System and method for controlling central processing unit power based on inferred workload parallelism

摘要

A method of dynamically controlling power within a multicore CPU is disclosed and may include receiving a degree of parallelism in a workload of a zeroth core and determining whether the degree of parallelism in the workload of the zeroth core is equal to a first wake condition. Further, the method may include determining a time duration for which the first wake condition is met when the degree of parallelism in the workload of the zeroth core is equal to the first wake condition and determining whether the time duration is equal to a first confirm wake condition. The method may also include invoking an operating system to power up a first core when the time duration is equal to the first confirm wake condition.

主权项

1. A method of dynamically controlling power within a multicore central processing unit (CPU), the method comprising:
determining a number of operating system scheduler ready-to-run queues that are available for scheduling on the multicore CPU; calculating a degree of parallelism in a workload of a zeroth core of the multicore CPU, wherein the degree of parallelism comprises a number of ready-to-run threads in the available operating system scheduler ready-to-run queues; determining that the number of ready-to-run threads exceeds a first threshold value for a first time duration that exceeds a second threshold value, wherein the first threshold value is set for triggering a core wake and the second threshold value is set for confirming the core wake; calculating a system load value for the multicore CPU based on a run queue value of the zeroth core and an operating frequency of the zeroth core; and when the number of ready-to-run threads exceeds the first threshold value for the first time duration that exceeds the second threshold value and the system load value meets a predetermined condition, powering up a first core of the multicore CPU to execute at least one of the available operating system scheduler ready-to-run queues, wherein the system load value indicates a number of powered cores for operation of the multicore CPU.