Julia High Performance - Second Edition

SIMD parallelization (AVX2, AVX512)

Single Instruction, Multiple Data (SIMD) is method of parallelizing computation within the CPU, whereby a single operation is performed on many data elements simultaneously. Modern CPU architectures contain instruction sets that can do this, operating on many variables at once.

On Intel processors, these types of instructions have been progressively implemented using names such as SSE, AVX2, and AVX512. Each of these implementations add on extra functionality, but also allow operations on wider data. SIMD was first implemented in older Intel processors, with the name SSE, which went through multiple versions. Most Intel and AMD processors from the last decade implement the AVX2 instruction set, which provides ...