Apple Inc.

Pradeep Kanapathipillai

Hari Kannan

Gerard Richard Williams

Mridul Agarwal

Richard F Russo

Stephan G Meier

Brian P Lilly

John H Mylius

Perumal R Subramoniam

Bernard Joseph Semeria

Andrew J Beaumont-Smith

James Vash

Aditya Kesiraju

Tyler J Huberty

Kulin N Kothari

Rajat Goel

Sandeep Kumar Gupta

Daniel C Murray

David James Williamson

Peter J Bannon

Gideon N Levinsky

Honkai Tam

John H Kelm

Jeff Gonion

Christopher M Tsay

Mahnaz P Sadoughi-Yarandi

Jeffry E Gonion

Joshua P De Cesare

Greg M Hess

Shingo Suzuki

Brian Robert Mestan

Po-Yung Chang

Conrado Blasco

Shih-Chieh Wen

Chetana Nagendra Keltcher

Joshua R Smith

Ran A Chachick

Charles E Tucker

Peter G Soderquist

Sean M Reynolds

Michael J Swift

Richard H Larson

Boris S Alvarez-Heredia

Raghava Rao V Denduluri

Matthew R Johnson

Edvin Catovic

Pankaj Raghuvanshi

Pankaj Lnu

Ming-Ta Hsu

Zhaoming Hu

Gerard R Williams Iii

Hari S Kannan

Techniques are disclosed relating to data synchronization barrier operations. A system includes a first processor that may receive a data barrier operation request from a second processor include in the system. Based on receiving that data barrier operation request from the second processor, the first processor may ensure that outstanding load/store operations executed by the first processor that are directed to addresses outside of an exclusion region have been completed. The first processor may respond to the second processor that the data barrier operation request is complete at the first processor, even in the case that one or more load/store operations that are directed to addresses within the exclusion region are outstanding and not complete when the first processor responds that the data barrier operation request is complete.

DSB operation with excluded region

In an embodiment, a coprocessor may include a bypass indication which identifies execution circuitry that is not used by a given processor instruction, and thus may be bypassed. The corresponding circuitry may be disabled during execution, preventing evaluation when the output of the circuitry will not be used for the instruction. In another embodiment, the coprocessor may implement a grid of processing elements in rows and columns, where a given coprocessor instruction may specify an operation that causes up to all of the processing elements to operate on vectors of input operands to produce results. Implementations of the coprocessor may implement a portion of the processing elements. The coprocessor control circuitry may be designed to operate with the full grid or partial grid, reissuing instructions in the partial grid case to perform the requested operation. In still another embodiment, the coprocessor may be able to fuse vector mode operations.

Coprocessors with bypass optimization, variable grid architecture, and fused vector operations

A method and system for compressing and decompressing data is disclosed. A compression command may initiate the prefetching of first data, which may be stored in a first buffer. Multiple words of the first data may be read from the first buffer and used to generate a plurality of compressed packets, each of which includes a command specifying a type of packet. The compressed packets may be combined into a group and multiple groups may be combined and stored in a second buffer. A decompression command may initiate the prefetching of second data, which is stored in the first buffer. A portion of the second data may be read from the first buffer and used to generate a group of compressed packets. Multiple output words may be generated dependent upon the group of compressed packets.

Hardware compression and decompression engine

A system and method for efficiently transferring address mappings and data access permissions corresponding to the address mappings. A computing system includes at least one processor and memory for storing a page table. In response to receiving a memory access operation comprising a first address, the address translation unit is configured to identify a data access permission based on a permission index corresponding to the first address, and access data stored in a memory location of the memory identified by a second address in a manner defined by the retrieved data access permission. The address translation unit is configured to access a table to identify the data access permission, and is configured to determine the permission index and the second address based on the first address. A single permission index may correspond to different permissions for different entities within the system.

Unified address translation

Systems, apparatuses, and methods for retaining architected state for relatively frequent switching between sleep and active operating states are described. A processor receives an indication to transition from an active state to a sleep state. The processor stores a copy of a first subset of the architected state information in on-die storage elements capable of retaining storage after power is turned off. The processor supports programmable input/output (PIO) access of particular stored information during the sleep state. When a wakeup event is detected, circuitry within the processor is powered up again. A boot sequence and recovery of architected state from off-chip memory are not performed. Rather than fetch from a memory location pointed to by a reset base address register, the processor instead fetches an instruction from a memory location pointed to by a restored program counter of the retained subset of the architected state information.

Architected state retention for a frequent operating state switching processor

A system and method for efficiently performing maintenance on a cache. In various embodiments, control logic in a cache controller or elsewhere receives an indication for invalidating a range of virtual-to-physical mappings in a given translation lookaside buffer (TLB). The logic determines a first latency to invalidate entries of the TLB based on a number of addresses in the range and a number of supported page sizes simultaneously stored in the TLB. The logic determines a second latency based on a number of entries in the TLB. If the first latency is greater, then the logic traverses through each TLB entry and invalidates TLB entries storing a virtual address within the range. If the first latency is smaller, then the logic traverses through each address in the range and invalidates TLB entries storing a virtual address within the range.

