Abstract: An apparatus includes one or more interface circuits, an interconnect, a memory controller, a memory bridge, a packet DMA circuit, and a switch. The memory controller, the memory bridge, and the packet DMA circuit are coupled to the interconnect. Each interface circuit is coupled to a respective interface to receive packets and/or coherency commands from the interface. The switch is coupled to the interface circuits, the memory bridge, and the packet DMA circuit. The switch is configured to route the coherency commands from the interface circuits to the memory bridge and the packets from the interface circuits to the packet DMA circuit. The memory bridge is configured to initiate corresponding transactions on the interconnect in response to at least some of the coherency commands. The packet DMA circuit is configured to transmit write transactions on the interconnect to the memory controller to store the packets in memory.

Abstract: A random number generator circuit includes a primary circuit configured to generate a value within a first range and a secondary circuit configured to generate a value within a second range. A detector circuit detects whether or not the value from the primary circuit is within the desired output range for the random number generator circuit, and selects either the value from the primary circuit or the value from the secondary circuit in response. The second range is the desired output range and the first range encompasses the second range. In one embodiment, the primary circuit has complex harmonics but may generate values outside the desired range. The secondary circuit may have less complex harmonics, but may generate values only within the desired range. In one implementation, the random number generator circuit is used to generate a replacement way for a cache.

Abstract: A system may include two or more agents, at least some of which may cache data. In response to a read transaction, a caching agent may snoop its cached data and provide a response in a response phase of the transaction. Particularly, the response may include an exclusive indication used to represent both exclusive and modified states within that agent. In one embodiment, the agent responding exclusive may be responsible for providing the data for a read transaction, and may transmit an indication of which of the exclusive or modified state that agent had the data in concurrent with transmitting the data.

Abstract: A mask generator circuit includes at least first and second mask generator circuits coupled to receive most significant and least significant sections of the pointer and to generate masks therefrom, and a plurality of circuits each configured to generate a region of the output mask from the mask generator circuit. The mask generated from the most significant bits section of the pointer (the most significant bits (MSB) mask) includes bits corresponding to various regions of the output mask. The plurality of circuits receive the MSB mask and the least significant bits (LSB) mask generated from the least significant bits section of the pointer and generate the output mask therefrom.

Abstract: A cache is configured to receive direct access transactions. Each direct access transaction explicitly specifies a cache storage entry to be accessed in response to the transaction. The cache may access the cache storage entry (bypassing the normal tag comparisons and hit determination used for memory transactions) and either read the data from the cache storage entry (for read transactions) or write data from the transaction to the cache storage entry (for write transactions). The direct access transactions may, for example, be used to perform testing of the cache memory. As another example, direct access transactions may be used to perform a reset of the cache (by writing known data to each cache entry). In embodiments employing error checking and correction (ECC) mechanisms, direct access write transactions could also be used to recover from uncorrectable ECC errors, by overwriting the failing data to eliminate the errant data.

Abstract: A processor supports instruction pointer (IP) relative addressing in at least one operating mode of the processor. For example, in some implementations, IP relative addressing is supported in an operating mode or modes in which the address size is greater than 32 bits (e.g. up to 64 bits). In some embodiments, the displacement may be limited to less than the address size (e.g. 32 bits, in one implementation) when such operating modes are active. Code density may be higher than if the displacements were expanded, and flexibility in the placement of variables in memory may be achieved. For example, static variables may be placed in memory with flexibility, and IP relative addressing may be used to locate the static variables.

Abstract: A messaging scheme that conserves system memory bandwidth during a memory read operation in a multiprocessing computer system is described. A source processing node sends a memory 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. The target node transmits a read response to the source node containing the requested data and also concurrently transmits a probe command to one or more of the remaining nodes in the multiprocessing computer system. In response to the probe command each remaining processing node checks whether the processing node has a cached copy of the requested data. If a processing node, other than the source and the target nodes, finds a modified cached copy of the designated memory location, that processing node responds with a memory cancel response sent to the target node and a read response sent to the source node.

Abstract: An indirect branch predictor includes a buffer storing branch target addresses corresponding to previously executed indirect branch instructions. The buffer is indexed with an index derived from history information corresponding to previously predicted indirect branch instructions and from the PC of the particular indirect branch instruction being predicted. In one embodiment, the buffer may be tagless and/or direct mapped. In various embodiments, the indirect branch target predictor may generate the index to the buffer using one or more techniques to improve the accuracy of the prediction: (i) offsetting the history information from the various previously predicted indirect branch instructions; (ii) weighting the history information based on the age of the previously predicted indirect branch instructions; and/or (iii) reversing the bit order of the PC of the particular indirect branch instruction being predicted.

Abstract: An apparatus is contemplated, including a router and a memory controller. The router is configured to route a write request and write data to the memory controller. The memory controller is coupled to receive the write request and the write data. If the write data is a number of bytes less than a minimum number of bytes writeable to a memory to which the memory controller is capable of being coupled, the memory controller is configured to read first data from the memory. The first data comprises the minimum number of bytes and includes bytes to be updated with the write data. The memory controller is configured to return the first data to the router as a read response. The router is configured to return the first data to the memory controller.

