Patternia Performance - v0.8.0

Overview

v0.8.0 introduces a tiered variant dispatch model and related runtime/layout optimizations for large-branch workloads.

---

Core Algorithms

1. Tiered Variant Dispatch

Variant dispatch is selected by alternative count:

• hot_inline for small alt counts
• warm_segmented for medium alt counts
• cold_compact for large alt counts

This reduces hot-path code pressure while preserving fast dispatch for common small/medium cases.

2. Compact Alt Map for Large Variant Sets

For large variant spaces, dispatch uses:

• active_index -> compact slot map
• miss -> fallback path
• hit -> dense trampoline table

This avoids scanning irrelevant alternatives and keeps large dispatch data compact.

3. Per-Alt Case Range Narrowing (Typed Variant)

Typed variant dispatch computes case ranges per active alternative, then scans only the relevant range instead of the full case set.

4. Hot/Cold Path Separation

Large-branch dispatch work is moved into non-inline cold helpers to reduce instruction footprint in hot entry paths.

---

Implementation Pointers

• include/ptn/core/common/optimize.hpp: dispatch tiers and metadata builders
• include/ptn/core/common/eval.hpp: tier selection and runtime dispatch execution

---

Notes

• This page records algorithm strategy for v0.8.0.
• API-level changes are documented in: v0.8.0