Method and apparatus for increasing computer performance through asynchronous memory block initialization

• US 6,601,153 B1
• Filed: 12/31/1999
• Issued: 07/29/2003
• Est. Priority Date: 12/31/1999
• Status: Expired due to Term

• Current Assignee: Unisys Corporation
• Original Assignee: Unisys Corporation
• Inventors: Engelbrecht, Kenneth L., Mikkelsen, Hans C., Ward, Wayne D.
• Primary Examiner(s): Peikari, B. James
• Application Number:
  US09/476,022
• Time in Patent Office: 1,306 Days
• Field of Search: 707/200, 707/206, 709/100, 709/102, 709/103, 712/219, 711/103, 711/165, 711/166
• US Class Current: 711/166
• CPC Class Codes: G06F 12/08 in hierarchically structured memory systems, e.g. virtual memory systems
• CPC Class Codes: G06F 12/0891 using clearing, invalidating or resetting means
• CPC Class Codes: Y10S 707/99953 Recoverability

Abstract

A system and method for increasing processing performance in a computer system by asynchronously performing system activities that do not conflict with normal instruction processing, during inactive memory access periods. The computer system includes at least one instruction processor to process instructions of an instruction stream, and a memory to store data. One or more inactive data blocks in the memory are identified, and a list of addresses corresponding to the identified inactive data blocks is generated. Available computing cycles occurring during processing in the computer system are identified, such as processing stalls and idle memory write periods. The inactive data blocks associated with the list of addresses are initialized to a predetermined state, during the available computing cycles. Addresses corresponding to those initialized data blocks are then made available to the computing system to facilitate use of the data blocks.

Claims

1. A method for increasing processing performance in a computer system having at least one instruction processor to process instructions of an instruction stream, and having a memory to store data, the method comprising:

identifying one or more inactive data blocks in the memory;

generating a list of addresses corresponding to the identified inactive data blocks;

identifying available computing cycles during processing in the computer system; and

initializing, via the at least one instruction processor that is used to process instructions of the instruction stream, the one or more inactive data blocks associated with the list of addresses to a predetermined state during one or more of the available computing cycles, wherein the initializing is performed asynchronously with respect to processing of the instruction stream by the at least one instruction processor.

2. The method of claim 1, wherein identifying available computing cycles comprises identifying available memory write cycles when a write interface to memory is idle.

3. The method of claim 2, wherein identifying available memory write cycles comprises determining when the write interface between a first level cache in the instruction processor and a second level cache in the computer system is temporarily idle.

4. The method of claim 2, wherein identifying available memory write cycles comprises identifying available entries in a memory write buffer.

5. The method of claim 1, wherein identifying available computing cycles comprises identifying temporary suspensions of instruction stream processing.

6. The method of claim 5, wherein identifying temporary suspensions of instruction stream processing comprises identifying memory latencies due to data access delays.

7. The method of claim 5, wherein identifying temporary suspensions of instruction stream processing comprises identifying data dependencies causing idle processing cycles.

8. The method of claim 1, further comprising storing one of the addresses of the list into a register of the instruction processor to identify which of the corresponding inactive data blocks is to be cleared.

9. The method of claim 1, further comprising initializing one or more of the inactive data blocks using the instructions of the instruction stream when an initialized inactive data block is required and no available computing cycles can be identified.

10. The method of claim 1, wherein generating a list of addresses comprises generating a queue of the addresses corresponding to the identified inactive data blocks.

11. The method of claim 1, wherein generating a list of addresses comprises generating a linked list of the addresses corresponding to the identified inactive data blocks.

12. The method of claim 1, wherein identifying one or more inactive data blocks in the memory comprises identifying one or more inactive pages in a secondary storage associated with the memory.

13. The method of claim 1, wherein identifying one or more inactive data blocks in the memory comprises identifying one or more inactive cache lines associated with a cache memory.

14. The method of claim 1, wherein identifying one or more inactive data blocks in the memory comprises identifying one or more inactive memory segments, wherein the memory segments comprise a predetermined block size of memory.

15. The method of claim 1, wherein initializing the inactive data blocks comprises clearing a subset of the inactive data block currently being cleared, each time the available computing cycles are identified, until the inactive data block is completely cleared.

16. The method of claim 15, wherein the subset of the inactive data block is a predetermined fixed number of bytes.

17. The method of claim 15, wherein the subset of the inactive data block is a variable number of bytes dependent on a number of the available computing cycles identified.

18. The method of claim 1, wherein initializing the inactive data blocks comprises writing memory locations associated with the inactive data blocks to a logic state of zero.

