International Business Machines Corporation

Apple Inc.

Qualcomm Incorporated

Brian Robert Mestan

Michael Lance Karm

David Mui

Peter G Soderquist

Gideon N Levinsky

David S Levitan

Ronald P Hall

John D Pape

Balaram Sinharoy

Giles R Frazier

Lydia Mai Do

Mauricio J Serrano

Richard William Doing

Mridul Agarwal

Jason Alan Cox

Kimberly Marie Fernsler

Benjamin W Stolt

Jeffrey Richard Summers

Kevin Neal Magill

Deepak Limaye

Dung Q Nguyen

Susan Elizabeth Eisen

Albert James Van Norstrand, Jr

Suresh Kumar Venkumahanti

Brian R Konigsburg

William Elton Burky

Pradeep Kanapathipillai

Nikhil Gupta

Amit Kumar

Joshua R Smith

Jiajin Tu

Michael O Klett

A reservation station that includes primary and secondary storage circuits is disclosed. The primary storage circuit may include multiple full entries, while the secondary storage circuit includes multiple store-data entries. A given store-data entry of the multiple store-data entries stores a subset of the information stored in a given full entry of the multiple full entries. In response to a determination that a store address associated with a particular store operation, stored in particular full entry of the multiple full entries, has been available for use for a threshold number of cycles without store data associated with the particular store operation being available for use, a control circuit may transfer the particular store operation to a particular store-data entry of the multiple store-data entries.

Reservation station with primary and secondary storage circuits for store operations

In an embodiment, a processor comprises an atomic predictor circuit to predict whether or not an atomic operation will complete successfully. The prediction may be used when a subsequent load operation to the same memory location as the atomic operation is executed, to determine whether or not to forward store data from the atomic operation to the subsequent load operation. If the prediction is successful, the store data may be forwarded. If the prediction is unsuccessful, the store data may not be forwarded. In cases where an atomic operation has been failing (not successfully performing the store operation), the prediction may prevent the forwarding of the store data and thus may prevent a subsequent flush of the load.

Atomic operation predictor to predict whether an atomic operation will complete successfully

Systems, apparatuses, and methods for performing efficient translation lookaside buffer (TLB) invalidation operations for splintered pages are described. When a TLB receives an invalidation request for a specified translation context, and the invalidation request maps to an entry with a relatively large page size, the TLB does not know if there are multiple translation entries stored in the TLB for smaller splintered pages of the relatively large page. The TLB tracks whether or not splintered pages for each translation context have been installed. If a TLB invalidate (TLBI) request is received, and splintered pages have not been installed, no searches are needed for splintered pages. To refresh the sticky bits, whenever a full TLB search is performed, the TLB rescans for splintered pages for other translation contexts. If no splintered pages are found, the sticky bit can be cleared and the number of full TLBI searches is reduced.

Reducing translation lookaside buffer searches for splintered pages

Techniques are disclosed relating to controlling cache replacement. In some embodiments, a computing system performs multiple searches of a data structure, where one or more of the searches traverse multiple links between elements of the data structure. The system may cache, in a traversal cache, traversal information that is usable by searches to skip one or more links traversed by one or more prior searches. The system may store tracking information that indicates a location in the traversal cache at which prior traversal information for a first search is stored. The system may select, based on the tracking information, an entry in the traversal cache for new traversal information generated by the first search. The selection may override a default replacement policy for the traversal cache, e.g., to select the location in the traversal cache to replace the prior traversal information with the new traversal information.

Cache replacement based on traversal tracking

Systems, apparatuses, and methods for limiting translation lookaside buffer (TLB) searches using active page size are described. A TLB stores virtual-to-physical address translations for a plurality of different page sizes. When the TLB receives a command to invalidate a TLB entry corresponding to a specified virtual address, the TLB performs, for the plurality of different pages sizes, multiple different lookups of the indices corresponding to the specified virtual address. In order to reduce the number of lookups that are performed, the TLB relies on a page size presence vector and an age matrix to determine which page sizes to search for and in which order. The page size presence vector indicates which page sizes may be stored for the specified virtual address. The age matrix stores a preferred search order with the most probable page size first and the least probable page size last.

