Profiles and debugs CUDA kernels using nsys, ncu, cuobjdump, and cuda-gdb.
npx skills add https://github.com/rdromer2/mambavision_tf_kernel --skill cuda
Please help me install this Agent Skill. Skill: cuda Source: https://github.com/rdromer2/mambavision_tf_kernel/tree/main/.agents/skills/cuda Command: npx skills add https://github.com/rdromer2/mambavision_tf_kernel --skill cuda
Streamlines CUDA kernel development, debugging, and performance profiling.
Run a baseline profiling session on your CUDA kernel to identify hotspots and begin iterative optimization.
High-intent search queries and answers about installing and using this skill.
To profile a CUDA kernel, establish baseline timing and run a profiling session using Nsight Systems (nsys) to identify hotspots, then apply Nsight Compute (ncu) for targeted kernel analysis and iterative optimization.
Debugging CUDA kernels involves using cuda-gdb for interactive debugging and compute-sanitizer for memory error checks, supplemented by printf-based tracing to isolate logic errors and verify memory access patterns.
Yes, you can inspect CUDA binaries using cuobjdump to analyze kernel memory access patterns and verify optimizations, which is essential for structured profiling and improving kernel performance.
Nsight Systems (nsys) provides system-wide profiling to locate hotspots, whereas Nsight Compute (ncu) performs targeted kernel analysis to optimize memory access and compute patterns for measurable improvements.
Your CUDA kernel optimization lacks measurable improvements if you skip establishing a baseline timing or fail to follow a structured profiling loop with targeted analysis using nsys and ncu.
flashinfer-aiExtract GPU kernel schemas from SGLang models into deduplicated Definition JSON files.
yo-stevenClassify SOL% bottlenecks and diagnose CUDA kernel underperformance from Nsight Compute output.
m0atAnalyze CUDA kernel performance bottlenecks with Nsight Systems and Nsight Compute.
wojonsAnalyze system and application performance with Linux kernel-level profiling tools.
