Sasha Goldshtein — The Vector in Your CPU: Exploiting SIMD for Superscalar Performance

Sasha Goldshtein, Sela Group — The Vector in Your CPU: Exploiting SIMD for Superscalar Performance Конференция DotNext 2015 Moscow Москва, For more than ten years modern processors have been equipped with vector registers and instructions that can make certain algorithms four to eight times faster. In the past, it took special language intrinsics or intimate knowledge of assembly language to use them. In this talk you will learn how to take advantage of vector instructions in .NET applications using the newly released library and the RyuJIT compiler that’s part of .NET 4.6. We will explore a variety of algorithms that can benefit from vectorization, ranging from simple 4-8x-wins like matrix multiplication and dot product, all the way to algorithms that must be refactored to benefit from vectorization because they contain inter-iteration dependencies or complex control flow.