SIMD Everything
Using the full parallel power of the modern chip instead of sequential scalar processing.
Performance
We treat the hardware as a sacred resource. High performance is not an "optimization" phase; it is a core structural requirement from day one.
Mandates
Cache-First Design: Design for memory hierarchy, not just logic.
Branch-Optimized: Code that respects the CPU's prediction logic.
No Latency Hiding: Solve the problem, don't just buffer it.
Physical Reality
Using the full parallel power of the modern chip instead of sequential scalar processing.
Pre-allocated arenas and predictable lifetimes instead of random heap fragmentation.
If it doesn't contribute to the core mission, it is removed. No "just-in-case" overhead.
The developer stays in control of the metal. No layers of indifference allowed.