Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0806—Multiuser, multiprocessor or multiprocessing cache systems
  • G06F12/0842—Multiuser, multiprocessor or multiprocessing cache systems for multiprocessing or multitasking
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0844—Multiple simultaneous or quasi-simultaneous cache accessing
  • G06F12/0846—Cache with multiple tag or data arrays being simultaneously accessible
  • G06F12/0848—Partitioned cache, e.g. separate instruction and operand caches
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0864—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches using pseudo-associative means, e.g. set-associative or hashing

Definitions

• Disclosed embodiments are directed to cache allocation in shared caches. More particularly, exemplary embodiments are directed to allocation of cache lines in shared partitioned caches in multi-threaded processors.
• the manifestation of multi-threading in processors may occur at different levels or at differing degrees of process granularity.
• the various threads of a multi-threaded processor may share one or more levels of caches and memory.
• the threads may have different data attributes like "streaming data,” “localized data,” “multiple-use data,” “shared,” “non-shared,” etc.
• Simple allocation and replacement schemes like least recently used (LRU) are unsuitable because the different contexts created by the threads and data attributes specific to threads must be accounted for.
• cache lines of a shared cache that have been populated by streaming data may be replaced by localized data or multiple-use data of another thread. Such scenarios are unacceptable and may lead to severe degradation of processing speed and efficiency.
• Some shared caches may include a static partitioning to keep data from different threads separate.
• static partitioning the threads get statically locked to a particular cache partition, which leads to additional drawbacks.
• a data attribute of a thread may dynamically transition from streaming data to localized data or vice-versa.
• static partitioning that locks threads too allocate in predefined portions of a cache, cannot account for such dynamic transitions of data attributes in threads.
• an exemplary embodiment is directed to a method of updating a shared cache comprising: receiving an address for a cache entry associated with a processing thread to be allocated in the cache; determining attributes associated with the address; setting a configuration register corresponding to the processing thread, with cache allocation information based on the attributes; determining a cache partition based on a partitioning register; and allocating the cache entry in the cache based on the configuration register and the cache partition.
• Another exemplary embodiment is directed to a method of cache allocation in a multithreaded processing system comprising: receiving an address associated with data, for a processing thread of the multi-threaded processing system, to be allocated in the cache; determining page attributes associated with the address; partitioning the cache into two or more portions; and allocating the data into one or more portions of the cache based at least on the page attributes.
• Yet another exemplary embodiment is directed to a multi-threaded processing system comprising: a cache; a processing thread; a memory management unit for determining attributes associated with an address for a cache entry associated with the processing thread to be allocated in the cache; a configuration register comprising cache allocation information based on the determined attributes; a partitioning register comprising partitioning information for partitioning the cache into two or more portions; and logic for allocating the cache entry into one of the portions of the cache based on the configuration register and the partitioning register.
• Yet another exemplary embodiment is directed to a multi-threaded processing system comprising a cache; a processing thread; memory management means for determining attributes associated with an address for a cache entry associated with the processing thread to be allocated in the cache; configuration means for storing cache allocation information based on the determined attributes; partitioning means for storing partitioning information for partitioning the cache into two or more portions; and means for allocating the cache entry into one of the portions of the cache based on the configuration means and the partitioning means.
• Yet another exemplary embodiment is directed to a non- transitory computer-readable storage medium comprising code, which, when executed by a multi-threaded processor, causes the multi-threaded processor to perform operations for updating a shared cache
