triton-operator-performance-eval

Name: triton-operator-performance-eval
Availability: InStock
Author: Ascend

Diagnose Triton kernel performance bottlenecks on Ascend NPUs using msprof data.

31|1|Updated Mar 28, 2026

Search Tags:#performance #profiling #kernel #triton #ascend #npu #msprof

Ready to run

npx skills add https://github.com/Ascend/agent-skills --skill triton-operator-performance-eval

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: triton-operator-performance-eval
Source: https://github.com/Ascend/agent-skills/tree/main/skills/triton-operator-performance-eval
Command: npx skills add https://github.com/Ascend/agent-skills --skill triton-operator-performance-eval

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

通过使用 msprof 数据帮助开发者识别和诊断 Ascend NPU 上 Triton 核函数的性能瓶颈，从而实现有效优化。

Core Features & Use Cases

基于 msprof/ msprof op 的函数级与算子级性能采集，提供全面的硬件利用情况分析。
瓶颈诊断与优化建议，将 Memory-Bound/Compute-Bound 等分类输出，辅助定位改进方向。
适用场景广泛，可用于不同输入形状、精度和 Triton 内核实现的性能对比与调优。

Quick Start

Run msprof on your Triton kernel to collect function-level and operator-level performance data for Ascend NPUs.

Frequently Asked Questions about triton-operator-performance-eval

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

FAQPage Schema

How do I diagnose Triton kernel performance bottlenecks on Ascend NPUs?▼

Diagnose Triton kernel performance bottlenecks on Ascend NPUs by analyzing msprof data to evaluate function-level and operator-level metrics, classifying issues as Memory-Bound or Compute-Bound, and generating optimization suggestions.

What msprof data is needed for Ascend NPU Triton kernel profiling?▼

Ascend NPU Triton kernel profiling requires msprof tooling data covering function-level and operator-level metrics, alongside access to specific kernel names for msprof op commands to accurately evaluate hardware utilization across diverse shapes and precisions.

Can I evaluate Triton kernel performance across different input shapes and precisions?▼

Evaluate Triton kernel performance across different input shapes and precisions by applying msprof profiling data to identify memory usage and compute efficiency variations, enabling direct performance comparison and targeted optimization for diverse workload configurations.

How does msprof op profiling classify Triton kernel bottlenecks?▼

msprof op profiling classifies Triton kernel bottlenecks by analyzing operator-level metrics to determine if performance limitations are Memory-Bound or Compute-Bound, outputting structured metrics that directly guide memory usage and compute efficiency optimization strategies.

What limitations exist when profiling Triton kernels on Ascend NPUs?▼

Profiling Triton kernels on Ascend NPUs requires prerequisite msprof tooling and explicit access to kernel names for msprof op execution, meaning unregistered or unsupported custom kernels cannot be automatically profiled without providing exact naming identifiers.