[an error occurred while processing the directive]
RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2
2020/05/27 16:11:15

Standard of parallel programming of OpenCL 3.0: new features and the supported devices

On April 27, 2020 on an action of International Workshop on OpenCL the consortium Khoronos Grop submitted preliminary specifications of OpenCL™ 3.0 standard.

OpenCL (Open Computing Language) is the standard for cross-platform parallel programming of different accelerators opened, not requiring runtime royalty fees which can be found in supercomputers, cloud servers, personal computers, mobile devices and the built-in platforms. The previous version of the standard - OpenCL 2.1 - was submitted in March, 2015.

Due to use of parallel computings it is possible to increase significantly the speed of work and a response of a broad spectrum of applications in many segments of the market, including professional tools for creativity, the scientific and medical software, processing of machine vision and also training and work of neuronets. Also practical examples of problems which can be solved using technologies of parallel computings are: computing chemistry, Data Science, bioinformation science, modeling of flows of liquids and gases, weather forecasting.

Стандарт OpenCL широко используется компаниями во многих сегментах рынка: создание content, machine learning, in industrial calculations

Khronos Group - industrial consortium which purpose is development of open standards of programming interfaces (API) in the field of creation and reproduction of dynamic graphics and a sound on a broad spectrum of platforms and devices with support of hardware acceleration is engaged in development of the standard. It includes more than 100 companies among which AMD, Apple, ARM, Google, Intel, Nvidia and others.

OpenCL 3.0 was created so that zheleza producers could provide broad functionality, flexibility of deployment for developers, giving to OpenCL-compatible solutions an opportunity to be focused on the functionality corresponding to their target markets.

File:Aquote1.png
OpenCL 3.0 is an important step forward in the aspiration to provide higher performance and innovations in the extending range of computing platforms and applications.
writes in Twitter Balaji Calidas, the director of developments in Qualcomm.
File:Aquote2.png

Changes which are included in the new version of the standard mainly serve that API could be used for work with new the iron type, such as FPGA, specialized processors for the built-in systems, processors of mobile devices, the system of boundary calculations (when data processing happens directly on IoT devices) that can be useful, for example in robotics, the automated production lines, at the massive distributed analytics. Earlier versions of the standard generally were guided by central processors and discrete video cards.

Вычисления OpenCL могут выполняться на CPU, графических processors, big range of dedicated chips

The new solutions included in OpenCL 3.0 became support of the asynchronous DMA expansions (Direct Memory Access), for transfer of complex structures of memory, for example images and 2D/3D structures. Support of the SPIR-V 1.3 language (the intermediate language for parallel computings and diagrams developed by Khronos Group) is in addition entered that will allow to increase interactions between OpenCL and API Vulkan (the application programming interface for 2D, 3D of graphics from Khronos Group).

OpenCL 3.0 is developed so that completely compatible only to the OpenCL 1.2 specifications, all other functions become optional. All OpenCL 2.X functions are in coordination defined in the new unified specification, and the current implementations of OpenCL 2.X which are updated to OpenCL 3.0 can continue to deliver the existing functionality with full backward compatibility.

File:Aquote1.png
Our many clients want to use the GPU programming language which works on all devices, and with growth of deployment in the systems of boundary calculations and mobile devices this requirement increases. OpenCL is the only solution for access to various silicon acceleration, and many key program stacks use OpenCL/SPIR-V as a backend. We are very glad that OpenCL 3.0 will promote even more widespread introduction in the industry as it convinces our clients that their last and future investments into OpenCL are justified.
told Vincent Hindriksen, the founder and the CEO of Stream HPC.
File:Aquote2.png

The new generation of the standard has compatibility both with the previous generations, and with other standards and languages
File:Aquote1.png
OpenCL is the most widespread open standard for heterogeneous parallel programming of the low level which is widely used in applications, libraries, mechanisms and compilers which need to cover the most different processors. OpenCL 2.X provides considerable functionality, but OpenCL 1.2 proved as the basic level necessary for all suppliers and the markets. OpenCL 3.0 integrates strictly organized optionality into monolithic specification 2.2, increasing flexibility of deployment that will allow OpenCL to raise a bar on everywhere available functionality in future specifications of a core.
told Neil Trevett, the vice president of NVIDIA, the president of Khronos Group and the chairman of the working group OpenCL in the press release.
File:Aquote2.png

OpenCL can be applied together with high-level API that can simplify development

After release of the OpenCL 3.0 specifications in Khronos Group within several months will collect proposals of market participants, on changes which can be added to this development environment, final specifications will be issued after that. The consortium hopes for support of the new standard by players of the market.

File:Aquote1.png
Intel resolutely supports the cross-architectural standards existing in a computing ecosystem such as OpenCL 3.0 and SYCL. The unified programming models based on standards will allow our developers to increase efficiency and to show creativity in the forthcoming release of our new architecture of Xe GPU.
told Jeff McVeigh, the vice president of Intel for architecture, graphics and the software.
File:Aquote2.png