摘要:
Embodiments that dynamically conserve power in non-uniform cache access (NUCA) caches are contemplated. Various embodiments comprise a computing device, having one or more processors coupled with one or more NUCA cache elements. The NUCA cache elements may comprise one or more banks of cache memory, wherein ways of the cache are vertically distributed across multiple banks. To conserve power, the computing devices generally turn off groups of banks, in a sequential manner according to different power states, based on the access latencies of the banks. The computing devices may first turn off groups having the greatest access latencies. The computing devices may conserve additional power by turning of more groups of banks according to different power states, continuing to turn off groups with larger access latencies before turning off groups with the smaller access latencies.
摘要:
Embodiments that dynamically reorganize data of cache lines in non-uniform cache access (NUCA) caches are contemplated. Various embodiments comprise a computing device, having one or more processors coupled with one or more NUCA cache elements. The NUCA cache elements may comprise one or more banks of cache memory, wherein ways of the cache are horizontally distributed across multiple banks. To improve access latency of the data by the processors, the computing devices may dynamically propagate cache lines into banks closer to the processors using the cache lines. To accomplish such dynamic reorganization, embodiments may maintain “direction” bits for cache lines. The direction bits may indicate to which processor the data should be moved. Further, embodiments may use the direction bits to make cache line movement decisions.
摘要:
Embodiments of the present invention address deficiencies of the art in respect to memory management and provide a method, system and computer program product for dynamic optimization of DRAM controller page policy. In one embodiment of the invention, a memory module can include multiple different memories, each including a memory controller coupled to a memory array of memory pages. Each of the memory pages in turn can include a corresponding locality tendency state. A memory bank can be coupled to a sense amplifier and configured to latch selected ones of the memory pages responsive to the memory controller. Finally, the module can include open page policy management logic coupled to the memory controller. The logic can include program code enabled to granularly change open page policy management of the memory bank responsive to identifying a locality tendency state for a page loaded in the memory bank.
摘要:
Embodiments that that distribute replacement policy bits and operate the bits in cache memories, such as non-uniform cache access (NUCA) caches, are contemplated. An embodiment may comprise a computing device, such as a computer having multiple processors or multiple cores, which has cache memory elements coupled with the multiple processors or cores. The cache memory device may track usage of cache lines by using a number of bits. For example, a controller of the cache memory may manipulate bits as part of a pseudo least recently used (LRU) system. Some of the bits may be in a centralized area of the cache. Other bits of the pseudo LRU system may be distributed across the cache. Distributing the bits across the cache may enable the system to conserve additional power by turning off the distributed bits.
摘要:
Embodiments that dynamically conserve power in non-uniform cache access (NUCA) caches are contemplated. Various embodiments comprise a computing device, having one or more processors coupled with one or more NUCA cache elements. The NUCA cache elements may comprise one or more banks of cache memory, wherein ways of the cache are vertically distributed across multiple banks. To conserve power, the computing devices generally turn off groups of banks, in a sequential manner according to different power states, based on the access latencies of the banks. The computing devices may first turn off groups having the greatest access latencies. The computing devices may conserve additional power by turning of more groups of banks according to different power states, continuing to turn off groups with larger access latencies before turning off groups with the smaller access latencies.
摘要:
Embodiments that that distribute replacement policy bits and operate the bits in cache memories, such as non-uniform cache access (NUCA) caches, are contemplated. An embodiment may comprise a computing device, such as a computer having multiple processors or multiple cores, which has cache memory elements coupled with the multiple processors or cores. The cache memory device may track usage of cache lines by using a number of bits. For example, a controller of the cache memory may manipulate bits as part of a pseudo least recently used (LRU) system. Some of the bits may be in a centralized area of the cache. Other bits of the pseudo LRU system may be distributed across the cache. Distributing the bits across the cache may enable the system to conserve additional power by turning off the distributed bits.
摘要:
A method and a system for utilizing less recently used (LRU) bits and presence bits in selecting cache-lines for eviction from a lower level cache in a processor-memory sub-system. A cache back invalidation (CBI) logic utilizes LRU bits to evict only cache-lines within a LRU group, following a cache miss in the lower level cache. In addition, the CBI logic uses presence bits to (a) indicate whether a cache-line in a lower level cache is also present in a higher level cache and (b) evict only cache-lines in the lower level cache that are not present in a corresponding higher level cache. However, when the lower level cache-line selected for eviction is also present in any higher level cache, CBI logic invalidates the cache-line in the higher level cache. The CBI logic appropriately updates the values of presence bits and LRU bits, following evictions and invalidations.
摘要:
A design structure embodied in a machine readable storage medium for designing, manufacturing, and/or testing a design for dynamic optimization of DRAM controller page policy is provided. The design structure can include a memory module, which can include multiple different memories, each including a memory controller coupled to a memory array of memory pages. Each of the memory pages in turn can include a corresponding locality tendency state. A memory bank can be coupled to a sense amplifier and configured to latch selected ones of the memory pages responsive to the memory controller. Finally, the module can include open page policy management logic coupled to the memory controller. The logic can include program code enabled to granularly change open page policy management of the memory bank responsive to identifying a locality tendency state for a page loaded in the memory bank.
摘要:
An adaptive prediction threshold scheme for dynamically adjusting prediction thresholds of entries in a Pattern History Table (PHT) by observing global tendencies of the branch or branches that index into the PHT entries. A count value of a prediction state counter representing a prediction state of a prediction state machine for a PHT entry is obtained. Count values in a set of counters allocated to the entry in the PHT are changed based on the count value of the entry's prediction state counter. The prediction threshold of the prediction state machine for the entry may then be adjusted based on the changed count values in the set of counters, wherein the prediction threshold is adjusted by changing a count value in a prediction threshold counter in the entry, and wherein adjusting the prediction threshold redefines predictions provided by the prediction state machine.
摘要:
The illustrative embodiments provide a method, a computer program product, and an apparatus for managing a cache. A probability of a future request for data to be stored in a portion of the cache by a thread is identified for each of the number of threads to form a number of probabilities. The data is stored with a rank in a number of ranks in the portion of the cache responsive to receiving the future request from the thread in the number of threads for the data. The rank is selected using the probability in the number of probabilities for the thread.