• the non-transitory computer-readable storage medium comprising: code for receiving an address for a cache entry associated with a processing thread to be allocated in the cache; code for determining attributes associated with the address; code for setting a configuration register with cache allocation information based on the attributes; code for determining a cache partition; and code for allocating the cache entry in the cache based on the configuration register and the cache partition.
• Yet another exemplary embodiment is directed to a non-transitory computer-readable storage medium comprising code, which, when executed by a multi-threaded processor, causes the multi-threaded processor to perform operations for cache allocation in a multi-threaded processing system
• the non-transitory computer-readable storage medium comprising: code for receiving an address associated with data, for a processing thread of the multi-threaded processing system, to be allocated in the cache; code for determining page attributes associated with the address; code for partitioning the cache into two or more portions; and code for allocating the data into one or more portions of the cache based at least on the page attributes.
• FIG. 1 is a block diagram of processing system 100 configured to implement cache partition and cache allocation according to exemplary embodiments.
• FIG. 2 illustrates a table comprising configuration register settings for cache allocation.
• FIG. 3 illustrates a schematic diagram of a logical implementation of cache allocation and cache partition in an exemplary data cache.
• FIG. 4 illustrates an exemplary operational flow of a method of allocating a cache according to exemplary embodiments.
• FIG. 5 is a block diagram of a particular illustrative embodiment of a wireless device that includes a multi-core processor configured according to exemplary embodiments.
• Exemplary embodiments are suitable for multi-threaded processors with shared caches.
• Embodiments may be configured to allow flexible allocation of data in partitioned caches.
• software can control allocation of data in a statically partitioned cache by overriding the static partition, and allocating data into a partition determined by software.
• software may allocate date in the entire cache, without being subjected to statically defined partitions of the cache that may be, for example, thread specific.
• Multi-threaded processor 102 may be configured to process two or more threads (e.g., threads TO, Tl, T2, and T3), which may be implemented in hardware or software.
• the two or more threads may access LI caches such as D-cache 104 and I- cache 106 which may be partitioned according to exemplary embodiments discussed herein.
• LI caches such as D-cache 104 and I- cache 106 which may be partitioned according to exemplary embodiments discussed herein.
• One or more levels of caches, such as L2 cache may also be present in the memory hierarchy of processor 100, culminating in shared main memory 108.
• Memory management unit (MMU) 110 may be in communication with the various caches and main memory to serve functions such as address translation from virtual addresses generated by multi-threaded processor 102 to physical address for accessing the caches and main memory 108. Suitable variations of processing system 100 will be recognized by skilled persons, without deviating from the scope of disclosed embodiments.
• D-cache 104 may be partitioned in various ways in order to control placement of data within the cache space according to particular needs.
• D-cache 104 may be partitioned into a main portion and an auxiliary portion.
• the main portion may be dedicated to data processing needs of threads from multi-threaded processor 102, while the auxiliary portion may be dedicated to streaming data which may be received for example from a direct memory access (DMA) engine (not shown).
• DMA direct memory access
• the relative or absolute sizes and placements of the main and auxiliary portions may be set statically.
• the cache may be statically partitioned into main and auxiliary portions in a selected ratio, such as, 50:50, 25:75, 75:25, etc.
• the cache partitioning may specify the number of ways in a main and auxiliary portion.
• the cache partitioning may specify 6 ways in the main and 2 ways in the auxiliary portion (corresponding to a 75:25 proportion).
• the cache partitioning may be dynamic.
• the relative proportions of the main and auxiliary portions may be based on the processing threads.
• Exemplary embodiments may include a configuration register per thread of multithreaded processor 102, wherein the configuration register may control cache allocation corresponding to the thread.
• the configuration register may be set by software.
• Each thread may have associated data attributes such as page attributes associated with addresses of data in the thread. These data attributes may be used to determine allocation of data in D-cache 104.
