AdaptMemBench: Application-Specific MemorySubsystem Benchmarking

TL;DR

AdaptMemBench is a flexible benchmarking framework that emulates application-specific memory access patterns to evaluate and optimize memory subsystem performance for scientific applications.

Contribution

It introduces a configurable, kernel-independent framework for application-specific memory benchmarking using polyhedral code generation.

Findings

Effective in characterizing memory performance for various access patterns

Useful as a testbed for memory optimization strategies

Demonstrated with case studies on computational kernels

Abstract

Optimizing scientific applications to take full advan-tage of modern memory subsystems is a continual challenge forapplication and compiler developers. Factors beyond working setsize affect performance. A benchmark framework that exploresthe performance in an application-specific manner is essential tocharacterize memory performance and at the same time informmemory-efficient coding practices. We present AdaptMemBench,a configurable benchmark framework that measures achievedmemory performance by emulating application-specific accesspatterns with a set of kernel-independent driver templates. Thisframework can explore the performance characteristics of a widerange of access patterns and can be used as a testbed for potentialoptimizations due to the flexibility of polyhedral code generation.We demonstrate the effectiveness of AdaptMemBench with casestudies on commonly used computational kernels such as triadand multidimensional stencil patterns.