Reconfigurable processor and method

• US 9,043,805 B2
• Filed: 09/23/2010
• Issued: 05/26/2015
• Est. Priority Date: 10/23/2009
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kwon, Kwon Taek, Min, Kyoung June, Jung, Seok Yoon
• Primary Examiner(s): Truong, Camquy
• Application Number:
  US12/923,487
• Publication Number:
  US 20110099555A1
• Time in Patent Office: 1,706 Days
• US Class Current: 718/108
• CPC Class Codes: G06F 15/7875 for multiple contexts
• CPC Class Codes: G06F 9/461 Saving or restoring of program or task context
• CPC Class Codes: G06F 9/462 with multiple register sets

Abstract

Disclosed are a reconfigurable processor and processing method, a reconfiguration control apparatus and method, and a thread modeler and modeling method. A memory area of a reconfigurable processor may be divided into a plurality of areas, and a context enabling a thread process may be stored in respective divided areas, in advance. Accordingly, when a context switching is performed from one thread to another thread, the other thread may be executed by using information stored in an area corresponding to the other thread.

Claims

1. A reconfigurable processor comprising:

a plurality of memories divided into a plurality of different areas, the different divided areas corresponding to respective different threads from a plurality of threads and with sizes proportional to the respective different threads, with corresponding areas of the different divided areas storing context information of the respective different threads; and

a plurality of function units to perform a corresponding calculation based on received context information stored in an area, from the different divided areas, corresponding to a respective thread indicated to be executed,

wherein the plurality of memories include a plurality of local register files each divided into a plurality of local register file areas, where each local register file area of a local register file is respectively configured to collectively store information for different threads from a particular function unit configured to execute the different threads, with the particular function unit being paired with the local register file distinguished from a different particular function unit paired with a different local register file, and

wherein the plurality of memories receive identification information of the respective thread to be executed and make available to the plurality of function units the area, the area being mapped to the identification information.

2. The reconfigurable processor of claim 1, wherein the plurality of function units perform the calculation by using the context information stored in the area, as mapped to by the identification information, upon receiving the identification information of the respective thread to be executed.

3. The reconfigurable processor of claim 1, wherein the plurality of function units stop execution of the respective thread, and perform another calculation by using context information stored in a different area, from different divided areas of a corresponding paired local register file, being mapped to by identification information of a different thread, when receiving the identification information of the different thread.

4. The reconfigurable processor of claim 1, further comprising:

a counter divided into a plurality of counter areas corresponding to the plurality of threads, to perform counting in each counter area corresponding to a thread being executed; and

a configuration memory to provide information about a configuration enabling execution of the thread being executed to the plurality of function units, based on a result of the counting.

5. The reconfigurable processor of claim 1, wherein a number of the plurality of different areas is adjustable based on a number of the respective different threads.

6. The reconfigurable processor of claim 1, wherein the plurality of memories further include a global register file, separate from the plurality of local register files, to store output information of different threads having been performed by at least one of the plurality of function units.

7. The reconfigurable processor of claim 6, wherein the plurality of function units and the plurality of local register files are part of a configuration array, distinguished from the global register file, controllable to represent different configurations of processing elements.

8. A reconfigurable processor comprising:

a plurality of memories divided into a plurality of different areas, the different divided areas corresponding to respective different threads from a plurality of threads and with sizes proportional to the respective different threads, with corresponding areas of the different divided areas storing context information of the respective different threads; and

a plurality of function units to perform a corresponding calculation based on received context information stored in an area, from the different divided areas, corresponding to a respective thread indicated to be executed,

wherein the plurality of memories comprise;

at least one global register file divided into a plurality of global register file areas corresponding to the respective different threads from the plurality of threads, to provide, to the plurality of function units, an input value loaded to an area of the plurality of divided global register file areas corresponding to the respective thread to be executed; and

a plurality of local register files each divided into a plurality of local register file areas corresponding to the respective different threads from the plurality of threads, to each store, in a respective area of each of the plurality of divided local register file areas corresponding to the respective thread to be executed, a respective output value of a calculation performed by a respective function unit, and

wherein the plurality of memories receive identification information of the respective thread to be executed and make available to the plurality of function units the area, the area being mapped to the identification information.

