C++ Zero Cost
Cost Conf 2023;
 Moscow & Belgrade;
29/07; Conference on Applied C++

YfD */
Watch broadcast

Broadcast recording

C++ Zero Cost Conf —
Yandex Conference on Applied C++

Top-notch practicing developers from global companies will deliver presentations, sharing real-life case studies and results.

This year, the conference will be held simultaneously in two countries, with online streaming from each. You can come and join us in Moscow or Belgrade — or watch the stream from the comfort of your home in any country.

Watch broadcast

Workshops 26 & 27 July

We want to give to developers as much practical knowledge as possible. That’s why, besides the conference, we organize several workshops — master classes on working with C++.

*All users registered at C++ Conf will be able to participate in them.

Deploying ML models using C++
26 July, 19:00 MSK

19:00 MSK 26 July
Pavel Filonov Independent Consultant
ex-Kaspersky, ex-Positive Technologies

Watch recording The video will be coming soon

When developing products that involve functionality based on machine learning, there are two important stages: model training and its application.

During the model training process, the Python language and various frameworks are often used to solve machine learning tasks.

However, when it comes to applying the model, there might arise a necessity to integrate a pre-trained ML model into C++ code.

In this workshop, we will explore several practical examples of such integration, using both existing C++ libraries and personally written implementation code when it is the simplest option.

We will take examples from the talk ’Train in Python, Apply in C++’ and create working and tested modules in C++ based on them.

  •  Key ideas and terms of machine learning.
  •  Simplest linear regression model.
  •  Implementation of model predictions in C++.
  •  Approaches to testing and metrics comparison.
  •  CatBoost gradient boosting libraries.
  •  КCode generation or C API integration.
  •  Batch processing and performance considerations.
  •  НNeural network models and vectorization.
  •  Implementation of a neural network in C++ using matrix operations.
  •  TensorFlow library (C API) and TensorFlow Lite (C++ API).
  •  Model conversion and ONNX library.
  •  Conclusions and next steps.
  •  Install Docker
  •  Pre-download the Docker image with all the required dependencies:
    docker pull sdukshis/cppml
  •  Learn how to mount the directory with source code inside the container and perform builds using CMake:
    cd <hello, world>
    docker run --rm $(pwd):/usr/src/app -ti sdukshis/cppml
    cmake -B build
    cmake --build build

Deduction guides: Impossible is possible.
27 July, 19:00 MSK

19:00 MSK 27 July
Ruslan Arutyunyan Intel, Senior Middleware Development Engineer

Watch recording The video will be coming soon

The presentation will provide the practical example of writing Deductions guides for the class that takes the Callable object as an argument.

We are supposed to go through the steps I’ve made to learn how these tasks can be solved. I will demonstrate how to fetch function signature from the object type and then use it as a developer likes, how to leverage the additional information about the required signature of a function, and how to make more convenient user interface based on above.

The presentation is going to tell the quite funny story of how impossible becomes possible just in two days.

The attendees are expected to know what is Class Template Argument Deduction (CTAD) and basics about how to write the Deduction Guides.

Program

Conference Opening

12:00
Ilya Shishkov Yandex Eats, Senior Software Developer

Смотреть на YouTube

Briefly about the structure of compilers and why VKontakte has its own

12:10
Alexander Kirsanov VKontakte, Team Lead of the KPHP

The speaker will explain the inner workings of compilers at a basic level, including scary words like lexers, parsers, and AST trees. Then, they will talk about KPHP, the compiler used internally at VKontakte, which powers the entire backend of the social network. Finally, Alexander will highlight a set of C++ tricks for optimization and memory saving that are used in complex use cases.

Смотреть на YouTube

Scalable vectorization in RISC‑V

13:00
Konstantin Vladimirov Syntacore, Head of Development Tools and Compilers Group

The scalable architecture of RISC-V introduces several interesting innovations. One of the most unusual features is scalable vectorization, which allows for runtime adjustments of the VPU settings.

