cuda-kernel-refine

One change per cycle: baseline → profile → classify → optimize → verify → compare → loop. Multiple simultaneous changes make it impossible to attribute improvement. Revert on regression.

Discover the project's benchmark infrastructure before running anything:

Run with enough iterations for stable results — coefficient of variation <5%.

使用 NVIDIA 分析工具（nsys、ncu）進行迭代 CUDA 核心最佳化。在優化內核、提高吞吐量、減少頻寬、分析屋頂線、比較基准或調查寄存器壓力和占用率時使用。 來源：trevors/dot-claude。