The problem of finding a character in a large file is solved using both the built-in string.Contains method and a manual for-loop.
A performance test shows that string.Contains is five times faster than a manual for-loop, introducing the concept of vectorized operations.
The concept of Single Instruction, Multiple Data (SIMD) is explained using an analogy of grocery shopping to contrast it with loops and multi-core processing.
Modern CPUs use special instruction sets like AVX and SVE with wide registers to perform SIMD operations, which .NET exposes through high and low-level APIs.
A conceptual approach for checking if an array is sorted is designed by loading overlapping data chunks into two vectors for pairwise comparison.
The IsSorted method is implemented using the Vector<T> API, covering Vector.LoadUnsafe, Vector.GreaterThan, and strategies for handling array boundaries.
Using Benchmark.NET, the SIMD implementation of IsSorted is shown to be about three times faster than the standard loop due to instruction overhead.
SIMD offers significant performance gains but increases code complexity, making it a trade-off best suited for performance-sensitive applications.
Jobs that call for the skills explored in this talk.