Hi everyone! I'm writing a book on performance engineering, and a few days ago, I finished a draft of one of its main crown jewels: the SIMD programming chapter.

The main findings that are published:

• You can compute array sums and other reductions such as the minimum 2x faster than std::accumulate or an auto-vectorized loop would.
• You can sometimes copy and assign memory 2x faster than memcpy and memset respectively.
• You can search array elements 10x times faster than std::find or manually.
• You can count a value in an array 1.5x faster than std::count or manually.
• You can calculate population counts of large vectors ~2x faster than with the intrinsic.
• You can filter arrays 6-7x faster, which translates to e. g. a 6-7x faster quicksort (to be published).
• You can calculate the index of the minimum element ~10x faster than std::min_element or manually.
• UPD: you can calculate the prefix sum of an array ~2.5x faster than std::partial_sum or manually.

All speedup numbers are architecture-specific and may be different (usually larger) on CodeForces servers.

Enjoy — and as always, I'm happy to hear any comments and suggestions.