgarianTee for blocking the race on GER2 (port 8300)?

done

The frametimes are jumping around a lot btw, not sure why

mhh, maybe the GL on metal driver isnt good enough yet atleast we can exclude overheating with the M1 xDD

CPU wise it seems to compare to a Ryzen 5500u GPU wise it beats all APUs for all laptops that are benchmarked, there is one laptop with a nVIDIA card (rtx 2060) thats still alot faster, but since the M1 doesnt have a native OpenGL driver its not fair comparison anyway

looks good for me

Image attachment

17:03

spikes are probably due to chat messages

Hello everyone, I have a small question. There is a button on the website for subscribing to the new maps. I wanted to ask if it is possible to use this as a webhook for DC if so how ?

with the histograms

Image attachment

Darius - ;v

Hello everyone, I have a small question. There is a button on the website for subscribing to the new maps. I wanted to ask if it is possible to use this as a webhook for DC if so how ?

I don't think so. It's just an RSS feed, maybe you can google for some discord bot for rss feeds

what exactly did you test? just general frame times? (@Chairn)

unknown.png

17:14

or any specific test

17:14

or is that the log from deen?

Jupstar ✪

CPU wise it seems to compare to a Ryzen 5500u GPU wise it beats all APUs for all laptops that are benchmarked, there is one laptop with a nVIDIA card (rtx 2060) thats still alot faster, but since the M1 doesnt have a native OpenGL driver its not fair comparison anyway

Should probably say cpu-wise single-threaded. But even that is unexpected, would have guessed much faster based on this: https://www.cpubenchmark.net/compare/Apple-M1-Pro-10-Core-3200-MHz-vs-AMD-Ryzen-5-5500U/4580vs4141

17:14

looks like my log

Jupstar ✪

or is that the log from deen?

yup, log from deens

what is the 2nd graph here? (@Chairn)

unknown.png

17:15

or the bottom

sb = len(glob('*.log'))
fig, axes = plt.subplots(2, sb)
for i, f in enumerate(glob('*.log')):
    with open(f) as file:
        text = file.readlines()
    times = list()
    for line in text:
        m = re.match(r'Frametime (?P<frametime>\d+) us', line)
        frametime = int(m['frametime'])
        times.append(frametime)
    axes[0,i].plot(np.arange(len(times)), times)
    axes[0,i].set_title(f)
    axes[0,i].set_ylim(0, 15000)
    axes[1,i].hist(times[2:], bins=100)

17:15

top row is just basic plot, bottom row is histograms (without the first 2 times that are outliers)

true, but our test is also very floating point heavy (@deen)

Should probably say cpu-wise single-threaded. But even that is unexpected, would have guessed much faster based on this: https://www.cpubenchmark.net/compare/Apple-M1-Pro-10-Core-3200-MHz-vs-AMD-Ryzen-5-5500U/4580vs4141

17:17

ok nice (@Chairn)

top row is just basic plot, bottom row is histograms (without the first 2 times that are outliers)

Hi who have things for stop lag please?

Hi, i have fiber

I took a profile (using sample) during Multeasymap run

ecore.sample.txt

823.67 KB

17:24

looks like mostly the Wait during CGraphics_Threaded::Swap. If I remove that it's much better

17:24

but crashes when I resize etc

:o

17:24

RenderTilemap

17:24

do you run opengl 1?

nope, modern opengl is on

yeah bit annoying, but RenderTilemap sounds wrong tho

17:25

should only happen in GL 1.x

17:26

weird

[18:26:28][gfx]: GPU vendor: Apple
[18:26:28][gfx]: GPU renderer: Apple M1 Pro
[18:26:28][gfx]: GPU version: 4.1 Metal - 76.3
[18:26:28][client]: version 15.8.1 on macos arm64
[18:26:28][server]: git revision hash: 054e7528dd7ebda1
[18:26:35][console]: Value: 3
[18:26:37][console]: Value: 3
[18:26:41][console]: Value: 0

for me its not called

I guess IsTileBufferingEnabled() returns false

oh, thats even weirder

17:28

i think for gl 3.3 it should just return true

No, only 3.3.4 and higher

17:28

and I have 3.3.0

m_OpenGLTileBufferingEnabled = m_IsNewOpenGL || m_pBackend->HasTileBuffering();

bool CGraphicsBackend_SDL_OpenGL::IsModernAPI(EBackendType BackendType)
{
    if(BackendType == BACKEND_TYPE_OPENGL)
        return (g_Config.m_GfxOpenGLMajor == 3 && g_Config.m_GfxOpenGLMinor == 3) || g_Config.m_GfxOpenGLMajor >= 4;