Konstantin will cover the following topics:

  • The basics of vectorization, using a simple example of fixed vectorization with AVX intrinsics.
  • The same example, but with scalable vectorization in RISC-V.
  • The speaker will then discuss the challenges of scalable vectorization:
    • Tail and mask-agnostic behavior and situations where the compiler considers undefined behavior (UB).
    • The impact on ABI and why sometimes fixing a benchmark requires a commit to libunwind.

Смотреть на YouTube

Перерыв

13:50

Not on the first try: simplifying C++ code with DSL

14:40
Ilya Shishkov Yandex Eats, Senior Software Developer

I will share an example of solving a real problem in one of the backend services of Yandex.Eats. The task is to calculate the durations of individual stages of an order in Eats: the courier’s arrival at the restaurant, preparation, delivery of the prepared dish, and so on. In addition to calculating the actual value, we need to log the formula used for the calculation, as well as the values of all variables in that formula. This is necessary for further analysis and debugging. Initially, it was programmed in a straightforward manner: there was separate code for calculating the value and separate code for generating the textual representation of the formula andI set a goal to make it possible to simply write the formula and have everything else generated automatically. saving the variable values. Such code had many drawbacks: it was difficult to understand and modify.

In my presentation, I will not only show the final solution I came up with but also the path I took to reach it. You will see how different features of C++ are applied to solve a real problem.

Смотреть на YouTube

The concept of undefined behavior

15:30
Roman Rusyaev Team Lead in Compiler Department

The presentation will demonstrate the motivation behind introducing undefined behavior in the C and C++ languages. Using specific examples, we will examine why certain behaviors are specified as undefined by the language standards. As we know, one of the main motivations for unspecified behavior is efficient code generation by the compiler, which includes loop optimizations, utilizing object layout information, and more. I will demonstrate the reason for introducing type aliasing rules and the consequences of violating them by the programmer. Also, we will discuss rules regarding signed and unsigned integer overflow in loops, sequence points, and the order of evaluation of expressions. After the presentation, the audience will gain a fundamental understanding of the necessity of the concept of unspecified behavior in programming languages.

Смотреть на YouTube

Break

16:20

Developing Applications Using C++20 Coroutines

17:10
Konstantin Vukolov Yandex, Senior Software Developer

  • How C++20 coroutines differ from traditional fibers
  • How to write your own library of stackless coroutines
  • What are the pitfalls when writing such asynchronous code

Смотреть на YouTube

News
from Working Group 21

18:00
Anton Polukhin Yandex Go, Head of the Common Components Development Group
We will discuss the news from the meeting of the International Committee for the Standardization of the C++ language and the interesting updates for C++26. Antonwill answer questions and talk about the plans for C++26 and C++29.

Смотреть на YouTube

Closure

18:50

Afterparty

19:00

Conference Opening

11:00
Sergey Fedorov Yango Deli Tech, Head of Warehouse Robots Development
Yulia Golovchanskaya Playrix, Programming Team Lead

Смотреть на YouTube

Class layout

11:10
Miloš Anđelkovic Pure Storage, Senior Software Engineer

This is a talk about how the non static member variables are laid out in memory. Part that is not covered is the case with virtual inheritance as it complicates things and this should only take an hour.

We start out with the Standard layout type, what it is, why it exists and what are it’s requirements.

Then the exploration of the language rules how the different blocks of variables with the same access can be moved around and how that was changed in C++23.

Next the inheritance and how it impacts the layout of the class with the mention of empty base class optimization. As a pair to inheritance the same (similar example) using composition (member variables) and no_unique_address attribute. This followed with examples using the clangs -cc1 -fdump-record-layouts which will generate reports for the layouts of the classes. As part of this mention the Subobjects.

Alignment requirements of the operating system and C++ and padding. Brief mention of alignof and alignas operators and many align_ functions in C++, but just a mention as there will be no time for details.

Finally a simple benchmark showing difference in performance between the structs that have the same member variables but one has a «good» and the other «bad» layout.

