Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A variable latency cache memory is disclosed. A cache subsystem includes a pipeline control circuit configured to initiate cache memory accesses for data. The cache subsystem further includes a cache memory circuit having a data array arranged into a plurality of groups, wherein different ones of the plurality of groups have different minimum access latencies due to different distances from the pipeline control circuit. A plurality of latency control circuits configured to ensure a latency is bounded to a maximum value for a given access to the data array, wherein a given latency control circuit is associated with a corresponding group of the plurality of groups. The latency for a given access may thus vary between a minimum access latency for a group closest to the pipeline control circuit to a maximum latency for an access to the group furthest from the pipeline control circuit.

Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A cache may include multiple request handling pipes, each of which may further include multiple request buffers, for storing device requests from one or more processors to one or more devices. Some of the device requests may require to be sent to the devices according to an order. For a given one of such device requests, the cache may select a request handling pipe, based on an address indicated by the device request, and select a request buffer, based on the available entries of the request buffers of the selected request handling pipe, to store the device request. The cache may further use a first-level and a second-level token stores to track and maintain the device requests in order when transmitting the device requests to the devices.

Abstract: A system may include a plurality of processors and a coprocessor. A plurality of coprocessor context priority registers corresponding to a plurality of contexts supported by the coprocessor may be included. The plurality of processors may use the plurality of contexts, and may program the coprocessor context priority register corresponding to a context with a value specifying a priority of the context relative to other contexts. An arbiter may arbitrate among instructions issued by the plurality of processors based on the priorities in the plurality of coprocessor context priority registers. In one embodiment, real-time threads may be assigned higher priorities than bulk processing tasks, improving bandwidth allocated to the real-time threads as compared to the bulk tasks.

Abstract: A cache may include multiple request handling pipes, each of which may further include multiple request buffers, for storing device requests from one or more processors to one or more devices. Some of the device requests may require to be sent to the devices according to an order. For a given one of such device requests, the cache may select a request handling pipe, based on an address indicated by the device request, and select a request buffer, based on the available entries of the request buffers of the selected request handling pipe, to store the device request. The cache may further use a first-level and a second-level token stores to track and maintain the device requests in order when transmitting the device requests to the devices.

Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A system may include a plurality of processors and a coprocessor. A plurality of coprocessor context priority registers corresponding to a plurality of contexts supported by the coprocessor may be included. The plurality of processors may use the plurality of contexts, and may program the coprocessor context priority register corresponding to a context with a value specifying a priority of the context relative to other contexts. An arbiter may arbitrate among instructions issued by the plurality of processors based on the priorities in the plurality of coprocessor context priority registers. In one embodiment, real-time threads may be assigned higher priorities than bulk processing tasks, improving bandwidth allocated to the real-time threads as compared to the bulk tasks.

Abstract: A scalable cache coherency protocol for system including a plurality of coherent agents coupled to one or more memory controllers is described. The memory controller may implement a precise directory for cache blocks from the memory to which the memory controller is coupled. Multiple requests to a cache block may be outstanding, and snoops and completions for requests may include an expected cache state at the receiving agent, as indicated by a directory in the memory controller when the request was processed, to allow the receiving agent to detect race conditions. In an embodiment, the cache states may include a primary shared and a secondary shared state. The primary shared state may apply to a coherent agent that bears responsibility for transmitting a copy of the cache block to a requesting agent. In an embodiment, at least two types of snoops may be supported: snoop forward and snoop back.

Abstract: A system may include a plurality of processors and a coprocessor. A plurality of coprocessor context priority registers corresponding to a plurality of contexts supported by the coprocessor may be included. The plurality of processors may use the plurality of contexts, and may program the coprocessor context priority register corresponding to a context with a value specifying a priority of the context relative to other contexts. An arbiter may arbitrate among instructions issued by the plurality of processors based on the priorities in the plurality of coprocessor context priority registers. In one embodiment, real-time threads may be assigned higher priorities than bulk processing tasks, improving bandwidth allocated to the real-time threads as compared to the bulk tasks.

