Implement hotspot detection.
Automatically identifying time intervals and call stacks where work concentrates.
In a timeline + flame graph UI, it would surface:
- Tallest stacks (most cumulative time) across CPU/GPU tracks.
- Longest self-time nodes (exclusive time) within stacks.
- High-frequency events (many short events that add up).
- Synchronization stalls (CPU waiting on GPU or vice versa).
- Correlation hotspots (CPU and GPU peaks aligned in time).
- Outliers (events much longer than typical for same kernel/function).
Visually, it’s the “hottest” regions: tallest, widest, or densest bands on the flame graph, and the heaviest spans on the timeline.
There should be an option to automatically find and highlight these areas based on one of the listed heuristics.
Implement hotspot detection.
Automatically identifying time intervals and call stacks where work concentrates.
In a timeline + flame graph UI, it would surface:
Visually, it’s the “hottest” regions: tallest, widest, or densest bands on the flame graph, and the heaviest spans on the timeline.
There should be an option to automatically find and highlight these areas based on one of the listed heuristics.