gale-gpu-perf

What problem does it solve?

This Skill prevents GPU changes in gale-gpu from becoming correct-but-slow by enforcing device residency, persistent handles, and mandatory profiling before the work is considered done.

Core Features & Use Cases

• Performance Discipline: Catches one-shot kernel loads, per-step host-device ping-pong, and other GPU utilization killers.
• Production Readiness: Verifies that simulation state stays on the device and that time stepping is driven without unnecessary host synchronization.
• Profiling Gate: Requires real measurements with nsys, ncu, and nvidia-smi so kernel changes are judged by wall-clock behavior, not assumptions.
• Use Case: When a CUDA kernel, launch wrapper, or GPU integrator path changes, use this Skill to confirm the code is fast, resident, and sweep-ready.

Quick Start

Use this skill whenever you modify GPU code in gale-gpu and report the profiling results before declaring the change complete.