tilegym-cutile-autotuning
Originally fromnvidia/skills
Installation
SKILL.md
CuTile Autotuning
Add autotuning to CuTile kernels using the exhaustive_search API with tune-once/cache/direct-launch pattern.
Instructions
Follow the decision tree to classify the kernel, design a search space, implement the tune-once/cache/launch pattern, and validate performance.
- Classify — use the Decision Tree to determine search dimensions (occupancy-only vs full tile search)
- Design search space — select the matching template from
references/kernel-type-templates.md; prune to ≤ 30 configs in the final code via arch filters (directed exploration probes may temporarily exceed this — see Design Philosophy) - Implement — add
exhaustive_search+ cache +ct.launchfollowing the Step-by-Step Workflow; handle in-place writes with split-buffer if needed - Test — run correctness with autotune enabled and with
DISABLE_AUTOTUNE=1 - Validate — A/B benchmark against fixed best-known config; see
references/search-strategies.md - Shrink — prune dead-weight configs that never win, targeting ≤ 8 configs per architecture to minimize compilation cost (Step 10)