domains / families / software-engineering / tilegym-cutile-autotuning

tilegym-cutile-autotuning

separate

Use 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.

When to use

• adding autotuning to CuTile kernels
• modifying CuTile autotuning code
• optimizing CuTile autotuning code
• debugging CuTile autotuning code
• designing parameter space for tile sizes and occupancy
• implementing tune-once/cache/launch pattern
• handling in-place writes with split-buffer
• validating performance with A/B benchmarking

When not to use

• tasks not involving CuTile autotuning
• code without exhaustive_search, replace_hints, hints_fn, or cuda.tile.tune
• filenames not containing autotune
• non-CuTile kernel optimization

Skill page

Source