Abstract: An apparatus is disclosed which includes a converter circuit and a noise suppression circuit. The converter circuit has a dynamic logic input, and is configured to generate a static logic output on an output node responsive to the dynamic logic input. The noise suppression circuit is coupled to receive a clock signal and is coupled to the output node. Responsive to a first phase of the clock signal, a precharge of a dynamic logic circuit generating the dynamic logic input occurs. The noise suppression circuit is configured to actively drive the static logic output on the output node responsive to the first phase. In some embodiments, the noise suppression circuit may reduce the noise sensitivity of the static logic output during the precharge phase, and may not impede operation of the converter circuit during the evaluate phase.

Abstract: A method and apparatus for improving the performance of a superscalar, superpipelined processor by identifying and processing instructions for performing addressing operations is provided. The invention heuristically determines instructions likely to perform addressing operations and assigns those instructions to specialized pipes in a pipeline structure. The invention can assign such instructions to both an execute pipe and a load/store pipe to avoid the occurrence of “bubbles” in the event execution of the instruction requires the calculation capability of the execute pipe. The invention can also examine a sequence of instructions to identify an instruction for performing a calculation where the result of the calculation is used by a succeeding load or store instruction. In this case, the invention controls the pipeline to assure the result of the calculation is available for the succeeding load or store instruction even if both instructions are being processed concurrently.

Abstract: A computer system is presented which implements a system and method for conveying packets between a coherent processing subsystem and a non-coherent input/output (I/O) subsystem. The processing subsystem includes a first processing node coupled to a second processing node via a coherent communication link. The first processing node includes a host bridge which translates packets moving between the processing subsystem and the I/O subsystem. The I/O subsystem includes an I/O node coupled to the first processing node via a non-coherent communication link. The I/O node may embody one or more I/O functions (e.g., modem, sound card, etc.). The coherent and non-coherent communication links are physically identical. For example, the coherent and non-coherent communication links may have the same electrical interface and the same signal definition. The host bridge translates non-coherent packets from the I/O node to coherent packets, and transmits the coherent packets to the second processing node.

Abstract: An apparatus is described in which the locations of a buffer are used to store a plurality of control packets received in a node, wherein the plurality of control packets belong to a plurality of virtual channels. The number of locations assigned to each virtual channel may be dynamically allocated. The number of locations allocated to each virtual channel may be determined by an update circuit. Count values corresponding to the number of locations allocated to each virtual channel may then be stored within a programmable storage, such as a register, for example. The count values may be subsequently copied into a slave register and incremented and decremented as locations become available and notifications corresponding to the available locations are sent, respectively.

Abstract: A processor supports a first processing mode in which the address size is greater than 32 bits. The address size may be nominally indicated as 64 bits, although various embodiments of the processor may implement any address size which exceeds 32 bits, up to and including 64 bits, in the first processing mode. The first processing mode may be established by placing an enable indication in a control register into an enabled state and by setting a first operating mode indication and a second operating mode indication in a segment descriptor to predefined states. Other combinations of the first operating mode indication and the second operating mode indication may be used to provide compatibility modes for 32 bit and 16 bit processing compatible with the x86 processor architecture (with the enable indication remaining in the enabled state). To call code operating in the first processing mode from the 32 bit or 16 bit code, a call gate descriptor is defined which occupies two entries in a segment descriptor table.

Abstract: A processor includes execution resources for handling a first memory operation and a concurrent second memory operation. If one of the memory operations is misaligned, the processor may allocate the execution resources for the other memory operation to that memory operation. In one embodiment, the older memory operation proceeds if misalignment is detected. The younger memory operation is retried and may be reexecuted at a later time. If the older memory operation is misaligned, the execution resources provided for the younger operation may be allocated to the older memory operation. If only the younger memory operation is misaligned, the younger memory operation may be the older memory operation during a subsequent reexecution and may thus be allocated the execution resources to allow the memory operation to complete.

Abstract: A cache is configured to select a cache block for eviction in response to detecting a cache miss. The cache transmits the address of the cache block as a write transaction on an interface to the cache, and the cache captures the address from the interface and reads the cache block from the cache memory in response to the address. The read may occur similar to other reads in the cache, detecting a hit in the cache (in the cache storage location from which the cache block is being evicted). The write transaction is initiated before the corresponding data is available for transfer, and the use of the bus bandwidth to initiate the transaction provides an open access time into the cache for reading the evicted cache block.

Abstract: A processor employs a store to load forward (STLF) predictor which may indicate, for dispatching loads, a dependency on a store. The dependency is indicated for a store which, during a previous execution, interfered with the execution of the load. Since a dependency is indicated on the store, the load is prevented from scheduling and/or executing prior to the store. The STLF predictor is trained with information for a particular load and store in response to executing the load and store and detecting the interference. Additionally, the STLF predictor may be untrained (e.g. information for a particular load and store may be deleted) if a load is indicated by the STLF predictor as dependent upon a particular store and the dependency does not actually occur. In one implementation, the STLF predictor records at least a portion of the PC of a store which interferes with the load in a first table indexed by the load PC.