Understanding Allocator Impact on Runtime Performance
Typical users rely on existing tools to understand the performance of their code. However, no tool is perfectly suited for all applications, hardware, or analysis. Moreover, constructing a strong case that explains the impact of a design choice on runtime performance occasionally requires abstracting the results of a performance comparison into easy-to-interpret performance metrics while eliding or minimizing unrelated factors.
This talk recounts the challenges encountered while attempting to analyze and delineate the performance impact of employing local memory allocators. Challenges included 1) choosing the correct tool, 2) working around the resolution and precision constraints of the instrumentation and profiling tools, and 3) designing effective presentation and visualization material.
Parsa Amini
Parsa Amini is interested in high-performance computing applications and high-productivity parallel C++ codes. He has been involved in developing the HPX runtime system and HPX applications. Parsa received his Ph.D. degree in Computer Science from Louisiana State University and has been investigating the costs of using assumptions in C++ programs since.