Abstract: Embodiments of the present invention provide a hardware accelerator that assists a host database system in processing its queries. The hardware accelerator comprises special purpose processing elements that are capable of receiving database query/operation tasks in the form of machine code database instructions, execute them in hardware without software, and return the query/operation result back to the host system.

Abstract: Embodiments of the present invention provide a hardware accelerator that assists a host database system in processing its queries. The hardware accelerator comprises special purpose processing elements that are capable of receiving database query/operation tasks in the form of machine code database instructions, execute them in hardware without software, and return the query/operation result back to the host system.

Abstract: Embodiments of the present invention provide a hardware accelerator that assists a host database system in processing its queries. The hardware accelerator comprises special purpose processing elements that are capable of receiving database query/operation tasks in the form of machine code database instructions, execute them in hardware without software, and return the query/operation result back to the host system.

Abstract: Embodiments of the present invention provide a hardware accelerator that assists a host database system in processing its queries. The hardware accelerator comprises special purpose processing elements that are capable of receiving database query/operation tasks in the form of machine code database instructions, execute them in hardware without software, and return the query/operation result back to the host system.

Abstract: Switching data packets includes forwarding first data packets to a first port that is associated with a first domain. Second data packets are forwarded to a second port that is associated with a second domain. The first data packets are managed using a first management subsystem of a service processor card, and the second data packets are managed using a second management subsystem of the service processor card.

Abstract: A multiple-way cache memory having a plurality of cache blocks and associated tag arrays includes a select circuit that stores way select values for each cache block. The way select values selectively disable one or more cache blocks from participating in cache operations by forcing tag comparisons associated with the disabled cache blocks to a mismatch condition so that the disabled cache blocks will not be selected to provide output data. The remaining enabled cache blocks may be operated as a less-associative cache memory without requiring cache addressing modifications.

Abstract: The resources of a partitioned cache memory are dynamically allocated between two or more processors on a multi-processor unit (MPU). In one embodiment, the MPU includes first and second processors, and the cache memory includes first and second partitions. A cache access circuit selectively transfers data between the cache memory partitions to maximize cache resources. In one mode, both processors are active and may simultaneously execute separate instruction threads. In this mode, the cache access circuit allocates the first cache memory partition as dedicated cache memory for the first processor, and allocates the second cache memory partition as dedicated cache memory for the second processor. In another mode, one processor is active, and the other processor is inactive. In this mode, the cache access circuit allocates both the first and second cache memory partitions as cache memory for the active processor.

Abstract: A multiple-way cache memory having a plurality of cache blocks and associated tag arrays includes a select circuit that stores way select values for each cache block. The way select values selectively disable one or more cache blocks from participating in cache operations by forcing tag comparisons associated with the disabled cache blocks to a mismatch condition so that the disabled cache blocks will not be selected to provide output data. The remaining enabled cache blocks may be operated as a less-associative cache memory without requiring cache addressing modifications.

Abstract: A single-chip central processing unit (CPU) includes a processing core and a complete cache-coherent I/O system that operates asynchronously with the processing core. An internal communications protocol uses synchronizers and data buffers to transfer information between a clock domain of the processing core and a clock domain of the I/O system. The synchronizers transfer control and handshake signal between clock domains, but the data buffer transfers data without input or output synchronization circuitry for data bits. Throughput for the system is high because the processing unit has direct access to I/O system so that no delays are incurred for complex mechanisms which are commonly employed between a CPU and an external I/O chip-set. Throughput is further increased by holding data from one DMA transfer in the data buffer for use in a subsequent DMA transfer.

Abstract: A memory controller for a memory subsystem of a computer system connects to a processor bus. The memory controller is for use with memory devices such as RDRAM or DDR SDRAM that allow for multiple open pages. Memory references are remapped by an address mapper and processed by a page tracking buffer to keep track of open pages in the memory devices. The controller also has a state machine, and an interface to memory devices. The page tracking buffer has a row address content addressable memory for determining when a reference is in an open page, and a bank content addressable memory for determining when a reference is to the same bank as an open page. The controller closes open pages of a bank prior to opening new pages in that bank. The page tracking buffer has fewer lines than the product of the maximum number of memory devices times the maximum number of simultaneously open pages of each device, but provides for tracking any page of any of the memory devices.

Abstract: A multiprocessor system includes a plurality of central processing units (CPUs) connected to one another by a system bus. Each CPU includes a cache controller to communicate with its cache, and a primary memory controller to communicate with its primary memory. When there is a cache miss in a CPU, the cache controller routes an address request for primary memory directly to the primary memory via the CPU as a speculative request without access the system bus, and also issues the address request to the system bus to facilitate data coherency. The speculative request is queued in the primary memory controller, which in turn retrieves speculative data from a specified primary memory address. The CPU monitors the system bus for a subsequent transaction that requests the specified data in the primary memory. If the subsequent transaction requesting the specified data is a read transaction that corresponds to the speculative address request, the speculative request is validated and becomes non-speculative.

