Experimental API

The experimental public layer reports performance and graphics capabilities for qualification. It does not expose the profiler implementation, raw OpenGL timer queries or mutable metric buffers.

Types

Type	Role
PerformanceMode	OFF, CPU or CPU_GPU request
PerformanceMetric	Named measurement vocabulary
PerformanceSnapshot	Read-only captured session data
MetricStatistics	Aggregated sample/call/fps/percentile/threshold statistics
GraphicsCapabilities	Read-only renderer/vendor/version and feature report
GpuTimerPolicy	Safe backend-selection policy
GpuTimerBackend	Effective GPU timing mechanism
GpuTimerArchitecture	Detected architecture family used by policy

Measurement levels

flowchart LR
  O[OFF] --> C[CPU<br/>wall-time samples]
  C --> G[CPU_GPU<br/>CPU plus supported GPU timing]
  G --> S[PerformanceSnapshot]
  S --> Q[Benchmark/qualification report]

CPU_GPU is a request, not a guarantee. Inspect getEffectiveMode(), hasGpuTimings(), the selected backend and diagnostics before interpreting values.

Interpretation rules

• CPU wall time includes host scheduling effects; GPU elapsed time measures a different execution domain.
• Percentiles describe the retained bounded sample window, not an infinite history.
• Capability flags say what the active context reports, not that a feature passed visual qualification.
• Benchmark evidence must record version/commit, Processing/Java, OS, CPU/GPU, resolution, routes, warm-up, duration and metric mode.

Use BenchmarkTool for reproducible qualification. Do not build scene behavior that depends on an experimental metric name.