Task preemption method in deep learning accelerator system and related system
CN119440748APending Publication Date: 2025-02-14MEDIATEK INC
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202411056587.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-08-03
- Filing Date
- 2024-08-02
- Publication Date
- 2025-02-14
AI Technical Summary
Technical Problem
The task preemption mechanism in existing deep learning accelerator (DLA) systems is extremely inefficient, resulting in insufficient resource utilization when sharing the same DLA hardware between multiple deep learning tasks.
Method used
By implementing a task preemption method on DLA hardware, it includes responding to the stop command of the DLA driver when executing the first task, pausing the task after completing the current operation and sending an interrupt request, receiving the second task and executing the second before restoring the first task Task.
Benefits of technology
It improves the resource utilization rate of DLA hardware, reduces task waiting time, and improves the sharing efficiency between multiple tasks.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN119440748A_ABST
Abstract
Deep learning accelerator (DLA) hardware performs task preemption. The DLA hardware executes a first task, the first task being a task using a multi-layer neural network. In response to a stop command from the DLA driver to stop executing the first task, the DLA hardware completes a current operation of the neural network and sends an interrupt request (IRQ) to the DLA driver. The DLA hardware then receives a second task from the DLA driver. The second task is executed before the first task is recovered to continue to be executed.
Need to check novelty before this filing date? Find Prior Art