A whirlwind introduction to dataflow graphs (2018)
a day ago
- #machine-model
- #performance-optimization
- #dataflow-graphs
- The article discusses the importance of understanding dataflow graphs for performance optimization in code.
- It emphasizes the need for a predictive performance model rather than relying solely on descriptive profiling.
- The author introduces a simplified machine model to analyze compute-bound loops and predict their behavior.
- Dataflow graphs are presented as a fundamental tool for understanding dependencies and optimizing performance.
- Examples include summing integers in an array versus a linked list, highlighting the impact of memory access patterns.
- Loop unrolling is explored as a technique to reduce overhead and improve throughput.
- The article concludes by stressing the generality of dataflow analysis across different scales and contexts.