This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| OpenCL 2.0 | |
|---|---|
| Name | OpenCL 2.0 |
| Developer | Khronos Group |
| Released | 2013 |
| Latest release | 2013 |
| Operating system | Microsoft Windows, Linux, macOS, Android |
| Programming language | C (programming language), C++ |
| License | proprietary standard with royalty-free conformance |
OpenCL 2.0 is a 2013 revision of the heterogeneous compute standard maintained by the Khronos Group that extends task-parallel and data-parallel programming across devices such as central processing unit, graphics processing unit, and accelerators. It builds on earlier work by the Khronos Group and adopters such as Intel Corporation, AMD, and NVIDIA Corporation to provide new memory, concurrency, and device compilation capabilities intended to simplify heterogeneous application design. The specification influenced implementations in products from ARM Holdings, Qualcomm, and independent contributors in the open source community including projects from Linaro and Google.
OpenCL 2.0 formalizes a set of capabilities for offloading compute kernels to heterogeneous devices, integrating concepts from parallel frameworks used by vendors like Intel Corporation and standards bodies such as IEEE. The release emphasizes improved programmability by adding features that reduce host-device synchronization, echoing trends seen in proposals from Microsoft and research from institutions like Massachusetts Institute of Technology and Stanford University. It aims for wider hardware portability across platforms supported by vendors such as AMD, NVIDIA Corporation, ARM Holdings, and embedded suppliers like Imagination Technologies.
Major additions in OpenCL 2.0 include Shared Virtual Memory, generic address space handling, and device-side enqueue, among others. Shared Virtual Memory (SVM) provides a model comparable to features promoted by Intel Corporation and research groups at University of California, Berkeley, enabling pointer sharing between host and device. Device-side enqueue allows kernels to launch work without host intervention, a capability influenced by task-graph approaches used at Google and in academic proposals from ETH Zurich. Other features such as enhanced atomics and pipe objects reflect design directions similar to concurrency work at Microsoft Research and IBM Research.
The architecture continues the model of platforms composed of hosts, compute devices, compute units, and processing elements, terms familiar to implementers at AMD and NVIDIA Corporation. OpenCL 2.0 introduces hierarchical memory interactions that align with hardware mechanisms found in architectures from ARM Holdings and Intel Corporation. The programming model supports heterogeneous kernels written in variants of C (programming language), with offline or runtime compilation strategies used by projects like LLVM and vendor toolchains from NVIDIA Corporation (CUDA toolchain parallels) and Intel Corporation (compiler frameworks). Work-items, work-groups, and command queues remain central, while device-side enqueue and SVM alter the synchronization and task decomposition patterns used in applications from companies such as Autodesk and Adobe Systems.
Language changes include support for nested parallelism, generic address space qualifiers, and revised built-in functions, echoing feature sets in contemporary language efforts at ISO/IEC and compiler innovations from GCC and Clang. The API gained new calls for SVM allocation and coordination, command queue properties, and pipe management, resembling extensions found in vendor SDKs from Intel Corporation and ARM Holdings. The specification's updated memory model was informed by formal memory models discussed at conferences hosted by ACM and IEEE Computer Society, aligning API semantics with expectations set by standards such as those developed by ISO committees.
Implementations arrived in both proprietary and open projects: vendors including Intel Corporation delivered drivers supporting SVM and coarse-grained SVM, while ARM Holdings and embedded vendors implemented subsets appropriate to their MMU capabilities. Distributions and platforms like Ubuntu, Fedora Linux, and Android incorporated SDKs and drivers from companies such as Qualcomm and Imagination Technologies. Academic and open source efforts leveraged LLVM and projects under the stewardship of Khronos Group to produce conformance tests and reference implementations used in labs at Massachusetts Institute of Technology and University of Cambridge.
Performance characteristics vary across hardware: implementations on devices from NVIDIA Corporation often favored driver-level optimizations while integrated GPUs from Intel Corporation showed gains through SVM-aware optimizations. Backward compatibility with prior versions encouraged by Khronos Group enabled many existing kernels to run unchanged, though performance portability required vendor tuning similar to the cross-platform challenges faced by Mozilla's Web standards and Apache Software Foundation projects. Vendors like AMD provided profiling tools and extensions to expose hardware counters familiar to developers from Intel Corporation and NVIDIA Corporation ecosystems.
The memory model introduced by OpenCL 2.0 tightened semantics for atomics, memory consistency, and SVM interactions, guided by formal work from groups associated with ACM SIGPLAN and IEEE. Implementers had to reconcile device memory protection schemes from processor architectures by ARM Holdings, Intel Corporation, and embedded suppliers such as Qualcomm to prevent pointer abuse and ensure isolation analogous to protections in Linux and Android. Conformance tests and security considerations were coordinated via the Khronos Group ecosystem and collaborators in industry labs at IBM Research and Microsoft Research.
Category:Application programming interfaces