Intel Corporation

Erik Hallnor

Zhen Fang

Martin Guy Dixon

Krishnakanth Venkata Sistla

Ravishankar Iyer

Yen-Cheng Liu

Alexander Gendler

Jeremy J Shrall

Vivek Garg

Dean A Mulla

Stephen R Van Doren

Li Zhao

Ashish V Choubal

Bahaa Fahim

Robert P Adler

Donald K Newell

Rameshkumar G Illikkal

Rupin H Vakharwala

Jeffrey D Chamberlain

Larisa Novakovsky

Stanley Steve Kulick

Jeffrey L Miller

Ramadass Nagarajan

Wim Heirman

Ibrahim Hur

Hemant G Rotithor

Stijn Eyerman

Zhongying Zhang

Vinit Mathew Abraham

Matthew Robert Erler

Kimberly C Weier

Kristof Du Bois

Yves Vandriessche

Nitin B Gupte

Robert J Toepfer

Antonio Juan

Kevin B Theobald

Steven Zhang

Ezra N Harrington

Embodiment of this disclosure provides a mechanism to store cache lines in dedicated cache of an idle core. In one embodiment, a multi-core processor comprising a first core, a second core, a first cache, a second cache, a third cache, and a cache controller unit is provided. The cache controller is operatively coupled to at least the first cache, the second cache, and the third cache. The cache controller is to evict a first line from the first cache, wherein the first core is in an active state. Responsive to the evicting of the first line, the first line is stored in the third cache. Responsive to storing the first line, a second line is evicted from the third cache. Responsive to evicting the second line, the second line is stored in the second cache when the second core is in an idle state.

Storing cache lines in dedicated cache of an idle core

A processor includes a cache-side address monitor unit corresponding to a first cache portion of a distributed cache that has a total number of cache-side address monitor storage locations less than a total number of logical processors of the processor. Each cache-side address monitor storage location is to store an address to be monitored. A core-side address monitor unit corresponds to a first core and has a same number of core-side address monitor storage locations as a number of logical processors of the first core. Each core-side address monitor storage location is to store an address, and a monitor state for a different corresponding logical processor of the first core. A cache-side address monitor storage overflow unit corresponds to the first cache portion, and is to enforce an address monitor storage overflow policy when no unused cache-side address monitor storage location is available to store an address to be monitored.

Scalably mechanism to implement an instruction that monitors for writes to an address

Embodiment of this disclosure provides a hierarchical structure of ordering points. In some embodiments, the hierarchical structure includes a single primary ordering point (POP) and at least one (or more) auxiliary order point (AOP) of a processing device. In one implementation, the processing device includes one or more cores; and a coherency circuit, operatively coupled to the cores. The processing device is to receive a plurality of memory access requests to be ordered by a first ordering point of the processing device. The processing device determines whether to stop the first ordering point based on a system event. Responsive to determining that the first ordering point is stopped, a second ordering point of the processing device is identified. Thereupon, a memory access request of the plurality of memory access requests is provided to the second ordering point.

Supporting hierarchical ordering points in a microprocessor system

Traffic control logic is provided to support a plurality of channels on a link. A plurality of reserved credit counters is provided to each identify reserved flow control credits for a corresponding one of the plurality of channels. Further, a shared credit counter is provided to identify shared flow control credits to be shared between two or more of the plurality of virtual channels.

Shared flow control credits

In an example, a shared uncore memory fabric of a system-on-a-chip (SoC) is configured to provide real-time power management. The SoC may include a power management agent to inform the shared fabric that the processing cores and peripherals will be idle for a time, and to negotiate a power-saving state. The uncore fabric may also include a local power manager that detects when no access requests have been received for a time, such as when cores are operating from cache. The shared fabric may then unilaterally enter a power-saving state, and remain in that state until an access request is received. In the power-saving state, power and/or clocks are gated, and the fabric's state is stored in retention cells. When a new access request is received, an ungated controller may handle preliminary processing while the local power manager restores the state and powers up the shared fabric.

Power management in an uncore fabric

In one embodiment, a processor includes a core to execute instructions and a core perimeter logic coupled to the core. The core perimeter logic may include a fabric interface logic coupled to the core. In turn, the fabric interface logic may include a first storage to store state information of the core when the core is in a low power state, and enable an inter-die interconnect coupled between the core and an uncore to be maintained in an active state during entry of the core into a low power state. Other embodiments are described and claimed.

Masking a power state of a core of a processor

A processing device and method for cache control including tracking updates to the line state of a cache superline are described. In response to a request pertaining to a superline, a cache controller of the processing device can perform one or more read-modify-write (RMW) operations to (a) a line state vector of a line state array and (b) a counter of the line state array. Based on a determination that one or more requests to the superline have completed, the line state vector from the line state array can be written to a tag array. The cache controller can track pending line state updates to a superline outside of the tag array, and a line state update can occur in the cache controller, rather than awaiting completion of all outstanding operations on a superline. Updates to multiple line states can be maintained simultaneously, and up-to-date ECCs computed.

Method and apparatus for cache line state update in sectored cache with line state tracker

An apparatus, method, and system are disclosed. In one embodiment the apparatus includes a cache memory, which a number of sets. Each of the sets in the cache memory have several cache lines. The apparatus also includes at least one process resource table. The process resource table maintains a cache line occupancy count of a number of cache lines. Specifically, the cache line occupancy count for each cache line describes the number of cache lines in the cache storing information utilized by a process running on a computer system. Additionally, the process resource table stores the occupancy count of less cache lines than the total number of cache lines in the cache memory.

Monitoring cache usage in a distributed shared cache

Domain-based cache management methods and systems, including domain event based priority demotion ('EPD'). In EPD, priorities of cached data blocks are demoted upon one or more domain events, such as upon encoding of one or more macroblocks of a video frame. New data blocks may be written over lowest priority cached data blocks. New data blocks may initially be assigned a highest priority. Alternatively, or additionally, one or more new data blocks may initially be assigned one of a plurality of higher priorities based on domain-based information, such as a relative position of a requested data block within a video frame, and/or a relative direction associated with a requested data block. Domain-based cache management may be implemented with one or more other cache management techniques, such as least recently used techniques. Domain-based cache management may be implemented in associative caches, including set associative caches and fully associative caches, and may be implemented with indirect indexing.

Domain-based cache management, including domain event based priority demotion

A proposal for power management control of an interconnect structure based on power state transition control. The power state transition is based on generating early warning signals and an idle timeout value setting based on response time and detection of subsequent requests.

Power state transition initiation control of memory interconnect based on early warning signal, memory response time, and wakeup delay

Growing community of inventors

Beaverton, OR, United States of America

Erik Hallnor