ncu-analysis

Analyze CUDA kernel performance using NVIDIA Nsight Compute (NCU). Use when the user wants to: (1) profile a CUDA kernel with NCU to generate .ncu-rep files, (2) analyze NCU reports programmatically using the ncu_report Python API, (3) extract specific metrics from NCU profiles (duration, throughput, occupancy, cache hit rates), (4) compare two kernel profiles side-by-side, (5) identify performance bottlenecks in GPU kernels, or (6) work with .ncu-rep files in any way.