Fast transfer of workload between multiple processors

• US 10,083,068 B2
• Filed: 06/22/2016
• Issued: 09/25/2018
• Est. Priority Date: 03/29/2016
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Kagan, Oleg
• Primary Examiner(s): KIM, DONG U
• Application Number:
  US15/189,868
• Publication Number:
  US 20170286186A1
• Time in Patent Office: 825 Days
• Field of Search: None
• CPC Class Codes: G06F 2209/509 Offload
• CPC Class Codes: G06F 9/4812 by interrupt, e.g. masked
• CPC Class Codes: G06F 9/4843 by program, e.g. task dispatcher, supervisor, operating system
• CPC Class Codes: G06F 9/52 Program synchronisation; Mutual exclusion, e.g. by means of semaphores
• CPC Class Codes: G06F 9/524 Deadlock detection or avoidance

Abstract

Techniques and systems for prescheduling an alternative CPU as soon as a need for a task is detected by a primary CPU are disclosed. A process includes detecting, by a producer thread running on a first CPU, an external interrupt, acquiring, by the producer thread, a spinlock, and in response to acquiring the spinlock, sending, by the producer thread, an event to a consumer thread that is to run on a second CPU. Upon receiving the event by the consumer thread, the consumer thread acquires the spinlock, which “stalls” the consumer thread in a loop until the spinlock is released by the producer thread. While the consumer thread is “waking up” in response to receiving the event from the producer thread, the producer thread creates a task, publishes the task, and releases the spinlock, thereby causing the consumer thread to exit the loop, retrieve the task, and execute the task.

Claims

1. A method comprising:

detecting, by a producer thread running on a first processor, an external interrupt;

acquiring, by the producer thread, a spinlock;

in response to acquiring the spinlock, sending, by the producer thread, an event to the consumer thread;

while the consumer thread wakes up;

creating, by the producer thread, a task, and

publishing, by the producer thread, the created task;

in response to publishing the created task, releasing, by the producer thread, the spinlock;

in response to releasing the spinlock, retrieving, by a consumer thread running on a second processor, the published task; and

executing, by the consumer thread, the retrieved task.

2. The method of claim 1, further comprising:

prior to retrieving the task, receiving, by the consumer thread, the event; and

in response to receiving the event, acquiring, by the consumer thread, the spinlock.

3. The method of claim 2, wherein the consumer thread acquires the spinlock prior to the producer thread releasing the spinlock, and wherein acquiring, by the consumer thread, the spinlock comprises iteratively checking, by the consumer thread, a value of a spinlock parameter associated with the spinlock.

4. The method of claim 2, wherein the consumer thread acquires the spinlock after the producer thread has already released the spinlock.

5. The method of claim 1, wherein:

the consumer thread is one of multiple consumer threads in a pool of consumer threads;

the producer thread is one of multiple producer threads in a pool of producer threads;

the task is one of multiple tasks; and

the event comprises a semaphore having a count equal to a number of the multiple consumer threads in the pool of consumer threads.

6. The method of claim 1, wherein publishing the task comprises storing information to be utilized for executing the task in shared memory that is accessible to the second processor via the consumer thread.

7. The method of claim 1, further comprising performing, by the consumer thread, loading operations while the producer thread creates the task and publishes the task, wherein the loading operations:

prepare the consumer thread to execute the task; and

are performed based at least in part on receipt of the event by the consumer thread.

8. A system comprising:

multiple processors including a first processor and a second processor; and

memory storing computer-executable instructions that, when executed by at least one processor of the multiple processors, cause the at least one processor to perform operations comprising;

executing, by the second processor, an instruction causing the second processor to wait for an event;

executing a producer thread on the first processor based at least in part on an interrupt detected by the first processor;

when the interrupt is detected, setting, by the producer thread, a parameter associated with a synchronization primitive to a first value;

in response to setting the parameter to the first value, sending, by the producer thread, the event to the second processor;

executing a consumer thread on the second processor based at least in part on the second processor receiving the event;

performing, by the consumer thread, loading operations;

while the consumer thread performs the loading operations;

creating, by the producer thread, a task,

publishing, by the producer thread, the task for consumption, and

in response to publishing the task for consumption, setting, by the producer thread, the parameter to a second value;

in response to setting the parameter to the second value, retrieving, by the consumer thread, the task; and

executing, by the consumer thread, the task.

9. The system of claim 8, the operations further comprising:

iteratively checking, by the consumer thread, the parameter associated with the spinlock to determine whether the parameter has been set to the second value; and

in response to the producer thread setting the parameter to the second value, acquiring, by the consumer thread, the spinlock.

10. The system of claim 8, wherein:

the consumer thread is one of multiple consumer threads in a pool of consumer threads;

the producer thread is one of multiple producer threads in a pool of producer threads;

the task is one of multiple tasks; and

the event comprises a semaphore having a count equal to a number of the multiple consumer threads in the pool of consumer threads.

11. The system of claim 10, wherein the synchronization primitive comprises a spinlock, the operations further comprising, in response to the producer thread setting the parameter to the second value, acquiring, by the pool of consumer threads, the spinlock in a shared manner.

12. The system of claim 10, wherein:

the multiple processors include a third processor;

a first subset of the multiple consumer threads execute on the second processor; and

a second subset of the multiple consumer threads execute on the third processor.

20. The system of claim 8, wherein setting, by the producer thread, the parameter to the second value comprises the producer thread releasing the synchronization primitive, wherein retrieving, by the consumer thread, the task, is performed in response to the consumer thread acquiring the synchronization primitive, and wherein the producer thread has released the synchronization primitive before the consumer thread attempts to acquire the synchronization primitive.

13. A method comprising:

detecting, by a producer thread running on a first processor, an external interrupt;

setting, by the producer thread, in response to the external interrupt, a parameter associated with a spinlock to a first value;

in response to setting the parameter to the first value, sending, by the producer thread, an event to a consumer thread running on a second processor;

performing, by the consumer thread, loading operations;

while the consumer thread performs the loading operations;

creating, by the producer thread, a task,

publishing, by the producer thread, the created task, and

in response to publishing the created task, setting, by the producer thread, the parameter to a second value;

setting, by the consumer thread, the parameter to the first value;

retrieving, by the consumer thread, the published task; and

executing, by the consumer thread, the retrieved task.

14. The method of claim 13, wherein the consumer thread performs the loading operations in response to receiving the event from the producer thread.

15. The method of claim 13, further comprising iteratively checking, by the consumer thread and prior to the producer thread setting the parameter to the second value, the parameter to determine whether the parameter has been set to the second value, wherein the consumer thread sets the parameter to the first value in response to determining that the parameter has been set to the second value.

16. The method of claim 13, wherein:

the consumer thread is one of multiple consumer threads in a pool of consumer threads;

the producer thread is one of multiple producer threads in a pool of producer threads;

the task is one of multiple tasks; and

the event comprises a semaphore having a count equal to a number of the multiple consumer threads in the pool of consumer threads.

17. The method of claim 16, wherein:

a first subset of the multiple consumer threads execute on the second processor; and

a second subset of the multiple consumer threads execute on a third processor.

18. The method of claim 13, wherein publishing the task comprises storing information to be utilized for executing the task in shared memory that is accessible to the second processor via the consumer thread.

19. The method of claim 13, wherein:

setting the parameter to the first value acquires the spinlock; and

setting the parameter to the second value releases the spinlock.