Stefanus Du Toit: “Hourglass Interfaces for C++ APIs”
Pedro Ramalhete, Andreia Correia: “How to Make Your Data Structures Wait-Free for Reads”
Gor Nishanov: “await 2.0: Stackless Resumable Functions”
Fedor Pikus: “Where did My Performance Go? (Scaling Visualization in Concurrent C++ Programs)”
Łukasz Mendakiewicz: “Viewing the World Through Array-Shaped Glasses”
Stefanus Du Toit: “Hourglass Interfaces for C++ APIs“
C++ provides a much richer set of abstractions than C. Classes, templates, overloading, and other core C++ features can be leveraged for more readable syntax, better compile time typechecking, more open ended genericity and improved modularity. On the flip side, C89 still boasts some advantages over C++, especially when viewed through a pragmatic lens. C ABIs on many platforms have been stable for decades, practically every language supports binding to C code through foreign function interfaces, and including nearly any C89 header has a negligible effect on compile time on modern computers. The Hourglass pattern provides the best of both worlds. It’s a way to structure libraries that retains the pragmatic benefits of C89 while still providing C++’s richness both at an interface and implementation level. It makes providing bindings from other languages to C++ libraries easier, and insulates from ABI issues such as incompatibilities between debug and release variants of runtimes. This talk provides an overview of the pattern, teaches practical techniques for its implementation using C++98 and C++11, and shares experience from using the pattern in real world projects.
Speaker’s bio: Stefanus Du Toit is a Software Lead at Thalmic Labs, where he enables amazing gestural experiences using the Myo armband. Stefanus previously worked as a Software Architect and Software Development Manager at Intel Corporation, and co-founded RapidMind (acquired by Intel), a startup that targeted GPUs and other processors using standard C++. Stefanus served on the C++ standards committee as Project Editor for C++14 and as Committee Secretary. He lives in Kitchener-Waterloo, Ontario, Canada and holds a BMath CS from the University of Waterloo.
Pedro Ramalhete, Andreia Correia: “How to Make Your Data Structures Wait-Free for Reads“
In this talk we will describe a new concurrency control algorithm with Blocking write operations and Wait-Free Population Oblivious read operations, which we named the Left-Right algorithm. We will show a new pattern where this algorithm is applied, which requires using two instances of a given resource, and can be used for any data structure, allowing concurrent access to it similarly to a Reader-Writer lock, but in a non-blocking manner for reads, including safe memory management without needing a GC.
Speaker’s bio: Pedro Ramalhete has a PhD in Particle Physics for research done at CERN. Most of the research itself required programming in C++, and he also had a major participation in the coding of the experiment’s data acquisition and decoding software in C and C++. Pedro is currently working at Cisco writing networking software, with emphasis on real-time concurrent synchronization techniques, and wait-free/lock-free data structures.
Gor Nishanov: “await 2.0: Stackless Resumable Functions“
When dealing with the task/future based asynchronous model, developers need to write lambda-heavy continuation code with potentially multiple nesting lambdas. This makes the code less readable and not very pleasant to write. Stackless resumable functions allow developers to use more natural imperative coding style and provide performance benefits that are not possible with library-only solutions.
Speaker’s bio: Gor Nishanov is a Principal Software Design Engineer on the Microsoft C++ team. He works on the ‘await’ feature. Prior to joining the C++ team, Gor was working on distributed systems in Windows Clustering team.
Fedor Pikus: “Where did My Performance Go? (Scaling Visualization in Concurrent C++ Programs)“
High performance is one of the main reasons programmers choose C++ for their applications. If you are writing in C++, odds are you need every bit of computing power your hardware can provide. Today, this means writing multi-threaded programs to effectively utilize the multiple CPU cores that the hardware manufacturers keep adding. Everyone knows that writing multi-threaded programs is hard. Writing correct multi-threaded programs is even harder. Only after spending countless hours debugging race conditions and weird intermittent bugs do many programmers learn that writing efficient multi-threaded programs is harder yet. Have you ever wanted to see what are all your threads doing when they should be busy computing? This talk will show you how. We begin by studying several techniques necessary for collecting large amounts of data from the running program very efficiently, with little overhead and minimal disruption to the program itself. We will cover efficient thread-safe memory management and efficient thread-safe disk I/O. Along the way we dabble in lock-free programming just enough to meet our needs, lest the subject will spiral into an hour-long talk of its own. With all these techniques put together, we can collect information about what each thread is doing, which threads are computing and what exactly, and which threads are slacking off waiting on locks, and do it at the time scale of tens of microseconds if necessary. Then we process the collected data and create a timeline that shows exactly what the program was doing at every moment in time.
Łukasz Mendakiewicz: “Viewing the World Through Array-Shaped Glasses“
It’s agreed among experts that the most performant data structure in C++ is an array. Or a vector. Or a dynarray. Indeed, until recently there was no standardized approach in C++ to view these types in a uniform manner. It was even murkier when the data had logically more than one dimension. This talk is an introduction to the new features proposed for C++17 in N3851 bringing all contiguous data into harmony and lifting it to higher dimensions: index, bounds, array_view and more. Attendees will also learn how indexable algorithms differ from the traditional elemental ones, and what does it mean for parallelism.
Speaker’s bio: Łukasz Mendakiewicz is a software engineer at Microsoft, where he focuses on the customer experience with parallel programming models for C++. He is especially interested in GPGPU acceleration, and puts this passion to work on C++ AMP. He holds an M.S. in Computer Science from AGH UST in Krakow, Poland.