Skill Claude CodeCodex
Draft, debug, and measure native CUDA C++ kernels and host launch workflows. Use for nvcc compilation, launch contracts, compute-sanitizer, Nsight timelines, focused kernel profiling, and generated-code inspection of hand-written CUDA; for TileLang programs use the TileLang debugging/profiling skills instead.
Skill Claude CodeCodex
Diagnose TileLang compilation, host argument, memory-liveness, runtime, and numerical failures. Use for failing kernels, buffer lifetime or shared-memory reuse analysis, generated-code and TVM-pass inspection, lowering traces, layout diagnostics, and minimal reproducers.
Skill Claude CodeCodex
Tune a correct TileLang kernel using measured bottlenecks, tile sizes, layouts, pipelining, fusion, and autotuning. Use when improving kernel performance; use profiling to establish measurements and debugging to resolve incorrect output.
Skill Claude CodeCodex
Measure TileLang kernel and workflow performance, choose benchmark or timeline tools, and compare implementations fairly. Use for latency, throughput, profiler selection, and establishing optimization baselines on TileLang-generated code; use the cuda skill for hand-written CUDA C++ sources.
Skill Claude CodeCodex
Implement and validate forward and backward TileLang kernels integrated with torch.autograd.Function. Use for custom differentiable operators and gradient mismatches; use the writing or debugging skills for forward-only kernels.
Skill Claude CodeCodex
Inspect a PyTorch baseline with torch.compile graphs/code and torch.profiler before drafting a custom kernel, then compare TileLang and PyTorch timelines, forward/backward regions, and allocations. Use for baseline analysis and attribution; use dobench for controlled latency comparisons.
Skill Claude CodeCodex
Draft TileLang operators using JIT, tensor contracts, tiled computation, and reusable kernel patterns. Use for new kernels, ports, dynamic shapes, and choosing between eager and PrimFunc-factory programming styles.
Skill Claude CodeCodex
Describe model or operator HIR with TileFoundry, inspect topology and placement, run static cost analysis, visualize authored dataflow, and check a runtime implementation against its evaluator. Use for TileFoundry requests and high-level fusion, sharding, or memory-placement design before or alongside TileLang kernel…
Skill Claude CodeCodex
Run a bounded TileLang implementation and optimization task from a PyTorch reference or existing kernel through baseline analysis, a written plan, candidate edits, lowering inspection, validation, measurement, and promotion. Use for iterative kernel work and reproducible handoffs; use a specialist directly for a…
Skill Claude CodeCodex
Write human-facing TileLang design documents for proposed or completed kernel implementations. Use when explaining an algorithm, execution mapping, memory flow, pipeline, tensor lifetimes, compiler evidence, tradeoffs, validation, or measured results to a reviewer. For machine-checked HIR modeling and static cost…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: