Samsung Electronics Co., Ltd.

Arm Limited

Karthik Sundaram

Rama S Gopal

Alexander Cole Shulyak

Arun Radhakrishnan

Yasuo Ishii

Joseph Michael Pusdesris

Paul Edward Kitchin

Murali S Chinnakonda

Kevin Michael Lepak

Abhishek Raja

James David Dundas

Michael Brian Schinzler

Adrian Montero

Brian C Grayson

Balaji Vijayan

Anoop Ramachandra Iyer

Prefetch generation circuitry generates requests to prefetch data to a cache, where the prefetch generation circuitry is configured to initiate a producer prefetch to request return of producer data having a producer address and to initiate at least one consumer prefetch to request prefetching of consumer data to the cache, the consumer data having an address derived from the producer data returned in response to the producer prefetch. Training circuitry updates, based on executed load operations, a training table indicating candidate producer-consumer relationships being trained for use by the prefetch generation circuitry in generating the producer/consumer prefetches. Replacement control circuitry controls replacement of candidate producer-consumer relationships based on a producer-data-consumer-operand (PD-CO) match-based replacement policy criterion, which depends on whether a PD-CO match condition, indicative of the producer data for a producer load matching an address operand of a consumer load, is satisfied for existing/new candidate producer-consumer relationships.

Replacement control for candidate producer-consumer relationships trained for prefetch generation

An apparatus comprises memory management circuitry to perform a translation table walk for a target address of a memory access request and to signal a fault in response to the translation table walk identifying a fault condition for the target address, prefetch circuitry to generate a prefetch request to request prefetching of information associated with a prefetch target address to a cache; and faulting address prediction circuitry to predict whether the memory management circuitry would identify the fault condition for the prefetch target address if the translation table walk was performed by the memory management circuitry for the prefetch target address. In response to a prediction that the fault condition would be identified for the prefetch target address, the prefetch circuitry suppresses the prefetch request and the memory management circuitry prevents the translation table walk being performed for the prefetch target address of the prefetch request.

Faulting address prediction for prefetch target address

Indirect prefetch circuitry initiates a producer prefetch requesting return of producer data having a producer address and at least one consumer prefetch to request prefetching of consumer data having a consumer address derived from the producer data. A producer prefetch filter table stores producer filter entries indicative of previous producer addresses of previous producer prefetches. Initiation of a requested producer prefetch for producer data having a requested producer address is suppressed when a lookup of the producer prefetch filter table determines that the requested producer address hits against a producer filter entry of the table. The lookup of the producer prefetch filter table for the requested producer address depends on a subset of bits of the requested producer address including at least one bit which distinguishes different chunks of data within a same cache line.

Producer prefetch filter

Data processing apparatuses and methods of processing data are disclosed. The operations comprise: storing copies of data items; and storing, in a producer pattern history table, a plurality of producer-consumer relationships, each defining an association between producer load indicator and a plurality of consumer load entries, each consumer load entry comprising a consumer load indicator and one or more usefulness metrics. Further steps comprise: initiating, in response to a data load from an address corresponding to the producer load indicator in the producer pattern history table and when at least one of the corresponding one or more usefulness meets a criterion, a producer prefetch of data to be prefetched for storing as a local copy; and issuing, when the data is returned, one or more consumer prefetches to return consumer data from a consumer address generated from the data returned by the producer prefetch and a consumer load indicator of a consumer load entry.

Data processing apparatus and method for generating prefetches

A high-frequency and low-power L1 cache and associated access technique. The method may include inspecting a virtual address of an L1 data cache load instruction, and indexing into a row and a column of a way predictor table using metadata and a virtual address associated with the load instruction. The method may include matching information stored at the row and the column of the way predictor table to a location of a cache line. The method may include predicting the location of the cache line within the L1 data cache based on the information match. A hierarchy of way predictor tables may be used, with higher level way predictor tables refreshing smaller lower level way predictor tables. The way predictor tables may be trained to make better predictions over time. Only selected circuit macros need to be enabled based on the predictions, thereby saving power.

