Common optimize technique
Vectorization（矢量化）
Before we understand the vectorization, we can see a common secnario.
We have a array that has 100 float numbers, we want to calculate square of every data. If we use traditional computer, we need to calculate it one by one. For each calculation of data, an instruction needs to be excuted, it is a time-consuming process. As shown in the following figure.
Vectorization is an optimize technique, it uses SIMD（Single Instruction, Multiple Data）instruction set of modern processors to utilize one instruction to judge multiple data at the same time. It reduces the iteration times and overhead of commute instruction. As shown in the following figure.
How can vectorization technique speed up the process of excutation ?
Vectorization technique relies on SIMD instruction set, this instruction set provide special instruction and register to judge multiple data items at the same time. e.g.
- Intel
- SSE（Streaming SIMD Extensions）
- AVX（Advanced Vector Extensions）
It is x86 micro processor instruction set extension which was introduced by Intel.
- ARM
- NEON（Advanced SIMD）
How can computer get the SIMD instruction from original instruction ?
Before computer excutes the code, it need a compiler to transform original code to SIMD instruction. So it asks us to follow some rules to help the compiler to vectorize original code.