Смотреть на YouTube

YDB Actor System: the history of optimizations that allowed us to outperform competitors

12:00
Vladislav Kuznetsov Nebius (YDB), Senior Software Engineer

YDB has been using and developing its own implementation of the actor system in C++ for almost 10 years. During this time, we have examined and analyzed our implementation from various angles and perspectives. However, six months ago, we decided to compare YDB with its competitors — CockroachDB and YugabyteDB — and once again, we discovered many interesting insights. Our team looked at benchmarks, YCSB benchmarks, and FlameGraph under load. We identified the most CPU-intensive areas, experimented with the code, and optimized the actor system and its usage. It is worth mentioning that optimizing the work with protobuf also yielded noticeable results. We tried thinLTO and PGO techniques as well. As a result, through iterative improvements, we found that in the YCSB test, YDB is the fastest distributed SQL database. In this talk, we will mainly focus on the actor system: from a general overview to interesting and non-obvious optimizations that have yielded results. We will also discuss the process of identifying performance issues.

Article with YCSB benchmark results:
Medium
Habr

Смотреть на YouTube

Break

12:50

Another Crafted Hash Table

13:40
Andrey Aksyonov Avito/Sphinx
  • Why bother writing your own craft solution in the 2020s and why did I do it?
  • How it was acceptable for us to lose to std::unordered_map, but unexpectedly, we didn’t manage to lose.
  • What craft is all about: Knowing your tasks and data.
  • What zero-cost means and why small things like numbers and strings should be sufficient for keys.
  • Understandable tricks that can be achieved with craft when you decide that small things are enough.
  • Other possible tricks in general, but usually you have to fire the person who performs them.
  • Results of non-scientific but real benchmarks.

Смотреть на YouTube

gRPC + userver = 🔥❤️

14:30
Alexander Malkov Yandex Go, C++ Developer

gRPC is a high-performance modern framework that is gaining significant popularity.

It is already being used at Yandex in conjunction with the userver framework. Alexander will showcase the benefits of using this RPC framework, demonstrate how easy it is to use gRPC with userver, and provide examples where gRPC can be advantageous over standard HTTP REST APIs. Additionally, the speaker will unveil the magic of coroutines under the hood.

Смотреть на YouTube

Break

15:20

Columnar in-memory storage: memory_resource and template

16:10
Alexander Borgardt Tech Lead duckstax.com

This season, we will dive even deeper into the internals of columnar in-memory storage and discuss why std struggles with memory_resource and different levels of caching. We will explore how to pack data in a way that leverages memory_resource and templates to achieve more efficient algorithms. We will also examine various approaches to designing column APIs with the use of memory_resource.

We will touch upon many aspects of building algorithms for columnar storage and examine it from a benchmarking perspective.

Смотреть на YouTube

Parsing Integers with SIMD

17:00
Sergey Slotin Lecturer, Tinkoff Education, ex‑Yandex

Modern CPUs have an extremely powerful but underutilized feature called SIMD («single instruction, multiple data»): special instructions that perform the same operation on a block of 16, 32, or 64 bytes of data in one go, resembling normal thread-based parallel computing and usually yielding a proportional speedup over scalar code. But unlike thread-based parallelism, SIMD adds the constraint of using the same operation for all data points, which can make applying SIMD to non-trivially-parallel problems tricky and require creative use of the instruction set.

In this talk, we will explore this technology by focusing on the toy problem of parsing a series of non-negative integers from stdin. Starting by discussing standard library implementations and their inefficiencies, we will write our own low-level implementation, iteratively optimize it using SIMD, and eventually develop a version that is 50-100 times faster than scanf(). Then, we will discuss how this solution can be extended to the more general case without sacrificing performance, and how these techniques are used in real-world high-performance parsers like simdjson.

Смотреть на YouTube

Closing

17:50

Afterparty

18:00

Anton Polukhin

Yandex Go, Head of the Common Components Development Group

Representative of Russia in the C++ Working Group of the International Organization for Standardization(WG21 ISO). Author of many accepted proposals to the C++ language standard.

