Processing device and method for managing tasks thereof
A neural processing device and a method for managing tasks thereof are provided. The neural processing device includes a neural core configured to perform a task and generate a completion signal for completion of the task, a core global configured to transfer task information for the task to the neural core and receive the completion signal of the task from the neural core, and a task manager configured to generate and transmit the task information to the core global, receive the completion signal from the core global, generate a completion report, and transmit the completion report.
1 . A neural processor comprising:
a core global configured to generate task information corresponding to a task descriptor;
a neural core configured to receive the task information from the core global, perform a task according to the task information, and generate a completion signal indicating completion of the task; and
a task manager configured to monitor an execution time of the task performed by the neural core, generate a timeout detection signal based on the execution time, and generate a timeout report based on the timeout detection signal, wherein
the core global is further configured to receive, from the neural core, the completion signal of the task,
the task manager is further configured to receive the completion signal from the core global and generate a completion report, and
the timeout report is included in the completion report and is transmitted to a command processor.
2 . The neural processor according to claim 1 , wherein
the task comprises a first task for which the completion report is configured to be generated and a second task for which the completion report is configured not to be generated, and
the task manager is further configured to check in a first task descriptor corresponding to the first task into a first check-in buffer and check in a second task descriptor corresponding to the second task into a second check-in buffer.
3 . The neural processor according to claim 2 , wherein
upon generation of a timeout report for the second task, the task manager is further configured to modify the second task descriptor such that the completion report is generated.
4 . The neural processor according to claim 1 , wherein
the task manager is further configured to determine whether to generate the timeout detection signal based on a comparison between the execution time and a threshold time.
5 . The neural processor according to claim 4 , wherein
the threshold time is individually set according to the task.
6 . A neural processor comprising:
a core global configured to generate task information corresponding to a task descriptor;
a neural core configured to receive the task information from the core global, perform a task according to the task information, and generate a completion signal indicating completion of the task; and
a task manager configured to monitor an execution time of the task performed by the neural core, generate a timeout detection signal based on the execution time, and generate a timeout report based on the timeout detection signal, wherein
the task manager comprises:
a task passage configured to generate the task descriptor for the task and transmit the task information corresponding to the task descriptor to the core global; and
a done passage configured to check in the task descriptor from the task passage and generate a completion report based on the completion signal received from the core global.
7 . The neural processor according to claim 6 , wherein
the done passage comprises:
a check-in buffer configured to receive and check in the task descriptor, receive the completion signal, and associate the task descriptor with the completion signal;
a dependency setter configured to check out the task descriptor based on the completion signal and generate a check-out report corresponding to a check-out of the task descriptor;
a timeout monitor configured to generate the timeout detection signal and generate the timeout report based on the timeout detection signal; and
a reporting managing module configured to receive the check-out report, generate the completion report, and transmit the completion report to a command processor.
8 . The neural processor according to claim 7 , wherein
the task passage is further configured to transmit the task descriptor to a queue, generate a transfer-done report, and transmit the transfer-done report to the reporting managing module, and
the transfer-done report is included in the completion report and is transmitted to the command processor.
9 . The neural processor according to claim 8 , wherein
the done passage further comprises an event monitor configured to generate an event report based on an event detection signal.
10 . The neural processor according to claim 9 , wherein
the reporting managing module comprises:
a reporting manager configured to generate the completion report;
an event report queue configured to receive the event report and transmit the received event report to the reporting manager;
a timeout report queue configured to receive the timeout report and transmit the received timeout report to the reporting manager;
a check-out report queue configured to receive the check-out report and transmit the received check-out report to the reporting manager; and
a transfer report queue configured to receive the transfer-done report and transmit the received transfer-done report to the reporting manager.
11 . The neural processor according to claim 10 , wherein
the reporting manager is further configured to monitor at least one of a status of the timeout report queue, a status of the check-out report queue, and a status of the event report queue.
12 . A method for monitoring a task, performed by a neural processor, the method comprising:
by a core global of the neural processor, generating task information corresponding to a task descriptor and providing the task information to a neural core of the neural processor;
by a task manager of the neural processor, monitoring an execution time of the task performed by the neural core and generating a timeout detection signal based on the execution time;
by the task manager, generating a timeout report based on the timeout detection signal;
by the core global, receiving, from the neural core, a completion signal indicating completion of the task;
by the task manager, receiving the completion signal from the core global and generating a completion report; and
by the task manager, transmitting the completion report including the timeout report to a command processor of the neural processor.
13 . The method according to claim 12 , further comprising generating, by the task manager, a check-out report based on the completion signal for the task received from the neural core, wherein
the generating the completion report comprises:
generating, by the task manager, the completion report based on the check-out report and the timeout report.
14 . The method according to claim 13 , wherein
the task comprises a first task for which the completion report is configured to be generated and a second task for which the completion report is configured not to be generated, and
the generating the task information corresponding to the task descriptor and the providing the task information to the neural core comprises:
checking in a first task descriptor corresponding to the first task into a first check-in buffer and checking in a second task descriptor corresponding to the second task into a second check-in buffer.
15 . The method according to claim 14 , wherein
the generating the completion report comprises modifying the second task descriptor such that a completion report is generated upon generation of a timeout report for the second task.
16 . The method according to claim 13 , wherein the generating the timeout detection signal comprises determining whether to generate the timeout detection signal based on a comparison between the execution time and a threshold time.
17 . The method according to claim 16 , wherein
whether to generate the timeout report and a threshold value for the execution time are individually set according to the task.
18 . The method according to claim 13 , further comprising generating, by the task manager, an event report based on an event detection signal.
19 . The method according to claim 18 , wherein
the generating the completion report comprises:
by the task manager, storing the timeout report in a timeout report queue, storing the check-out report in a check-out report queue, and storing an event report, provided from an event detection device, in an event report queue, and
by the task manager, monitoring at least one of a status of the timeout report queue, a status of the check-out report queue, and a status of the event report queue.