Internals

This section documents the internal architecture of stablebear. It is intended for contributors and advanced users who want to understand how the library works under the hood.

• CUDA block execution engine
  • File structure
  • Block decomposition
    • Algorithm
    • Memory budget and nSplitsHint
      • Minimum block side (GPU occupancy floor)
    • Higher-rank tensors
    • Triangle skip modes
  • Double-buffered pipeline
  • Streaming function data
  • Result scatter
  • BlockOp policy
  • PCF-specific implementation
    • PCF data layout
    • Rectangle iteration kernel
  • Operation data flows
  • Adding a new function type
• Deterministic random generation
  • Design goal
  • Architecture
  • Engine: xoroshiro128++
  • Seeding via make_engine
    • splitmix64
  • RandomGenerator
  • Walk integration
    • Why flat indexing works
  • Concrete example
  • Global and explicit generators
  • Switching engines
  • References