Qualcomm Incorporated

George Patsilaras

Andrew Edmund Turner

Bohuslav Rychlik

Subbarao Palacharla

Simon Peter William Booth

Feng Wang

Jeffrey Shabel

Kedar Bhole

Moinul H Khan

Wesley James Holland

Matthew Levi Severson

Anwar Q Rohillah

Jeffrey Gemar

Alain Artieri

Girish Bhat

Vipul Gandhi

Engin Ipek

Younghoon Kim

Hiral Nandu

Jeong-Ho Woo

Yanru Li

Christopher E Koob

Prajakt Kulkarni

Serag Gadelrab

Vinod Chamarty

Aris Balatsos

Ali Iranli

Raghu Sankuratri

Sean Sweeney

Simon Booth

Isaac Berk

Shengqi Yang

Kris Tiri

Myil Ramkumar

Sparsh Singhai

Tarek Zghal

Farrukh Hijaz

Johnny Jone Wai Kuan

Fahad Ali

Rakesh Kumar Gupta

Jaemoon Kim

Giwon Kim

Goran Goran

Zhenbiao Ma

Yen-Kuan Wu

Hamza Omar

Bhupinder Singh Pabla

Christopher Edward Koob

Pavan Kumar Thirunagari

Can Hankendi

Serag Monier Gadelrab

Sean David Sweeney

A method includes generating a plurality of future reference picture lists associated with a plurality of future pictures of a set of pictures, wherein the set of pictures includes a current picture and the plurality of future pictures, and the plurality of future pictures follow the current picture in coding order, determining, based on information derived from the plurality of future reference picture lists associated with the plurality of future pictures, whether to write the current picture in a dedicated chip memory or whether to write the current picture in a non-dedicated system memory, and writing the current picture in the dedicated chip memory or the non-dedicated system memory based on the determination of whether to write the current picture in the dedicated chip memory or whether to write the current picture the non-dedicated system memory.

Caching pictures for video coding

Aspects presented herein relate to methods and devices for data or graphics processing including an apparatus, e.g., a graphics processing unit (GPU). The apparatus may configure an address range in a cache. The apparatus may also obtain a request to access data in the cache, where the request to access the data includes an address in the cache that maps to a set index in a plurality of set indexes, where an address value for the set index corresponds to a portion of the address. Further, the apparatus may select an updated address value for the set index, where the updated address value is associated with an updated address within the address range, and the updated address value corresponds to a portion of the updated address. The apparatus may also allocate the data in the request to the updated address value for the set index.

Physical address based set partitioning

Providing content-aware cache replacement and insertion policies in processor-based devices is disclosed. In some aspects, a processor-based device comprises a cache memory device and a cache controller circuit of the cache memory device. The cache controller circuit is configured to determine a plurality of content costs for each of a plurality of cached data values in the cache memory device, based on a plurality of bit values of each of the plurality of cached data values. The cache controller circuit is configured to identify, based on the plurality of content costs, a cached data value of the plurality of cached data values associated with a lowest content cost as a target cached data value. The cache controller circuit is also configured to evict the target cached data value from the cache memory device.

Providing content-aware cache replacement and insertion policies in processor-based devices

Various embodiments include methods and devices for implementing a criterion aware cache replacement policy by a computing device. Embodiments may include updating a staling counter, writing a value of a local counter to a system cache in association with a location in the system cache for with data, in which the value of the local counter includes a value of the staling counter when (i.e., at the time) the associated data is written to the system cache, and using the value of the local counter of the associated data to determine whether the associated data is stale.

Multimedia compressed frame aware cache replacement policy

Integrated circuits (ICs) employ subsystem shared cache memory for facilitating extension of low-power island (LPI) memory. An LPI subsystem and primary subsystems access a memory subsystem on a first access interface in a first power mode and the LPI subsystem accesses the memory subsystem by a second access interface in the low power mode. In the first power mode, the primary subsystems and the LPI subsystem may send a subsystem memory access request including a virtual memory address to a subsystem memory interface of the memory subsystem to access either data stored in an external memory or a version of the data stored in a shared memory circuit. In the low-power mode, the LPI subsystem sends an LPI memory access request including a direct memory address to an LPI memory interface of the memory subsystem to access the shared memory circuit to extend the LPI memory.

Integrated circuits (IC) employing subsystem shared cache memory for facilitating extension of low-power island (LPI) memory and related methods

Various embodiments include methods and devices for virtual cache coherency. Embodiments may include receiving a snoop for a physical address from a coherent processing device, determining whether an entry for the physical address corresponding to a virtual address in a virtual cache exists in a snoop filter, and sending a cache coherency operation to the virtual cache in response to determining that the entry exists in the snoop filter.

Space and time cache coherency

The energy consumed by data transfer in a computing device may be reduced by transferring data that has been encoded in a manner that reduces the number of one '1' data values, the number of signal level transitions, or both. A data destination component of the computing device may receive data encoded in such a manner from a data source component of the computing device over a data communication interconnect, such as an off-chip interconnect. The data may be encoded using minimum Hamming weight encoding, which reduces the number of one '1' data values. The received data may be decoded using minimum Hamming weight decoding. For other computing devices, the data may be encoded using maximum Hamming weight encoding, which increases the number of one &#x201c;1&#x201d; data values while reducing the number of zero &#x201c;0&#x201d; values, if reducing the number of zero values reduces energy consumption.

