Skip to content

Add skills / collection with aie-kernel-opt guide#3162

Open
hunhoffe wants to merge 2 commits into
mainfrom
aie-kernel-opt-skill
Open

Add skills / collection with aie-kernel-opt guide#3162
hunhoffe wants to merge 2 commits into
mainfrom
aie-kernel-opt-skill

Conversation

@hunhoffe

Copy link
Copy Markdown
Collaborator

Introduce a top-level skills/ directory holding Agent Skills: self-contained, on-demand guides that capture measured, codebase-specific knowledge for coding agents and humans alike.

The first skill, aie-kernel-opt, is a standalone guide to optimizing AIE / Peano-compiled kernels: a measure-first methodology (baseline, bit-exact gate, ablation, verify-in-.o) plus a priority-ordered catalog of concrete optimization levers (loop hints, compile-time constants, divide elimination, branch-splitting, epilogue vectorization, operand-layout pre-pack, wide packing, wider mmul, DMA layout offload) and the constraint each one carries. Mechanisms are written generically across AIE generations; performance deltas are framed as illustrative optimization headroom on example kernels.

I derived this skill from my work in #3142 I'm sure it could be improved when applied to a wider variety of kernels.

hunhoffe and others added 2 commits June 10, 2026 15:59
Introduce a top-level skills/ directory holding Agent Skills: self-contained,
on-demand guides that capture measured, codebase-specific knowledge for coding
agents and humans alike.

The first skill, aie-kernel-opt, is a standalone guide to optimizing AIE /
Peano-compiled kernels: a measure-first methodology (baseline, bit-exact gate,
ablation, verify-in-.o) plus a priority-ordered catalog of concrete
optimization levers (loop hints, compile-time constants, divide elimination,
branch-splitting, epilogue vectorization, operand-layout pre-pack, wide
packing, wider mmul, DMA layout offload) and the constraint each one carries.
Mechanisms are written generically across AIE generations; performance deltas
are framed as illustrative optimization headroom on example kernels.

Co-Authored-By: Claude Opus 4 (1M context) <noreply@anthropic.com>
@hunhoffe hunhoffe marked this pull request as ready for review June 11, 2026 16:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant