Optimization for Modern Processors
- Characteristics of modern processors
- Modern processors use increased parallelism to get high performance on DSP tasks
- Several different paths to achieve this goal:
- Allowing many parallel operations to be encoded in each instruction
- Issuing multiple instructions per cycle—superscalar, VLIW (very long instruction word) architectures
- Adding SIMD (single instruction, multiple data) capabilities
- Given the current architectural landscape, what optimization techniques are effective?