Data re-encoding for energy-efficient data transfer in a computing device

Memory system features may promote cache coherency where first and second memory clients may attempt to work on the same data. A second client cache system may provide a read request for data and associated metadata. The metadata element may be detected in a first client cache system. The first client cache system may write or flush, such as to a system memory, one or more cache lines containing the metadata and associated data and invalidate the flushed cache lines. The second client cache system may receive the data and metadata, such as from the system memory, completing or fulfilling the read request.

Marshalled data coherency

In one aspect, space in a tile-unaware cache associated with an address aperture may be managed in different ways depending on whether a processing component initiating an access request through the aperture to a tile-based memory is tile-unaware or tile-aware. Upon a full-tile read by a tile-aware process, data may be evicted from the cache, or space may not be allocated. Upon a full-tile write by a tile-aware process, data may be evicted from the cache. In another aspect, a tile-unaware process may be supplemented with tile-aware features by generating a full tile of addresses in response to a partial-tile access. Upon a partial-tile read by the tile-unaware process, the generated addresses may be used to pre-fetch data. Upon a partial-tile write, the addresses may be used to evict data. Upon a bit block transfer, the addresses may be used in dividing the bit block transfer into units of tiles.

System and method for mixed tile-aware and tile-unaware traffic through a tile-based address aperture

An intelligent tile-based prefetching solution executed by a compression address aperture services linearly addressed data requests from a processor to memory stored in a memory component having a tile-based address structure. The aperture monitors tile reads and seeks to match the tile read pattern to a predefined pattern. If a match is determined, the aperture executes a prefetching algorithm uniquely and optimally associated with the predefined tile read pattern. In this way, tile overfetch is mitigated while the latency on first line data reads is reduced.

System and method for intelligent tile-based prefetching of image frames in a system on a chip

Various aspects include methods for implementing retaining high locality data in a higher level cache memory on a computing device. Various aspects may include receiving a cache access request for a first cache line in the higher level cache memory indicating a locality of the first cache line, determining whether the access request indicates high locality, and setting a high locality indicator of the first cache line in response to determining that the cache access request indicates high locality. Various aspects may include determining whether a lower level cache memory hit counter of a first cache line of a first cache exceeds a lower level cache locality threshold, setting a high locality indicator of the first cache line in response to determining that the lower level cache memory hit counter exceeds the lower level cache locality threshold and resetting the lower level cache memory hit counter of the first cache.

Performance by retaining high locality data in higher level cache memory

Methods, devices, and non-transitory process-readable storage media for compacting data within cache lines of a cache. An aspect method may include identifying, by a processor of the computing device, a base address (e.g., a physical or virtual cache address) for a first data segment, identifying a data size (e.g., based on a compression ratio) for the first data segment, obtaining a base offset based on the identified data size and the base address of the first data segment, and calculating an offset address by offsetting the base address with the obtained base offset, wherein the calculated offset address is associated with a second data segment. In some aspects, the method may include identifying a parity value for the first data segment based on the base address and obtaining the base offset by performing a lookup on a stored table using the identified data size and identified parity value.

Cache line compaction of compressed data segments

Systems and methods relate to managing shared resources in a multithreaded processor comprising two or more processing threads. Danger levels for the two or more threads are determined, wherein the danger level of a thread is based on a potential failure of the thread to meet a deadline due to unavailability of a shared resource. Priority levels associated with the two or more threads are also determined, wherein the priority level is higher for a thread whose failure to meet a deadline is unacceptable and the priority level is lower for a thread whose failure to meet a deadline is acceptable. The two or more threads are scheduled based at least on the determined danger levels for the two or more threads and priority levels associated with the two or more threads.

Bandwidth/resource management for multithreaded processors

Aspects include computing devices, systems, and methods for dynamically partitioning a system cache by sets and ways into component caches. A system cache memory controller may manage the component caches and manage access to the component caches. The system cache memory controller may receive system cache access requests and reserve locations in the system cache corresponding to the component caches correlated with component cache identifiers of the requests. Reserving locations in the system cache may activate the locations in the system cache for use by a requesting client, and may also prevent other client from using the reserved locations in the system cache. Releasing the locations in the system cache may deactivate the locations in the system cache and allow other clients to use them. A client reserving locations in the system cache may change the amount of locations it has reserved within its component cache.

Method and apparatus for a shared cache with dynamic partitioning

Aspects include computing devices, systems, and methods for implementing a cache memory access requests for data smaller than a cache line and eliminating overfetching from a main memory by combining the data with padding data of a size of a difference between a size of a cache line and the data. A processor may determine whether the data, uncompressed or compressed, is smaller than a cache line using a size of the data or a compression ratio of the data. The processor may generate the padding data using constant data values or a pattern of data values. The processor may send a write cache memory access request for the combined data to a cache memory controller, which may write the combined data to a cache memory. The cache memory controller may send a write memory access request to a memory controller, which may write the combined data to a memory.

Power aware padding

Growing community of inventors

Del Mar, CA, United States of America

George Patsilaras