2018 Fiscal Year Final Research Report
Parallel Computation Theory for Memory Machine Models and Next Generation GPGPU Architecture
Project/Area Number |
26280002
|
Research Category |
Grant-in-Aid for Scientific Research (B)
|
Allocation Type | Partial Multi-year Fund |
Section | 一般 |
Research Field |
Theory of informatics
|
Research Institution | Hiroshima University |
Principal Investigator |
Nakano Koji 広島大学, 工学研究科, 教授 (30281075)
|
Co-Investigator(Kenkyū-buntansha) |
高藤 大介 広島大学, 工学研究科, 助教 (00314732)
伊藤 靖朗 広島大学, 工学研究科, 准教授 (40397964)
|
Project Period (FY) |
2014-04-01 – 2019-03-31
|
Keywords | Parallel Algorithms / GPGPU / Memory Machine Models |
Outline of Final Research Achievements |
The GPU (Graphics Processing Unit) is a specialized circuit designed to accelerate computation for building and manipulating images. Latest GPUs are designed for general purpose computing and can perform computation in applications traditionally handled by the CPU. The main purpose of this research is to propose appropriate theoretical models for GPU computing, develop efficient parallel algorithms based on the theoretical models, and evaluate the performance. We have developed theoretical models, Discrete Memory Machine model, Unified Memory Machine model, and Hierarchical Memory Machine model which capture the essence of memory access to the shared memory and the global memory of the GPU. Based on these models, we have developed many efficient algorithms on the GPU. In particular, we have developed a new technique that we call SKSS (Single Kernel Soft Synchronization) technique. We have shown that this technique can be applied to accelerate dynamic programming algorithms on the GPU.
|
Free Research Field |
情報工学
|
Academic Significance and Societal Importance of the Research Achievements |
GPUの理論的モデルを提案することにより,理論研究者がGPU向けアルゴリズムを研究するためのベースを提供することができた.これまでは,並列アルゴリズムの理論研究者にとってGPU向け並列アルゴリズムの実装作業は困難であったが,これにより,GPU上での並列処理技法の研究が容易に行えるようになった.また,このモデルをベースに研究代表者らはGPUのいくつかの具体的なアルゴリズム手法,例えば,SKSS (Single Kernel Soft Synchronization)などを提案し,その有効性をGPUへの実装実験により実証することができた.
|