HOPPERFISH: Holistic Profiling with Portable Extensible and Robust Framework Intended for Systems with Heterogeneity

HOPPERFISH (Holistic Profiling with Portable Extensible and Robust Framework Intended for Systems with Heterogeneity), a holistic profiling framework that unifies analysis across the application, runtime, microarchitecture, and hardware layers to streamline robust feature correlation in heterogeneous computing systems.

HOPPERFISH Workflow

Detailed Overview

HOPPERFISH is an open-source cross-layer profiling framework for heterogeneous SoCs, enabling synchronized monitoring across applications, runtime, microarchitecture, and hardware. It delivers lightweight, accurate, and portable profiling for real-time anomaly detection and machine learning applications. Equipped with capabilities for real-time monitoring and cross-layer analysis, HOPPERFISH offers:

Cross-layer profiling integrated into CEDR runtime, unifying application, runtime, hardware, and microarchitecture features.
Thread-safe, timer-triggered sampling for online monitoring without halting execution.
Portable feature collection pipeline across CPUs, GPUs, and FPGAs.
Modular and extensible design supporting platform-specific events while staying hardware-agnostic.
Runtime instrumentation to expose task queues, PE utilization, and low-overhead sampling not available in vendor profilers.