Abstract: The resources of a partitioned cache memory are dynamically allocated between two or more processors on a multi-processor unit (MPU). In one embodiment, the MPU includes first and second processors, and the cache memory includes first and second partitions. A cache access circuit selectively transfers data between the cache memory partitions to maximize cache resources. In one mode, both processors are active and may simultaneously execute separate instruction threads. In this mode, the cache access circuit allocates the first cache memory partition as dedicated cache memory for the first processor, and allocates the second cache memory partition as dedicated cache memory for the second processor. In another mode, one processor is active, and the other processor is inactive. In this mode, the cache access circuit allocates both the first and second cache memory partitions as cache memory for the active processor.

Abstract: The main cache of a processor in a multiprocessor computing system is coupled to receive writeback data during writeback operations. In one embodiment, during writeback operations, e.g., for a cache miss, dirty data in the main cache is merged with modified data from an associated write cache, and the resultant writeback data line is loaded into a writeback buffer. The writeback data is also written back into the main cache, and is maintained in the main cache until replaced by new data. Subsequent requests (i.e., snoops) for the data are then serviced from the main cache, rather than from the writeback buffer. In some embodiments, further modifications of the writeback data in the main cache are prevented. The writeback data line in the main cache remains valid until read data for the cache miss is returned, thereby ensuring that the read address reaches the system interface for proper bus ordering before the writeback line is lost.

Abstract: An instruction fetch unit for fetching instructions from an instruction cache of a processor. The fetch unit includes a next fetch address mechanism generating predicted next fetch addresses, the next fetch address mechanism generating a next fetch address for a fetch bundle over at least two cycles of the processor. The next fetch address mechanism determines the next fetch address based on whether a control transfer instruction from an intermediate set of fetched instructions is taken.

Abstract: A method and apparatus for performing multiple branch predictions per cycle is disclosed. The method and apparatus according to the present invention determine, within one fetch cycle, which instructions in a plurality of fetch instructions are branches and whether such branches are taken or not taken thereby finding the oldest taken branch, which has a target address that is fetched within the same fetch cycle.

Abstract: In a superscalar processor, multiple instructions are executed in parallel to obtain multiple execution results, and the multiple execution results are stored in a working register file. Each execution result in the working register file has at least one status bit associated therewith which identifies the execution result as valid data. The multiple execution results contained in the working register data then retired by changing the status bits associated with each execution result to identify the execution result as final data. In this manner, the speculative data is retired as the final data without data movement of the speculative data, thus reducing a number of ports needed in the superscalar processor.

Abstract: A method for repairing a pipeline in response to a branch instruction having a branch, includes the steps of providing a branch repair table having a plurality of entries, allocating an entry in the branch repair table for the branch instruction, storing a target address, a fall-through address, and repair information in the entry in the branch repair table, processing the branch instruction to determine whether the branch was taken, and repairing the pipeline in response to the repair information and the fall-through address in the entry in the branch repair table when the branch was not taken.

Abstract: Two-way set associative data is stored in a cache memory array. An odd set data bank stores odd number sets of the two-way set associative data, where the two ways of each odd number set are aligned horizontally within the odd set data bank. An even set data bank stores even number sets of the two-way set associative data, where the two ways of each even number set are aligned horizontally within the even set data bank. Also, the odd set data bank is aligned horizontally with the even set data bank such that each odd number set is aligned horizontally with a next even number set. The horizontally aligned ways are interleaved for data path width reduction. Set and way selection circuits extract lines of data from the array. The array may be structurally implemented by single-ported RAM cells.

Abstract: One embodiment of the present invention provides a system for predicting an address of an instruction following a branch instruction in a computer instruction stream. This system concurrently performs a fast single-cycle branch prediction operation to produce a first predicted address, and a more-accurate multiple-cycle branch prediction operation to produce a second predicted address. The system assumes that the first predicted address is correct and proceeds with a subsequent instruction fetch operation using the first predicted address. If the first predicted address is the same as the second predicted address, the subsequent instruction fetch operation is allowed to proceed using the first predicted address. Otherwise, the subsequent fetch operation is delayed so that it can proceed using the second predicted address.

Abstract: One embodiment of the present invention provides a system for predicting an address of an instruction following a branch instruction in a computer instruction stream. This system receives a current address specifying an address of a current instruction. It uses this current address (or possibly a preceding address) to generate a first select signal, which is used to select a first predicted address of an instruction following the current instruction in the computer instruction stream. At the same time the system generates a second select signal, which takes more time to generate than the first select signal but achieves a more accurate selection for a predicted address of the instruction following the current instruction. The system assumes that the first predicted address is correct and proceeds with a subsequent instruction fetch operation using the first predicted address. Next, the system compares the first select signal with the second select signal.