Request ordering in a cache

A method, comprising:

• receiving, at a cache, requests from a processor for accessing different devices, wherein at least some of the requests are required to be sent to the different devices according to a temporal order, wherein the cache includes two or more request handling pipes respectively comprising (a) two or more request buffers configured to store the requests from the processor for accessing the different devices and (b) a first-level token store;
• for an individual one of the at least some requests from the processor:
• selecting, using bank selection circuit of the cache, one request handling pipe and one request buffer to store the individual request;
• storing, at the first-level token store of the selected one request handling pipe, a value indicating the selected one request buffer of the selected one request handling pipe in which the individual request is stored;
• and
• storing, at a second-level token store, a value indicating the selected one request handling pipe in which the individual request is stored;
• and
• performing, using a request selection circuit of the cache:
• receiving an indication of a request handling pipe selected out of the two or more request handling pipes based on the value at the second-level token store that stores an oldest request out of the at least some requests stored at the two or more request handling pipes;
• for respective ones of the two or more request handling pipes, receiving an indication of a request buffer selected out of the two or more request buffers of the respective ones of the request handling pipes based on the values at the first-level token stores of the respective ones of the request handling pipes that stores an oldest request out of a subset of the at least some requests stored at the two or more request buffers of the respective ones of the request handling pipes;
• and
• sending the oldest request stored at the indicated request buffer of the indicated request handling pipe to one of the devices.

A system, comprising:

• one or more processors;
• and
• a cache configured to be shared by the one or more processors, comprising:
• two or more request handling pipes respectively comprising:
• a first-level token store;
• and
• two or more request buffers configured to store requests from a processor for accessing different devices, wherein at least some of the requests are required to be sent to the different devices according to a temporal order;
• a second-level token store;
• a bank selection circuit configured to:
• for an individual one of the at least some requests from the processor:
• select one request handling pipe and one request buffer of the selected one request handling pipe to store the individual request;
• store, at the first-level token store of the selected one request handling pipe, a value indicating the selected one request buffer of the selected one request handling pipe in which the individual request is stored;
• and
• store, at the second-level token store, a value indicating the selected one request handling pipe in which the individual request is stored;
• and
• a request selection circuit configured to:
• receive an indication of a request handling pipe selected out of the two or more request handling pipes based on the value at the second-level token store that stores an oldest request out of the at least some requests stored at the two or more request handling pipes;
• for respective ones of the two or more request handling pipes, receive an indication of a request buffer selected out of the two or more request buffers of the respective ones of the request handling pipes based on the values at the first-level token stores of the respective ones of the request handling pipes that stores an oldest request out of a subset of the at least some requests stored at the two or more request buffers of the respective ones of the request handling pipes;
• and
• send the oldest request stored at the indicated request buffer of the indicated request handling pipe to one of the devices.