Universal floating point matrix processor hardware structure based on FPGA (field programmable gate array)

Abstract

The invention discloses a universal floating point matrix processor hardware structure based on an FPGA (field programmable gate array). The universal floating point matrix processor hardware structure comprises a plurality of matrix operation accelerators, a plurality of shared matrix buffer memories, a main processor, an instruction dispatcher, a direct access controller, an arbiter, an external memory and a memory management unit, the shared matrix buffer memories provide universal memory spaces and support high-speed communication on chips between the accelerators, the main processor transmits instructions, dispatches operation and calculates addresses, the instruction dispatcher inspects whether the instructions have occupying conflict or not, the direct access controller finishes data transmission, the arbiter distributes data, the external memory provides buffer, and the memory management unit provides an access interface. The hardware structure can realize 'matrix-matrix' operation on a universal platform, software programs can be modified on the main processor according to different applications, a matrix operation accelerator structure can be conveniently inserted to support various matrix operations, the universal floating point matrix processor hardware structure has high design flexibility and has the advantages of lower energy consumption and higher calculated performance as compared with an embedded processor performing the matrix operations in the past.

Description

technical field [0001] The invention relates to the technical fields of computers and electronic information, in particular to a general floating-point matrix processor hardware structure based on FPGA (Field-Programmable Gate Array, Field Programmable Gate Array). Background technique [0002] In recent years, mobile working platforms such as drones and mobile robots have become one of the research hotspots. These devices have high flexibility and are widely used in disaster relief, geological survey and other scenarios. A large number of matrix operations are required in the application of the mobile work platform, which has high requirements on the speed and power consumption of matrix operations. For example, the Kalman filter is a method widely used in mobile robot localization, and its operation process includes a series of matrix operations. In order to ensure the real-time positioning of the robot, these matrix operations need to be performed at high speed; at the ...

AI Technical Summary

Problems solved by technology

Continuous status checking increases the workload of the main processor, while the gap between the main processor waiting for the accelerator to complete the calculation results in a waste of computing resources

Images