Specification for OpenCL 1.2 released
by admin ·
The Khronos Group has published the OpenCL open standard in version 1.2. The downward-compatible new specification brings more speed and some new features.
The Khronos Group is an industry consortium of vendors and users of graphics hardware and software. In December 2008, it turned OpenCL 1.0 before, “the first open and platform-independent standard for” heterogeneous parallel programming.
This version was OpenCL 1.1 in June 2010. OpenCL specified interfaces that allow time-intensive computations can be distributed across multiple CPUs and GPUs. Many drivers and libraries now implement the specification.
OpenCL is an open standard that can be used by anyone free of charge. The new version 1.2 includes enhanced features compared to version 1.1 and will allow more flexibility and higher speed. About thirty companies were involved in the new version.
The specification now provides that instruments can be divided into sub-units in order to directly influence the allocation of tasks to specific execution units. The mechanism also makes it possible to reserve a portion of the execution units for high-priority tasks, or use shared resources such as caches better.
OpenCL programs can now be compiled separately and linked, similar to conventional compilers. Thus the creation of libraries is made possible, which can be used by other developers. Support for pictures has been extended by one-dimensional images as well as arrays of one-and two-dimensional images.
Such images can now be generated using an OpenGL extension also from OpenGL textures. Built-in kernel, representing the skills of specialized hardware allow now the use of these devices with OpenCL.
These include certain video encoders, and digital signal processors. Surfaces can now also be shared with DirectX 9 or 11.
The conformity tests for OpenCL has been greatly expanded in version 1.2. The specifications, on-line reference manuals and reference cards are available on the website OpenCL available.