• a table comprising cache allocation directives for a first thread (e.g. running on multi-threaded processor 102) based on the first thread's configuration register CP 200 is illustrated.
• CP 200 is set to a value of "00"
• the allocation/placement of data in D-cache 104 is controlled by associated data attributes.
• any statically determined allocation of data to defined partitions of D-cache 104 will be ignored and the data attributes will dictate where data will be placed.
• MMU 110 may be configured to assist in the cache allocation by parsing the address associated with the data for data attributes and using these attributes to set the configuration register CP accordingly.
• another mode could designate "no allocation” wherein allocation of the associated data in D-cache 104 is skipped and the data is treated as uncached.
• Yet another mode could designate a "most recently used (MRU)" allocation, wherein the data is allocated to the portion (e.g. main or auxiliary) which was most recently used for data allocation in D-cache 104 for the first thread.
• MRU most recently used
• FIG. 3 an exemplary implementation of allocation of D-cache 104 using thread specific configuration registers is illustrated.
• Three configuration registers CP 300_0, 300_1, and 300_2 are shown, corresponding to threads TO, Tl, and T2 respectively.
• an expanded view of D-cache 104 comprises tag array 302, state array 304, and data array 306.
• D-cache partitioning register DP 310 configured to control partitioning across tag array 302, state array 304, and data array 306 of D-cache 104 will now be discussed in further detail.
• D-cache partitioning register DP 310 may be configured to hold specific partitioning information for D-cache 104.
• a static partition of D-cache 104 may be implemented and DP 310 may be set with information pertaining to relative proportions of main and auxiliary portions of D-cache 104.
• DP 310 may be set as part of system configuration of processing system 100.
• the static partitioning may be common to each processing thread, or may be thread- specific. Specific static partitions may be selected by a programmer.
• D-cache 104 may be dynamically partitioned in a thread specific manner. Combinations of static and dynamic partitions are also possible and may be implemented by suitably configuring partitioning register DP 310.
• DP 310 may be used to partition tag array 302 into main tag array portion 302a and auxiliary tag array portion 302b.
• state array 304 may be partitioned into main state array portion 304a and auxiliary state array portion 304b; and data array 306 may be partitioned into main data array portion 306a and auxiliary data array portion 306b.
• Physical address (PA) of data which needs to be allocated in D-cache 104 may be parsed into tag bits, PA [tag bits] 316a, PA [set bits] 316b, and PA [data array bits] 316c.
• PA [tag bits] 316a may be compared against tags present in tag array 302 (which may be implemented as a content addressable memory (CAM)) using comparators 314.
• Comparators 314 may also validate a positive result (hit) with corresponding state bits read out from state array 304 using PA [set bits] 316b to ensure that the hitting tag array entry is valid.
• replacement way logic 308 may assist in allocating the entry in data array 306 corresponding to the PA, by taking into account modes of configuration register CP.
• thread specific configuration registers CP 300_0, 300_1 and 300_2 with the above described mechanism of writing to D-cache 104 will now be explained.
• thread mux 312 may be configured to select between configuration registers CP 300_0, 300_1 and 300_2 corresponding to the selected thread.
• thread mux 312 will select configuration register CP 300_0 which is shown to hold the value "00.” As previously described, this corresponds to the first mode, wherein existing partitioning of D-cache 104 will be ignored and allocation of D-cache 104 will be based on data attributes. Accordingly, following the above-described mechanism of writing to D-cache 104, if there is a miss, replacement way logic 308 will assist in writing data into main data array portion 306a or auxiliary data array portion 306b at an allocated location derived from PA [data array bits] 316c, and overriding any allocation which may be determined by partitioning information derived from DP 310.
• the data may belong to thread Tl, thus causing thread mux 312 to select configuration register CP 300_1 which has been shown to hold the value "01".
• this value corresponds to the second mode, wherein the data will be allocated to the main portion of D-cache 104.
• the location of the main portion within D-cache 104 will be determined based on DP 310. Accordingly, following the above-described mechanism of writing to D-cache 104, on a miss, data will be written to main data array portion 306a at a location derived from PA [data array bits] 316c.
