Algorithm Automatic Tuning Method for Heterogeneous Computer Architecture

Abstract

The invention provides an automatic algorithm tuning method for a heterogeneous computer system structure. A plurality of versions of an algorithm are stored in the heterogeneous computer system structure, the method comprises the steps that step 1), for a to-be-executed algorithm, an optimal algorithm version is determined according to the efficiency of the algorithm versions in the algorithm versions which are applicative for nodes and can be obtained; step 2), the optimal algorithm version is executed. The automatic algorithm tuning method for the heterogeneous computer system structure combines computer system structure features and data features participating in calculation to perform automatic tuning on the algorithm, user intervention is not needed, a user does not need to know eachversion of the algorithm, and reduces the threshold for the user to use a heterogeneous cluster.

Description

technical field [0001] The invention relates to the field of computer architecture, in particular to an algorithm automatic tuning technology in a heterogeneous environment. Background technique [0002] Over the past few decades, computer architecture has gone through the CPU-dominated phase and the many-core processor-dominated phase, and now it has entered the era of heterogeneous clusters. The architecture of each computing node (hereinafter referred to as node) in a heterogeneous cluster may be different. For example, some nodes contain Intel MIC, some nodes contain GPU, and some nodes support different SIMD instruction sets. Computing capabilities vary by architecture. [0003] The emergence of complex computer architectures, especially heterogeneous clusters, has brought new challenges to algorithm optimization and transplantation. At present, most algorithm libraries or toolkits do not provide different optimized algorithm versions for different computer architectur...

Images