mosra/magnum

Lightweight and modular C++11/C++14 graphics middleware for games and dataviz

Vladimír Vondruš

@mosra

An appointed successor can manage your public repositories after presenting a death certificate then waiting for 7 days or presenting an obituary then waiting for 21 days.

does github know something i don't?! :scream:

Janos95

@Janos95

haha thats pretty morbid

Konstantinos Chatzilygeroudis

@costashatz

Hahahha

Jonathan Hale

@Squareys

does github know something i don't?! :scream:

I think all bigger open source projects are getting this :P

Janos95

@Janos95

just noticed this :

"echo "#include <algorithm>" | gcc-head -std=c++11 -P -E -x c++ - | wc -l" returns 11765

while

"echo "#include <algorithm>" | gcc-head -std=c++20 -P -E -x c++ - | wc -l" returns a whopping 45222

compiling an empty file including the algorithm header takes 0.15 sec (with c++11) vs 0.9 sec (with c++20)

Vladimír Vondruš

@mosra

hah, i thought you'd compare that to gcc 9

Janos95

@Janos95

I chased the includes and it turns out that <algorithm> transitively includes <ranges> which is 30 loc and contains lots of template stuff

Vladimír Vondruš

@mosra

because this happened for gcc 10, so i expected improvements: https://twitter.com/wakomeup/status/1196097492456214529

#include <ranges> is a fireable offense, then

Janos95

@Janos95

yes I can live without <ranges> in largish projects, but abondoning std::sort etc is not really an options :/

Vladimír Vondruš

@mosra

damn, right

i was just about to ask what exactly do you need from there

isn't it in some utility / internal header on gcc? so i could do another Utility/StlMath.h "deshittifier" but for <algorithm>

personally even the 0.15 sec was too much for me and so <algorithm> is included almost nowhere in magnum

another option: don't use c++20 :D

Janos95

@Janos95

yes I think one can include

#include <bits/stl_algobase.h>
#include <bits/stl_algo.h>

not sure about other platforms

another option: don't use c++20 :D

but I already changed all my enable_if's to requires clauses :laughing:

Janos95

@Janos95

personally even the 0.15 sec was too much for me and so <algorithm> is included almost nowhere in magnum

should I compare compile times?

Vladimír Vondruš

@mosra

of magnum? :D

Vladimír Vondruš

@mosra

you might need to force the -std in CMAKE_CXX_FLAGS, otherwise it'll use 11

Janos95

@Janos95

with DCMAKE_CXX_STANDARD=11 it is 17.43 seconds and with DCMAKE_CXX_STANDARD=20 it is 22.13 seconds. Very unscientific, average of two parallel debug compiles, but variance was quit low, about 0.5 seconds maybe.

but interestingly it is about 1.5 slower than my system gcc, maybe I messed up compiling gcc, so everything with a grain of salt.

you might need to force the -std in CMAKE_CXX_FLAGS, otherwise it'll use 11

ah ok, I'll check one moment.

Vladimír Vondruš

@mosra

wow, 17 seconds, that's better than i remember

Janos95

@Janos95

yeah it was parallel with 14 cores

Vladimír Vondruš

@mosra

ahah okay

people with fast machines

was it with all tests?

or just the libs

libs alone i remember were around 20 seconds last time i checked, on a 8-core machine

Janos95

@Janos95

ah no, no tests. Good point.

Vladimír Vondruš

@mosra

but nobody builds tests except for me, so without them it's the more realistic measurement

with tests it's about 850 ninja targets i'm at 90 seconds and it's getting unbearable

Janos95

@Janos95

a full build with all test takes 26 seconds on 14 cores, one of these newer amd processors will probably eat magnum for breakfast ;). On the other hand it's kind of a bummer if one is not able to work on a laptop anymore...

Vladimír Vondruš

@mosra

that's insane

here the tests take majority of build time

while in your case the extreme parallelism makes it 17 secs for the libs and just ~9 for the tests?

Burak Canik

@fauder

https://android.googlesource.com/platform/frameworks/base/+/refs/heads/master/packages/SystemUI/src/com/android/systemui/glwallpaper/ImageProcessHelper.java#129 wait actually what the fuck

...

Vladimír Vondruš

@mosra

:fire: :phone: :fire:

Janos95

@Janos95

while in your case the extreme parallelism makes it 17 secs for the libs and just ~9 for the tests?

no I cannot reproduce this. There must have been something wrong now its more like 17sec for magnum and 33sec for the tests. This is also a lot more reasonable comparing to your numbers :)

Burak Canik

@fauder

I should use a Renderbuffer instead of a Texture2D if I want the buffer to be MSAA'd right ?

Burak Canik

@fauder

Got it working with MSAA.

I wonder what is the equivalent of a renderbuffer in DirectX

https://computergraphics.stackexchange.com/a/4432

I should study this

Where communities thrive

People

Repo info

Activity