17:29

oh

17:29

misread, you're right

17:30

ok, i ran without my config and it actually default initialized to opengl 1.5

17:30

Weird:

[2022-01-18 18:29:44][sdl]: SDL version 2.0.20 (compiled = 2.0.20)
[2022-01-18 18:29:44][gfx]: Created OpenGL 3.0 context.
[2022-01-18 18:29:44][gfx]: unable to create OpenGL context: Failed creating OpenGL context at version requested
[2022-01-18 18:29:44][sdl]: SDL version 2.0.20 (compiled = 2.0.20)
[2022-01-18 18:29:44][gfx]: Created OpenGL 2.1 context.
[2022-01-18 18:29:44][opengl]: Vendor string: Apple
[2022-01-18 18:29:44][opengl]: Version string: 2.1 Metal - 76.3
[2022-01-18 18:29:44][sdl]: SDL version 2.0.20 (compiled = 2.0.20)
[2022-01-18 18:29:44][gfx]: Created OpenGL 1.5 context.
[2022-01-18 18:29:44][opengl]: Vendor string: Apple
[2022-01-18 18:29:44][opengl]: Version string: 2.1 Metal - 76.3
[2022-01-18 18:29:44][gfx]: GPU vendor: Apple
[2022-01-18 18:29:44][gfx]: GPU renderer: Apple M1 Pro
[2022-01-18 18:29:44][gfx]: GPU version: 2.1 Metal - 76.3

17:31

But 3.3 works, which I have in my config, why are we not trying that initially?

[2022-01-18 18:31:02][gfx]: Created OpenGL 3.3 context.
[2022-01-18 18:31:02][opengl]: Vendor string: Apple
[2022-01-18 18:31:02][opengl]: Version string: 4.1 Metal - 76.3
[2022-01-18 18:31:02][gfx]: GPU vendor: Apple
[2022-01-18 18:31:02][gfx]: GPU renderer: Apple M1 Pro
[2022-01-18 18:31:02][gfx]: GPU version: 4.1 Metal - 76.3

3.3 was buggy on some of learaths mac

17:33

but 3.0 should be default, but i think learath also had problems with that on Mac

17:33

but he was then defaulted to opengl 2.1 i think

17:33

ah yeah

17:33

ur log says the same

17:33

but fails there too for some reason

17:34

i dunno, but i think the mesa d3d12 also only targets opengl 3.3

17:34

its possible that there is an extension we use that isnt supported for the metal wrapper

17:35

i guess that's why it goes to 1.x gl 3.3 on the other hand doesnt use any extension

17:35

its a core profile

4.1 seems to work fine

1.22 MB

multeasymap.log

406.32 KB

17:44

Should we maybe try to get the highest working opengl version? wouldn't that be better or doesn'T matter?

i mean i'd love it obviously, would make 2.1 and 3.0 useless again, they are bloat anyway xd

17:50

i think mind also had lag spikes with gl 3.3 not 3.0 for an unknown reason i wish zink would work as good on windows as on linux, then we would secretly run vulkan xd

17:51

now the results are really good (@deen)

4.1 seems to work fine

17:55

wow the client on pts is really slow, what did we change lately, i almost get double the FPS with a current version xd

strange, I'm not sure

18:02

I only hear from people that fps always drop, never that they increase

my benchmark said 408 fps on openbenchmarking (https://openbenchmarking.org/result/2011232-FI-DDNETJ17015) now i get 682fps with the client from the website and 745 fps with custom lto+Ofast client (even on 1440p, instead of full hd as on the benchmarking website)

resolution also doesn't matter for me, seems like it's mostly cpu bound

too bad, would have liked to compare the M1 vs laptop chips :P

18:06

the M1 has 388 fps

18:06

yeah thats true

18:07

oh that was rainymore

my benchmark said 408 fps on openbenchmarking (https://openbenchmarking.org/result/2011232-FI-DDNETJ17015) now i get 682fps with the client from the website and 745 fps with custom lto+Ofast client (even on 1440p, instead of full hd as on the benchmarking website)

18:07

maybe that was with the weird bug

18:07

but still, my CPU never beats a ryzen 5900x or alder lake so easily

18:08

something is really wrong with the version on pts xD

I can put new version on pts I guess?

https://openbenchmarking.org/result/2108220-PTS-ZINKROUN40 520 FPS with AMD Ryzen 9 5900X 12-Core And i get 682 fps with Ryzen 7 3800X lmao

I think pts builds from source code, is it maybe a Debug build or something?

