Fill partitioning of a shared cache
View Patent ↗Fill partitioning of a shared cache is described. In an embodiment, all threads running in a processor are able to access any data stored in the shared cache; however, in the event of a cache miss, a thread may be restricted such that it can only store data in a portion of the shared cache. The restrictions to storing data may be implemented for all cache miss events or for only a subset of those events. For example, the restrictions may be implemented only when the shared cache is full and/or only for particular threads. The restrictions may also be applied dynamically, for example, based on conditions associated with the cache. Different portions may be defined for different threads (e.g. in a multi-threaded processor) and these different portions may, for example, be separate and non-overlapping. Fill partitioning may be applied to any on-chip cache, for example, a L1 cache.
1. A method of managing memory in a processor, the method comprising:
allowing unrestricted access to any data stored in a shared cache in the processor;
in response to a cache miss event associated with a thread running in the processor, using a fill partitioning policy to select a location within a defined portion of the shared cache; and storing content of a memory location associated with the cache miss event in the selected location in the shared cache; and
dynamically switching on and off use of the fill partitioning policy to select a location based on one or more of: thread ID and cache conditions,
wherein the size of the defined portion of the shared cache allocated to the thread is adjusted dynamically while operations are in progress.
2. The method according to claim 1 , further comprising using the fill partitioning policy following each cache miss event.
3. The method according to claim 1 , further comprising using the fill partitioning policy to select the location within the defined portion of the shared cache only when the shared cache is full.
4. The method according to claim 1 , wherein using a fill partitioning policy to select a location within a defined portion of the shared cache comprises using a modified replacement algorithm arranged to select a location within the defined portion associated with the thread.
5. The method according to claim 1 , wherein using a fill partitioning policy to select a location within a defined portion of the shared cache comprises:
selecting a mask based on an identifier of the thread, the mask defining a portion of the shared cache.
6. The method according to claim 5 , further comprising, if the shared cache is full, applying the selected mask to a replacement algorithm.
7. The method according to claim 5 , further comprising, if the shared cache is not full, applying the selected mask to filter available locations within the shared cache.
8. The method according to claim 5 , wherein the selecting comprises selecting the mask from a set of stored pre-defined masks.
9. The method according to claim 8 , wherein each mask in the set of stored pre-defined masks defines a different portion of the shared cache.
10. The method according to claim 1 , wherein the cache conditions comprise one of: cache performance, cache load, a rate of cache misses for a thread, a comparison of rates of cache misses for different threads, an amount of data that a thread is storing to a cache, and a number of fills for a thread which evict another thread's data from a cache.
11. The method according to claim 1 , further comprising switching fill partitioning on and off for threads on an individual basis.
12. A processor comprising:
a shared cache; and
memory management logic arranged to manage access to memory including the shared cache and to allow unrestricted access to any data stored in the shared cache, wherein the memory management logic comprises fill partitioning logic arranged, in response to a cache miss event associated with a thread running in the processor, to use a fill partitioning policy to select a location within a defined portion of the shared cache; and to store content of a memory location associated with the cache miss event in the selected location in the shared cache,
wherein the memory management logic is further arranged to dynamically switch on and off use of the fill partitioning logic based on one or more of: thread ID and cache conditions, and
wherein the size of the defined portion of the shared cache allocated to the thread is adjusted dynamically while operations are in progress.
13. The processor according to claim 12 , wherein the fill partitioning policy is used following each cache miss.
14. The processor according to claim 12 , wherein the fill partitioning policy is only used to select the location when the shared cache is full.
15. The processor according to claim 12 , wherein using a fill partitioning policy to select a location within a defined portion of the shared cache comprises using a modified replacement algorithm arranged to select a location within the defined portion associated with the thread.
16. The processor according to claim 12 , wherein using a fill partitioning policy to select a location within a defined portion of the shared cache comprises:
selecting a mask based on an identifier of the thread, the mask defining a portion of the shared cache.
17. The processor according to claim 12 , wherein the processor is a multi-threaded processor.
18. The processor according to claim 12 , further comprising multiple processing cores.
19. The processor according to claim 12 , wherein the shared cache is a L1 cache.