Advanced Micro Devices Corporation

Ati Technologies Ulc

Khaled Hamidouche

Michael Wayne LeBeane

Brandon K Potter

Walter B Benton

Hari S Thangirala

Bradford Michael Beckmann

Sergey Blagodurov

Shaizeen Dilawarhusen Aga

Joseph Lee Greathouse

Johnathan R Alsop

Nicholas P Malaya

Michael L Chu

Kishore Punniyamurthy

Furkan Eris

Richard David Sodke

Non-blocking processing system are described. In accordance with the described techniques, a pending range store receives, at a start of a bulk memory operation, a pending memory range of the bulk memory operation. A logic unit includes at least one of check conflict logic or check address logic. The logic unit detects a conflicting memory access based on a target address of the pending memory range conflicting with a memory access request separate from the bulk memory operation. The logic unit performs at least a portion of the bulk memory operation associated with the target address before the memory access request is allowed to proceed.

Non-blocking parallel bulk memory operations

A device includes a plurality of processing elements (PEs). A symmetric memory is allocated in each of the plurality of PEs. The device includes a switch connected to the plurality of PEs. The switch is to: receive, from a first processing element (PE) of the plurality of PEs, a message that includes a buffer offset, compute, based on the buffer offset, a first memory address of a first buffer in a first symmetric memory of the first PE and a second memory address of a second buffer in a second symmetric memory of a second PE of the plurality of PEs, and initiate, based on the first memory address and the second memory address, a direct memory access operation to access the first buffer and the second buffer.

Low latency offloading of collectives over a switch

A framework disclosed herein extends a relaxed, scoped memory model to a system that includes nodes across a commodity network and maintains coherency across the system. A new scope, cluster scope, is defined, that allows for memory accesses at scopes less than cluster scope to operate on locally cached versions of remote data from across the commodity network without having to issue expensive network operations. Cluster scope operations generate network commands that are used to synchronize memory across the commodity network.

Efficient memory-semantic networking using scoped memory models

A processing system includes a first processor couplable to a first memory and a second memory. In response to a page migration trigger for a page in the first memory, the first processor is configured to, responsive to the page being a read-only page storing code for execution, initiate migration of the page to a code cache portion of a second memory associated with a second processor and shared by multiple processes executing at the second processor, and to configure each process of a set of processes executing at the second processor to access and execute the code from the code cache portion.

Systems and methods for reducing instruction code memory footprint for multiple processes executed at a coprocessor

An approach is provided for coalescing network commands in a GPU that implements a SIMT architecture. Compatible next network operations from different threads are coalesced into a single network command packet. This reduces the number of network command packets generated and issued by threads, thereby increasing efficiency, and improving throughput. The approach is applicable to any number of threads and any thread organization methodology, such as wavefronts, warps, etc.

Network command coalescing on GPUs

A method of training a neural network includes, at a local computing node, receiving remote parameters from a set of one or more remote computing nodes, initiating execution of a forward pass in a local neural network in the local computing node to determine a final output based on the remote parameters, initiating execution of a backward pass in the local neural network to determine updated parameters for the local neural network, and prior to completion of the backward pass, transmitting a subset of the updated parameters to the set of remote computing nodes.

Optimized asynchronous training of neural networks using a distributed parameter server with eager updates

Systems, apparatuses, and methods for generating network messages on a parallel processor are disclosed. A system includes at least a parallel processor, a general purpose processor, and a network interface unit. The parallel processor includes at least a plurality of compute units, a command processor, and a cache. A thread within a kernel executing on a compute unit of the parallel processor generates a network message and stores the network message and a corresponding indication in the cache. In response to detecting the indication of the network message in the cache, the command processor processes and conveys the network message to the network interface unit without involving the general purpose processor.

GPU networking using an integrated command processor

A method includes storing a first portion of a sparse triangular matrix in a local memory and launching a kernel for executing a set of workgroups. The first portion includes a plurality of row blocks, and each workgroup in the set of workgroups is associated with one of the plurality of row blocks. The method also includes, for each workgroup in the set of workgroups, solving the row block. The row block is solved by, for each row segment of a first subset of row segments in the row block, calculating a partial sum for the row segment based on one or more matrix elements in the row segment, and writing the partial sum to a remote memory of a first remote processing unit prior to terminating the kernel.

Optimized and scalable sparse triangular linear systems on networks of accelerators

Systems, apparatuses, and methods for performing network packet templating for graphics processing unit (GPU)-initiated communication are disclosed. A central processing unit (CPU) creates a network packet according to a template and populates a first subset of fields of the network packet with static data. Next, the CPU stores the network packet in a memory. A GPU initiates execution of a kernel and detects a network communication request within the kernel and prior to the kernel completing execution. Responsive to this determination, the GPU populates a second subset of fields of the network packet with runtime data. Then, the GPU generates a notification that the network packet is ready to be processed. A network interface controller (NIC) processes the network packet using data retrieved from the first subset of fields and from the second subset of fields responsive to detecting the notification.

Growing community of inventors

Austin, TX, United States of America

Khaled Hamidouche