Apple Inc.

Advanced Micro Devices Corporation

Nikhil Gupta

Aditya Kesiraju

Brett Stanley Feero

Kulin N Kothari

Viney Gautam

Stephan G Meier

Mridul Agarwal

Gideon N Levinsky

Tyler J Huberty

Ian D Kountanis

Marius Evers

Marvin A Denman

Conrado Blasco

James David Dundas

Debjit Das Sarma

Deepankar Duggal

Francesco Spadini

Muawya M Al-Otoom

Amit Kumar

Wei-Yu Chen

Douglas R Williams

Ethan R Schuchman

Sahil Arora

Pankaj Lnu

James R Hakewill

Mridul Agarawal

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

An apparatus includes a cache memory circuit, and a hierarchal store queue circuit that further includes a primary queue and a secondary queue. The hierarchal store queue circuit may be configured to write incoming store requests to the primary queue in response to the primary queue currently having capacity, and to write incoming store requests to the secondary queue in response to the primary queue currently not having capacity. The hierarchal store queue circuit may be further configured to commit store requests to the cache memory circuit from the primary queue but not from the secondary queue. In response to a determination that the primary queue currently has capacity, the hierarchal store queue circuit may perform a transfer of at least one store request from the secondary queue to the primary queue.

Hierarchical store queue circuit

An apparatus includes a processor core that includes an instruction decode circuit and a control circuit. The instruction decode circuit is configured to decode instructions, including a plurality of store instructions used to store information in a memory hierarchy. The control circuit is configured, after a particular store instruction is decoded, to preserve store information related to the particular store instruction, including a first program counter value for the particular store instruction. In response to decoding a subsequent load instruction with a corresponding second program counter value, the control circuit is configured to determine, using the first and second program counter values, whether a dependency has been established between the subsequent load instruction and the particular store instruction. In response to a determination that the dependency has been established, the control circuit is configured to use the preserved store information to perform the subsequent load instruction.

Program counter zero-cycle loads

In an embodiment, a processor includes a buffer in an interface unit. The buffer may be used to accumulate coprocessor instructions to be transmitted to a coprocessor. In an embodiment, the processor issues the coprocessor instructions to the buffer when ready to be issued to the coprocessor. The interface unit may accumulate the coprocessor instructions in the buffer, generating a bundle of instructions. The bundle may be closed based on various predetermined conditions and then the bundle may be transmitted to the coprocessor. If a sequence of coprocessor instructions appears consecutively in a program, the rate at which the instructions are provided to the coprocessor (on average) at least matches the rate at which the coprocessor consumes the instructions, in an embodiment.

Coprocessor operation bundling

In an embodiment, a processor comprises a reservation station that issues a first load operation for execution, a store queue, and a replayed load buffer coupled in parallel with the reservation station. During execution of the first load operation, the store queue detects that the first load operation hits on a first store operation in the store queue that lacks store data and causes a replay of the first load operation. The replayed load buffer captures an identifier of the first load operation and the first store operation based on the replay of the first load operation, wherein the replayed load buffer monitors the reservation station for issuance of a first store data operation corresponding to the first store operation and issues the first load operation for reexecution based on the issuance of the first store data operation.

Buffer for replayed loads in parallel with reservation station for rapid rescheduling

In an embodiment, a processor includes a plurality of prefetch circuits configured to prefetch data into a data cache. A primary prefetch circuit may be configured to generate first prefetch requests in response to a demand access, and may be configured to invoke a second prefetch circuit in response to the demand access. The second prefetch circuit may implement a different prefetch mechanism than the first prefetch circuit. If the second prefetch circuit reaches a threshold confidence level in prefetching for the demand access, the second prefetch circuit may communicate an indication to the primary prefetch circuit. The primary prefetch circuit may reduce a number of prefetch requests generated for the demand access responsive to the communication from the second prefetch circuit.

Secondary prefetch circuit that reports coverage to a primary prefetch circuit to limit prefetching by primary prefetch circuit

In an embodiment, at least one CPU processor and at least one coprocessor are included in a system. The CPU processor may issue operations to the coprocessor to perform, including load/store operations. The CPU processor may generate the addresses that are accessed by the coprocessor load/store operations, as well as executing its own CPU load/store operations. The CPU processor may include a memory ordering table configured to track at least one memory region within which there are outstanding coprocessor load/store memory operations that have not yet completed. The CPU processor may delay CPU load/store operations until the outstanding coprocessor load/store operations are complete. In this fashion, the proper ordering of CPU load/store operations and coprocessor load/store operations may be maintained.

Coprocessor memory ordering table

In an embodiment, a processor includes a load/store unit that executes load/store operations. The load/store unit may implement a two-level load queue. One of the load queues, referred to as a load retirement queue (LRQ), may track load operations from initial execution to retirement. Ordering constraints may be enforced using the LRQ. The other load queue, referred to as a load execution queue (LEQ), may track loads from initial execution to forwarding of data. Replay may be managed by the LEQ. In an embodiment, the LEQ may be smaller than the LRQ, which may permit the management of replay while still meeting timing requirements. Additionally, the larger LRQ may permit more load operations to be pending (not retired) in the processor, widening the window for out of order execution and supporting potentially higher processor performance.

Processor with multiple load queues including a queue to manage ordering and a queue to manage replay

A method and system for prefetching instructions is disclosed. A signature may be updated in response to determining that a received instruction initiates a transition in function. A plurality of entries stored in a memory may be searched in response to determining the signature has been updated and a prefetch operation may be initiated in response to determining that the signature matches a given entry of the plurality of entries.

Callgraph signature prefetch

A prefetch circuit may include a memory, each entry of which may store an address and other prefetch data used to generate prefetch requests. For each entry, there may be at least one 'quality factor' (QF) that may control prefetch request generation for that entry. A global quality factor (GQF) may control generation of prefetch requests across the plurality of entries. The prefetch circuit may include one or more additional prefetch mechanisms. For example, a stride-based prefetch circuit may be included that may generate prefetch requests for strided access patterns having strides larger than a certain stride size. Another example is a spatial memory streaming (SMS)-based mechanism in which prefetch data from multiple evictions from the memory in the prefetch circuit is captured and used for SMS prefetching based on how well the prefetch data appears to match a spatial memory streaming pattern.

Prefetch circuit with global quality factor to reduce aggressiveness in low power modes

A processor, a device, and a non-transitory computer readable medium for performing branch prediction in a processor are presented. The processor includes a front end unit. The front end unit includes a level 1 branch target buffer (BTB), a BTB index predictor (BIP), and a level 1 hash perceptron (HP). The BTB is configured to predict a target address. The BIP is configured to generate a prediction based on a program counter and a global history, wherein the prediction includes a speculative partial target address, a global history value, a global history shift value, and a way prediction. The HP is configured to predict whether a branch instruction is taken or not taken.

Bandwidth increase in branch prediction unit and level 1 instruction cache

A method and apparatus are provided for increasing the accuracy of a branch predictor. A branch prediction table provides a first instance of a branch prediction value associated with an instruction being speculatively executed a first time; and provides a second instance of the branch prediction value associated with the instruction being speculatively executed a second rime. The first instance of the branch prediction value may be subsequently revised after the instruction associated with the first instance of the branch prediction value is retired. Information regarding whether that branch instruction was accurately predicted may then be used to update the branch prediction table and the second instance of the branch prediction value.

Growing community of inventors

Santa Clara, CA, United States of America

Nikhil Gupta