Auto-tuning full applications: A case study

Abstract

In this paper, we take a concrete step towards materializing our long-term goal of providing a fully automatic end-to-end tuning infrastructure for arbitrary program components and full applications. We describe a general-purpose offline auto-tuning framework and apply it to an application benchmark, SMG2000, a semi-coarsening multigrid on structured grids. We show that the proposed system first extracts computationally-intensive loop nests into separate executable functions, a code transformation called outlining. The outlined loop nests are then tuned by the framework and subsequently integrated back into the application. Each loop nest is optimized through a series of composable code transformations, with the transformations parameterized by unbound optimization parameters that are bound during the tuning process. The values for these parameters are selected using a search-based auto-tuner, which performs a parallel heuristic search for the best-performing optimized variants of the outlined loop nests. We show that our system pinpoints a code variant that performs 2.37 times faster than the original loop nest. When the full application is run using the code variant found by the system, the application's performance improves by 27%.

Identifying computational phases from inter-process communication traces of HPC applications

“…For example, Process 7 will send to and receive from processes 2, 3,4,6,8,10,11,12,18,19,20,21,22,23,24,26,27, and 28.…”

“…At a high-level, SMG2000 performs three distinct phases to solve the problem as reported in [24]. These phases are Initialization, Setup and Solve.…”

Automatic Tuning of the Fast Multipole Method Based on Integrated Performance Prediction

“…Compiler-based approaches are mostly independent from specific applications. In [18], whole applications are tuned by the compiler by extracting the most expensive loops from the source code and applying transformations such as tiling, unrolling, or permutation. The approach presented in [19] uses compiled binaries of applications to derive models of their execution times.…”

SEEDS: a software engineer's energy-optimization decision support framework

“…In contrast to common compiler optimizations, autotuning approaches often take into account details about the specific application being optimized and the environment where it will execute. Such approaches have been applied to specific types of software (e.g., computer algebra libraries [51] and high performance computing [39,49] as well as for general purpose languages and platforms (e.g., [45]). …”