Sambanova Systems, Inc.

Apple Inc.

Advanced Micro Devices Corporation

Oracle International Corporation

Nathan Francis Sheeley

Raghu Prabhakar

Matheen Musaddiq

Sitanshu Gupta

Sumti Jairath

Scott Layson Burson

Pramod Nataraja

Ajit Punj

Paul Joseph Jordan

Ram Sivaramakrishnan

Erik Schlanger

Matthew B Smittle

Adrian J Isles

Mingran Wang

Mark H Nodine

Bill K Kwan

Weihang Fan

Eric Swartzendruber

Maulik Desai

Irfan Waheed

Amitabh Menon

Nicolas Xavier Pena

Mark Greenberg

Patrick Peters

Chin-Chia Kuo

Darshan Dhimantkumar Gandhi

Munsefar Khaleque

A compiler transforms a high-level program into configuration data for a coarse-grained reconfigurable (CGR) data processor with an array of CGR units. The compiler includes a method that identifies a skip buffer in a dataflow graph, determines limitations associated with the array, and searches for a lowest cost implementation topology and stage depth. At least three topologies are considered, including a cascaded buffer topology, a hybrid buffer topology, and a striped buffer topology. The lowest cost implementation topology and stage depth are based on the size of the buffered data (usually, the size of a tensor), the depth of the skip buffer, and the array's limitations. The hybrid buffer topology includes multiple sections of parallel memory units. The data travels between memory units in one section to adjacent memory units in a next section without intervening reorder buffers.

Skip buffer splitting

A method of processing partitions of a tensor in a target order includes receiving, by a reorder unit and from two or more producer units, a plurality of partitions of a tensor in a first order that is different from the target order, storing the plurality of partitions in the reorder unit, and providing, from the reorder unit, the plurality of partitions in the target order to one or more consumer units. In an example, the one or more consumer units process the plurality of partitions in the target order.

Tensor partitioning and partition access order

The technology disclosed relates to matrix multiplication where the multiplier can be a sparse matrix. In particular, a multiplication device includes first circuitry configured to obtain the multiplicand matrix and an index of columns of the multiplier matrix and to generate an intermediate matrix that has one row per entry in the index copied from a respective row of the multiplicand matrix based on a value of a corresponding entry in the index. The device also includes second circuitry configured to receive the intermediate matrix from the first circuitry, obtain non-zero values of the multiplier matrix and a list of a number of non-zero entries per row of the multiplier matrix, and generate a product matrix as a result of multiplies of the non-zero values of the multiplier matrix and the intermediate matrix.

Sparse matrix multiplier in hardware and a reconfigurable data processor including same

An integrated circuit includes a plurality of configurable units, each configurable unit having two or more corresponding sections. The plurality of configurable units is arranged in a serial arrangement to form a chain of sections of the configurable units. A data bus is connected to the plurality of configurable units which communicates data at a clock rate. The chain of sections is to receive and write a series of tensors at the clock rate at a first end section of the chain of sections, and sequentially propagate the series of tensors through individual sections within the chain of sections at the clock rate. The chain of sections is to output the series of tensors at a second end section of the chain of sections. The chain of sections is to also output the series of tensors at an intermediate section of the chain of sections.

Multi-headed multi-buffer for buffering data for processing

An apparatus includes an execution unit, an instruction queue, and a control circuit. The control circuit may be configured to activate a plurality of processor threads. Each of the plurality of processor threads may include a respective plurality of instructions. The instruction queue may be configured to issue at least one instruction included in the plurality of processor threads to the execution unit at a first rate. The control circuit may also be configured to track, for a particular processor thread, a period of time from activating the particular processor thread. The instruction queue may be further configured to limit issue of a next instruction for at least one other processor thread to a second rate, based on a comparison of the period of time to a threshold amount of time. The second rate may be lower than the first rate.

Hardware mechanism to mitigate stalling of a processor core

This disclosure describes a method for accomplishing sequential logical equivalence verification using a hierarchical piecewise approach. Initially, the method provides a reference semiconductor design and a second semiconductor design with logic edits relative to it. The method submits both to formal verification to check the reference design against the second semiconductor design with all edits disabled. The semiconductor design is partitionedand associated input constraints. The edits are further groupedand ordered. The invention also discovers a set of dependencies of the logic editsand checks that the ordering of groups obeys the dependencies. Each group of edits is further submitted to formal verificationand any input constraints assumed for any partitions are verified in their enclosing partition. Finally, the method reports success if formal verification succeeds on each group of logic edits and on each set of input constraints

Method for piecewise hierarchical sequential verification

A video processing apparatus and methodology use a combination of a processor and a video decoding hardware block to decode video data by using a reference block cache memory to perform motion compensation decode operations in the video decoding hardware block. To improve the cache hit rate, each memory access for required reference block(s) is used to fetch one or more additional reference blocks which can be used to improve the cache hit rate with future motion compensation operations. Speculative fetch control logic selects the additional reference blocks by using a frequency history table to accumulate compared motion vector information for a current motion compensation block with motion vector information from previously processed motion compensation blocks.

Macroblock cache

Growing community of inventors

Austin, TX, United States of America

Nathan Francis Sheeley