Data-Oriented C++ in Scientific Programming

Instructors: Jonas Latt

What you'll learn

Software architecture using an object-oriented and a data-oriented approach
Efficient programming using language-specific features in C++
GPU and multi-core CPU software development in C++

Skills you'll gain

Details to know

Shareable certificate

Add to your LinkedIn profile

Assessments

54 assignments

Taught in English

There are 5 modules in this course

Learn how to write efficient, maintainable C++ code for data-intensive applications in this hands-on course. Key application areas include scientific simulation software, statistical data processing, and computer graphics.

You’ll begin by exploring how modern C++ supports high-efficiency programming and review the principles of a central piece of hardware in data-intensive applications: computer memory. Next, the course shifts to software design. You’ll examine performance shortcomings of traditional object-oriented programming and dive into data-oriented perspective, which improves performance by structuring code around the data itself. You’ll learn how this paradigm overcomes bottlenecks and leads to more scalable, high-performance solutions. Finally, you’ll apply these principles to modern multi-threaded systems. Through hands-on experience you will learn to use C++'s built-in parallel features to target both multi-core CPUs and GPUs.

To follow this MOOC, you should be familiar with the fundamentals of the C++ language. In this first module however, you will find a refresher on the most important concepts needed to complete the course. You will (re)discover variable types (stack variables, references, smart pointers), containers and algorithms, classes and polymorphism. If you already master these notions but your knowledge predates the C++11 standard, we still encourage you to follow this module for up-to-date recommendations on the efficient use of C++. In particular, have a look at Lesson 4 if you are not yet familiar with the notion of ranges in C++20 and later versions.

What's included

17 videos1 reading11 assignments

17 videosTotal 119 minutes

MOOC trailer2 minutes
Welcome to this MOOC4 minutes
Arrays, vectors, and initialization8 minutes
Scope of variables and memory allocation10 minutes
Transfer of data from and into a function11 minutes
More about containers11 minutes
Algorithms and lambda functions11 minutes
Ranges: an introduction6 minutes
Ranges: more data-oriented examples 5 minutes
Views11 minutes
Smart pointers in C++ : basic concepts8 minutes
Smart pointers in C++ : vectors of smart pointers5 minutes
Structures and classes : Creation and initialization9 minutes
Structures and classes : function objects4 minutes
Polymorphic classes : general ideas7 minutes
Polymorphic classes : abstract and derived classes at work4 minutes
Let's wrap up1 minute

1 readingTotal 30 minutes

📢 Important preliminary information: Assessments and subtitles30 minutes

11 assignmentsTotal 114 minutes

Do you need Module 1 ?30 minutes
Arrays, Vectors, Variable Initialization8 minutes
Scope of variables and memory allocation5 minutes
Transfer of data from and into a function10 minutes
Containers10 minutes
Algorithms and lambda functions8 minutes
Ranges8 minutes
Views8 minutes
Smart pointers6 minutes
Structures and classes6 minutes
Polymorphic classes15 minutes

Part of the work of processing data efficiently consists in reducing the amount of computations and the complexity of the execution path. A complex execution path is often implemented to express design choices in the organization of software components. In the form of a compile-time execution framework, C++ offers a way to offload this complexity to the compilation stage and keep the execution environment simple and efficient.

What's included

12 videos1 reading10 assignments

12 videosTotal 98 minutes

Presentation of the module3 minutes
Introduction to templates11 minutes
Abbreviated function templates9 minutes
Template specialization and integer arguments9 minutes
Constant expressions12 minutes
Compile-time loops6 minutes
Compile-time polymorphism using templates10 minutes
Loop unrolling9 minutes
Variants and visitors8 minutes
Runtime polymorphism11 minutes
Interface-based development8 minutes
Let's wrap up2 minutes

1 readingTotal 10 minutes

Variants and Visitors (complement)10 minutes

10 assignmentsTotal 127 minutes

Using templates6 minutes
Using abbreviated syntax for function templates6 minutes
Integers as template arguments20 minutes
Which expressions can be evaluated at compile-time?6 minutes
Compile-time polymorphism using templates15 minutes
Loop unrolling7 minutes
Implement the variant-visitor strategy15 minutes
From dynamic to variant-based polymorphism15 minutes
Extend a library with templated polymorphism7 minutes
Final quiz30 minutes

Memory is one of the most critical resources of a computing platform. In this module, you will learn how memory is used in the C++ language and how memory access works on different types of machines. You will translate this knowledge into code that is at the same time clearly structured and efficient.

What's included

15 videos1 reading13 assignments

15 videosTotal 127 minutes

Presentation of the module3 minutes
Memory organization11 minutes
Virtual memory9 minutes
Memory management unit6 minutes
Memory latency11 minutes
Latency vs bandwidth8 minutes
Cache memory11 minutes
Cache latency in practice11 minutes
Data prefetching13 minutes
Vector instructions14 minutes
RAM vs CPU9 minutes
Definitions: FLOPS and AI5 minutes
Roofline performance model10 minutes
Roofline performance model in practice3 minutes
Let's wrap up3 minutes

1 readingTotal 10 minutes

Memory size comparison10 minutes

13 assignmentsTotal 170 minutes

Memory organization10 minutes
Virtual memory10 minutes
Memory management unit5 minutes
Memory latency10 minutes
Latency vs bandwidth15 minutes
Cache memory10 minutes
Memory size comparison10 minutes
Cache latency in practice10 minutes
Data prefetching15 minutes
Vector instructions15 minutes
Performance metrics15 minutes
Roofline performance model15 minutes
Final quiz30 minutes

Learn to think about your program in a way that puts the critical hardware resource, memory, in the center of your thoughts. Apply these principles to discrete-event systems, a problem solving strategy with applications in all applied areas of science. Understand that data-oriented programming is not opposed to object-oriented programming but rather offers a way to model your objects in a way that accounts for your data structure, your algorithms, and your hardware.

What's included

8 videos9 assignments

8 videosTotal 40 minutes

Presentation of the module2 minutes
Linear memory layout6 minutes
Array of Structures vs Structure of Arrays6 minutes
Complex Agent Based Model6 minutes
From Object Oriented to Data Oriented5 minutes
Scalable molecular dynamics simulation5 minutes
Data oriented solution8 minutes
Let's wrap up2 minutes

9 assignmentsTotal 103 minutes

Memory Layout Estimation4 minutes
Recognizing SoA and AoS6 minutes
Reading the example code15 minutes
Object modeling vs. data modeling10 minutes
Code transformation15 minutes
Efficiency analysis25 minutes
What data layout?3 minutes
Implementing the example5 minutes
Final quiz20 minutes

This module introduces you to an amazing feature available in the C++17 and onward: the ability to parallelize an algorithm with almost no effort, using C++ parallel algorithms. You will learn that you can gain a manifold speedup with simple changes by running your code on multi-core CPUs and on GPUs. You will dive once more into the working principles of computer memories to understand how to reach substantial performance improvements, and you will link the concepts of data-oriented programming to efficient paralellism.