Marvell Asia Pte., Ltd.

Cavium, Inc.

Sifive, Inc.

Marvellworld Trade Ltd.

David Kravitz

Wilson Parkhurst Snyder, Ii

David Albert Carlson

Daniel Edward Dever

Bradley Gene Burgess

Richard Eugene Kessler

Shubhendu Sekhar Mukherjee

Edward J McLellan

Manan Salvi

Alexandre Solomatnikov

Rabin Sugumar

Bryan Chin

Nitin Dhiroobhai Godiwala

Gerald Lampert

Andrew Hanselman

Systems and methods are disclosed for transferring data between a memory system and a vector register file. For example, a system may include a vector pipeline including a vector physical register file; a load store unit; one or more pipeline stages configured to decode a vector memory instruction to obtain a macro-operation and dispatch the macro-operation to both the load store unit and the vector pipeline, and a baler circuitry, including a buffer with entries. The vector pipeline is configured to crack the macro-operation into multiple micro-operations. The baler circuitry is configured to implement the multiple micro-operations to transfer data between one or more selected entries of the buffer and respective registers of the vector physical register file. The load store unit is configured to implement the macro-operation to transfer data between one or more addresses in a memory system and the one or more selected entries of the buffer.

Vector load store operations in a vector pipeline using a single operation in a load store unit

Systems and methods are disclosed for transferring an operand between a vector pipeline and a scalar pipeline. For example, some methods may include transferring an operand from a scalar pipeline to a scalar-to-vector buffer responsive to the scalar pipeline executing a first micro-op, wherein the scalar-to-vector buffer includes an entry having a width equal to a width of a scalar register of the scalar pipeline and a data store configured to store an indication mapping the entry to the first micro-op; updating the data store to include the indication mapping the entry to the first micro-op; identifying, by the vector pipeline in response to execution of a second micro-op and in dependence on the indication mapping the entry to the first micro-op, the entry storing the operand; and transferring the operand from the entry in the scalar-to-vector buffer to the vector pipeline responsive to the vector pipeline executing the second micro-op.

Transfer buffer between a scalar pipeline and vector pipeline

Apparatus and methods which bundle micro-operations with respect to a vector instruction, dynamically allocate register blocks for a vector instruction, and track the registers using valid bits. A method includes decoding, by a decoder, a vector instruction having a length multiplier of at least two into a number of micro-operations less than the length multiplier, allocating, by an issue queue, an issue queue entry to each of the number of micro-operations and executing, by the issue queue with execution units, each of the number of micro-operations a number of times from the issue queue entry to collectively match the length multiplier.

Bundling and dynamic allocation of register blocks for vector instructions

An instruction execution circuit operable to reduce two or more micro-operations into one by producing multiple permutation and merge results in one execution cycle. The execution circuit includes a permutation and merge switching fabric and a bank of multiplexers. For a fetched instruction, a decoder decodes an opcode to generate a set of control indications used to control the multiplexers to select bytes from the respective inputs that are destined for each of the multiple results. In this manner, multiple permutation results can be output from the execution circuits in one micro-operation.

Pair merge execution units for microinstructions

Examples of a carry chain for performing an operation on operands each including elements of a selectable size is provided. Advantageously, the carry chain adapts to elements of different sizes. The carry chain determines a mask based on a selected size of an element. The carry chain selects, based on the mask, whether to carry a partial result of an operation performed on corresponding first portions of a first operand and a second operand into a next operation. The next operation is performed on corresponding second portions of the first operand and the second operand, and, based on the selection, the partial result of the operation. The carry chain stores, in a memory, a result formed from outputs of the operation and the next operation.

Carry chain for SIMD operations

A method of implementing a processor architecture and corresponding system includes operands of a first size and a datapath of a second size. The second size is different from the first size. Given a first array of registers and a second array of registers, each register of the first and second arrays being of the second size, selecting a first register and corresponding second register from the first array and the second array, respectively, to perform operations of the first size. This allows a user, who is interfacing with the hardware processor through software, to provide data of the datapath bit-width instead of the register bit-width. Advantageously, the user is agnostic to the size of the registers.

