3.6.1 : What is vectorization ?
It is a copy-paste of the computing part of a CPU (ALU) to enable several computing at the same time.Figure 2 : Comparison between scalar (classical) computation and vectorization.
Different vectorized architectures exist :
- SSE2 : compute 2 floating point single precision computation at the same time
- SSE4 : compute 4 floating point single precision computation at the same time
- AVX : compute 8 floating point single precision computation at the same time
- AVX 512 : compute 16 floating point single precision computation at the same time