domains / families / software-engineering / tilegym-improve-cutile-kernel-perf

tilegym-improve-cutile-kernel-perf

separate

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.

When to use

• optimize cutile kernel
• improve kernel perf
• tune cutile performance
• make kernel faster
• iteratively benchmark and refine a cuTile GPU kernel
• systematic profiling
• bottleneck analysis
• IR comparison

When not to use

• changing kernel functional semantics
• modifying test infrastructure
• modifying benchmark harness

Skill page

Source