Chair of the the C++ Working Group of the International Organization for Standardization. Develops and supports the standard GCC library, Boost libraries. Author of Boost libraries PFR, TypeIndex, DLL, StackTrace and active maintainer of Any, Conversion, LexicalCast, Variant.

Author of the book Boost C++ Application Development Cookbook.

Andrey Aksyonov

Avito/Sphinx

Codes code, commands commands, delivers deliveries. Works at Avito, oversees search infrastructure, works on the Sphinx search engine. Has experienced a variety of tasks and roles.

Pavel Filonov

Independent Consultant, ex-Kaspersky, ex-Positive Technologies

Participated in the development of products such as MaxPatrol SIEM, Kaspersky MLAD, MDR as a C++ developer and Data Scientist. He is an instructor for C++ and MLOps courses at OTUS.

Ruslan Arutyunyan

Intel, Senior Middleware Development Engineer

Ruslan is a senior middleware development engineer who has been working in a team specialized in libraries development for concurrency and parallelism. He was a technical leader of two libraries used for autonomous driving. Currently, Ruslan has been working on various Intel products, with main focus on oneDPL.

He is also a member of the C++ standardization working group with a focus on the C++ standard library and parallelism, and with several accepted proposals.

Yulia Golovchanskaya

Playrix, Programming Team Lead

C++ developer with experience in the field of Computer Vision and GameDev. Participated in the development of software for 3D city reconstruction, quality control in factories, and road condition analysis algorithms for traffic control systems. Taught practical C++ to students at the Higher School of Economics (HSE) in Saint Petersburg. Currently a team lead of a technical team in GameDev, specializing in feature development.

Ilya Shishkov

Yandex Eats, Senior Software Developer

Has been working as a C++ developer at Yandex for 10 years. Experienced in high-load backend and applications, as well as in creating online courses.

Created an online specialization in C++ «Belts in C++», as well as the online course «Algorithmic Foundation for Programmers».

Speaker at various conferences:
C++ Russia, SECR, KnowledgeConf. Runs the Telegram channel «Algorithmic Foundatio».

Sergey Fedorov

Yango Deli Tech, Head of Warehouse Robots Development

Has been writing in C++ since 1999. Has done various tasks: from automating bakeries and crematoriums to developing banking and stock exchange software and game development. Currently, he is the Head of Warehouse Robots Development at Yango. Loves open source and puzzling tasks.

Alexander Kirsanov

VKontakte, Team Lead of the KPHP

Works on the development of the KPHP compiler within VKontakte, as well as a number of other open-source projects related to languages and tooling. Frequently speaks at conferences on various directions and topics, leads student educational projects, and believes that the team leader should write more code than anyone else.

Roman Rusyaev

Team Lead in Compiler Department

Developer of compilers for various hardware architectures.

Vladislav Kuznetsov

Nebius (YDB), Senior Software Engineer

6 years in Yandex — 6 years in YDB. Grew and developed together, participated in the development of the storage layer for data storage and enhancement of a proprietary file system. Recently, he has been optimizing, speeding up, and maximizing transaction processing. His main interest is code efficiency. He enjoys exploring systems, identifying performance problems, and speeding everything up.

Alexander Malkov

Yandex Go, C++ Developer

Before working at Yandex, he was the CTO of custom software development.

He has a deep love for technology and the technological platform, which is why he came to develop userver and stayed. He enjoys learning new things and sharing them with others, and also coding a lot. He has over 10 years of overall development experience.

Konstantin Vladimirov

Syntacore, Head of Development Tools and Compilers Group

He has been involved with compilers since 2010, and programming in C++ since 2004. He has spent the last year working on the extensible RISC-V architecture.

Konstantin Vukolov

Yandex, Senior Software Developer

He has been involved in development under the conditions of code optimization for 8 years, is passionate about high-load systems, efficient code, and the application of the latest technologies in heavy industry.

