Method and apparatus for initiating CPU data prefetches by an external agent

• US 20060085602A1
• Filed: 10/15/2004
• Published: 04/20/2006
• Est. Priority Date: 10/15/2004
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Blankenship, Robert G., Huggahalli, Ramakrishna, Tetrick, Raymond S., Batson, Brannon J.
• Granted Patent: US 7,360,027 B2
• Time in Patent Office: Days
• US Class Current: 711/137
• CPC Class Codes: G06F 12/0862 with prefetch
• CPC Class Codes: G06F 2212/6028 Prefetching based on hints or prefetch instructions
• CPC Class Codes: G06F 9/345 of multiple operands or results addressing multiple banks G06F12/06
• CPC Class Codes: G06F 9/383 Operand prefetching cache prefetching G06F12/0862

Abstract

An arrangement is provided for an external agent to initiate data prefetches from a system memory to a cache associated with a target processor, which needs the data to execute a program, in a computing system. When the external agent has data, it may create and issue a prefetch directive. The prefetch directive may be sent along with system interconnection transactions or sent as a separate transaction to devices including the target processor in the system. When receiving and recognizing the prefetch directive, a hardware prefetcher associated with the target processor may issue a request to the system memory to prefetch data to the cache. The target processor can access data in the cache more efficiently than it accesses data in the system memory. Some pre-processing may also be associated with the data prefetch.

Claims

1. An apparatus for initiating CPU data prefetches, comprising:

a target agent to execute a program;

an external source agent to supply at least a portion of data for the target agent to execute the program, and when having new data, to create and issue a prefetch directive; and

a hardware prefetcher associated with the target agent to issue a request to prefetch the new data from a memory to a target cache when the prefetch directive is received.

2. The apparatus of claim 1, wherein the target cache stores at least a portion of data needed for the target agent to execute the program.

3. The apparatus of claim 1, wherein the memory stores data used by at least one of the target agent and the external source agent.

4. The apparatus of claim 1, further comprising a routing support platform to facilitate routing of the prefetch directive from the external source agent to the target agent and data passage among the target agent, the target cache, the data source agent, and the memory.

5. The apparatus of claim 1, wherein the target agent has faster access to data in the target cache than to data in the memory.

6. The apparatus of claim 1, wherein the prefetch directive is sent to agents including at least the target agent along with interconnection transactions existing at least between the target agent and the external source agent.

7. The apparatus of claim 1, wherein the prefetch directive is sent as a separate transaction to agents, including at least the target agent.

8. The apparatus of claim 7, wherein the prefetch directive is sent to at least the target agent at least one of when the external source agent has the new data, and after the external source agent has the new data but before the target agent reads the new data.

9. The apparatus of claim 1, further comprising a requesting agent to send properties of the new data to the external source agent.

10. The apparatus of claim 9, wherein the properties of the new data comprise memory address and direct cache access (DCA) preferences, the DCA preferences including a DCA enablement indicator, an identification of at least one of the target agent and the target cache, and target cache replacement attributes.

11. The apparatus of claim 1, wherein the prefetch directive comprises:

a message opcode to identify the prefetch directive;

a source address to indicate where in the memory data is to be prefetched; and

routing information to help route the prefetch directive to the hardware prefetcher and route data to at least one of the memory and the target cache.

12. The apparatus of claim 11, wherein the prefetch directive further comprises preprocessing information to instruct the hardware prefetcher to preprocess the prefetched data before delivering the data into the target cache.

13. The apparatus of claim 11, wherein the prefetch directive further comprises attributes indicating at least one of the level of the target cache that receives the prefetched data, the replacement and eviction policy of the cache, and the state of a new data block placed in the cache.

14. The apparatus of claim 1 is represented by data stored in a computer readable medium, the data comprising a functional description of the apparatus.

15. The apparatus of claim 14, wherein the data representing the apparatus comprises a hardware description language code.

16. The apparatus of claim 14, wherein the data representing the apparatus comprises data representing a plurality of mask layers string physical data representing the presence or absence of material at various locations of each of the plurality of mask layers.

17. A method for initiating CPU data prefetches, comprising:

executing a program by a target agent;

obtaining new data by an external source agent, the new data needed for the target agent to run the program;

creating and issuing a prefetch directive when the new data is obtained by the external source agent; and

issuing a request to prefetch the new data into a target cache, when the prefetch directive is received.

18. The method of claim 17, further comprising receiving properties of the new data, the properties comprising memory address and direct cache access (DCA) preferences, the DCA preferences including a DCA enablement indicator, an identification of at least one of the target agent and the target cache, and target cache replacement attributes.

19. The method of claim 17, wherein the target cache stores at least a portion of data needed for the target agent to execute the program, the target agent has faster access to data in the target cache than access to data in a memory that stores data used by at least one of the target agent and the external source agent.

20. The method of claim 17, further comprising writing the new data into a memory.

21. The method of claim 17, further comprising sending the prefetch directive to agents including at least the target agent along with interconnection transactions existing at least between the target agent and the external source agent.

22. The method of claim 17, further comprising preprocessing the prefetched data according to instructions included in the prefetch directive before delivering the data into the target cache and reading the new data from the target cache by the target agent.

23. The method of claim 17, further comprising sending the prefetch directive as a separate transaction to agents, including at least the target agent.

24. The method of claim 23, further comprising sending the prefetch directive to at least the target agent at least one of when the external source agent has the new data, and after the external source agent has the new data but before the target agent reads the new data.

25. The method of claim 17, wherein the prefetch directive comprises:

a message opcode to identify the prefetch directive;

a source address to indicate where in the memory data is to be prefetched; and

routing information to help route the prefetch directive to the hardware prefetcher and route data to at least one of the memory and the target cache.