• the data may belong to thread T2, thus causing thread mux 312 to select configuration register CP 300_2, which has been shown to hold the valuelO".
• this value corresponds to the third mode, wherein the data will be allocated to the auxiliary portion of D-cache 104.
• the location of the auxiliary portion within D-cache 104 will be determined based on DP 310. Accordingly, following the above-described mechanism of writing to D-cache 104, on a miss, data will be written to auxiliary data array portion 306b at a location derived from PA [data array bits] 316c.
• D-cache 104 may be statically partitioned.
• the subsequent allocation of the data may be controlled by appropriately setting the corresponding configuration register CP, based on attributes.
• cache allocation may be for any suitable cache entry such as a cache line or a cache block comprising data or instructions.
• the fine grained control of allocation of data/instructions in exemplary embodiments may improve management of shared caches in software, and thereby improve performance of processing system 100.
• an embodiment can include a method of updating a cache comprising: receiving an address for a cache entry associated with a processing thread to be allocated in the cache (e.g. PA for data associated with thread TO in FIG. 2) - Block 402; determining attributes (e.g. page attributes) associated with the address - Block 404; setting a configuration register with cache allocation information based on the attributes (e.g. configuration register CP for thread TO in FIG. 2) - Block 406; determining a cache partition based on a partitioning register (e.g. DP 310) - Block 408; and allocating the cache entry in the cache (e.g. into main or auxiliary portion) based on the configuration register and the cache partition - Block 410.
• a partitioning register e.g. DP 310
• a software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
• An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor.
• FIG. 5 a block diagram of a particular illustrative embodiment of a wireless device that includes a multi-core processor configured according to exemplary embodiments is depicted and generally designated 500.
• the device 500 includes a digital signal processor (DSP) 564, which may include various blocks illustrated in FIG. 1, such as, processor 102, MMU 110, D-Cache 104, and I-Cache 106 coupled to memory 532 as shown.
• DSP digital signal processor
• FIG. 5 also shows display controller 526 that is coupled to DSP 564 and to display 528.
• Coder/decoder (CODEC) 534 e.g., an audio and/or voice CODEC
• CDEC Coder/decoder
• wireless controller 540 which may include a modem
• Speaker 536 and microphone 538 can be coupled to CODEC 534.
• FIG. 5 also indicates that wireless controller 540 can be coupled to wireless antenna 542.
• DSP 564, display controller 526, memory 532, CODEC 534, and wireless controller 540 are included in a system-in-package or system-on-chip device 522.
• input device 530 and power supply 544 are coupled to the system-on-chip device 522.
• display 528, input device 530, speaker 536, microphone 538, wireless antenna 542, and power supply 544 are external to the system-on-chip device 522.
• each of display 528, input device 530, speaker 536, microphone 538, wireless antenna 542, and power supply 544 can be coupled to a component of the system-on-chip device 522, such as an interface or a controller.
• FIG. 5 depicts a wireless communications device
• DSP 564 and memory 532 may also be integrated into a set-top box, a music player, a video player, an entertainment unit, a navigation device, a personal digital assistant (PDA), a fixed location data unit, or a computer.
• a processor e.g., DSP 564 may also be integrated into such a device.
• an embodiment of the invention can include a computer readable media embodying a method for cache allocation. Accordingly, the invention is not limited to illustrated examples and any means for performing the functionality described herein are included in embodiments of the invention.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Memory System Of A Hierarchy Structure (AREA)

Priority Applications (3)

Applications Claiming Priority (2)

Publications (1)

Family

ID=48446688

Family Applications (1)

Country Status (5)

Cited By (1)

Families Citing this family (14)

Citations (3)

Family Cites Families (22)

• 2012
  • 2012-05-08 US US13/466,359 patent/US9824013B2/en not_active Expired - Fee Related
• 2013
  • 2013-05-08 CN CN201380023546.2A patent/CN104272278B/zh not_active Expired - Fee Related
  • 2013-05-08 WO PCT/US2013/040040 patent/WO2013169836A1/en not_active Ceased
  • 2013-05-08 JP JP2015511642A patent/JP6125617B2/ja active Active
  • 2013-05-08 EP EP13723374.8A patent/EP2847684A1/en not_active Ceased
• 2017
  • 2017-03-30 JP JP2017067172A patent/JP6279787B2/ja not_active Expired - Fee Related

Patent Citations (3)

Also Published As

Similar Documents

Legal Events