Miloš Anđelkovic

Pure Storage, Senior Software Engineer

Miloš got acquainted with C++ back in his high school days, and ever since using two for loops to print a pyramid of stars, he knew he wanted to be a software engineer. During his studies, he got introduced to a range of other programming languages, but he found C++ to be the most enjoyable to work with (yeah, he knows). His professional career started as a teaching assistant, which he found quite fulfilling and hopes to return to one day. Since then, he has been working on software in various industries, including gaming. Additionally, he is a member of the Serbian C++ User Group and serves as the organizer of CPP Prague, a user group for programming enthusiasts in Prague.

Sergey Slotin

Lecturer, Tinkoff Education, ex‑Yandex

Sergey Slotin is a computer scientist and a software engineer specializing in algorithm design and performance optimization.

He has designed some of the world’s fastest algorithm implementations, including binary search, B-tree, integer factorization, integer parsing, as well as a range of SIMD-based array processing algorithms. Active as an educator, his writing includes «Algorithmica» — one of the most popular Russian-language resources about algorithms and data structures, and «Algorithms for Modern Hardware», a book on high performance computing.

Alexander Borgardt

Tech Lead duckstax.com

For over 10 years, he has been researching new approaches and creating solutions in the fields of adtech, fintech, and ML.

His areas of interest include the development of distributed systems, parallel and concurrent information processing.

He contributes to the improvement of the world through open source projects. You can find his work at https://duckstax.com/.

Anton Polukhin

Yandex Go, Head of the Common Components Development Group

Representative of Russia in the C++ Working Group of the International Organization for Standardization(WG21 ISO). Author of many accepted proposals to the C++ language standard.

Chair of the the C++ Working Group of the International Organization for Standardization. Develops and supports the standard GCC library, Boost libraries. Author of Boost libraries PFR, TypeIndex, DLL, StackTrace and active maintainer of Any, Conversion, LexicalCast, Variant.

Author of the book Boost C++ Application Development Cookbook.

Ilya Shishkov

Yandex Eats, Senior Software Developer

Has been working as a C++ developer at Yandex for 10 years. Experienced in high-load backend and applications, as well as in creating online courses.

Created an online specialization in C++ «Belts in C++», as well as the online course «Algorithmic Foundation for Programmers».

Speaker at various conferences:
C++ Russia, SECR, KnowledgeConf. Runs the Telegram channel «Algorithmic Foundatio».

Sergey Fedorov

Yango Deli Tech, Head of Warehouse Robots Development

Has been writing in C++ since 1999. Has done various tasks: from automating bakeries and crematoriums to developing banking and stock exchange software and game development. Currently, he is the Head of Warehouse Robots Development at Yango. Loves open source and puzzling tasks.

Sergey Vidyuk

Yandex Infrastructure, Senior Developer

Works on developing a build system for Yandex’s monorepository. Has 10 years of experience in developing mobile mapping applications (Navitel, then 2GIS). In his hobby time, he develops his own future/promise library, portable_concurrency.

Loves C++, large acyclic graphs, and hiding mutexes from business logic in multithreaded code.

Roman Lebedev

AWS, Senior Software Engineer

10+ years of experience in C++ development. Has worked in various fields: AV content processing, ad targeting, HFT. Spent the last 5 years working in the efficiency direction, first at Yandex Taxi, then at Lavka. Currently working on the development of AWS Lambda.

Anna Ilyakhina

Yandex, Backend DevRel

Has been side by side with IT professionals for more than 8 years. Builds relationships with communities: C++, DevOps, Database. Also works with open-source projects: ClickHouse, CatBoost, YDB, userver, YTsaurus, etc.

Organizes Yandex events, partners integrations (HighLoad Armenia, HighLoad Serbia), works with speakers at internal and external conferences, accompanies authors on «Habr», develops internal and external communities, works on online projects and DevRel strategy.

Former Mother of DevRel at «Leroy Merlin».

Mikhail Borisov

Yandex, Team Lead

Come on over,
it’s gonna be fun!;
Apply to
Contact the organizer
All materials are available under license CC BY-NC 3.0 with required reference to © Yandex LLC.
© 2019–2023, Yandex LLC
Tue Aug 01 2023 11:12:23 GMT+0300 (Moscow Standard Time)