公开(公告)号：US20220207643A1

公开(公告)日：2022-06-30

申请号：US17134904

申请日：2020-12-28

Applicant: ADVANCED MICRO DEVICES, INC.

Inventor： SOORAJ PUTHOOR ,  BRADFORD BECKMANN ,  NUWAN JAYASENA ,  ANTHONY GUTIERREZ

IPC: G06T1/20 ,  G06F9/30 ,  G06F9/54 ,  G06F9/38

Abstract: Implementing heterogenous wavefronts on a graphics processing unit (GPU) is disclosed. A schedule assigns heterogeneous wavefronts for execution on a compute unit of a processing device. The heterogeneous wavefronts include different types of wavefronts such as vector compute wavefronts service-level wavefronts that vary in resource requirements and instruction sets. As one example, heterogenous wavefronts may include scalar wavefronts and vector compute wavefronts that execute on scalar units and vector units, respectively. Distinct sets of instructions are executed for the heterogenous wavefronts on the compute unit. Heterogenous wavefronts are processed in the same pipeline of the processing device.

公开(公告)号：US20210157756A1

公开(公告)日：2021-05-27

申请号：US16693638

申请日：2019-11-25

Applicant: ADVANCED MICRO DEVICES, INC.

Inventor： ANTHONY GUTIERREZ

IPC: G06F13/362 ,  G06F11/30 ,  G06F11/34

Abstract: Data transfer acceleration includes receiving, by a data transfer accelerator in a first node of a plurality of nodes, from a second node of the plurality of nodes, a request for data in a second state, wherein the second node stores an instance of the data in a first state; generating a message including one or more operations to transform the data from the first state to the second state; and sending the message to the second node in response to the request.

公开(公告)号：US20240169019A1

公开(公告)日：2024-05-23

申请号：US17991493

申请日：2022-11-21

Applicant: Advanced Micro Devices, Inc.

Inventor： ANTHONY GUTIERREZ ,  ALI ARDA EKER

IPC: G06F17/16

CPC classification number: G06F17/16

Abstract: A technical solution to the technical problem of how to improve performance when performing SpMV multiplication uses sparse matrix row similarity to schedule SpMV multiplication operations. CSR representation metadata is generated for a CSR representation and indicates the locations of non-zero values in the rows of the corresponding sparse matrix or the cache locations of column data needed for SpMV multiplication operations. The CSR representation metadata is used to determine the similarity of rows in the sparse matrix based upon Cosine similarity, Jaccard similarity, Locality Sensitive Hashing (LSH) that approximates Jaccard similarity, or other measures of similarity. The row similarity is used to schedule SpMV multiplication operations to increase data locality, reduce cache misses, reduce time stalling on memory accesses, and reduce bandwidth consumption. Implementations include the use of similarity thresholds to schedule SpMV multiplication operations on particular threads and processing elements and load balancing to further improve performance.