18:11

hm, no: https://openbenchmarking.org/innhold/290edb92c7444057df0e2309845b73875c0c3d32 (see install.sh)

ok downloaded 15.3.2 from the website

18:12

its indeed insanly slow

Oh well, I'll just upload a new version to pts

ok nice

I forgot how to upload to PTS

:D i dont even remember if michael did it for you

18:16

you posted a forum link someday, where you asked him

only running at 40% CPU during benchmark btw, that seems weird too, would expect 100% at least

18:55

Managed it:

Create ~/.phoronix-test-suite/test-profiles/def/ddnet-1.2.0
$ ./phoronix-test-suite install def/ddnet-1.2.0
$ ./phoronix-test-suite benchmark def/ddnet-1.2.0
$ cd ~/.phoronix-test-suite/test-profiles/def/ddnet-1.2.0
$ zip -9r /Users/deen/.phoronix-test-suite/openbenchmarking.org/ddnet-1.2.0.zip *
$ ./phoronix-test-suite upload-test-profile def/ddnet-1.2.0

18:56

At least to https://openbenchmarking.org/test/def/ddnet, now have to ask Michael to put it to pts

19:02

Also the proper tests now work on macos, had some bash/zsh problems there I guess

i guess that comes from the waitforidle call Tho tests are messured in fps anyway, but could ofc still test to remove it (@deen)

only running at 40% CPU during benchmark btw, that seems weird too, would expect 100% at least

19:19

i even get higher fps with gfx_asyncrender_old 0 in the cpu bound test, probs when it misses the GPU not idling and does a silent update(one without a render to the screen)

[ddnet/ddnet] Pull request opened: #4614 Use LTO when possible

Not much of a difference but doesn't cost much either: https://openbenchmarking.org/result/2201183-DEF-NONLTO6564 https://openbenchmarking.org/result/2201183-DEF-LTO6334352

Checklist

[x] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[ ] Considered possible null pointers and out of bounds array indexing

...

[ddnet/ddnet] Pull request closed: #4614 Use LTO when possible

[ddnet/ddnet] Issue opened: #4615 Android Build Error

When I tried building DDNet for android, using the command- scripts/android/cmake_android.sh all DDNet Release I got the error:

> Task :compileReleaseJavaWithJavac FAILED
/home/-/-/ddnet-master/build_android/src/main/java/tw/DDNet/NativeMain.java:3: error: package org.libsdl.app does not exist
import org.libsdl.app.SDLActivity;

I followed the instructions at [https://github.com/ddnet/ddnet/tree/7e697963cf9f7657a22d74fdd0f85032f73baa2e/scripts/android](https://github.com/ddnet/ddn...

[ddnet/ddnet] Pull request opened: #4616 Close find handle in fs_file_time on windows

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memcheck](https://github.com/ddnet/ddnet/#using-addresssanitizer--u...

has anyone ever experimented with increasing the amount of ticks per second in ddrace? is it even possible

yes but not possible without changing pyhsics

[ddnet:master] 2 new commits

2f17fa9 close find handle in fs_file_time on windows - Robyt3 28884fa Merge #4616 - bors[bot]

22:28

[ddnet/ddnet] Pull request closed: #4616 Close find handle in fs_file_time on windows

Jupstar ✪

yes but not possible without changing pyhsics

it would just be floating point physics and the like, right?

probably also other stuff

22:29

freeze jumps for example would need more speed (freeze jumps where you hit the ground)

i feel like higher tick would feel a lot smoother

22:30

oh right

the problem is stuff like gravity for example that is added, and later other stuff is multiplied

freeze jumps where you dj just as you enter freeze would get more precise

you can get close with 2nd sqrt and stuff but its never the same

22:31

and yeah tw also rounds all physics by converting it to ints or smth

22:31

but that could theoretically be improved i guess

22:33

i even still have the source code it seems xd

https://github.com/Jupeyy/ddnet/commit/adc8209b56a2619f5d01d919ae324ed5fe031cbb

200 ticks · Jupeyy/ddnet@adc8209

22:49

but i havent tested it since years xd

22:50

also 200 ticks were overkill, made the physics round too much, i think you standed in the air for a few ticks xD

[ddnet-web:master] 1 new commit

9ae16d5 Add donation by Destoros - def-

Exported 176 message(s)