26. The method of claim 25, wherein the prefetch directive further comprises attributes indicating at least one of the level of the target cache that receives the prefetched data, the replacement and eviction policy of the cache, and the state of a new data block placed in the cache.

27. An article comprising a machine readable medium having stored thereon a prefetch directive, the prefetch directive comprising:

a first field including a message opcode to identify the prefetch directive;

a second field including a source address to indicate where in the memory data is to be prefetched; and

a third field including routing information to help route the prefetch directive to a target and data to at least one of a memory and a target cache, the target associated with the target cache.

28. The article of claim 27, wherein the prefetch directive further comprises preprocessing information to instruct the target to preprocess the prefetched data before delivering the data into the target cache.

29. The article of claim 27, wherein the prefetch directive further comprises attributes indicating at least one of the level of the target cache that receives the prefetched data, the replacement and eviction policy of the cache, and the state of a new data block placed in the cache.

30. The article of claim 27, wherein the target receives and recognizes the prefetch directive and issues a request to the memory to prefetch the data from the memory to the target cache.

31. A processor, comprising:

a prefetcher to fetch data; and

a prefetch directive receiving mechanism, coupled to the prefetcher, to recognize and receive an external prefetch directive message from another agent, the external prefetch directive message indicating a first address and signaling to the prefetcher to cause the prefetcher to fetch data associated with the first address.

32. The processor of claim 31, further comprising a prefetch directive decoding mechanism, coupled to the prefetcher and the prefetch directive receiving mechanism, to decode information included in the external prefetch directive message.

33. The processor of claim 31, further comprising a preprocessing mechanism to preprocess the fetched data according to information included the external prefetch directive message before delivering the fetched data to a receiving cache of the processor.

34. The processor of claim 31, wherein the prefetcher fetches data associated with the first address from a memory, the first address being a memory address of the data.

35. The processor of claim 31, further comprising a prefetch directive sending mechanism to send an external prefetch directive message when the processor has data to send to a target agent.

36. The processor of claim 31, wherein the external prefetch directive message comprises:

a message opcode to identify the external prefetch directive message;

a source address to indicate where in the memory the data is to be prefetched; and

routing information to help route the external prefetch directive message to the prefetcher and route the data to a receiving cache of the processor.

37. The processor of claim 36, wherein the external prefetch directive message further comprises preprocessing information to instruct the prefetcher to preprocess the fetched data before delivering the data into the receiving cache.

38. The processor of claim 36, wherein the external prefetch directive message further comprises attributes indicating at least one of the level of the receiving cache, the replacement and eviction policy of the receiving cache, and the state of a new data block placed in the receiving cache.

39. A computing system, comprising:

a target processor to execute a program;

an external agent to supply at least a portion of data for the target agent to execute the program, and when having new data, to initiate data prefetch from a memory to a cache, the cache associated with the target processor; and

a supporting platform to provide interfaces between the target processor and the external agent and to facilitate data move among the target processor, the external agent, the cache, and the memory;

wherein the target processor has faster access to data in the cache than access to data in the memory.

40. The computing system of claim 39, wherein the external agent initiates a data prefetch from the memory to the cache by creating and issuing a prefetch directive.

41. The computing system of claim 39, wherein the external agent comprises another processor in the computing system.

42. The computing system of claim 39, wherein the external agent comprises a chipset in the computing system.

43. The computing system of claim 39, wherein the external agent comprises an I/O device, the I/O device being at least one of included in the computing system and coupled to the computing system.

44. The computing system of claim 39, further comprising a hardware prefetcher, associated with the target processor, to issue a request to prefetch the new data from the memory to the cache when receiving and recognizing the prefetch directive.

45. The computing system of claim 39, wherein the prefetch directive is sent to agents including at least the target processor along with interconnection transactions existing at least between the target processor and the external agent.

46. The computing system of claim 39, wherein the prefetch directive is sent as a separate transaction to agents, including at least the target processor.

47. The computing system of claim 39, wherein the prefetch directive is sent to at least the target processor at least one of when the external agent has the new data, and after the external agent has the new data but before the target processor reads the new data.

48. The computing system of claim 39, wherein the prefetch directive comprises:

a message opcode to identify the prefetch directive;

a source address to indicate where in the memory data is to be prefetched; and

routing information to help route the prefetch directive to a hardware prefetcher and route data to at least one of the memory and the cache.

49. The computing system of claim 48, wherein the prefetch directive further comprises preprocessing information to instruct the hardware prefetcher to preprocess the prefetched data before delivering the data into the cache.

50. The computing system of claim 48, wherein the prefetch directive further comprises attributes indicating at least one of the level of the cache that receives the prefetched data, the replacement and eviction policy of the cache, and the state of a new data block placed in the cache.

51. An article comprising a machine readable medium having stored thereon data which, when accessed by a processor in conjunction with simulation routines, provides functionality of a model including:

a target agent to execute a program;

the external source agent to supply at least a portion of data for the target agent to execute the program, and when having new data, to create and issue a prefetch directive; and

a hardware prefetcher associated with the target agent, when receiving and recognizing the prefetch directive, to issue a request to prefetch the new data from a memory to a target cache;

wherein the target agent has more efficient access to data in the target cache than access to data in the memory.

52. The article of claim 51, wherein the prefetch directive is sent to agents including at least the target agent along with interconnection transactions existed at least between the target agent and the external source agent.

53. The article of claim 51, wherein the prefetch directive is sent as a separate transaction to agents, including at least the target agent.