Sorting Algorithms: Speed Is Found In The Minds of People - Andrei Alexandrescu - CppCon 2019

If you're interested in these types of implementation/optimization details, Andrei Alexandrescu gave a great talk at this year's CppCon about practical sorting, e.g. when to break out of quicksort to something else: [VIDEO]

The article is C++, so not sure why you are bringing up AOT/JIT since it doesn't apply here. And clearly there is a large gain possible by doing a calculation rather than a conditional with a high entropy result, otherwise, the article's mentioned speedup wouldn't have happened. A PGO feedback loop isn't going to do anything on a high entropy conditional. I saw a great talk at CppCon 2019 by Andrei Alexandrescu about speeding up std::sort. He does this same trick among others to speed up the C++ std library sort routine. [VIDEO]

In his talk "Speed Is Found In The Minds of People" [0], Andrei Alexandrescu tries to make a sort really fast for medium sized vector and then closes on thoughts on this precise distinction. If this question intress you, it is worth watching. [0]: [VIDEO]

I enjoyed Alexandrescu’s CppCon 2019 talk “Speed Is Found In The Minds of People” About optimizing generic sorting routines for modern hardware. [VIDEO]

Well, this argument goes both ways, a generic implementation can never leverage special properties of your data. See i.e. Alexandrecu's talk [0] at about 1:05ff. If as a company you have special use cases or guidelines, you may easily benefit from libraries using these properties. [0] [VIDEO]

I'm not sure about much analysis and theory that explicitly captures branch prediction, but in last year's cppcon keynote Andrei offered a possible metric [0] where he takes into account the average "distance" between two subsequent array accesses. It's basically, (C(n) + M(n) + kD(n)) / n Where M is moves, C is comparisons, k is a some constant, and D is "distance". [0]: [VIDEO]

Do you have some more detaile about your app? I use perf, sysprof, trace32, visual studio profiler for profiling, but this highly depends on your environment. These assorted links might be interesting to you: - [LINK] - [VIDEO] - [LINK] - [LINK]

Not exactly a lecture series, but Andrei Alexandrescu has had some fantastic keynote presentations at CppCon. I don't even program in C++ any more, but I've watched these more than once, they're entertaining, approachable, and inspiring: Speed Is Found In The Minds of People [VIDEO] std::allocator Is to Allocation what std::vector Is to Vexation [LINK] An actual lecture series I've enjoyed watching is "The WE-Heraeus International Winter School on Gravity and Light", which covers General Relativity using modern mathematical approaches: [LINK] Again, I haven't done physics in 20 years, but I've found it surprisingly easy to follow because it is so clearly presented.

Here's an excellent example of what GP said: [VIDEO] (long video, skip to 26:00 if you are in a hurry) The key is that code isn't running in a vacuum. It's operating on data (or controlling systems, etc). A smaller amount of code may be operating on the data in an inefficient manner, whereas a larger amount may be doing it more efficiently. In the above talk, this contrast is very stark, because the "more code" version does some stuff and then does the exact thing the "less code" version does, yet is faster. This kind of thing is common enough (albeit less stark than in the above example), that "less code is faster" is not a great "default" assumption to make.

I really enjoyed “Speed Is Found In The Minds Of People”, a talk by Andrei Alexandrescu. At first he is just showing simple stuff, but stay with it, he goes deep. [VIDEO]

The FAANGS certainly have people doing this kind of work. For example, Andrei Alexandrescu has a bunch of keynote talks at CPPCon where he talks about 1-4% speed optimisations he makes to the std library at Facebook. [VIDEO] I also like the talk by Nicholas Ormrod “The strange details of std::string at Facebook": [LINK]