Inspur Suzhou Intelligent Technology Co., Ltd

Yuanli Wang

Guoqiang Mei

Jiangwei Wang

A graph data processing method includes: acquiring target graph data to be processed; compiling statistics on the target graph data according to a first preset rule, so as to divide the target graph data into a plurality of graph data blocks and determine a boundary value and weight of each of the plurality of graph data blocks; and storing the boundary value and weight of each of the plurality of graph data blocks in a corresponding memory according to a second preset rule, so as to schedule the target graph data during a graph calculation process by use of the boundary values and the weights.

Graph data processing method, apparatus, and device, and medium

A field programmable gate array (FPGA) dynamic reconfiguration method, apparatus, device and readable storage medium are provided. The technical solution includes: performing board support package (BSP) flat compilation on a target project to obtain a static region; performing BSP generation and reconfiguration information compilation on the target project to obtain static information; revising the static region using the static information to obtain reconfiguration compilation version projects that meet timing and correspond to different reconfiguration compilation parameters, respectively; importing a preset heterogeneous acceleration kernel to the reconfiguration compilation version projects and then performing static compilation to obtain clock frequencies corresponding to the reconfiguration compilation version projects, respectively; and determining a target reconfiguration compilation version project with a clock frequency meeting performance requirements using the clock frequencies, and obtaining a dynamic reconfiguration compilation version project file. The dynamic reconfiguration compilation version project file obtained in this technical solution ensures that the static region can meet the timing, and also enables an operating clock of the heterogeneous acceleration kernel to meet the performance requirements for heterogeneous acceleration.

Growing community of inventors

Jiangsu, China

Yuanli Wang