GPU parallel compute resource allocation method and device

Abstract

The invention relates to the field of computer technology, and particularly provideds a GPU parallel compute resource allocation method and apparatus, The methods include: a computing task determinedby a computing process is obtained, Aa layered directed acyclic graph model is obtained by adopting a two-layer parallel computing model to process the computational task under the initial parameters,a layered directional acyclic graph model is used to compute the computational task by using a two-layer parallel computational model, according to the number of a plurality of preset parameters andthe number of calculation elements corresponding to each preset parameter, the thread block is labeled according to the preset parameters and the calculation elements corresponding to each preset parameter, and the calculation elements corresponding to each preset parameter are allocated according to the labels of each thread block, so that each thread block calculates the allocated calculation elements. By using the above method, the parallel computing efficiency can be effectively improved.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a GPU parallel computing resource configuration method and device. Background technique [0002] With the rapid development of computer technology, the traditional CPU design process has gradually reached the physical limit, and the growth rate of computing power has lagged far behind "Moore's Law". The improvement of computing power gradually turns to rely on new parallel computing technologies such as multi-core and many-core. In recent years, GPU, as an advanced many-core heterogeneous parallel computing device, has been widely used to accelerate large-scale computing-intensive tasks such as climate simulation, protein folding, and deep learning. Taking the GPU supporting the NVIDIA CUDA architecture as an example, when processing computing tasks, the computing tasks are organized into one or more Kernel operating system kernels containing a large number of threads....

AI Technical Summary

Problems solved by technology

[0003] The inventor found through research that for massive computing tasks, Kernel stream parallelism is only coarse-grained parallelism at the task level, and the parallelism between different Kernels is limited, and the actual acceleration effect is poor

In particular, for computing tasks with a high degree of process serialization (such as sparse matrix factorization, sparse triangular equation solving and other sparse algorithms based on directed graphs), the parallelism inside the above-mentioned Kernel is very low, while the parallelism between Kernels is still low. Limited, there are a lot of vacant computing resources in actual computing, making the actual parallel efficiency very low

Images