| title | date | lastmod |
|---|---|---|
Data Level Parallelism |
2022-11-08 |
2022-11-21 |
The same operation is performed on multiple data values concurrently in multiple processing units. Can reduce the Instruction Count to enhance performance
Different types of hardware can support different levels of data parallelism.
- Allow sequential thinking yet achieves parallel speedup
- Reduced energy usage
- More efficient parallel efficiency
One vector instruction can perform N computations, where N is the vector length.
- Reduces the number of instructions: less branching
- Less execution time with lower instruction count
- Simpler design as there is no requirement for data dependency check since each execution is independent
- Array processor works more like a true parallel system, with each processor able to run same instruction on different data
- Vector processor works in more of a pipelined fashion.
