Event Processor

The Event Processor module serves as the intermediary stage that preprocesses raw event data collected by the Event Handler. It applies custom transformation and analysis routines using modular processor functions such as:

• AccelProf::tensor_info_process

• AccelProf::kernel_info_process

This design supports both CPU- and GPU-based data preprocessing, allowing:

• Normalization of event fields

• Filtering and aggregation of traces

• Device-side analysis, enabled through GPU offloading for high-throughput scenarios

GPU-based preprocessing is supported via patched APIs such as sanitizerPatchModule, allowing device-resident functions (marked with __device__) to process large volumes of fine-grained traces efficiently.