The performance of multithreaded applications on multicore architectures
depends largely on locality and communication. However, most performance
analyses are architecture-dependent, and hence insights gleaned from an
application's behavior on one platform may not apply when the application is
run on another. In contrast, architecture-independent metrics allow a
program's performance to be analyzed across a range of architectures without
incurring the overhead of repeated profiling and analysis. We propose
multicore-aware reuse distance, which captures the inherent locality
properties of an application along with the impact of inter-thread data
interactions. We then show how statistical sampling and parallelization can
speed this analysis up by orders of magnitude with minimal loss of accuracy,
enabling the use of privatized O(1) data structures, reduced
synchronization, and sampling rates as low as one in a million.