skills.rest
xchang1121
xchang1121COMMUNITY
Explore all skills by this author →
GitHub

triton-ascend-matmul

Optimize Triton matrix-multiplication kernels on Ascend hardware.

6|1|Updated Apr 19, 2026
Search Tags:#optimization#kernel#tiling#triton#ascend#swizzle#matmul
One-click install
npx skills add https://github.com/xchang1121/AutoResearch-CC-hook --skill triton-ascend-matmul
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: triton-ascend-matmul
Source: https://github.com/xchang1121/AutoResearch-CC-hook/tree/main/skills/triton-ascend/guides/triton-ascend-matmul
Command: npx skills add https://github.com/xchang1121/AutoResearch-CC-hook --skill triton-ascend-matmul

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Provides a structured guide to optimize matrix-multiplication kernels on Ascend hardware using Triton, enabling higher throughput and lower latency for dense compute workloads.

Core Features & Use Cases

  • Targeted optimization for matmul-like operators: matmul, mm, bmm, linear, gemm, outer_product, einsum (including matrix multiply), and conv converted to matmul.
  • Techniques include Cube Core utilization, tiling strategies, Swizzle optimization, and support for high-dimensional tensor shapes.
  • Use Case: Apply when core compute is dominated by large matrix multiplications and you need to maximize performance on Ascend with Triton.

Quick Start

Apply this guide to optimize a matmul kernel on Ascend using Triton and validate performance on a representative workload.

Frequently Asked Questions about triton-ascend-matmul

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I optimize matmul kernels on Ascend hardware with Triton?▼

Optimize matmul kernels on Ascend by applying Triton tiling strategies, Cube Core utilization, and swizzle optimization to maximize throughput and reduce latency for dense compute workloads.

What is the best way to use Cube Core for matrix multiplication in Triton?▼

Use Cube Core for matrix multiplication in Triton by selecting hardware-specific core counts and tile sizes to maximize compute density and throughput on Ascend devices.

Can I optimize high-dimensional tensor shapes like einsum and bmm using this Triton guide?▼

Yes, you can optimize high-dimensional tensor shapes like einsum and bmm using this Triton guide, which provides tiling and swizzle techniques tailored for higher-dim matrix multiplication.

Does Triton support swizzle optimization for conv converted to matmul on Ascend?▼

Yes, Triton supports swizzle optimization for conv converted to matmul on Ascend, enabling efficient memory access patterns and improved latency for convolutional compute workloads.

Why does my Triton matmul kernel have high latency on Ascend?▼

High latency in Triton matmul kernels on Ascend often results from suboptimal tiling strategies or underutilized Cube Cores, requiring hardware-specific tile size and backend configuration adjustments.

Related Skills

Explore More Software Engineering Skills→
mindspore-aimindspore-ai
258

triton-ascend-example-matmul

Demonstrate Triton Ascend matmul with 2D tiling and K-loop accumulation.

Official
Advanced
xchang1121xchang1121
6

triton-ascend-performance-improvement

Apply optimization patterns to Triton Ascend kernels for improved performance.

Community
Advanced
xchang1121xchang1121
6

triton-ascend-example-matmul

Implement a Triton Ascend matmul example with 2D tiling and K-loop accumulation.

Community
Advanced
AscendAscend
31

triton-operator-design

Generate Triton operator design documents for Ascend NPU.

Official
Advanced