AMD AI Analyzer - VS Code Extension

Version: 1.9.20261001 Publisher: AMD License: SEE LICENSE IN LICENSE

A Visual Studio Code extension for analyzing AI models with AMD AI Analyzer. This extension provides comprehensive tools for visualizing, analyzing, and optimizing AI workloads for AMD hardware.

Overview

The AMD AI Analyzer extension integrates AI model analysis capabilities directly into Visual Studio Code. It enables developers to:

Analyze AI model performance and partitioning
Visualize computation graphs (original, optimized, and mapped)
Examine NPU operator insights and hardware timing
View detailed performance metrics and timeline analysis
Manage multiple analysis sessions from a unified navigator view

Features

🔍 Analysis Views

Folder Summary: Model, compiler, and build details for an analysis folder, with a browsable file list and built-in viewer for logs and reports
Partition Summary: Detailed breakdown of model operators and partitioning decisions
NPU Insights Summary: Hardware-specific operator analysis for Neural Processing Units
Performance Summary: Comprehensive performance metrics and bottleneck identification
Memory Map: L1, L2, L3, and PM load memory usage with layer details and spilling insights
Timeline View: Hardware execution timeline with timing information

📊 Graph Visualization

Original Graph: Visualize the unmodified AI model computation graph
Optimized Graph: View the graph after optimization passes
Mapped Graph: See how operations map to hardware accelerators
Partition Graph: Understand model partitioning across execution targets

🎯 Workspace Integration

Context Menu Integration: Right-click folders in the Explorer to analyze with AI Analyzer
Navigator Tree View: Dedicated sidebar for managing analysis sessions and reports
Multi-folder Support: Analyze multiple model folders simultaneously
Remote & SSH Support: Open and view reports seamlessly over SSH and remote development connections
Theme-aware Reports: Summary, graph, Memory Map, and timeline views follow your VS Code light or dark theme

⚙️ CustomOps Generation

Operator Selection: Select individual operators from the CPU Because table in Partition Summary and Partition Graph
Custom Operator Generation: Generate custom operators using Copilot AI assistance
Vitis AI Integration: Seamless integration with Vitis AI configuration for operator customization

🤖 Chat Agent Integration

Profiling Data in Chat: Make your profiling data available to chat agents, so you can ask questions about timing, operators, memory, and partitioning in plain language
Guided Setup: The extension offers to set this up the first time you open a folder, and the offer can be taken up later from the Command Palette
Uses Your Environment: Runs from the same Python environment you already configured — nothing extra to install or configure by hand

🛠️ Development Features

Output Logging: Comprehensive logs in dedicated output channel
Environment Configuration: Flexible Python environment setup
Auto Port Management: Automatic port allocation for analyzer backend (8000-8100)
Restart Server: Manually restart the AI Analyzer server from a clickable icon next to the status bar label

Requirements

System Requirements

VS Code Version: 1.102.0 or higher
Operating System: Linux, macOS, or Windows
Python Environment: Python 3.12 with AMD AI Analyzer package installed

Python Environment Setup

Ryzen AI

https://ryzenai.docs.amd.com/en/latest/inst.html

Ryzen AI Analyzer

https://ryzenai.docs.amd.com/en/latest/ai_analyzer.html

The extension requires a Python environment with the aianalyzer package installed. The environment should be accessible via:

Conda environment activation script
Virtual environment activation script
Direct Python executable path

Configuration

Extension Settings

This extension contributes the following settings:

`aianalyzer.activateScript`

e.g., /path/to/env/bin/activate

Type: string
Default: ``
Description: Path to the activation script for AI Analyzer Python environment

`aianalyzer.mcp.enabled`

Type: boolean
Default: false
Description: Makes AI Analyzer profiling data available to chat agents. Turned on for you when you accept the setup offer, and can be turned off again at any time.

First-Time Setup

Open VS Code Command Palette (Ctrl+Shift+P or Cmd+Shift+P)
Run command: AI Analyzer: Select Environment Folder
Browse to your Python environment folder containing the aianalyzer package
The extension will validate and configure the environment

Usage

Analyzing a Folder

Right-click any folder in the VS Code Explorer
Select "Open in AI Analyzer"
The analyzer will launch and populate the AI Analyzer view

Method 2: Navigator View

Click the AI Analyzer icon in the Activity Bar (left sidebar)
Click the "Open Folder" button (folder icon in view title)
Select a folder to analyze

Viewing Reports

Once analysis completes, reports appear in the AI Analyzer navigator tree:

Expand the folder node to see available reports
Click any report to open it in the editor
Multiple reports can be open simultaneously

Available Commands

Access these commands via Command Palette (Ctrl+Shift+P):

AI Analyzer: Open Folder - Manually select a folder to analyze
AI Analyzer: Open in AI Analyzer - Analyze a folder (also available by right-clicking a folder in the Explorer)
AI Analyzer: Open Folder Summary - Open a folder summary report (also available from the navigator tree)
AI Analyzer: Open Memory Map - Open a memory map report (also available from the navigator tree)
AI Analyzer: Show AI Analyzer View - Open the navigator sidebar
AI Analyzer: Show Logs - View extension output logs
AI Analyzer: Select Environment Folder - Configure Python environment
AI Analyzer: Set Up MCP Server - Make profiling data available to chat agents
AI Analyzer: Restart Server - Restart the AI Analyzer server (also available from the status bar)
AI Analyzer: Generate Custom Operator - Generate a custom operator for the operators selected in a graph

Report-specific commands (partition summary, graphs, timeline, memory map, and others) are registered for programmatic use and are normally opened by selecting a report in the navigator tree.

Navigator View Features

Refresh: Reload the tree view (refresh icon in title bar)
Close Folder: Remove a folder from analysis (close icon on folder nodes)
Active Report Tracking: Currently open report is highlighted in the tree

Generating CustomOps

Open the Partition Graph or Partition Summary report
Locate the CPU Because table in Partition Summary showing operators and their characteristics
Select an operator you want to customize
Use the custom operator generation feature to create an optimized implementation
Copilot AI will assist in generating CustomOps code based on operator requirements

Asking Chat Agents About Your Reports

The first time you open a folder, the extension offers to make your profiling data available to chat agents. You can also set this up at any time:

Open the Command Palette (Ctrl+Shift+P or Cmd+Shift+P)
Run command: AI Analyzer: Set Up MCP Server
Confirm when your editor asks whether to trust and start the server

Once it is running, you can ask an agent questions such as "open this profiling folder and show me the top five NPU operators by time", and it will read the data directly from your reports. Turn it off again by clearing the AI Analyzer: Mcp: Enabled setting.

Known Issues

Port 8000 must be available or within range 8000-8100
Large model analyses may take several minutes to complete
Webview state is not persisted between VS Code restarts

Release Notes

Version 1.9.20261001

Folder Summary fixes: Selecting a file opens it in the editor again, and spacing and colors are corrected, including when following a light or dark theme
Timeline accuracy: NPU timelines now report the correct inference number
Clock frequency display: The NPU clock is labelled only when you set it yourself; a frequency the server resolved is shown without a caption

Version 1.9.20260921

Folder Summary webview: The Summary entry of an analysis folder now opens in a dedicated view with model, compiler, and build details, a browsable file list, and a built-in viewer for logs and reports
Open Folder Summary command: New AI Analyzer: Open Folder Summary command opens the same report from the command palette

Version 1.9.20260822

Chat agent integration: Profiling data from your reports can now be queried by chat agents, so you can ask about timing, operators, memory, and partitioning in plain language
Guided setup: The extension offers to set this up the first time you open a folder, and the new AI Analyzer: Set Up MCP Server command is available whenever you want it later

Version 1.9.0

Simplified configuration: Removed the experimental internal-features workspace setting; available reports and analysis content are determined by the backend server
Report panel reliability: Improved behavior when opening, closing, and switching between multiple report panels

Version 1.8.20260810

Theme-aware reports: Partition, NPU, and performance summaries, graph, timeline, and Memory Map views now follow your VS Code light or dark theme
Memory Map webview: Memory Map reports open in a dedicated view with L1, L2, L3, and PM load analysis (separate from summary reports)
Memory Map reliability: Fixed Memory Map reports that previously opened as a blank panel

Version 1.8.0

Remote & SSH Support: Reports now load correctly when working over SSH or remote development connections
Restart Server: New clickable icon next to the status bar to manually restart the AI Analyzer server
Simplified Configuration: The Vitis AI config path is now detected automatically
Improved reliability of timeline navigation, including breadcrumbs and inference selection
Corrected timeline view layout and spacing
The navigator tree now refreshes reliably when opening or closing folders

Version 1.7.0

Enhanced logging and debugging capabilities
Improved environment configuration workflow

Contributing

This extension is developed by AMD. For issues, feedback, or contributions, please contact the development team.

License

See LICENSE file for details.

Enjoy analyzing your AI models with AMD AI Analyzer! 🚀

aianalyzer

AMD