Limiting translation lookaside buffer searches using active page size

Techniques are disclosed relating to controlling cache replacement. In some embodiments, search control circuitry is configured to perform multiple searches of a data structure (e.g., page table walks) where searches traverse multiple links between elements of the data structure. In some embodiments, a traversal cache caches traversal information that is usable by searches to skip one or more links traversed by one or more prior searches. In some embodiments, tracking control circuitry stores tracking information in a first entry, where the tracking information indicates a location in the traversal cache at which prior traversal information for a first search is stored. In some embodiments, replacement control circuitry selects, based on the tracking information in the first entry of the tracking control circuitry, an entry in the traversal cache for new traversal information generated by the first search (which may include selecting the first entry to override a default replacement policy).

Systems, apparatuses, and methods for implementing translation lookaside buffer (TLB) striping to enable efficient invalidation operations are described. TLB sizes are growing in width (more features in a given page table entry) and depth (to cover larger memory footprints). A striping scheme is proposed to enable an efficient and high performance method for performing TLB maintenance operations in the face of this growth. Accordingly, a TLB stores first attribute data in a striped manner across a plurality of arrays. The striped manner allows different entries to be searched simultaneously in response to receiving an invalidation request which identifies a particular attribute of a group to be invalidated. Upon receiving an invalidation request, the TLB generates a plurality of indices with an offset between each index and walks through the plurality of arrays by incrementing each index and simultaneously checking the first attribute data in corresponding entries.

Translation lookaside buffer striping for efficient invalidation operations

Techniques are disclosed relating to filtering access to a content-addressable memory (CAM). In some embodiments, a processor monitors for certain microarchitectural states and filters access to the CAM in states where there cannot be a match in the CAM or where matching entries will not be used even if there is a match. In some embodiments, toggle control circuitry prevents toggling of input lines when filtering CAM access, which may reduce dynamic power consumption. In some example embodiments, the CAM is used to access a load queue to validate that out-of-order execution for a set of instructions matches in-order execution, and situations where ordering should be checked are relatively rare.

Content-addressable memory filtering based on microarchitectural state

Atomic operation predictor to predict if an atomic operation will successfully complete and a store queue to selectively forward data based on the predictor

A system and method for efficiently forwarding cache misses to another level of the cache hierarchy. Logic in a cache controller receives a first non-cacheable load miss request and stores it in a miss queue. When the logic determines the target address of the first load miss request is within a target address range of an older pending second load miss request stored in the miss queue with an open merge window, the logic merges the two requests into a single merged miss request. Additional requests may be similarly merged. The logic issues the merged miss requests based on determining the merge window has closed. The logic further prevents any other load miss requests, which were not previously merged in the merged miss request before it was issued, from obtaining a copy of data from the returned fill data. Such prevention in a non-coherent memory computing system supports memory ordering.

Managing serial miss requests for load operations in a non-coherent memory system

A system and method for efficiently supporting a cache memory hierarchy potentially using a zero size cache in a level of the hierarchy. In various embodiments, logic in a lower-level cache controller or elsewhere receives a miss request from an upper-level cache controller. When the requested data is non-cacheable, the logic sends a snoop request with an address of the memory access operation to the upper-level cache controller to determine whether the requested data is in the upper-level data cache. When the snoop response indicates a miss or the requested data is cacheable, the logic retrieves the requested data from memory. When the snoop response indicates a hit, the logic retrieves the requested data from the upper-level cache. The logic completes servicing the memory access operation while preventing cache storage of the received requested data in a cache at a same level of the cache memory hierarchy as the logic.

Processing memory accesses while supporting a zero size cache in a cache hierarchy

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

A method, system, and computer program product for instruction fetching within a processor instruction unit, utilizing a loop buffer, one or more virtual loop buffers, and/or an instruction buffer. During instruction fetch, modified instruction buffers coupled to an instruction cache (I-cache) temporarily store instructions from a single branch, backwards short loop. The modified instruction buffers may be a loop buffer, one or more virtual loop buffers, and/or an instruction buffer. The instruction fetch within the instruction unit of a processor retrieves the instructions for the short loop from the modified buffers during the loop cycles of the single branch, backwards short loop, rather than from the instruction cache.

Growing community of inventors

Austin, TX, United States of America

Brian Robert Mestan