Mythic, Inc.

Freescale Semiconductor,inc.

Arm Limited

Nxp Usa, Inc.

Sergio Schuler

Malav Parikh

Zainab Nasreen Zaidi

David Alan Fick

Bryant Sorensen

Raul Garibay

Alex Dang-Tran

Vimal Reddy

Adam Caughron

Paul Toth

Erik Schlanger

Michael D Snyder

Gary L Whisenhunt

Andrew Morten

Raul A Garibay, Jr

Natarajan Seshan

Yasuo Ishii

Scott C Johnson

Alex James Waugh

Klas Magnus Bruce

Geoffray Lacourba

Michael Brian Schinzler

Andrew John Turner

Chang Joo Lee

Trinh Huy Nguyen

Leick D Robinson

David C Holloway

Matt B Smittle

Scott Johnson

David M Thompson

Klas M Bruce

A system and method for automated data propagation and automated data processing within an integrated circuit includes an intelligence processing integrated circuit comprising at least one intelligence processing pipeline, wherein the at least one intelligence processing pipeline includes: a main data buffer that stores input data; a plurality of distinct intelligence processing tiles, wherein each distinct intelligence processing tile includes a computing circuit and a local data buffer; a token-based governance module, the token-based governance module implementing: a first token-based control data structure; a second token-based control data structure, wherein the first token-based control data structure and the second-token based control data operate in cooperation to control an automated flow of the input data and/or an automated processing of the input data through the at least one intelligence processing pipeline.

Systems and methods for implementing an intelligence processing computing architecture

A system and method for a computing tile of a multi-tiled integrated circuit includes a plurality of distinct tile computing circuits, wherein each of the plurality of distinct tile computing circuits is configured to receive fixed-length instructions; a token-informed task scheduler that: tracks one or more of a plurality of distinct tokens emitted by one or more of the plurality of distinct tile computing circuits; and selects a distinct computation task of a plurality of distinct computation tasks based on the tracking; and a work queue buffer that: contains a plurality of distinct fixed-length instructions, wherein each one of the fixed-length instructions is associated with one of the plurality of distinct computation tasks; and transmits one of the plurality of distinct fixed-length instructions to one or more of the plurality of distinct tile computing circuits based on the selection of the distinct computation task by the token-informed task scheduler.

Tile subsystem and method for automated data flow and data processing within an integrated circuit architecture

A data processing apparatus includes receive circuitry that receives an indication of a trigger block of instructions. Branch prediction circuitry provides, in response to the trigger block of instructions, branch predictions in respect of a branch within: a subsequent block of instructions subsequent to the trigger block of instructions in execution order, when in a 1-taken mode of operation and a later block of instructions subsequent to the subsequent block of instructions in execution order, when in a 2-taken mode of operation

Branch predictor triggering

A system and method for virtually addressing an array of accelerator tiles of a mixed-signal integrated circuit includes testing each of a plurality of distinct matrix multiply accelerator (MMA) tiles of a grid of MMA tiles, the grid of MMA tiles being defined by the plurality of distinct grid of MMA tiles being arranged in a plurality of rows and a plurality of columns along an integrated circuit, each of the plurality of distinct MMA tiles within the grid of MMA tiles having a distinct physical address on the integrated circuit; identifying one or more defective MMA tiles within the grid of MMA tiles based on the testing; and configuring the grid of MMA tiles with a plurality of virtual addresses for routing data to or routing data from one or more non-defective MMA tiles of grid of MMA tiles based on identifying the one or more defective MMA tiles.

Systems and methods for implementing redundancy for tile-based intelligence processing computing architecture

Aspects of the present disclosure relate to an interconnect comprising an interface to couple to a master device, the interface comprising buffer storage. The interface is configured to receive a request from the master device for data comprising a plurality of data blocks, the master device requiring the data blocks in a defined order. A data collator is configured to: receive the request; issue a data pull request to cause the interface to allocate buffer space in the buffer storage for buffering the requested data; and responsive to receiving a confirmation that the buffer space is allocated, provide the requested data to the buffer storage. The interface is configured to employ the buffer storage to enable re-ordering of the plurality of data blocks of the requested data, prior to outputting the plurality of data blocks to the master device; and output the plurality of data blocks to the master device in the defined order.

Apparatus and method for providing data to a master device

Systems and methods include an integrated circuit that includes a plurality of computing tiles, wherein each of the plurality of computing tiles includes: a matrix multiply accelerator, a computing processing circuit; and a flow scoreboard module; a local data buffer, wherein the plurality of computing tiles together define an intelligence processing array; a network-on-chip system comprising: a plurality of network-on-chip routers establishing a communication network among the plurality of computing tiles, wherein each network-on-chip router is in operable communication connection with at least one of the plurality of computing tiles and a distinct network-on-chip router of the plurality of network-on-chip routers; and an off-tile buffer that is arranged in remote communication with the plurality of computing tiles, wherein the off-tile buffer stores raw input data and/or data received from an upstream process or an upstream device.

A technique for scheduling execution of threads at a processor is disclosed. The technique includes executing a thread de-emphasis instruction of a thread that de-emphasizes the thread until the number of pending memory transactions, such as cache misses, associated with the thread are at or below a threshold. While the thread is de-emphasized, other threads at the processor that have a higher priority can be executed or assigned system resources. Accordingly, the likelihood of a stall in the processor is reduced.

Thread de-emphasis instruction for multithreaded processor

A data processing device and methods thereof are disclosed. The data processing device can operate in three different modes. In a first, N-bit mode, the data processing device performs memory accesses based on N-bit values and performs arithmetic operations using N-bit values. In a second, hybrid N-bit/M-bit mode, the data processing device performs memory accesses based on M-bit values, where M is less than N, and performs arithmetic operations using N-bit values. In a third, M-bit mode, the data processing device performs memory accesses based on M-bit values and performs arithmetic operations using M-bit values. The three modes provide for compatibility with a wide range of applications. Further operation in the M-bit mode can provide a power savings when implementing applications compatible with that mode.

Multiple address and arithmetic bit-mode data processing device and methods thereof

A processor has a fetch unit and a branch execution unit. The fetch unit has a branch predictor. The branch predictor has a branch target buffer and a branch direction predictor. A wake value is a number of instruction fetches that is predicted to be performed after a fetch of a branch. Thus, for a first branch, for example, a first wake number is predicted. A low power mode of the branch predictor is enabled for a duration of the first wake value in response to hit in the branch target buffer in which the hit is in response to the first branch.

Dynamic branch prediction using a wake value to enable low power mode for a predicted number of instruction fetches between a branch and a subsequent branch

Growing community of inventors

Austin, TX, United States of America

Sergio Schuler