Large-scale data parallel computation method with many-core structure

Abstract

The invention provides a computation method which utilizes a parallel processing chip with an isomorphic many-core to carry out large-scale data parallel processing, and the method utilizes a parallel processor with an isomorphic many-core structure to carry out parallel computation processing for large-scale data without logical relationship. Modules in an application sequentially run on an isomorphic many-core microprocessor, wherein each module is operated repeatedly, processes sub-modules with different data every time and is separated into a function which a plurality of threads executing on a many-core. The threads which correspond to each function are organized into a thread group according to a certain rule; the thread group is batch processing for the threads derived from the function; all the threads which are derived from the same function can access a global shared memory space on the many-core microprocessor; the accessing of each thread to a thread group shared memory space and the global shared memory space comprises reading data on the thread group memory space and the global shared memory space and writing an operation result in; and the threads which correspond to each function process different data simultaneously so as to achieve high parallelization.

Description

technical field [0001] The present invention relates to a method for parallel computing of large-scale data on a many-core processor architecture in terms of high-performance computing, and more specifically, relates to a method for using a device with a many-core processor architecture to perform large-scale parallel computing and processing of data without logical relations . Background technique [0002] With the rapid progress of semiconductor process technology and the continuous development of architecture, coupled with the commercial and scientific requirements for higher computing performance, processor functions, microprocessor architectures, and multi-core and many-core design technologies continue to innovate. The many-core microprocessor contains several processors inside, which provides more powerful processing capability and higher parallelism than a single core. [0003] In many-core microprocessors, processor cores often use shared storage to exchange shared...

AI Technical Summary

Problems solved by technology

[0004] Because of the above characteristics, when a many-core processor performs large-scale data processing, since each processor core in it has the same priority in accessing the shared storage space, the possibility of memory access conflicts is high, which will lead to memory access problems. Increased delay; and because of the unified addressing of the storage space, when the parallel compiler designed according to the shared storage method is not used, the compiled program may have address conflicts

[0005] In the field of intensive scientific computing, the most prominent features are the large amount of data and the strong repeatability of computing operations. The parallel computing method implemented on the many-core processor of the general architecture is not high in parallelism, and the reading and writing of data often lags behind. Therefore, it is urgent to find a parallel computing method on many-core processors that can reduce data transmission delay and further increase the parallelism of data processing.

Images