High-frequency and low-power L1 cache and associated access technique

A system and a method to cascade execution of instructions in a load-store unit (LSU) of a central processing unit (CPU) to reduce latency associated with the instructions. First data stored in a cache is read by the LSU in response a first memory load instruction of two immediately consecutive memory load instructions. Alignment, sign extension and/or endian operations are performed on the first data read from the cache in response to the first memory load instruction, and, in parallel, a memory-load address-forwarded result is selected based on a corrected alignment of the first data read in response to the first memory load instruction to provide a next address for a second of the two immediately consecutive memory load instructions. Second data stored in the cache is read by the LSU in response to the second memory load instruction based on the selected memory-load address-forwarded result.

Memory load to load fusing

According to one general aspect, a load unit may include a load circuit configured to load at least one piece of data from a memory. The load unit may include an alignment circuit configured to align the data to generate an aligned data. The load unit may also include a mathematical operation execution circuit configured to generate a resultant of a predetermined mathematical operation with the at least one piece of data as an operand. Wherein the load unit is configured to, if an active instruction is associated with the predetermined mathematical operation, bypass the alignment circuit and input the piece of data directly to the mathematical operation execution circuit.

Memory load and arithmetic load unit (ALU) fusing

A computing system includes: an instruction dispatch module module configured to receive a program instruction; and an address reordering module, coupled to the instruction dispatch module, configured to filter the program instruction when the program instruction is a hit in a cache-line in a prefetch filter. The computer system further includes: an instruction dispatch module configured to receive a program instruction; an address reordering module, coupled to the instruction dispatch module, configured to: allocate a tag in a tag module for the program instruction in a program order, allocate a virtual address in a virtual address module for the program instruction and out-of-order relative to the program order, and insert a pointer associated with the tag to link the tag to the virtual address.

Address re-ordering mechanism for efficient pre-fetch training in an out-of order processor

According to one general aspect, a method may include monitoring a plurality of pre-fetch cache requests associated with a data stream. The method may also include evaluating an accuracy of the pre-fetch cache requests. The method may further include, based at least in part upon the accuracy of the pre-fetch cache requests, adjusting a maximum amount of data that is allowably pre-fetched in excess of a data stream's current actual demand for data.

Adaptive mechanism to tune the degree of pre-fetches streams

According to one general aspect, an apparatus may include a cache pre-fetcher, and a pre-fetch scheduler. The cache pre-fetcher may be configured to predict, based at least in part upon a virtual address, data to be retrieved from a memory system. The pre-fetch scheduler may be configured to convert the virtual address of the data to a physical address of the data, and request the data from one of a plurality of levels of the memory system. The memory system may include a plurality of levels, each level of the memory system configured to store data.

Pre-fetch chaining

A computing system includes: an instruction dispatch module configured to receive a program instruction; an address reordering module, coupled to the instruction dispatch module, configured to filter the program instruction when the program instruction is a hit in a cache-line in a prefetch filter. The computer system further includes: an instruction dispatch module configured to receive a program instruction; an address reordering module, coupled to the instruction dispatch module, configured to: allocate a tag in a tag module for the program instruction in a program order, allocate a virtual address in a virtual address module for the program instruction in an out-of-order relative to the program order, and insert a pointer associated with the tag to link the tag to the virtual address.

Address re-ordering mechanism for efficient pre-fetch training in an out-of-order processor

A Fill Buffer (FB) based data forwarding scheme that stores a combination of Virtual Address (VA), TLB (Translation Look-aside Buffer) entry# or an indication of a location of a Page Table Entry (PTE) in the TLB, and a TLB page size information in the FB and uses these values to expedite FB forwarding. Load (Ld) operations send their non-translated VA for an early comparison against the VA entries in the FB, and are then further qualified with the TLB entry# to determine a 'hit.' This hit determination is fast and enables FB forwarding at higher frequencies without waiting for a comparison of Physical Addresses (PA) to conclude in the FB. A safety mechanism may detect a false hit in the FB and generate a late load cancel indication to cancel the earlier-started FB forwarding by ignoring the data obtained as a result of the Ld execution. The Ld is then re-executed later and tries to complete successfully with the correct data.

Growing community of inventors

Austin, TX, United States of America

Karthik Sundaram