Abstract: A technique for managing register assignments. The technique involves maintaining, in a register list memory circuit having entries that respectively correspond to physical registers, a list of register assignments that assign logical registers to the physical registers. The technique further involves maintaining, in a vector memory circuit having bits that respectively correspond to the physical registers, a valid vector that forms, in combination with the list of register assignments, a list of valid register assignments. Furthermore, the technique involves storing, for an instruction that is mapped by the data processor, a copy of the valid vector from the vector memory circuit to a silo memory circuit. Preferably, the processor using the technique has the ability to execute branches of instructions speculatively, and to recover if it is determined that the processor executed down an incorrect instruction branch.

Abstract: A system and method for transferring a data stream between devices having different clock domains. The method initiates a serial data stream between a transmitter and a receiver. The transmitter operates according to a first clock having a first clock rate, and the receiver operates according to a second clock having a second clock rate. A ratio between the second clock rate and the first clock rate is an integer number greater than or equal to one. A first state is provided over a serial line between the transmitter and the receiver. One or more start bits are provided over the serial line. The start bits indicate a second state different from the first state. One or more ratio bits are provided over the serial line after the start bit. The ratio bits indicate the ratio between the second clock rate and the first clock rate. The start bits are received. Using a transition between the first state and the second state evident in receiving each of the start bits, the ratio bits are received.

Abstract: A technique handles load instructions within a data processor that includes a cache circuit having a data cache and a tag memory indicating valid entries within the data cache. The technique involves writing data to the data cache during a series of four processor cycles in response to a first load instruction. Additionally, the technique involves updating the tag memory and preventing reading of the tag memory in response to the first load instruction during a first processor cycle in the series of processor cycles. Furthermore, the technique involves reading tag information from the tag memory during a processor cycle of the series of four processor cycles following the first processor cycle in response to a second load instruction.

Abstract: A bypass mechanism is disclosed for a computer system that executes load and store instructions out of order. The bypass mechanism compares the address of each issuing load instruction with a set of recent store instructions that have not yet updated memory. A match of the recent stores provides the load data instead of having to retrieve the data from memory. A store queue holds the recently issued stores. Each store queue entry and the issuing load includes a data size indicator. Subsequent to a data bypass, the data size indicator of the issuing load is compared against the data size indicator of the matching store queue entry. A trap is signaled when the data size indicator of the issuing load differs from the data size indicator of the matching store queue entry. The trap signal indicates that the data provided by the bypass mechanism was insufficient to satisfy the requirements of the load instruction.

Abstract: A processor employing a dependency link file. Upon detection of a load which hits a store for which store data is not available, the processor allocates an entry within the dependency link file for the load. The entry stores a load identifier identifying the load and a store data identifier identifying a source of the store data. The dependency link file monitors results generated by execution units within the processor to detect the store data being provided. The dependency link file then causes the store data to be forwarded as the load data in response to detecting that the store data is provided. The latency from store data being provided to the load data being forwarded may thereby be minimized. Particularly, the load data may be forwarded without requiring that the load memory operation be scheduled. Performance of the microprocessor may be increased due to the reduced load latency achievable in the above-mentioned cases.

Abstract: In a multiprocessing computer system, a cache-coherent data transfer scheme that also conserves the system memory bandwidth during a memory read operation is described. A source processing node sends a read command to a target processing node to read data from a designated memory location in a system memory associated with the target processing node. In response to the read command, the target processing node transmits a probe command to all the remaining processing nodes in the computer system regardless of whether one or more of the remaining nodes have a copy of the data cached in their respective cache memories. Probe command causes each node to maintain cache coherency by appropriately changing the state of the cache block containing the requested data and sending respective probe responses to the source node. Probe command also causes the node having an updated copy of the cache block to send the cache block to the source node through a read response.

Abstract: A processor employing a dependency link file. Upon detection of a load which hits a store for which store data is not available, the processor allocates an entry within the dependency link file for the load. The entry stores a load identifier identifying the load and a store data identifier identifying a source of the store data. The dependency link file monitors results generated by execution units within the processor to detect the store data being provided. The dependency link file then causes the store data to be forwarded as the load data in response to detecting that the store data is provided. The latency from store data being provided to the load data being forwarded may thereby be minimized. Particularly, the load data may be forwarded without requiring that the load memory operation be scheduled.

Abstract: A data caching system and method includes a data store for caching data from a main memory, a primary tag array for holding tags associated with data cached in the data store, and a duplicate tag array which holds copies of the tags held in the primary tag array. The duplicate tag array is accessible by functions, such as external memory cache probes, such that the primary tag remains available to the processor core. An address translator maps virtual page addresses to physical page address. In order to allow a data caching system which is larger than a page size, a portion of the virtual page address is used to index the tag arrays and data store. However, because of the virtual to physical mapping, the data may reside in any of a number of physical locations. During an internally-generated memory access, the virtual address is used to look up the cache. If there is a miss, other combinations of values are substituted for the virtual bits of the tag array index.

