Shared computing resource preemption method and device, user equipment and storage medium

Abstract

The invention provides a GPU multi-core processor shared computing resource preemption method and device, user equipment and a storage medium. The method comprises the steps that a management scheduling thread receives priorities of threads in a thread pool corresponding to service cores or threads from at least one service core or thread; the management scheduling thread judges whether the priorities of all the threads are the same or not; if so, a step of exiting the preemption processing flow is performed, otherwise, the step 3 is executed; the management scheduling thread acquires the lowest priority in the priorities corresponding to all the currently executed threads; the management scheduling thread judges whether the thread with the priority higher than the lowest priority does notacquire the computing resources or not; if not, the preemption processing flow quits; if yes, the step 5 is entered; according to the sequence of the priorities from low to high, the task of at leastone service core or thread corresponding to the lowest priority which is being executed is selected, and a suspension operation is initialized on the selected task.

Description

technical field [0001] The present invention relates to the technical field of graphic processors, in particular to a method and device for preempting shared computing resources of GPU multi-core processors, user equipment and storage media. Background technique [0002] The GPU adopts a streaming parallel computing mode. In order to increase the parallel processing efficiency of data and improve the full utilization of the underlying computing resources (SIMD / DIMD), a multi-core / multi-thread processor will be used to establish a data flow processing computing channel to perform computing tasks. command analysis, configure relevant information, and allow computing resources (SIMD / DIMD) to compute and process data. [0003] Each core / thread corresponds to a thread pool (such as figure 2 dotted line), there are multiple threads in the thread pool (one-to-one correspondence between the thread and the priority of the computing task being executed), each core / thread will select ...

AI Technical Summary

Problems solved by technology

Each core / thread executes the highest-priority computing task in its respective thread pool, and the high-priority tasks in the current thread pool will exhaust the computing resources (SIMD / DIMD), causing other core / thread pools to be higher than the current thread The priority task cannot be executed due to lack of computing resources

Images