skills.rest
FeRhodium
FeRhodiumCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

computation-analysis

Identify computation-intensive operators and assess CANN support for Ascend NPU performance.

7|Updated Jan 29, 2026
Search Tags:#optimization#analysis#profiling#bottlenecks#Ascend NPU#CANN
One-click install
npx skills add https://github.com/FeRhodium/ascend-migration --skill computation-analysis
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: computation-analysis
Source: https://github.com/FeRhodium/ascend-migration/tree/main/skills/computation-analysis
Command: npx skills add https://github.com/FeRhodium/ascend-migration --skill computation-analysis

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps identify computation-intensive operators and assesses CANN support to optimize Ascend NPU performance.

Core Features & Use Cases

  • Identify computation-heavy operators and their locations within model code (e.g., matmul, convolution, attention).
  • Assess CANN operator library support status and identify CPU fallback risks.
  • Propose practical optimization opportunities with torch_npu, such as AMP usage, operator fusion, and data layout improvements.
  • Provide a structured profiling approach and actionable guidance for performance validation.
  • Use Case: analyze a PyTorch model to locate bottlenecks and generate a profiling plan for Ascend NPU optimization.

Quick Start

Use the computation-analysis skill to identify computation-heavy operators in your Ascend NPU model and generate a profiling plan.

Frequently Asked Questions about computation-analysis

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

FAQPage Schema
How do I identify computation-intensive operators causing Ascend NPU bottlenecks?▼

To identify Ascend NPU bottlenecks, analyze your model source code to locate computation-intensive operators like matmul, convolution, and attention based on their execution weight.

What is the risk of CANN operator library fallback to CPU on Ascend NPU?▼

CANN fallback to CPU occurs when the operator library lacks support for specific model operations, degrading overall Ascend NPU performance by shifting heavy computation off the accelerator.

How do I optimize torch_npu performance using AMP and operator fusion?▼

Optimize torch_npu performance by applying Automatic Mixed Precision (AMP), leveraging operator fusion opportunities, and improving data layout to reduce computation overhead on the Ascend NPU.

What is the best way to profile Ascend NPU performance for a PyTorch model?▼

The best way to profile Ascend NPU performance is to generate a structured profiling plan that validates optimization steps and maps heavy operators within your PyTorch model codebase.

Does the computation-analysis skill work without CANN installed?▼

Analyzing CANN operator library support status is a core function, so assessing CPU fallback risks and optimization opportunities requires visibility into your CANN environment configuration.

Related Skills

Explore More Software Engineering Skills→
mindspore-aimindspore-ai
7

dist-op-analysis

Extract interface signatures, Primitive/ATen mappings, and sharding strategies for MindSpore mint and PyTorch ops.

Official
Advanced
FeRhodiumFeRhodium
7

training-migration

Migrate PyTorch training code to Ascend NPU with AMP and HCCL.

Community
Advanced
Happyverse-TeamHappyverse-Team

evaluate-presets

Run and benchmark Ralph's hat collection presets across backends.

Official
Intermediate
AscendAscend
31

simple-vector-triton-gpu-to-npu

Migrate Triton GPU kernels to Ascend NPU with 1D grid mapping.

Official
Advanced