9. The reconfigurable processor of claim 8, wherein the plurality of function units and the plurality of local register files are part of a configuration array, distinguished from the global register file, controllable to represent different configurations of processing elements, and the configuration array is distinguished from the global register file.

10. A reconfiguration control apparatus comprising:

a reconfigurable processor having a plurality of memories divided into a plurality of different areas, the different divided areas corresponding to respective different threads from a plurality of threads and with sizes proportional to the respective different threads, and to store context information of the respective different threads in corresponding areas of the different divided areas; and

a controlling unit to control a storing of the context information of the respective different threads in the corresponding different divided areas,

wherein the plurality of memories include a plurality of local register files each divided into a plurality of local register file areas, where each local register file area of a local register file is respectively configured to collectively store information for different threads from a particular function unit of the reconfigurable processor configured to execute the different threads, with the particular function unit being paired with the local register file distinguished from a different particular function unit of the reconfigurable processor paired with a different local register file, and

wherein the plurality of memories receive identification information of the respective thread to be executed and make available to the plurality of function units the area, the area being mapped to the identification information.

11. The reconfiguration control apparatus of claim 10, wherein:

the controlling unit provides, to the reconfigurable processor, identification information of a thread to be currently executed from among the respective different threads; and

the reconfigurable processor accesses an area, from different divided areas of a corresponding paired local register file, corresponding to the provided identification information, and executes the thread by using context information stored in the accessed area.

12. The reconfiguration control apparatus of claim 11, wherein:

the controlling unit commands context switching to another available thread, represented by another area of the different divided areas of the corresponding paired local register file, and to provide identification information of the other thread to the reconfigurable processor, upon a stall occurring with respect to the thread to be currently executed; and

the reconfigurable processor accesses the other area corresponding to the identification information of the other thread, and executes the other thread by using context information stored in the accessed other area of the corresponding paired local register file for the other thread.

13. The reconfiguration control apparatus of claim 10, further comprising:

a thread status register to store a status of each of the respective different threads, wherein the controlling unit updates the status of each of the respective different threads,

the status of each of the respective different threads indicating whether the thread is in a status where a stall occurs, a status where the stall is released, a status where execution of the thread is available, and a status where execution of the thread is completed.

14. The reconfiguration control apparatus of claim 10, wherein the reconfigurable processor comprises:

the plurality of memories, divided into the different divided areas, storing context information of the respective different threads in corresponding areas of the different divided areas; and

the plurality of function units to perform a corresponding calculation based on received context information stored in an area, from the different divided areas, corresponding to a thread to be currently executed.

15. The reconfiguration control apparatus of claim 10, wherein the plurality of memories further include a global register file, separate from the plurality of local register files, to store output information of different threads having been performed by at least one of the plurality of function units.

16. The reconfiguration control apparatus of claim 15, wherein the reconfigurable processor further comprises a configuration array, distinguished from the global register, representing differently configurable processing elements, the configuration array including the plurality of function units and plurality of local register files.

17. A reconfiguration control method comprising:

dividing a plurality of memories included in a reconfigurable processor into a plurality of different divided areas corresponding to respective different threads from a plurality of threads and with sizes proportional to the respective different threads; and

storing context information of the respective different threads in corresponding areas of the different divided areas,

wherein the plurality of memories include a plurality of local register files each divided into a plurality of local register file areas, where each local register file area of a local register file is respectively configured to collectively store information for different threads from a particular function unit of a reconfigurable processor configured to execute the different threads, with the particular function unit being paired with the local register file distinguished from a different particular function unit of the reconfigurable processor paired with a different local register file, and

wherein the plurality of memories receive identification information of the respective thread to be executed and make available to the plurality of function units the area the area being mapped to the identification information.

