How rocm/rocprofiler-systems works
Main Components
a. Core Components:
• Instrumentation Engine (Binary/Runtime)
• Sampling Engine
• Process Monitor
• Data Collection System
• Visualization Pipeline
• Python Integration Layer
• Configuration Management
b. External Interfaces:
• ROCm Platform Integration
• Hardware Counter Interface
• MPI Support
• GPU Metrics Collection
• CPU Metrics Collection
Architectural Layers
User Interface Layer:
• Command-line tools (rocprof-sys-*)
• Python API
• Configuration Interface
Core Processing Layer:
• Binary Instrumentation Engine
• Sampling Engine
• Process Monitor
• Data Collection Controllers
Integration Layer:
• ROCm Integration
• Hardware Counter Interface
• MPI Support
• Threading Support (Pthreads, OpenMP)
Data Management Layer:
• Perfetto Integration
• Timemory Integration
• Trace Collection
• Profile Storage
Hardware Interface Layer:
• CPU Metrics Collection
• GPU Metrics Collection
• System Metrics Collection
Key Design Elements
• Blue: User interface components
• Green: Core processing components
• Yellow: Integration components
• Orange: Data management components
• Red: Hardware interface components
b. Connection Types:
• Double lines: API boundaries
Specific Implementation Details
• Binary instrumentation flow (both runtime and rewrite modes)
• Sampling architecture with frequency control
• Process monitoring system
• Data collection pipeline
• Integration points with ROCm
• Python profiling mechanism
• Trace visualization pipeline
Component Relationships
Clear relationships between:
• Configuration system and all components
• Data collection and storage systems
• Hardware interfaces and metric collection
• User interfaces and core processing
• Integration layers and external systems