DYNAMIC WORKLOAD DISTRIBUTION APPARATUS FOR OPENCL

摘要

PURPOSE: An open computing language (OpenCL) dynamic work allocation device is provided to allocate work amount to each core in kernel start at a time and consider cache consistency in a process of collecting result values in kernel end, thereby guaranteeing the cache consistency of cores and efficiently allocating work. CONSTITUTION: A kernel performance start point determination unit (110) determines a performance start point of kernel which is composed of work groups with work items. A work allocation method determination unit (120) determines a work allocation method for the work items included in the work groups of the kernel. A work allocation unit (130) allocates work to sub-cores corresponding to the work allocation method. A work collection unit (140) collects work results of the sub-cores in kernel performance end and obtains a work result of maintaining cache consistency. [Reference numerals] (110) Kernel performance start point determination unit; (120) Work allocation method determination unit; (130) Work allocation unit; (140) Work collection unit; (AA) Sub-core