Vectorization
The transformation of scalar computation into operations that process multiple data elements per instruction or accelerator operation.
Performance Context
Vectorization converts scalar work into operations that process several elements per instruction or accelerator operation. Compilers may auto-vectorize suitable loops, while explicit intrinsics expose a chosen SIMD instruction set more directly. Contiguous memory, alias information, and loop dependencies strongly affect success.
Measurement Boundary
Vectorization is not automatically faster. Small problem sizes, gather/scatter access, branch-heavy code, alignment costs, or a memory-bandwidth bottleneck can erase the arithmetic advantage.
Related Performance Concepts
- SIMD
- Memory Bandwidth
- Loop Optimization
- Operator Fusion