METHOD AND APPARATUS FOR DECOUPLED RETRIEVAL OF CACHE MISS DATA

The invention provides a method of operating a cache memory so that operation is optimized. Instead of fetching data immediately upon a cache miss, the present invention continues with subsequent cache accesses. Decoupled from cache access, cache misses are fetched to cache. During operation, for each request in a sequence of data requests, it is determined if the requested data can be found in cache memory. If the data is not found in the cache, the next request in the sequence is processed without first retrieving the data pending from the earlier request. A miss list is generated for each of the requests in the sequence of requests whose data is not found in the cache. The data that is associated with the requests in the miss list is obtained from DRAM and used to satisfy the requests. Some cache lines may have one or more pending hits to data associated with the cache line. Those requests are kept in a pending hits list and processed in order as required. There may also be pending misses kept in a pending misses list where the list contains one or more pending misses to data associated with the cache line. A flag or indicator is set for a cache line when there are misses associated with the cache line.