Abstract: A first and a second local interrupt controller are disposed on a single integrated circuit. The first and second local interrupt controllers are coupled to controllably provide at least one interrupt request signal, respectively, to a first and second processor. An input/output (I/O) interrupt controller is also on the integrated circuit and coupled to receive an interrupt request from at least one input/output device. A communication circuit on the integrated circuit is coupled to the input/output interrupt controller and the first and second local interrupt controllers. The communication circuit provides for transfer of interrupt information between the first local interrupt controller, the second local interrupt controller and the input/output interrupt controller.

Abstract: A floating point unit capable of executing multiple instructions in a single clock cycle using a central window and a register map is disclosed. The floating point unit comprises: a plurality of translation units, a future file, a central window, a plurality of functional units, a result queue, and a plurality of physical registers. The floating point unit receives speculative instructions, decodes them, and then stores them in the central window. Speculative top of stack values are generated for each instruction during decoding. Top of stack relative operands are computed to physical registers using a register map. Register stack exchange operations are performed during decoding. Instructions are then stored in the central window, which selects the oldest stored instructions to be issued to each functional pipeline and issues them. Conversion units convert the instruction's operands to an internal format, and normalization units detect and normalize any denormal operands.

Abstract: A technique for implementing load-locked and store-conditional instruction primitives by using a local cache for information about exclusive ownership. The valid bit in particular provides information to properly execute load-locked and store-conditional instructions without the need for lock flag or local lock address registers for each individual locked address. Integrity of locked data is accomplished by insuring that load-locked and store-conditional instructions are processed in order, that no internal agents can evict blocks from a local cache as a side effect as their processing, that external agents update the context of cache memories first using invalidating probe commands, and that only non-speculative instructions are permitted to generate external commands.

Abstract: A floating point unit capable of executing multiple instructions in a single clock cycle using a central window and a register map is disclosed. The floating point unit comprises: a plurality of translation units, a future file, a central window, a plurality of functional units, a result queue, and a plurality of physical registers. The floating point unit receives speculative instructions, decodes them, and then stores them in the central window. Speculative top of stack values are generated for each instruction during decoding. Top of stack relative operands are computed to physical registers using a register map. Register stack exchange operations are performed during decoding. Instructions are then stored in the central window, which selects the oldest stored instructions to be issued to each functional pipeline and issues them. Conversion units convert the instruction's operands to an internal format, and normalization units detect and normalize any denormal operands.

Abstract: A floating point unit capable of executing multiple instructions in a single clock cycle using a central window and a register map is disclosed. The floating point unit comprises: a plurality of translation units, a future file, a central window, a plurality of functional units, a result queue, and a plurality of physical registers. The floating point unit receives speculative instructions, decodes them, and then stores them in the central window. Speculative top of stack values are generated for each instruction during decoding. Top of stack relative operands are computed to physical registers using a register map. Register stack exchange operations are performed during decoding. Instructions are then stored in the central window, which selects the oldest stored instructions to be issued to each functional pipeline and issues them. Conversion units convert the instruction's operands to an internal format, and normalization units detect and normalize any denormal operands.

Abstract: A floating point unit capable of executing multiple instructions in a single clock cycle using a central window and a register map is disclosed. The floating point unit comprises: a plurality of translation units, a future file, a central window, a plurality of functional units, a result queue, and a plurality of physical registers. The floating point unit receives speculative instructions, decodes them, and then stores them in the central window. Speculative top of stack values are generated for each instruction during decoding. Top of stack relative operands are computed to physical registers using a register map. Register stack exchange operations are performed during decoding. Instructions are then stored in the central window, which selects the oldest stored instructions to be issued to each functional pipeline and issues them. Conversion units convert the instruction's operands to an internal format, and normalization units detect and normalize any denormal operands.

Abstract: Use of an internal processor data bus is maximized in a system where external transactions may occur at a rate which is fractionally slower than the rate of the internal transactions. The technique inserts a selectable delay element in the signal path during an external operation such as a cache fill operation. The one cycle delay provides a time slot in which an internal operation, such as a load from an internal cache, may be performed. This technique therefore permits full use of the time slots on the internal data bus. It can, for, example, allow load operations to begin at a much earlier time than would otherwise be possible in architectures where fill operations can consume multiple bus time slots.

Abstract: A pipelined CPU executing instructions of variable length, and referencing memory using various data widths. Macroinstruction pipelining is employed (instead of microinstruction pipelining), with queueing between units of the CPU to allow flexibility in instruction execution times. A wide bandwidth is available for memory access; fetching 64-bit data blocks on each cycle. A hierarchical cache arrangement has an improved method of cache set selection, increasing the likelihood of a cache hit. A writeback cache is used (instead of writethrough) and writeback is allowed to proceed even though other accesses are suppressed due to queues being full. A branch prediction method employs a branch history table which records the taken vs. not-taken history of branch opcodes recently used, and uses an empirical algorithm to predict which way the next occurrence of this branch will go, based upon the history table.