DDraceNetwork - developer

why are there no good rust com bindings that i can find

05:35

is it in the existing winapi bindings

when crablang integration in ddnet?

06:45

imagine

xd

<Jupstar> @Learath2 @heinrich5991 https://quick-bench.com/q/hyy4fcz8FuGU_2HpTqHk3Uh8D9Y

Quick C++ Benchmarks

Quickly benchmark C++ runtimes

07:21

<Jupstar> completely ugly unoptimized shitty bad variadic template code

07:21

<Jupstar> build like sprintf

07:21

<Jupstar> still beats the c function easily

07:21

<Jupstar> so fuck the useless benchmarks in the outside world, if even this unoptimized, code using no SIMD can easily beat it...

what is ws-client?

<Jupstar> with gcc it's even better, i guess the constexpr simply sees through it or smth like that

07:30

<Jupstar> https://i.imgur.com/e12F4KT.png

fokkonaut

what is ws-client?

ChillerDragons super fast discord alternative

wow

07:31

zileradreghuhn 2.0

interesting

07:39

so naive variadic template code is fast, but stringstream is slow

<Jupstar> and lets be honest cpp standard manages to be slow no matter which feature

<fokkonaut> test

08:18

<fokkonaut> tf

ws-client

<Jupstar> @Learath2 @heinrich5991 https://quick-bench.com/q/hyy4fcz8FuGU_2HpTqHk3Uh8D9Y

I mean you keep drifting further and further away from the initial claim, how is a bunch of constexpr templates anything similar to stringstream which uses runtime polymorphism and hopes the compiler is smart enough to devirtualize them. What you implemented is closer to std::format than sstream

08:36

I mean it’s still an interesting benchmark, I’ll give you that

Learath2

https://www.zverovich.net/2020/06/13/fast-int-to-string-revisited.html here is a dude doing just int to string, which is a meh benchmark but even here std::ostringstream fell behing sprintf

Have u read ur own benchmark?

08:39

Here it is faster in the second one

08:39

So even the shitty CPP version is good enough

the initial claim was something like "stringstream is slow"

08:43

Jupstar said "you could probably implement a fast version of stringstream"

08:43

both seem to be true

It tells you, with stream reuse and on that specific platform it is. On the first platform it wasn’t. I still don’t understand why one benchmark makes your case but doesn’t make mine

<Jupstar> anyway, if i said cpp's string streams are fast, then sry

08:44

<Jupstar> i still claim that streams are theoretically faster, except if both, formatters and stream implementations are evaluated at compile time

08:44

<Jupstar> then it probably doesn't matter

Theoretically neither can be faster. Under the as-if rule they do the same thing they have the same optimal form for a sufficiently advanced alien compiler

whats the stringstream equivalent in rust?

08:45

ah

08:46

std::basic_iostream is like a write read trait

Jupstar ✪

they are the most efficient ones

^

Ryozuki

whats the stringstream equivalent in rust?

String

rly?

08:46

ah ye for strings

08:46

but its not utf8 right

If the compiler sees through all the virtual calls, stringstream should end up being equivalent to a recursive template, if my inbrain compiler is correct that is

if you want non-utf8, then Vec<u8>

simple question what's the point of &str if there's String

08:47

(I mean when using it)

&str can point to part of a String

what is std::string then

std::string is like Vec<u8>

typedef basic_string<char> string;

or String

so many layers

08:48

inheritance boo

<Jupstar> the typedef isnt really inheritance

08:49

<Jupstar> its a named generic

i know

08:49

but

08:49

this is

<Jupstar> xd

08:49

<Jupstar> zero cost

doubt

08:50

we actually know it's not zero-cost in practice ^^

08:50

see std::stringstream vs snprintf

<Jupstar> yeah, it's the lie you get told about cpp

08:51

<Jupstar> with static linking, lto, ofast, u might still get smth similar to zero cost

It’s zero cost to a sufficiently advanced compiler

unique_ptr is unfortunately also not entirely zero-cost

08:51

Box is, though!

pog

08:51

box is interwined deeper with the compiler in rust tho

08:51

u cant implement ur own box

not due to that

no?

your own box is also zero-cost

(unique_ptr is barely not zero-cost for an extremely specific reason)

the problem with std::unique_ptr is the move constructor which runs code

08:52

yes

08:52

it's almost zero-cost

https://youtu.be/rHIkrotSwcc not a bad listen over breakfast

CppCon

CppCon 2019: Chandler Carruth “There Are No Zero-cost Abstractions”

<Jupstar> cpp committee should go all in for performance

"There are no zero-cost abstractions" lie!

08:56

Option<&i32> is a zero-cost abstraction over C++'s int32_t *

<Jupstar> click bait xd

nice nice

08:59

https://blog.rust-lang.org/inside-rust/2023/05/29/1.70.0-prerelease.html

1.70.0 pre-release testing | Inside Rust Blog

Want to follow along with Rust development? Curious how you might get involved? Take a look!

08:59

i have yet again the urge to contribute to llvm

08:59

and fix the c api

09:00

but i have time to think about my life choices when compiling llvm

09:00

and end up not doing anything

<Jupstar> is the c api also flawed? thought u worked with the cpp API xd

i started working with the cpp api cuz the C api misses some features

09:00

mainly the data layout api

09:01

to get a type alignment and size, etc

09:01

https://mlir.llvm.org/docs/DataLayout/

Data Layout Modeling - MLIR

Multi-Level IR Compiler Framework

09:01

this

09:02

i also cannot transform mlir to llvm ir via the c api

09:02

btw* a project by the llvm author using mlir https://www.modular.com/mojo (edited)

Mojo

: Programming language for all of AI

Mojo combines the usability of Python with the performance of C, unlocking unparalleled programmability of AI hardware and extensibility of AI models.