Abstract: A cache, system and method for improving the snoop bandwidth of a cache directory. A cache directory may be sliced into two smaller cache directories each with its own snooping logic. By having two cache directories that can be accessed simultaneously, the bandwidth can be essentially doubled. Furthermore, a “frequency matcher” may shift the cycle speed to a lower speed upon receiving snoop addresses from the interconnect thereby slowing down the rate at which requests are transmitted to the dispatch pipelines. Each dispatch pipeline is coupled to a sliced cache directory and is configured to search the cache directory to determine if data at the received addresses is stored in the cache memory. As a result of slowing down the rate at which requests are transmitted to the dispatch pipelines and accessing the two sliced cache directories simultaneously, the bandwidth or throughput of the cache directory may be improved.

Abstract: A cache memory which loads two memory values into two cache lines by receiving separate portions of a first requested memory value from a first data bus over a first time span of successive clock cycles and receiving separate portions of a second requested memory value from a second data bus over a second time span of successive clock cycles which overlaps with the first time span. In the illustrative embodiment a first input line is used for loading both a first byte array of the first cache line and a first byte array of the second cache line, a second input line is used for loading both a second byte array of the first cache line and a second byte array of the second cache line, and the transmission of the separate portions of the first and second memory values is interleaved between the first and second data busses. The first data bus can be one of a plurality of data busses in a first data bus set, and the second data bus can be one of a plurality of data busses in a second data bus set.

Abstract: A cache coherent data processing system includes at least a first cache memory supporting a first processing unit and a second cache memory supporting a second processing unit. The first cache memory includes a cache array and a cache directory of contents of the cache array. In response to the first cache memory detecting on an interconnect a broadcast operation that specifies a request address, the first cache memory determines from the operation a type of the operation and a coherency state associated with the request address. In response to determining the type and the coherency state, the first cache memory filters out the broadcast operation without accessing the cache directory.

Abstract: A data processing system includes a plurality of processing units coupled for communication by a communication link and a configuration register. The configuration register has a plurality of different settings each corresponding to a respective one of a plurality of different link information allocations. Information is communicated over the communication link in accordance with a particular link information allocation among the plurality of link information allocations determined by a respective setting of the configuration register.

Abstract: A method and apparatus for preventing selection of Deleted (D) members as an LRU victim during LRU victim selection. During each cache access targeting the particular congruence class, the deleted cache line is identified from information in the cache directory. A location of a deleted cache line is pipelined through the cache architecture during LRU victim selection. The information is latched and then passed to MRU vector generation logic. An MRU vector is generated and passed to the MRU update logic, which is selects/tags the deleted member as a MRU member. The make MRU operation affects only the lower level LRU state bits arranged in a tree-based structure state bits so that the make MRU operation only negates selection of the specific member in the D state, without affecting LRU victim selection of the other members.

Abstract: A cache memory logically partitions a cache array having a single access/command port into at least two slices, and uses a first cache directory to access the first cache array slice while using a second cache directory to access the second cache array slice, but accesses from the cache directories are managed using a single cache arbiter which controls the single access/command port. In the illustrative embodiment, each cache directory has its own directory arbiter to handle conflicting internal requests, and the directory arbiters communicate with the cache arbiter. An address tag associated with a load request is transmitted from the processor core with a designated bit that associates the address tag with only one of the cache array slices whose corresponding directory determines whether the address tag matches a currently valid cache entry.

Abstract: A method, apparatus, and computer for identifying selection of a bad victim during victim selection at a cache and recovering from such bad victim selection without causing the system to crash or suspend forward progress of the victim selection process. Among the bad victim selection addressed are recovery from selection of a deleted member and recovery from use of LRU state bits that do not map to a member within the congruence class. When LRU victim selection logic generates an output vector identifying a victim, the output vector is checked to ensure that it is a valid vector (non-null) and that it is not pointing to a deleted member. When the output vector is not valid or points to a deleted member, the LRU victim selection logic is triggered to re-start the victim selection process.

Abstract: A cache memory logically partitions a cache array into at least two slices each having a plurality of cache lines, with a given cache line spread across two or more cache ways of contiguous bytes and a given cache way shared between the two cache slices, and if one a cache way is defective that is part of a first cache line in the first cache slice and part of a second cache line in the second cache slice, it is disabled while continuing to use at least one other cache way which is also part of the first cache line and part of the second cache line. In the illustrative embodiment the cache array is set associative and at least two different cache ways for a given cache line contain different congruence classes for that cache line. The defective cache way can be disabled by preventing an eviction mechanism from allocating any congruence class in the defective way. For example, half of the cache line can be disabled (i.e., half of the congruence classes).

Abstract: In a cache coherent data processing system including at least first and second coherency domains, a master performs a first broadcast of an operation within the cache coherent data processing system that is limited in scope of transmission to the first coherency domain. The master receives a response of the first coherency domain to the first broadcast of the operation. If the response indicates the operation cannot be serviced in the first coherency domain alone, the master increases the scope of transmission by performing a second broadcast of the operation in both the first and second coherency domains. If the response indicates the operation can be serviced in the first coherency domain, the master refrains from performing the second broadcast, so that communication bandwidth utilized to service the operation is reduced.

