Developers can access a new oneDNN Graph API, in addition to expanded features in oneMKL, oneAPI DPC++ (SYCL) Library, oneVPL, and Level Zero
The oneAPI Community Forum, led by Rod Burns, VP of Ecosystem at Codeplay Software*, announces that it has added new members to the steering committee. The additions include software experts Kevin Harms from Argonne National Labs and Penporn Koanantakool from Google. Kevin is a team lead for Performance Engineering within the Argonne Leadership Computing Facility (ALCF) and has been closely involved in recent collaborations to enable Nvidia and AMD hardware with oneAPI. Kevin received his Master of Science from Illinois Institute of Technology and his Bachelor of Science from Purdue University. Penporn has a Ph.D. from UC Berkeley and is a senior software engineer at Google working on machine learning infrastructure. She has collaborated with various industry partners to optimize TensorFlow on numerous hardware architectures, some of which rely on oneAPI components such as oneDNN. You can also read about further optimizations in this blog. Robert Cohn from Intel has accepted the role as oneAPI Spec Editor. Robert is Senior Principal Engineer at Intel and holds PhD computer science from Carnegie Mellon.
The oneAPI Community Forum groups, structure and governance is held in this open source repository.
The oneAPI Community Forum is also pleased to announce the release of the oneAPI 1.2 Specification. In this release, there are a number of new features as well as a major new release of the oneDNN specification, which includes the brand new “oneDNN Graph” bringing enhanced performance by enabling a larger scope of deep neural network (DNN) compute graph functionality. The release also includes updates and extensions to DPC++ (oneAPI’s open source SYCL implementation) for better management of contexts, queues, and memory management, and enhanced support for images. For oneMKL there are enhancements for the BLAS libraries with new routines. These include support for half/bfloat16, including dense matrix copy and transpose routines as well as updates for BLAS GEMM and GEMM batch. oneVPL added a new API for processing camera RAW data and support for more video color formats. Level Zero added a fabric topology discovery API, support for sRGB, support for image copy with pitch as well as clarifications on existing API.
Please find additional information on the different oneAPI spec elements below.
Get involved and join the oneAPI Community Forum
Learn about the latest oneAPI updates, industry initiatives and news. Check out our videos and podcasts. Visit our GitHub repo to review the spec and give feedback. Or join the conversation happening now on our Discord channel. Then get inspired, network with peers and participate in oneAPI events.
*Codeplay is an Intel company.