19. The method of claim 1, wherein initializing the inactive data blocks comprises writing the inactive data blocks to a predetermined data pattern that obscures a pre-clearing state of the inactive data blocks.

20. The method of claim 1, further comprising generating a second list of second addresses corresponding to the inactive data blocks which have been initialized and are available for use by the computing system.

21. The method of claim 20, wherein generating a second list of addresses comprises generating a queue of the second addresses.

22. The method of claim 20, wherein generating a second list of addresses comprises generating a linked list of the second addresses.

23. A method for increasing processing performance in a computer system having at least one instruction processor to process instructions of an instruction stream, and having a memory to store data, the method comprising:

identifying one or more inactive data pages in the memory;

storing a first list of first addresses corresponding to the identified inactive data pages in a first linked list;

identifying idle memory write cycles associated with a data write interface between the instruction processor and the memory;

clearing the one or more inactive data pages associated with the list of addresses to a predetermined state during one or more of the identified idle memory write cycles, wherein the at least one instruction processor that processes instructions of the instruction stream initiates the clearing via the data write interface independent of activity in the instruction stream; and

storing a second list of second addresses corresponding to the one or more inactive data pages that have been cleared in a second linked list.

24. The method of claim 23, further comprising passing one of the first addresses to the instruction processor, via the instruction stream, to identify which of the first addresses to be cleared.

25. The method of claim 24, further comprising generating an interrupt, at the instruction processor, upon completion of clearing the first address that was passed to the instruction processor.

26. The method of claim 25, wherein generating an interrupt comprises tagging a page-cleared indication onto an existing interrupt already generated by the processor.

27. The method of claim 23, further comprising retrieving one of the first addresses in the first linked list by the instruction processor.

28. The method of claim 27, further comprising executing an instruction in the instruction stream to notify the instruction processor of an address of the first linked list.

29. The method of claim 27, further comprising directing the instruction processor to store the second address corresponding to the inactive data page that has been cleared onto the second linked list.

30. The method of claim 29, wherein directing the instruction processor to store the second address comprises executing an instruction in the instruction stream to notify the instruction processor of an address of the second linked list.

31. The method of claim 30, wherein clearing the one or more inactive data pages associated with the list of addresses to a predetermined state comprises zeroing the inactive data pages by writing memory locations associated with the inactive data pages to a logic state of zero.

32. A system for asynchronously initializing inactive data blocks in a computing system, comprising:

a memory for storing instructions and data associated with an instruction stream, wherein the memory includes inactive data blocks currently unused by the computing system and having residual data therein;

a first queue to store one or more addresses corresponding to the inactive data blocks;

a second queue to store one or more addresses corresponding to the inactive data blocks that have been initialized; and

at least one instruction processor to process instructions of the instruction stream, the at least one instruction processor including a write interface coupled to the memory to allow data to be written to the memory, wherein the at least one instruction processor is configured and arranged to accept the addresses from the first queue, initialize the corresponding inactive data blocks via the write interface during idle memory write cycles between the at least one instruction processor and the memory, and to store the addresses corresponding to the inactive data blocks that have been initialized onto the second queue.

33. The system as in claim 32, wherein the instruction processor is configured and arranged to initialize the corresponding inactive data blocks via the write interface by transmitting a predetermined data pattern to the inactive data blocks from the instruction processor to the memory via the write interface during the idle memory write cycles.

34. The system as in claim 33, wherein the instruction processor is further configured and arranged to asynchronously transmit the predetermined data pattern via the write interface to overwrite the residual data of the inactive data blocks, independent of activity in the instruction stream.

35. The system as in claim 33, wherein the instruction processor is further configured and arranged to asynchronously transmit the predetermined data pattern by writing memory locations associated with the inactive data blocks to a logic state of zero.

36. The system as in claim 32, wherein the instruction processor includes a first level cache memory, and wherein the memory is a second level cache memory coupled to receive data from the first level cache memory via the write interface.

37. A system for asynchronously initializing inactive data blocks in a computing system, comprising:

a memory for storing instructions and data associated with an instruction stream, wherein the memory includes inactive data blocks currently unused by the computing system and having residual data therein, and wherein the memory further includes a first linked list to store one or more addresses corresponding to the inactive data blocks and a second linked list to store one or more addresses corresponding to the inactive data blocks that have been initialized; and

at least one instruction processor to process instructions of the instruction stream, the at least one instruction processor including a write interface coupled to the memory to allow data to be written to the memory, wherein the at least one instruction processor is configured and arranged to accept the addresses from the first linked list, initialize the corresponding inactive data blocks via the write interface during idle memory write cycles between the at least one instruction processor and the memory, and to store the addresses corresponding to the inactive data blocks that have been initialized onto the second linked list.