Machine Learning Enabled Scalable Performance Prediction of Scientific Codes

TL;DR

This paper introduces PPT-AMMP, a machine learning-based tool that predicts the runtime of scientific codes on various hardware platforms by analyzing code structure and hardware parameters, validated on physics benchmarks.

Contribution

The paper presents a novel performance prediction toolkit combining code analysis, memory modeling, and machine learning, enabling accurate runtime predictions for scientific applications.

Findings

Accurately predicts runtime of physics benchmarks.

Identifies hardware bottlenecks through sensitivity analysis.

Extends to performance prediction of SNAP application.

Abstract

We present the Analytical Memory Model with Pipelines (AMMP) of the Performance Prediction Toolkit (PPT). PPT-AMMP takes high-level source code and hardware architecture parameters as input, predicts runtime of that code on the target hardware platform, which is defined in the input parameters. PPT-AMMP transforms the code to an (architecture-independent) intermediate representation, then (i) analyzes the basic block structure of the code, (ii) processes architecture-independent virtual memory access patterns that it uses to build memory reuse distance distribution models for each basic block, (iii) runs detailed basic-block level simulations to determine hardware pipeline usage. PPT-AMMP uses machine learning and regression techniques to build the prediction models based on small instances of the input code, then integrates into a higher-order discrete-event simulation model of PPT running on Simian PDES engine. We validate PPT-AMMP on four standard computational physics benchmarks, finally present a use case of hardware parameter sensitivity analysis to identify bottleneck hardware resources on different code inputs. We further extend PPT-AMMP to predict the performance of scientific application (radiation transport), SNAP. We analyze the application of multi-variate regression models that accurately predict the reuse profiles and the basic block counts. The predicted runtimes of SNAP when compared to that of actual times are accurate.