Exploring SYCL SC

Khronos already has two safety-critical variants of APIs: OpenGL SC for graphics and Vulkan SC for graphics and low-level compute. In addition, OpenVX 1.3 has defined a safety-critical feature set for AI/Vision applications. However, in the safety space there is currently no high-level compute API to develop the complex algorithms of tomorrow. By implementing SYCL SC on top of Vulkan SC and building on safety certified drivers and hardware, the whole stack, all the way from hardware to application, can be safety certified. SYCL SC will also unlock the ecosystem of libraries and tools that already exist for SYCL for the safety-critical domain. Frameworks such as AI/ML frameworks can be built on top of SYCL SC in the future, to provide even higher levels of abstraction. This presentation will talk about the aims of the new standard, which are aligned with Vulkan SC. It will also touch on some initial design ideas, with a focus on deterministic rather than dynamic behavior. We suggest the removal of some SYCL features that are unlikely to be used in a safety context, to facilitate the safety certification of the runtime itself and the addition of extensions that provide functionality that is useful in a context where safety is critical, such as related to the timing of kernels. We will discuss the importance of Misra's C++ guidelines, particularly the upcoming Misra C++ 202x standard, for applications and hence APIs, whilst acknowledging the need to remain compatible with standard SYCL as much as possible. We set up the Exploratory Forum to collect feedback from potential runtime implementers, library developers, researchers and users on what their requirements for a high-level compute API in a safety-critical context are. The Exploratory Forum is open to non-Khronos-members under Khronos NDA and we actively encourage non-members to participate. Once a wide range of requirements has been collected the next step is the creation of a new Khronos group, which would work towards a specification. This presentation will describe the issues that organizations are facing that can be solved through the new standard and provoke discussion on how to develop an API that will meet the needs of the safety-critical community. After the presentation we invite the audience to join the Exploratory Forum to talk about their own requirements and experiences as well as collaborate to develop a framework for the new standard to be defined.