Tilegym Improve Cutile Kernel Perf
NVIDIA/TileGym/skills/tilegym-improve-cutile-kernel-perf
analysisOfficial
Official Provider SkillView repo
Iteratively optimize cuTile kernel performance through systematic profiling, bottleneck analysis, IR comparison, and targeted tuning. Covers tile sizes, occupancy, autotune configs, TMA, latency hints, persistent scheduling, num_ctas, flush_to_zero, and IR-level debugging. Use when asked to "optimize cutile kernel", "improve kernel perf", "tune cutile performance", "make kernel faster", or iteratively benchmark and refine a cuTile GPU kernel in the TileGym project.
Files11 files
BENCHMARK.md78 lines
Loading editor…
Install
RecommendedOne command — your agent picks it up automatically.
Select an AI agent above to see the install command.
or
Manual Install
More stepsDownload the archive and add the files to your project manually.
Skill details
Versionv1.0.0
AuthorNVIDIA
Categoryanalysis
Skill IDNVIDIA/TileGym/skills/tilegym-improve-cutile-kernel-perf
Files11 files
Related skills
Tilegym Adding Cutile KernelAdd a new cuTile GPU kernel operator to TileGym. Covers dispatch registration in ops.py, cuTile backend implementation, __init__.py exports, test creation, and benchmark in tests/benchmark. Use when adding, creating, or implementing a new cuTile operator/kernel in TileGym, or when asking how to register a new cuTile op.Tilegym Converting Cutile To JuliaConverts cuTile Python GPU kernels (@ct.kernel) to cuTile.jl Julia equivalents. Handles kernel syntax translation, 0-indexed to 1-indexed conversion, broadcasting differences, memory layout (row-major to column-major), type system mapping, and launch API differences. Use when converting, porting, or translating cuTile Python kernels to Julia cuTile.jl, or debugging/optimizing existing Julia cuTile translations.Tilegym Converting Cutile To TritonConverts cuTile GPU kernels (@ct.kernel) to Triton (@triton.jit). Handles standard in-repo conversion, debugging (cudaErrorIllegalAddress, shape mismatch, numerical mismatch), and mapping cuTile idioms (ct.load/ct.store, ct.Constant, ct.launch) to Triton equivalents. Covers dual-kernel layout flags (e.g. transpose=True/False + autotune grid via META) per translations/advanced-patterns.md. Use when converting, porting, or translating cuTile kernels to Triton, or debugging existing Triton translatTilegym Converting Cutile Triton To Cutile RsUse this skill to convert, port, or translate Triton-TileIR or cuTile-Python GPU kernels to cutile-rs (Rust). The orchestrator runs scripts/preflight.sh, then drives a bounded Agent A -> B -> D -> E pipeline (Agent C is diagnostic, Agent F optional), delegating all kernel/host/correctness/perf work to sub-agents and routing by each stage's single-line VERDICT.Tilegym Cutile AutotuningUse when adding, modifying, optimizing, or debugging CuTile autotuning code. Trigger signals: `exhaustive_search` / `replace_hints` / `hints_fn` / `cuda.tile.tune` in code, `autotune` in filenames, or correctness/performance issues in autotuned CuTile kernels. Covers: tune-once/cache/launch pattern, per-architecture configs (sm80–sm120), parameter space design (tile sizes, occupancy, num_ctas), and 7 common pitfalls with solutions.Tilegym Cutile OptimizingQuestion-gated router for the curated cuTile optimization wiki: kernel-family playbooks, measured techniques with caveats, performance patterns, and practical language knowledge for Blackwell and Hopper. Use when an agent has inspected the current kernel and has a concrete question that could change its plan, when measurements point to a specific bottleneck, or when a cuTile language or API detail is blocking implementation. Do not use as mandatory onboarding, as the source of the first design,