Abstract: An apparatus and method for data sharing optimizations. One embodiment of a processor comprises: a plurality of cores; a plurality of caches to store cachelines read from memory by the plurality of cores; cache management circuitry to manage coherency of the cachelines by performing snoops including probes and confirms, the cache management circuitry to: determine a number of outstanding probes and a number of outstanding confirms within a snoop monitoring window; and prioritize confirms over probes or prioritize probes over confirms based, at least in part, on the number of outstanding confirms relative to the number of outstanding probes.

Abstract: Techniques and mechanisms for efficiently providing access to cached data. In an embodiment, a cache coherency engine comprises circuitry to provide a snoop filter which stores entries each corresponding to a respective line of one or more caches. The one or more caches comprise a first cache which includes a first set, and the snoop filter includes a first plurality of sets which are each configured to be available to represent a line of the first set. In another embodiment, the one or more caches comprise multiple caches which each comprise a respective first set, wherein, for each set of the first plurality of sets, any line in the multiple caches which is to be represented by that each set is to be a line in the respective first sets of the multiple caches.

Abstract: Embodiments include a cache associated with at least a first core or functional circuit block; and a power controller to evaluate at least one of energy and power consumption associated with the first core or functional circuit block remaining in an active state, entering a first sleep state, or entering a second sleep state; the power controller to evaluate the at least one of energy and power consumption based, at least in part, on energy associated with flushing the plurality of cachelines when entering into the second sleep state, an expected sleep time, and power consumption of the first core or functional circuit block in the active state, the first sleep state, and the second sleep state; the power controller to cause the first core or functional circuit block to remain active, enter into the first sleep state, or the second sleep state based on a minimum energy consumption.

Abstract: An embodiment of an integrated circuit may comprise a core, and a cache controller coupled to the core, the cache controller including circuitry to identify data from a working set for dynamic inclusion in a next level cache based on an amount of re-use of the next level cache, send a shared copy of the identified data to a requesting core of one or more processor cores, and maintain a copy of the identified data in the next level cache. Other embodiments are disclosed and claimed.

Abstract: Methods and apparatus relating to leveraging system cache for performance cores are described. In an embodiment, a system cache stores one or more cachelines that are to be evicted from a processor cache. Logic circuitry determines whether to store the one or more cachelines in the system cache based at least in part on comparison of a threshold value with a hit rate associated with the one or more cachelines. Other embodiments are also disclosed and claimed.

Abstract: Techniques and mechanisms for efficiently providing access to cached data. In an embodiment, a cache coherency engine comprises circuitry to provide a snoop filter which stores entries each corresponding to a respective line of one or more caches. The one or more caches comprise a first cache which includes a first set, and the snoop filter includes a first plurality of sets which are each configured to be available to represent a line of the first set. In another embodiment, the one or more caches comprise multiple caches which each comprise a respective first set, wherein, for each set of the first plurality of sets, any line in the multiple caches which is to be represented by that each set is to be a line in the respective first sets of the multiple caches.

Abstract: Disclosed embodiments relate to a shared read request (SRR) using a common request tracker (CRT) as a temporary cache. In one example, a multi-core system includes a memory and a memory controller to receive a SRR from a core when a Leader core is not yet identified, allocate a CRT entry and store the SRR therein, mark it as a Leader, send a read request to a memory address indicated by the SRR, and when read data returns from the memory, store the read data in the CRT entry, send the read data to the Leader core, and await receipt, unless already received, of another SRR from a Follower core, the other SRR having a same address as the SRR, then, send the read data to the Follower core, and deallocate the CRT entry.

Abstract: An embodiment of an integrated circuit may comprise a core, and a cache controller coupled to the core, the cache controller including circuitry to identify data from a working set for dynamic inclusion in a next level cache based on an amount of re-use of the next level cache, send a shared copy of the identified data to a requesting core of one or more processor cores, and maintain a copy of the identified data in the next level cache. Other embodiments are disclosed and claimed.

Abstract: Methods and apparatus to implement multiple inference compute engines are disclosed herein. A disclosed example apparatus includes a first inference compute engine, a second inference compute engine, and an accelerator on coherent fabric to couple the first inference compute engine and the second inference compute engine to a converged coherency fabric of a system-on-chip, the accelerator on coherent fabric to arbitrate requests from the first inference compute engine and the second inference compute engine to utilize a single in-die interconnect port.

Abstract: Disclosed embodiments relate to a shared read request (SRR) using a common request tracker (CRT) as a temporary cache. In one example, a multi-core system includes a memory and a memory controller to receive a SRR from a core when a Leader core is not yet identified, allocate a CRT entry and store the SRR therein, mark it as a Leader, send a read request to a memory address indicated by the SRR, and when read data returns from the memory, store the read data in the CRT entry, send the read data to the Leader core, and await receipt, unless already received, of another SRR from a Follower core, the other SRR having a same address as the SRR, then, send the read data to the Follower core, and deallocate the CRT entry.

Abstract: Methods and apparatus to implement multiple inference compute engines are disclosed herein. A disclosed example apparatus includes a first inference compute engine, a second inference compute engine, and an accelerator on coherent fabric to couple the first inference compute engine and the second inference compute engine to a converged coherency fabric of a system-on-chip, the accelerator on coherent fabric to arbitrate requests from the first inference compute engine and the second inference compute engine to utilize a single in-die interconnect port.