Translation lookaside buffer invalidation by range

In an embodiment, a processor may implement an access map-pattern match (AMPM)-based prefetch circuit for a multi-level cache system. The access patterns that are matched to the access maps may include prefetches for different cache levels. Centralizing the generation of prefetches into one prefetch circuit may provide better observability and controllability of prefetching at various levels of the cache hierarchy, in an embodiment. Prefetches at different levels may be controlled individually based on the accuracy of those prefetches, in an embodiment. Additionally, in an embodiment, access patterns that are longer that a given threshold may have the granularity of the prefetches change so that more data is prefetched and the prefetches occur farther in advance, in some embodiments.

Unified prefetch circuit for multi-level caches

Systems, apparatuses, and methods for optimizing a load-store dependency predictor (LSDP). When a younger load instruction is issued before an older store instruction and the younger load is dependent on the older store, the LSDP is trained on this ordering violation. A replay/flush indicator is stored in a corresponding entry in the LSDP to indicate whether the ordering violation resulted in a flush or replay. On subsequent executions, a dependency may be enforced for the load-store pair if a confidence counter is above a threshold, with the threshold varying based on the status of the replay/flush indicator. If a given load matches on multiple entries in the LSDP, and if at least one of the entries has a flush indicator, then the given load may be marked as a multimatch case and forced to wait to issue until all older stores have issued.

Load/store dependency predictor optimization for replayed loads

Systems, apparatuses, and methods for committing store instructions out of order from a store queue are described. A processor may store a first store instruction and a second store instruction in the store queue, wherein the first store instruction is older than the second store instruction. In response to determining the second store instruction is ready to commit to the memory hierarchy, the processor may allow the second store instruction to commit before the first store instruction, in response to determining that all store instructions in the store queue older than the second store instruction are non-speculative. However, if it is determined that at least one store instruction in the store queue older than the second store instruction is speculative, the processor may prevent the second store instruction from committing to the memory hierarchy before the first store instruction.

Out of order store commit

A load-store unit having one or more banked queues is disclosed. In one embodiment, a load-store unit includes at least one queue that is subdivided into multiple banks. Although divided into multiple banks, the queue logically appears to software as a single queue. A first bank of the queue includes a first plurality of entries, with the second bank of the queue having a second plurality of entries, wherein each of the entries is arranged to store memory instructions. Each of the banks is associated with corresponding logic circuitry that controls one or more pointers for that bank. The pointer information may be exchanged between the logic circuits associated with the banks. Based on the pointer information that is exchanged, each bank may output (e.g., for retirement) one entry per cycle.

Load-store unit with banked queue

A processor includes an instruction issue circuit, and high-utilization and low-utilization execution unit circuits coupled to execute instructions received from the instruction issue unit. On average, utilization of the low-utilization execution unit circuit is lower than utilization of the high-utilization execution unit circuit. The processor also includes a retention circuit coupled to a different power domain than the low-utilization execution unit circuit, and a power management circuit. The power management circuit may be configured to detect that inactivity of the low-utilization execution unit circuit satisfies a threshold inactivity level; upon detecting that the threshold inactivity level is satisfied, cause architecturally-visible state of the low-utilization execution unit circuit to be copied to the retention circuit; and subsequent to copying of the architecturally-visible state to the retention circuit, cause the low-utilization execution unit circuit to enter a power-off state, where the retention circuit retains stored data during the power-off state.

Execution unit power management

Systems, apparatuses, and methods for modifying access permissions in a processor. A processor may include one or more permissions registers for managing access permissions. A first permissions register may be utilized to override access permissions embedded in the page table data. A plurality of bits from the page table data may be utilized as an index into the first permissions register for the current privilege level. An attribute field may be retrieved from the first permissions register to determine the access permissions for a given memory request. A second permissions register may also be utilized to set the upper and lower boundary of a region in physical memory where the kernel is allowed to execute. A lock register may prevent any changes from being made to the second permissions register after the second permissions register has been initially programmed.

Access permissions modification

Systems, methods, and apparatuses for reducing the load to load/store address latency in an out-of-order processor. When a producer load is detected in the processor pipeline, the processor predicts whether the producer load is going to hit in the store queue. If the producer load is predicted not to hit in the store queue, then a dependent load or store can be issued early. The result data of the producer load is then bypassed forward from the data cache directly to the address generation unit. This result data is then used to generate an address for the dependent load or store, reducing the latency of the dependent load or store by one clock cycle.

Reducing latency for pointer chasing loads

Growing community of inventors

Santa Clara, CA, United States of America

Pradeep Kanapathipillai