一种基于自适应负载分配的稀疏张量典范分解方法及系统
By preprocessing and uniformly distributing sparse tensors using an adaptive load distribution method, the problem of unbalanced load and data conflict in the sparse tensor canonical decomposition algorithm on GPU systems is solved, thus achieving efficient computation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2023-09-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing sparse tensor canonical decomposition algorithms suffer from low parallelism and uneven load distribution between threads on large-scale parallel systems such as GPUs, resulting in performance waste and data conflicts.
An adaptive load allocation method is adopted. By preprocessing the sparse tensor, the slices and fibers are evenly distributed. Based on the characteristics of the specific tensor and the computing platform, an appropriate load allocation strategy is selected, and combined with conflict resolution, the computational allocation is optimized.
It improves the computational efficiency of the sparse tensor canonical decomposition algorithm, makes full use of the computing platform performance, reduces data conflicts, and enhances parallelism and load balancing.
Smart Images
Figure CN117311971B_ABST