Abstract: A method and apparatus for performing data prefetch in a multiprocessor system are disclosed. The multiprocessor system includes multiple processors, each having a cache memory. The cache memory is subdivided into multiple slices. A group of prefetch requests is initially issued by a requesting processor in the multiprocessor system. Each prefetch request is intended for one of the respective slices of the cache memory of the requesting processor. In response to the prefetch requests being missed in the cache memory of the requesting processor, the prefetch requests are merged into one combined prefetch request. The combined prefetch request is then sent to the cache memories of all the non-requesting processors within the multiprocessor system. In response to a combined clean response from the cache memories of all the non-requesting processors, data are then obtained for the combined prefetch request from a system memory.

Abstract: A data processing system includes a plurality of processing units each having a respective point-to-point communication link with each of multiple others of the plurality of processing units but fewer than all of the plurality of processing units. Each of the plurality of processing units includes interconnect logic, coupled to each point-to-point communication link of that processing unit, that broadcasts requests received from one of the multiple others of the plurality of processing units to one or more of the plurality of processing units. The interconnect logic includes a partial response data structure including a plurality of entries each associating a partial response field with a plurality of flags respectively associated with each processing unit containing a snooper from which that processing unit will receive a partial response.

Abstract: A cache coherent data processing system includes a memory and at least first and second coherency domains that each include a respective one of first and second cache memories. A master in the first coherency domain selects a scope of an initial broadcast of an operation targeting a request address allocated to the memory from among a first scope including only the first coherency domain and a second scope including both the first and second coherency domains. The master selects the scope based, at least in part, upon whether the memory belongs to the first coherency domain and performs an initial broadcast of the operation within the cache coherent data processing system utilizing the selected scope.

Abstract: A cache coherent data processing system includes at least first and second coherency domains each including at least one processing unit. The first coherency domain includes a first cache memory and a second cache memory, and the second coherency domain includes a remote coherent cache memory. The first cache memory includes a cache controller, a data array including a data storage location for caching a memory block, and a cache directory. The cache directory includes a tag field for storing an address tag in association with the memory block and a coherency state field associated with the tag field and the data storage location. The coherency state field has a plurality of possible states including a state that indicates that the memory block is possibly shared with the second cache memory in the first coherency domain and cached only within the first coherency domain.

Abstract: A cache coherent data processing system includes at least first and second coherency domains each including at least one processing unit and a cache memory. The cache memory includes a cache controller, a data array including a data storage location for caching a memory block, and a cache directory. The cache directory includes a tag field for storing an address tag in association with the memory block and a coherency state field associated with the tag field and the data storage location. The coherency state field has a plurality of possible states including a state that indicates that the address tag is valid, that the storage location does not contain valid data, and that the memory block is possibly cached outside of the first coherency domain.

Abstract: A cache coherent data processing system includes at least first and second coherency domains coupled for communication. The first and second coherency domains each include a respective one of first and second cache memories. A master in the first coherency domain selects a scope of an initial broadcast of an operation from among a first scope including only the first coherency domain and a second scope including both the first and second coherency domains based, at least in part, upon a type of the operation. The master then performs an initial broadcast of the operation within the cache coherent data processing system utilizing the selected scope.

Abstract: A system and method for cache management in a data processing system having a memory hierarchy of upper memory and lower memory cache. A lower memory cache controller accesses a coherency state table to determine replacement policies of coherency states for cache lines present in the lower memory cache when receiving a cast-in request from one of the upper memory caches. The coherency state table implements a replacement policy that retains the more valuable cache coherency state information between the upper and lower memory caches for a particular cache line contained in both levels of memory at the time of cast-out from the upper memory cache.

Abstract: In a cache coherent data processing system including at least first and second coherency domains, a memory block is stored in a system memory in association with a domain indicator indicating whether or not the memory block is cached, if at all, only within the first coherency domain. A master in the first coherency domain determines whether or not a scope of broadcast transmission of an operation should extend beyond the first coherency domain by reference to the domain indicator stored in the cache and then performs a broadcast of the operation within the cache coherent data processing system in accordance with the determination.

Abstract: A cache memory logically associates a cache line with at least two cache sectors of a cache array wherein different sectors have different output latencies and, for a load hit, selectively enables the cache sectors based on their latency to output the cache line over successive clock cycles. Larger wires having a higher transmission speed are preferably used to output the cache line corresponding to the requested memory block. In the illustrative embodiment the cache is arranged with rows and columns of the cache sectors, and a given cache line is spread across sectors in different columns, with at least one portion of the given cache line being located in a first column having a first latency, and another portion of the given cache line being located in a second column having a second latency greater than the first latency.

Abstract: A data processing system includes a plurality of processing units, including at least a local master and a local hub, which are coupled for communication via a communication link. The local master includes a master capable of initiating an operation, a snooper capable of receiving an operation, and interconnect logic coupled to a communication link coupling the local master to the local hub. The interconnect logic includes request logic that synchronizes internal transmission of a request of the master to the snooper with transmission, via the communication link, of the request to the local hub.

Abstract: A system and method for re-ordering store operations from a processor core to a store queue. When a store queue receives a new processor-issued store operation from the processor core, a store queue controller allocates a new entry in the store queue. In response to allocating the new entry in the store queue, the store queue controller determines whether or not the new entry is dependent on at least one other valid entry in the store queue. In response to determining the new entry is dependent on at least one other valid entry in the store queue, the store queue controller inhibits requesting of the new entry to the RC dispatch logic until each valid entry on which the new entry is dependent has been successfully dispatched to an RC machine by the RC dispatch logic.