![[PukiWiki]](image/pukiwiki.png "[PukiWiki]")
However, it has the performance limit. So other processors are proposed, such as GPU specialized image processing, ASIC optimized for application, and RC combined flexibility of software with the high performance of hardware by processing. So in RC, improvement of flexibility and performance can be expected.
Accordingly, Adaptive Processors (APs) are proposed in RC. For characteristic, APs are separated instruction, GCFG (Global Configuration data) having the only data dependency and LCFG (Local Configuration data) having operation, immediate value and address. And they configure data path dynamically, and simple design flow. APs compiler generates Data Flow Graph (DFG) as configuration data. And GCFG and LCFG is mapped memory.
Resource of APs call Object space. Object space are configured computation objects, memory objects and data bus. APs configure data path from generated DFG, and the object can keep this configuration until calculation are finished.
However, huge continuous graph have not been performed, because APs resource are limited. So, DFG should be partitioned to apply the resource.
In previous work, due to use minimum partitioning with METIS (graph partitioning tool to solve k way cut problem), we can partition the graph to apply resource.
However, Number of graph partition have been increase in the case of exceed fixed number or small resource. So, Graph partitioning with METIS is not optimal to apply.
APs are required to change the constraint of METIS, or other algorithms.
This research improve graph partitioning method.
