Abstract: In one or more embodiments, a processor includes one or more circuits to flush instructions from an instruction pipeline on a selective basis responsive to detecting a branch misprediction, such that those instructions marked as being dependent on the branch instruction associated with the branch misprediction are flushed. Thus, the one or more circuits may be configured to mark instructions fetched into the processor's instruction pipeline(s) to indicate their branch prediction dependencies, directly or indirectly detect incorrect branch predictions, and directly or indirectly flush instructions in the instruction pipeline(s) that are marked as being dependent on an incorrect branch prediction.

Abstract: A Content-Terminated Direct Memory Access (CT-DMA) circuit autonomously transfers data of an unknown length from a source to a destination, terminating the transfer based on the content of the data. Filter criteria are provided to the CT-DMA prior to the data transfer. The filter criteria include pattern data that are compared to transfer data, and transfer termination rules for interpreting the comparison results. Data are written to the destination until the filter criteria are met. Representative filter criteria may include that one or more units of transfer data match pattern data; that one or more units of transfer data fail to match pattern data; or that one or more units of transfer data match pattern data a predetermined number of times.

Abstract: A system and method taught herein control data prefetching for a data cache by tracking prefetch hits and overall hits for the data cache. Data prefetching for the data cache is disabled based on the tracking of prefetch hits and data prefetching is enabled for the data cache based on the tracking of overall hits. For example, in one or more embodiments, a cache controller is configured to track a prefetch hit rate reflecting the percentage of hits on the data cache that involve prefetched data lines and disable data prefetching if the prefetch hit rate falls below a defined threshold. The cache controller also tracks an overall hit rate reflecting the overall percentage of data cache hits (versus misses) and enables data prefetching if the overall hit rate falls below a defined threshold.

Abstract: Efficient techniques are described for enforcing order of memory accesses. A memory access request is received from a device which is not configured to generate memory barrier commands. A surrogate barrier is generated in response to the memory access request. A memory access request may be a read request. In the case of a memory write request, the surrogate barrier is generated before the write request is processed. The surrogate barrier may also be generated in response to a memory read request conditional on a preceding write request to the same address as the read request. Coherency is enforced within a hierarchical memory system as if a memory barrier command was received from the device which does not produce memory barrier commands.

Abstract: A sliding-window, block-based Branch Target Address Cache (BTAC) comprises a plurality of entries, each entry associated with a block of instructions containing at least one branch instruction having been evaluated taken, and having a tag associated with the address of the first instruction in the block. The blocks each correspond to a group of instructions fetched from memory, such as an I-cache. Where a branch instruction is included in two or more fetch groups, it is also included in two or more instruction blocks associated with BTAC entries. The sliding-window, block-based BTAC allows for storing the Branch Target Address (BTA) of two or more taken branch instructions that fall in the same instruction block, without providing for multiple BTA storage space in each BTAC entry, by storing BTAC entries associated with different instruction blocks, each containing at least one of the taken branch instructions.

Abstract: A processing system may include a memory configured to store data in a plurality of pages, a TLB, and a memory cache including a plurality of cache lines. Each page in the memory may include a plurality of lines of memory. The memory cache may permit, when a virtual address is presented to the cache, a matching cache line to be identified from the plurality of cache lines, the matching cache line having a matching address that matches the virtual address. The memory cache may be configured to permit one or more page attributes of a page located at the matching address to be retrieved from the memory cache and not from the TLB, by further storing in each one of the cache lines a page attribute of the line of data stored in the cache line.

Abstract: Efficient techniques are described for tracking a potential invalidation of a data cache entry in a data cache for which coherency is required. Coherency information is received that indicates a potential invalidation of a data cache entry. The coherency information in association with the data cache entry is retained to track the potential invalidation to the data cache entry. The retained coherency information is kept separate from state bits that are utilized in cache access operations. An invalidate bit, associated with a data cache entry, may be utilized to represents a potential invalidation of the data cache entry. The invalidate bit is set in response to the coherency information, to track the potential invalidation of the data cache entry. A valid bit associated with the data cache entry is set in response to the active invalidate bit and a memory synchronization command. The set invalidate bit is cleared after the valid bit has been cleared.

Abstract: The search key and key fields of a CAM in a cache are encoded with a Hamming distance of at least two to increase the speed of the CAM by ensuring each mismatching match line is discharged by at least two transistors in parallel. Where the cache is physically tagged, the search key is a physical address. The page address portion of the physical address is encoded prior to being stored in a TLB. The page offset bits are encoded in parallel with the TLB access, and concatenated with the encoded TLB entry. If a page address addresses a large memory page size, a plurality of corresponding sub-page addresses may be generated, each addressing a smaller page size. These sub-page addresses may be encoded and stored in a micro TLB. The encoded key and key field are tolerant of single-bit soft errors.

Abstract: Embodiments include a cache controller adapted to determine whether a memory line for which the processor is to issue an address-only kill request resides in a fill buffer for the cache line in a shared state. If so, the cache controller may mark the fill buffer as not having completed bus transactions and issue the address-only kill request for that fill buffer. The address-only kill request may transmit to other processors on the bus and the other processors may respond by invalidating the cache entries for the memory line. Upon confirmation from the other processors, a bus arbiter may confirm the kill request, promoting the memory line already in that fill buffer to exclusive state. Once promoted, the fill buffer may be marked as having completed the bus transactions and may be written into the cache.

