Migrate operators into TileOPs with a phased kernel-to-op workflow.
npx skills add https://github.com/tile-ai/TileOPs --skill migrating-new-op
Please help me install this Agent Skill. Skill: migrating-new-op Source: https://github.com/tile-ai/TileOPs/tree/main/.claude/skills/migrating-new-op Command: npx skills add https://github.com/tile-ai/TileOPs --skill migrating-new-op
Provides a structured process for adding or migrating an operator into TileOPs, ensuring a clear kernel-to-op delivery path and a comprehensive validation checklist.
Confirm the target level (L1 or L2) and follow the required phase sequence (kernel, then op, then tests) to migrate an operator.
High-intent search queries and answers about installing and using this skill.
Migrating a CUDA operator into TileOPs requires following a phased workflow: requirement alignment, kernel-first migration using TileLang, op wiring, tests, and cleanup. You must implement the kernel before wiring the op interface.
The correct dependency direction for a new operator in TileOPs is strictly one-way: op depends on kernel. Synchronized exports are enforced to prevent stale interfaces between the op and kernel layers.
To validate a migrated operator in TileOPs, follow the testing phase after completing kernel migration and op wiring. The process enforces Level-based deliverables and a validation checklist to standardize implementation and tests.
Yes, you must confirm the target level (L1 or L2) before migrating a kernel to TileOPs. Level Confirmation is an enforced step that defines the required phase sequence and explicit deliverables before coding begins.
The best way to structure files for a new TileOPs operator is to follow the design-by-layer separation guidance with explicit file organization and synchronized exports. This maintains the strict one-way op-to-kernel dependency.
Yes, you can migrate a Triton kernel into TileOPs using the kernel-first migration phase. The workflow standardizes implementation by referencing existing TileLang kernels and TileOPs interfaces for the migration.
tile-aiValidate kernel and op format conformance for T.prim_func and TileLang macros.
m16khbRun the test suite and present four options for merging, PR, keeping, or discarding a branch.
mindspore-aiOptimize TileLang CUDA kernels and diagnose compilation and runtime issues.
mindspore-aiOptimize Triton CUDA kernels with performance strategies and API usage rules.
