Cache coherency in multiprocessor system
A processor includes a plurality of cache memories, and a plurality of processor cores, each associated with one of the cache memories. Each of at least some of the cache memories is associated with information indicating whether data stored in the cache memory is shared among multiple processor cores.
1. A processor, comprising:
a plurality of processor cores;
a plurality of cache memories associated with the processor cores, with at least some of the cache memories configurable to associate information indicating whether data stored in the cache memory is shared among multiple ones of the processor cores, with the data in the cache memory being directly accessible to processor cores sharing the data; and with the processor cores further comprising:
cache coherency circuitry that replaces cache lines stored at one or more addresses in the cache memory, using the associated information that determines whether data stored at the addresses are shared among the multiple processor cores, wherein the cache coherence circuitry uses write through core caches and coalescing write buffers to achieve high performance, and with the cache coherency circuitry using existing L2 caches as a private L2 and a distributed shared L3 cache, with shared addresses and memory locations allocated using a memory allocation function.
2. The processor of claim 1 wherein the processor cores implement a directory aware replacement policy, and the associated information is a number of directory shared pointers of shown by a cache line.
3. The processor of claim 1 wherein the circuitry is configured to replace data that is shared by fewer processor cores than other data stored in the cache memory.
4. The processor of claim 1 wherein the cache coherency circuitry is configured to replace data that is not shared by multiple processor cores before replacing data that is shared by multiple processor cores.
5. The processor of claim 1 wherein the one or more addresses correspond to a cache line.
6. The processor of claim 1 wherein a first processor core directly accesses shared data stored in a second cache memory of a second processor core without first accessing a first cache memory of the first processor core.
7. The processor of claim 1 wherein the associated information includes the number of processor cores sharing the data.
8. The processor of claim 1 wherein the associated information indicates whether data stored in the cache memory is shared among multiple processor cores is stored in the associated cache memory.
9. The processor of claim 1 , wherein the processor is on a single integrated circuit.
10. The processor of claim 2 , wherein cache coherency circuitry uses the associated information that includes data about the directory state in the cache, and uses a replacement state in order choose a cache line in the cache to evict.
11. A method for replacing addresses in one of multiple cache memories associated with respective processor cores, comprising:
storing information associated with the one of the multiple cache memories, the information indicating whether data stored in the one cache memory is shared among multiple processor cores, the data in the one cache memory being directly accessible to processor cores sharing the data; and with the processor cores further comprising:
replacing data stored at one or more addresses in the one cache memory according to a cache coherency protocol that uses a process dependent on associated information indicating whether data stored at the addresses are shared among multiple processor cores, wherein the cache coherence protocol is implemented fully in hardware and uses write through core caches and coalescing write buffers to achieve high performance, and treats existing L2 caches simultaneously as both a private L2 and a distributed shared L3 cache, with shared addresses and memory locations allocated using a special memory allocation function.
12. The method of claim 11 , further comprising:
executing a directory aware replacement policy and the information is a number of directory shared pointers of shown by a cache line.
13. The method of claim 12 , further comprising:
replacing data that is shared by fewer processor cores than other data stored in the cache memory.
14. The method of claim 12 , further comprising:
replacing data that is not shared by multiple processor cores before replacing data that is shared by multiple processor cores.
15. The method of claim 12 , wherein the one or more addresses correspond to a cache line.
16. The method of claim 12 , further comprising:
directly accessing by a first processor core shared data stored in a second cache memory of a second processor core without first accessing a first cache memory of the first processor core.
17. The method of claim 12 , wherein the information indicates the number of processor cores sharing the data.
18. The method of claim 12 , further comprising:
storing the information indicating whether data stored in the cache memory is shared among multiple processor cores in the associated cache memory.
19. The method of claim 12 , further comprising:
providing the multiple processor cores are on a single integrated circuit.
20. The method of claim 12 , wherein replacing further comprises:
using information about the directory state in the cache and a replacement state in order choose a cache line in the cache to evict.
21. A processor comprising:
a plurality of processor cores; and
a plurality of cache memories that are associated with the plurality of processor cores, with at least some of the cache memories configured to associate information indicating whether data stored in the respective cache memory is shared among multiple ones of the processor cores, with the data in the cache memory being directly accessible to those processor cores that share the data; and
cache coherency circuitry that replaces cache lines stored at one or more addresses in the cache memory, using the associated information that is a number of directory shared pointers of a cache line, the cache coherency circuitry determining based on the associated information whether data stored at the addresses are shared among the multiple one of the processor cores, with the cache coherence circuitry further comprising:
write through core caches and coalescing write buffers, and with the cache coherency circuitry using existing L2 caches as private L2 and a distributed shared L3 caches, with shared addresses and memory locations allocated by execution on the processor of a memory allocation function instruction.
22. The processor of claim 21 wherein the processor cores implement a directory aware replacement policy and the processor is on a single integrated circuit.