Implementing 128-bit SIMD operations on a 64-bit datapath

A method of implementing a processor architecture and corresponding system includes operands of a first size and a datapath of a second size. The second size is different from the first size. Given a first array of registers and a second array of registers, each register of the first and second arrays being of the second size, selecting a first register and corresponding second register from the first array and the second array, respectively, to perform operations of the first size. Advantageously, this allows a user, who is interfacing with the hardware processor through software, to provide data to the processor agnostic to the size of the registers and datapath bit-width of the processor.

Instructions are executed in a pipeline. Storage accessible to the pipeline stores branch prediction information characterizing results of branch instructions previously executed. A predicted branch result is provided, for at least some branch instructions, based on a selected predictor of multiple predictors. An actual branch result is provided based on an executed branch instruction, and the branch prediction information is updated based on the actual branch result. The predictors include: a first predictor that determines the predicted branch result based on at least a portion of the branch prediction information; and a second predictor that determines the predicted branch result independently from the branch prediction information.

Managing predictor selection for branch prediction

Instructions are executed in a pipeline of a processor, where each instruction is associated with a particular context. A first storage stores branch prediction information characterizing results of branch instructions previously executed. The first storage is dynamically partitioned into partitions of one or more entries. Dynamically partitioning includes updating a partition to include an additional entry by associating the additional entry with a particular subset of one or more contexts. A predicted branch result is determined based on at least a portion of the branch prediction information. An actual branch result provided based on an executed branch instruction is used to update the branch prediction information. Providing a predicted branch result for a first branch instruction includes retrieving a first entry from a first partition based at least in part on an identified first subset of one or more contexts associated with the first branch instruction.

Managing branch prediction information for different contexts

A method and a system embodying the method for work request arbitration, comprising receiving a work request, the work request indicating one or more groups from a plurality of groups; determining at least one of a plurality of parameters in accordance with the received work request; determining eligibility to provide work among the one or more groups that have work in a work queue in accordance with a first set of the plurality of parameters; and selecting one of the determined eligible groups to provide the work in accordance with a second set of the plurality of parameters is disclosed.

Method and an apparatus for work request arbitration in a network processor

One disclosed embodiment provides an integrated circuit that has a plurality of processors and a plurality of processor trace collection logic units. Each processor trace collection logic unit corresponds with, and is operatively coupled to, one of the processors. A separate filtering logic unit is operatively coupled to the plurality of processor trace collection logic units. In some embodiments of the integrated circuit, each processor trace collection logic unit is operative to continuously collect processor trace information from a corresponding operatively coupled processor. Each filtering logic unit is operative to monitor the continuous processor trace information for occurrence of a predetermined condition, and to store some of the processor trace information to memory in response to occurrence of that condition.

Apparatus and method for distributed instruction trace in a processor system

A network processor includes a schedule, sync and order (SSO) module for scheduling and assigning work to multiple processors. The SSO includes an on-deck unit (ODU) that provides a table having several entries, each entry storing a respective work queue entry, and a number of lists. Each of the lists may be associated with a respective processor configured to execute the work, and includes pointers to entries in the table. pointer is added to the list based on an indication of whether the associated processor accepts the WQE corresponding to the pointer.

Work request processor

A network processor includes a schedule, sync and order (SSO) module for scheduling and assigning work to multiple processors. The SSO includes an on-deck unit (ODU) that provides a table having several entries, each entry storing a respective work queue entry, and a number of lists. Each of the lists may be associated with a respective processor configured to execute the work, and includes pointers to entries in the table. A pointer is added to the list based on an indication of whether the associated processor accepts the WQE corresponding to the pointer.

Growing community of inventors

Cambridge, MA, United States of America

David Kravitz