18. The reconfiguration control method of claim 17, further comprising:

providing, to the reconfigurable processor, identification information of a thread to be currently executed from among the respective different threads;

accessing an area, from the different divided areas, corresponding to the provided identification information; and

executing the thread by using context information stored in the accessed area.

19. The reconfiguration control method of claim 18, further comprising:

commanding a context switch to another available thread, represented by another area of the different divided areas, and providing identification information of the other thread to the reconfigurable processor upon a stall occurring with respect to the thread to be currently executed;

accessing the other area corresponding to the identification information of the other thread; and

executing the other thread by using context information stored in the accessed other area for the other thread.

20. The reconfiguration control method of claim 19, further comprising:

changing a status, from plural stored statuses of each of the respective different threads, of the thread to be currently executed upon the stall occurring or being released.

21. A non-transitory computer readable receding medium comprising computer readable code to control at least one processing device to implement the method of claim 17.

22. The reconfiguration control method of claim 17, wherein the plurality of memories further include a global register file to store output information of different threads having been performed by at least one of the plurality of function units.

23. The reconfiguration control method of claim 22, wherein plural paired function units and corresponding local register files are in a configuration array, of the reconfigurable processor, representing differently configurable processing elements, the configuration array being distinguished from the global register file.

24. A thread modeling method comprising:

calling a plurality of Kernel functions, and setting input datasets corresponding to the plurality of Kernel functions as respective threads, respectively;

performing context switching from a thread of the respective threads currently set to be executed to another thread of the respective threads, while maintaining a local register file of each of plural respectively paired function units respectively configured to execute the respective threads and with sizes proportional to the respective threads, and executing the other thread upon a stall occurring with respect to the thread of the respective threads currently set to be executed; and

maintaining a global register file separate from the local register file,

wherein each local register file is configured to maintain, upon the stall, collective storage of information for different threads from a respectively paired function unit, of the plural function units, configured to execute the different threads in different divided areas of each local register file, with each divided area being mapped to a different thread, and

wherein the local resister files receive identification information of the respective thread to be executed and make available to the plurality of function units the area, the area being mapped to the identification information.

25. The thread modeling method of claim 24, further comprising storing, in a global register file, information of different threads having been performed by at least one of the plurality of function units.

26. The thread modeling method of claim 25, wherein plural paired function units and corresponding local register files are in a configuration array representing differently configurable processing elements, the configuration array being distinguished from the global register file.

27. The thread modeling method of claim 25, wherein the setting of the input datasets includes storing the input data sets in different areas of the global register respectively file mapped to different threads.

28. A thread modeling method comprising:

calling a Kernel function, and dividing input datasets corresponding to the called Kernel function;

setting the divided input datasets as respective threads;

performing context switching from a thread of the respective threads currently set to be executed to another thread of the respective threads, while maintaining a local register file of each of plural respectively paired function units respectively configured to execute the respective threads and with sizes proportional to the respective threads, corresponding to an input dataset of the divided input datasets different from an input dataset set as the thread currently set to be executed, and executing the other thread upon a stall occurring with respect to the thread of the respective threads currently set to be executed,

wherein each local register file is configured to maintain, upon the stall, collective storage of information for different threads from a respectively paired function unit, of the plural function units, configured to execute the different threads in different divided areas of each local register file, with each divided area being mapped to a different thread, and

wherein the local register files receive identification information of the respective thread to be executed and make available to the plurality of function units the area the area being mapped to the identification information.

29. The thread modeling method of claim 28, further comprising storing, in a global register file, information of different threads having been performed by at least one of the plurality of function units of the at least one processor.

30. The thread modeling method of claim 29, wherein plural paired function units and corresponding local register files are in a configuration array representing differently configurable processing elements, the configuration array being distinguished from the global register file.

31. The thread modeling method of claim 29, wherein the setting of the divided input datasets includes storing each of the divided input data sets in different divided areas of the global register file, wherein each the divided areas of the global register file is mapped to a different thread.