Abstract: Systems, apparatuses, and methods for performing coherence processing and memory cache processing in parallel are disclosed. A system includes a communication fabric and a plurality of dual-processing pipelines. Each dual-processing pipeline includes a coherence processing pipeline and a memory cache processing pipeline. The communication fabric forwards a transaction to a given dual-processing pipeline, with the communication fabric selecting the given dual-processing pipeline, from the plurality of dual-processing pipelines, based on a hash of the address of the transaction. The given dual-processing pipeline performs a duplicate tag lookup in parallel with a memory cache tag lookup for the transaction. By performing the duplicate tag lookup and the memory cache tag lookup in a parallel fashion rather than in a serial fashion, latency and power consumption are reduced while performance is enhanced.

Abstract: Various systems and methods for ensuring real-time snoop latency are disclosed. A system includes a processor and a cache controller. The cache controller receives, via a channel, cache snoop requests from the processor, the snoop requests including latency-sensitive and non-latency sensitive requests. Requests are not prioritized by type within the channel. The cache controller limits a number of non-latency sensitive snoop requests that can be processed ahead of an incoming latency-sensitive snoop requests. Limiting the number of non-latency sensitive snoop requests that can be processed ahead of an incoming latency-sensitive snoop request includes the cache controller determining that the number of received non-latency sensitive snoop requests has reached a predetermined value and responsively prioritizing latency-sensitive requests over non-latency sensitive requests.

Abstract: Systems, apparatuses, and methods for performing coherence processing and memory cache processing in parallel are disclosed. A system includes a communication fabric and a plurality of dual-processing pipelines. Each dual-processing pipeline includes a coherence processing pipeline and a memory cache processing pipeline. The communication fabric forwards a transaction to a given dual-processing pipeline, with the communication fabric selecting the given dual-processing pipeline, from the plurality of dual-processing pipelines, based on a hash of the address of the transaction. The given dual-processing pipeline performs a duplicate tag lookup in parallel with a memory cache tag lookup for the transaction. By performing the duplicate tag lookup and the memory cache tag lookup in a parallel fashion rather than in a serial fashion, latency and power consumption are reduced while performance is enhanced.

Abstract: A method and apparatus for selectively powering down a portion of a cache memory includes determining a power down condition dependent upon a number of accesses to the cache memory. In response to the detection of the power down condition, selecting a group of cache ways included in the cache memory dependent upon a number of cache lines in each cache way that are also included in another cache memory. The method further includes locking and flushing the selected group of cache ways, and then activating a low power mode for the selected group of cache ways.

Abstract: A mechanism for evicting a cache line from a cache memory includes first selecting for eviction a least recently used cache line of a group of invalid cache lines. If all cache lines are valid, selecting for eviction a least recently used cache line of a group of cache lines in which no cache line of the group of cache lines is also stored within a higher level cache memory such as the L1 cache, for example. Lastly, if all cache lines are valid and there are no non-inclusive cache lines, selecting for eviction the least recently used cache line stored in the cache memory.

Abstract: A method and apparatus for evicting cache lines from a cache memory includes receiving a request from one of a plurality of processors. The cache memory is configured to store a plurality of cache lines, and a given cache line includes an identifier indicating a processor that performed a most recent access of the given cache line. The method further includes selecting a cache line for eviction from a group of least recently used cache lines, where each cache line of the group of least recently used cache lines occupy a priority position less that a predetermined value, and then evicting the selected cache line.

Abstract: Systems and methods for reducing leakage power in a L2 cache within a SoC. The L2 cache is partitioned into multiple banks, and each bank has its own separate power supply. An idle counter is maintained for each bank to count a number of cycles during which the bank has been inactive. The temperature and leaky factor of the SoC are used to select an operating point of the SoC. Based on the operating point, an idle counter threshold is set, with a high temperature and high leaky factor corresponding to a relatively low idle counter threshold, and with a low temperature and low leaky factor corresponding to a relatively high idle counter threshold. When a given idle counter exceeds the idle counter threshold, the voltage supplied to the corresponding bank is reduced to a voltage sufficient for retention of data but not for access.

Abstract: An apparatus for processing cache requests in a computing system is disclosed. The apparatus may include a pending request buffer and a control circuit. The pending request buffer may include a plurality of buffer entries. The control circuit may be coupled to the pending request buffer and may be configured to receive a request for a first cache line from a pre-fetch engine, and store the received request in an entry of the pending request buffer. The control circuit may be further configured to receive a request for a second cache line from a processor, and store the request received from the processor in the entry of the pending request buffer in response to a determination that the second cache line is the same as the first cache line.