Rationales behind C++ atomics, Faster complex numbers for C++

Hi,

this time we will have two talks. One by Yegor Derevenets about the Rationales behind C++ atomics and one short talk by André Bergner on faster complex number for C++.

There are two reasons to be on time (before 20:00).
1.) We will have Pizza and Beer for you! ;)
2.) There is a big gate to the backyard that gets closed at 20:00h. In case you are late please call the number 030-6110351180 and somebody will come and open the gate for you. The meetup will be in the last of the yards before the river as shown in this picture.

http://photos1.meetupstatic.com/photos/event/d/9/9/6/600_436375702.jpeg

Abstracts

Faster complex numbers for C++

Complex numbers are important fundamental tool from mathematics enabling many problems to be written in a more generic form. The STL comes with an implementation to work with complex numbers in a natural way.

Motivated by some useful real world examples I will discuss benchmarks of std::complex and demonstrate how an alternative implementation based on expression templates outperforms std::complex and competes with hand-crafted C code.

Rationales behind C++ atomics

C++11 introduced special atomic types that allow the user to perform operations on memory atomically. In addition to that, these atomic operations allow the user to request a specific order in which they, as well as preceding and succeeding memory accesses, will hit the memory, i.e., will be observed by other threads.

The default order is sequential consistency: atomic operations are visible to all the threads in the same order, they are not reordered with other memory accesses in the same thread. C++ provides a way to request weaker guarantees: release, acquire, consume, relaxed memory order.

In the talk I will try to explain why having five (actually, six) different memory orders is a good idea, what they actually mean, what is the intuition, and how atomic memory accesses with these orders are compiled into x86 and Power assembly.