Methods and apparatus for canceling a memory data fetch
View Patent ↗According to the present invention, methods and apparatus are provided for increasing the efficiency of data access in a multiple processor, multiple cluster system. Mechanisms for transmitting memory cancels to memory controllers in the various clusters of a multiple cluster system are provided. In one example, memory cancels are transmitted between clusters when it is determined that a memory line associated with a probe is dirty. The memory cancel directs the memory controller to no longer proceed with a data fetch from main memory. In another example, memory cancels are transmitted at a home cluster based on information in a coherence directory in order to more quickly end a data fetch.
1. A computer system, comprising:
a home cluster including a first plurality of processors and a home cache coherence controller, the first plurality of processors and the home cache coherence controller interconnected in a point-to-point architecture;
a remote cluster including a second plurality of processors and a remote cache coherence controller, the second plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture;
wherein the remote cluster cache coherence controller is configured to receive a probe request associated with a memory line and send a memory cancel to the home cluster if a probed memory line associated with one of the second plurality of processors is dirty, wherein the memory cancel prevents an extraneous data fetch by a home cluster memory controller.
2. The computer system of claim 1 , wherein the memory line is dirty if the memory line is in the modified or owned state.
3. The computer system of claim 1 , wherein the memory cancel directs the home cluster memory controller to cancel a data fetch from main memory.
4. The computer system of claim 1 , wherein the probed memory line is determined to be dirty by the remote cluster.
5. A method for probing a system, the method comprising:
receiving a probe request associated with a memory line at a remote cluster, the remote cluster including a plurality of processors and a remote cache coherence controller, the plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture;
determining if the probed memory line associated with one of the plurality of processors is dirty; and
sending a memory cancel to a home cluster including a home cluster cache coherence controller, wherein the memory cancel prevents an extraneous data fetch by a home cluster memory controller.
6. The method of claim 5 , wherein the memory line is dirty if the memory line is in the modified or owned state.
7. The method of claim 5 , wherein the memory cancel directs the home cluster memory controller to cancel a data fetch from main memory.
8. The method of claim 5 , wherein the probed memory line is determined to be dirty by the remote cluster.
9. A cache coherence controller, the cache coherence controller comprising:
means for receiving a probe request associated with a memory line at a remote cluster cache coherence controller, the remote cluster including a plurality of processors and the remote cache coherence controller, the plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture;
means for determining if the probed memory line associated with one of the plurality of processors is dirty; and
means for sending a memory cancel to a home cluster including a home cluster cache coherence controller, wherein the memory cancel prevents an extraneous data fetch by a home cluster memory controller.
10. A computer system, comprising:
a home cluster including a first plurality of processors and a home cache coherence controller, the first plurality of processors and the home cache coherence controller interconnected in a point-to-point architecture;
a remote cluster including a second plurality of processors and a remote cache coherence controller, the second plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture;
wherein the home cluster cache coherence controller is configured to forward a probe to the remote cluster and send a memory cancel to a home cluster memory controller if the home cluster cache coherence controller determines that the memory line is dirty, wherein the memory cancel prevents an extraneous data fetch by a home cluster memory controller.
11. The computer system of claim 10 , wherein the memory line is dirty if the memory line is in the modified or owned state.
12. The computer system of claim 10 , wherein the home cluster cache coherence controller determines that the memory line associated with the probe in the remote cluster is dirty by referencing a coherence directory.
13. The computer system of claim 10 , wherein the home cluster cache coherence controller forwards a probe with a memory cancel indicator to the remote cluster.
14. The computer system of claim 13 , wherein the home cluster cache coherence controller further tracks the condition that no memory cancel will be received from the remote cluster.
15. The computer system of claim 13 , wherein the home cluster coherence controller sets the cancel bit in the read response from the remote cluster based on knowledge of a previously generated cancel response.
16. A method for probing remote nodes, the method comprising:
determining at a home cluster if the memory line associated with a request is dirty, a home cluster including a first plurality of processors and a home cache coherence controller, the first plurality of processors and the home cache coherence controller interconnected in a point-to-point architecture;
forwarding a probe to a remote cluster, the remote cluster including a second plurality of processors and a remote cache coherence controller, the second plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture; and
sending a memory cancel to the home cluster memory controller before a response to the probe is received from the remote cluster, wherein the memory cancel prevents an extraneous data fetch by a home cluster memory controller.
17. The method of claim 16 , wherein the memory line is dirty if the memory line is in the modified or owned state.
18. The method of claim 16 , wherein the home cluster cache coherence controller determines that the memory line associated with the probe in the remote cluster is dirty by referencing a coherence directory.
19. The method of claim 16 , wherein the home cluster cache coherence controller forwards a probe with a memory cancel indicator to the remote cluster.
20. The method of claim 19 , wherein the home cluster cache coherence controller further tracks the condition that no memory cancel will be received from the remote cluster.
21. The method of claim 19 , wherein the home cluster coherence controller sets the cancel bit in the read response from the remote cluster based on knowledge of a previously generated cancel response.
22. The method of claim 16 , wherein sending the memory cancel to the home cluster memory controller before a response is received from the remote cluster increases the likelihood that a memory fetch will not complete.
23. The method of claim 22 , wherein the memory fetch includes acquiring invalid data from memory and forwarding the invalid data to a home cluster processor originating the probe request.
24. The method of claim 23 , wherein preventing the unnecessary memory fetch of invalid data saves system resources.
25. A multiple processor system, comprising:
means for determining at a home cluster if the memory line associated with a request is dirty, a home cluster including a first plurality of processors and a home cache coherence controller, the first plurality of processors and the home cache coherence controller interconnected in a point-to-point architecture;
means for forwarding a probe to a remote cluster, the remote cluster including a second plurality of processors and a remote cache coherence controller, the second plurality of processors and the remote cache coherence controller interconnected in a point-to-point architecture; and
means for sending a memory cancel to the home cluster memory controller before a response to the probe is received from the remote cluster, wherein the memory cancel prevents an extraneous data fetch by a borne cluster memory controller.