Abstract: Data from a source domain operating at a first data rate is transferred to a FIFO in another domain operating at a different data rate. The FIFO buffers data before transfer to a sink for further processing or storage. A source side counter tracks space available in the FIFO. In disclosed examples, the initial counter value corresponds to FIFO depth. The counter decrements in response to a data ready signal from the source domain, without delay. The counter increments in response to signaling from the sink domain of a read of data off the FIFO. Hence, incrementing is subject to the signaling latency between domains. The source may send one more beat of data when the counter indicates the FIFO is full. The last beat of data is continuously sent from the source until it is indicated that a FIFO position became available; effectively providing one more FIFO position.

Abstract: A processor having a multistage pipeline includes a TLB and a TLB controller. In response to a TLB miss signal, the TLB controller initiates a TLB reload, requesting address translation information from either a memory or a higher-level TLB, and placing that information into the TLB. The processor flushes the instruction having the missing virtual address, and refetches the instruction, resulting in re-insertion of the instruction at an initial stage of the pipeline above the TLB access point. The initiation of the TLB reload, and the flush/refetch of the instruction, are performed substantially in parallel, and without immediately stalling the pipeline. The refetched instruction is held at a point in the pipeline above the TLB access point until the TLB reload is complete, so that the refetched instruction generates a “hit” in the TLB upon its next access.

Abstract: A register file is disclosed. The register file includes a plurality of registers and a decoder. The decoder may be configured to receive an address for any one of the registers, and disable a read operation to the addressed register if data in the addressed register is invalid.

Abstract: Intermediate results are passed between constituent instructions of an expanded instruction using register renaming resources and control logic. A first constituent instruction generates intermediate results and is assigned a PRN in a constituent instruction rename table, and writes intermediate results to the identified physical register. A second constituent instruction performs a look up in the constituent instruction rename table and reads the intermediate results from the physical register. Constituent instruction rename logic tracks the constituent instructions through the pipeline, and delete the constituent instruction rename table entry and returns the PRN to a free list when the second constituent instruction has read the intermediate results.

Abstract: A method of resolving simultaneous branch predictions prior to validation of the predicted branch instruction is disclosed. The method includes processing two or more predicted branch instructions, with each predicted branch instruction having a predicted state and a corrected state. The method further includes selecting one of the corrected states. Should one of the predicted branch instructions be mispredicted, the selected corrected state is used to direct future instruction fetches.

Abstract: A method of managing cache partitions provides a first pointer for higher priority writes and a second pointer for lower priority writes, and uses the first pointer to delimit the lower priority writes. For example, locked writes have greater priority than unlocked writes, and a first pointer may be used for locked writes, and a second pointer may be used for unlocked writes. The first pointer is advanced responsive to making locked writes, and its advancement thus defines a locked region and an unlocked region. The second pointer is advanced responsive to making unlocked writes. The second pointer also is advanced (or retreated) as needed to prevent it from pointing to locations already traversed by the first pointer. Thus, the pointer delimits the unlocked region and allows the locked region to grow at the expense of the unlocked region.

Abstract: A processor pipeline is segmented into an upper portion—prior to instructions going out of program order—and one or more lower portions beyond the upper portion. The upper pipeline is flushed upon detecting that a branch instruction was mispredicted, minimizing the delay in fetching of instructions from the correct branch target address. The lower pipelines may continue execution until the mispredicted branch instruction confirms, at which time all uncommitted instructions are flushed from the lower pipelines. Existing exception pipeline flushing mechanisms may be utilized, by adding a mispredicted branch identifier, reducing the complexity and hardware cost of flushing the lower pipelines.

Abstract: A conditional instruction architected to receive one or more operands as inputs, to output to a target the result of an operation performed on the operands if a condition is satisfied, and to not provide an output if the condition is not satisfied, is executed so that it unconditionally provides an output to the target. The conditional instruction obtains the prior value of the target (that is, the value produced by the most recent instruction preceding the conditional instruction that updated that target). The condition is evaluated. If the condition is satisfied, an operation is performed and the result of the operation output to the target. If the condition is not satisfied, the prior value is output to the target. Subsequent instructions may rely on the target as an operand source (whether written to a register or forwarded to the instruction), prior to the condition evaluation.

Abstract: A method of resolving simultaneous branch predictions prior to validation of the predicted branch instruction is disclosed. The method includes processing two or more predicted branch instructions, with each predicted branch instruction having a predicted state and a corrected state. The method further includes selecting one of the corrected states. Should one of the predicted branch instructions be mispredicted, the selected corrected state is used to direct future instruction fetches.

Abstract: A processing system and method of communicating within the processing system is disclosed. The processing system may include a bus; a memory region coupled to the bus; and a plurality of processing components having access to the memory region over the bus, each of the processing components being configured to perform a semaphore operation to gain access to the memory region by simultaneously requesting a read operation and a write operation to a semaphore location over the bus.

Abstract: A processor includes a conditional branch instruction prediction mechanism that generates weighted branch prediction values. For weakly weighted predictions, which tend to be less accurate than strongly weighted predictions, the power associating with speculatively filling and subsequently flushing the cache is saved by halting instruction prefetching. Instruction fetching continues when the branch condition is evaluated in the pipeline and the actual next address is known. Alternatively, prefetching may continue out of a cache. To avoid displacing good cache data with instructions prefetched based on a mispredicted branch, prefetching may be halted in response to a weakly weighted prediction in the event of a cache miss.