LTTng-HAS: Bringing LTTng tracing to HAS-based GPU runtimes

摘要

In this paper,we propose LTTng-HAS, a set of tools that allowfor the collection of a single, unifiedsoftware graphics processing unit (GPU) trace in ROCr, a Heterogeneous System Architecture(HAS)-based API and runtime. HAS is a cross-vendor standard facilitating the programming ofheterogeneous systems that include CPUs, GPUs, and possibly other types of devices. Ouropen-source solution is generic and easily adaptable to diverse GPU runtimes or APIs. UsingLinux Trace Toolkit Next Generation (LTTng), a highly efficient Linux tracer, it collects differenttypes of events over multiple executions of an application and aims to gather all the datainto a single trace, offering an easy way to generate GPU-related traces. Our instrumentationis achieved simply by preloading libraries, without recompiling the target application, whichmakes it flexible and easy to use. The resulting traces, which include API call stack information,GPU hardware metrics, command queue, and compute kernel profiling, are well adapted forpostprocessing and further analysis. Our solution also includes tracing data from the Linux kerneland proposes views for Trace Compass, an interactive trace visualizer.