# Tilelang Vector Skill

> TileLang npuir Vector 算子开发指南。用户提及逐元素、激活函数、归约、广播、sigmoid、rmsnorm、softmax 子流程、vadd/vmul/vexp/vcast/vbrc、向量精度或向量性能优化时必须使用本技能。默认输出必须优先采用 v 前缀 API，而非 npuir_xxx 形式。

- Skill: `tile-ai/tilelang-vector-skill` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add tile-ai/tilelang-vector-skill`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tile-ai/tilelang-vector-skill/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Integrations & APIs
- Author: tile-ai (https://skillmd.com/u/tile-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tile-ai/tilelang-vector-skill

---


# TileLang Vector Skill (npuir)

## Mandatory routing rule

Before answering, follow AGENTS.md section "Docs Auto Routing Rules (Mandatory)".

## Operator baseline rule (Mandatory)

- Before writing a new vector operator, first check examples/ and testing/npuir/.
- Prefer adapting an existing operator case rather than writing from scratch.

## API style policy

Mandatory default style:
- Prefer T.vadd, T.vsub, T.vmul, T.vdiv
- Prefer T.vexp, T.vln, T.vsqrt, T.vrsqrt, T.vrelu, T.vsigmoid
- Prefer T.vcast, T.vbrc, T.vcmp, T.vselect

Compatibility:
- T.npuir_add and friends are allowed only for compatibility with legacy code.

## Core workflow

1. Define shape and block strategy
2. Allocate UB or shared buffers based on mode
3. Copy in, compute with v-prefix APIs, copy out
4. Validate against torch reference

## References

- references/api-quickref.md
- references/examples.md
- references/troubleshooting.md

## Official docs to consult

- docs/Tilelang.language/数学操作/T.vadd.md
- docs/Tilelang.language/数学操作/T.vmul.md
- docs/Tilelang.language/数学操作/T.vexp.md
- docs/Tilelang.language/数据类型转换操作/T.vcast.md
- docs/Tilelang.language/shape操作/T.vbrc.md
- docs/Tilelang.language/规约操作/T.reduce.md

## Related skills

- tilelang-cube-skill
- tilelang-mixcv-skill
- tilelang-debug-helper