Experts from the Khronos SYCL Working Group and the SYCL Safety Critical (SYCL SC) Working Group delivered two informative tech talks at the 13th International Workshop on SYCL and OpenCL (IWOCL 2025). They discussed what’s new in the latest update of the SYCL 2020 specification, what value SYCL brings to developers for accelerated parallel computing, how SYCL helps enhance safety-critical systems across various industries and multi-vendor heterogeneous architectures, and more.
This blog will provide highlights from both sessions and, hence, an overview of the SYCL standard and its SYCL SC extension.
SYCL 2020: What’s New?
From [00:09:05] in the ‘SYCL State of the Union’ talk, Tom Deakin (the SYCL Working Group’s chairperson from the University of Bristol) sheds some light on the SYCL 2020 Revision 10 published by the Khronos Group in April ‘25. The latest update to the specification represents a refined and stabilized version of the SYCL 2020 standard, incorporating clarifications, bug fixes, and minor enhancements based on the community’s and implementers’ feedback.
Here are some highlighting features of SYCL 2020 Rev10:
-
Improved Compatibility with Modern C++ - The SYCL 2020 spec continues to align with evolving C++ standards, especially around features like sycl::atomic_ref, which mirrors C++ atomic operations while ensuring compatibility with SYCL's heterogeneous execution model.
-
Clarifications and Fixes – It provides numerous clarifications to existing APIs and behaviors to ensure consistency across implementations. It also fixes ambiguities in areas like memory management, kernel invocation, and device selection.
-
KHR SYCL Extensions - Rev 10 includes references to optional KHR extensions that enhance functionality, such as reduced queue submission overhead and support for new data types. These extensions enhance SYCL's capabilities while maintaining cross-vendor compatibility and are intended to be stable and widely supported. Example KHR extension available in rev10 includes sycl_khr_default_context that reduces overhead when constructing default queues and thus improves the performance for applications that create several queues.
→ More details about KHR SYCL extensions are available in Appendix F of the SYCL 2020 specification.
-
Conformance and Scope: The SYCL 2020 Conformance Test Suite, an open-source testing framework, clearly defines what is considered part of the SYCL 2020 conformance scope. It helps implementers understand what must be supported for a conformant SYCL implementation. It plays a critical role in maintaining cross-platform consistency, reliability, and interoperability for SYCL-based applications.
→ Take a deep dive into the SYCL Ecosystem – get started with SYCL, explore SYCL projects and research materials, get the latest feature updates, check out upcoming SYCL community events, and much more at sycl.tech.
→ Check out the SYCL Reference, a developer-friendly documentation that complements the official SYCL specification by offering a more accessible and practical way to explore SYCL's API.
Real-world Implementations of SYCL
At IWOCL this year, the SYCL developer community and members of the SYCL Working Group celebrated an important milestone – the 10th anniversary of the SYCL ecosystem. Originally conceived as a C++ programming model for OpenCL™, SYCL has evolved into a powerful, vendor-neutral standard that underpins some of the most demanding workloads in AI, HPC, scientific research, and desktop applications. Notably, its integration into tools (highlighted at IWOCL) demonstrates its growing impact across diverse domains.
→ Check out this Khronos blog: A Decade of Heterogeneous C++ Compute Acceleration with SYCL.
Intel® oneAPI DPC++/C++ Compiler is an industry-standard compiler based on LLVM* technology. As the world’s first SYCL 2020 conformant compiler, it enables single-source multi-vendor, cross-architecture programming in modern C++ across CPUs, GPUs, and other accelerators. It provides developers with advanced features such as enabling GPU offload and support for OpenMP* 6.0.
Other compilers implementing SYCL include AdaptiveCpp, triSYCL, SimSYCL, and many more. Refer to the session recording from [00:20:00] where Tom discusses some of the major feature-complete compilers that leverage SYCL.
The UXL Foundation encourages open source, multiarchitecture, accelerated parallel computing free from vendor lock-in. Its oneAPI specification-based libraries and projects are backed by the SYCL framework. Intel’s own implementations of the oneAPI libraries, its oneAPI-powered tools, including the Intel oneAPI DPC++/C++ Compiler and the Intel® DPC++ Compatibility Tool for automated CUDA* to C++ with SYCL code migration, and our AI framework optimizations also support SYCL to enable cross-platform programming.
→ Learn more and get started using our oneAPI developer toolkits.
The oneAPI Construction Kit, originally developed by Codeplay Software and now governed by the UXL Foundation, enables developers to leverage oneAPI and its SYCL-based components on various hardware platforms, including RISC-V, ARM, and other proprietary accelerators.
→ Check out the complete recording and presentation slides for the ‘SYCL State of the Union’ talk – take a deep dive into the SYCL framework!
SYCL Enhancing Safety-Critical Applications
In another session called the ‘SYCL SC State of the Union’ at IWOCL this year, Lukas Sommer from Codeplay Software talked about the Khronos SYCL SC Working Group and its efforts for creating and open source, SYCL-fueled standard for empowering safety-critical C++-based heterogeneous systems. The Khronos SYCL SC™ specification and working group introduced in March 2023 leverages the SYCL 2020 standard. It aims to bridge the gap between the low-level API, such as Khronos’ Vulkan* SC, and the high-level C++ language for streamlining system safety certifications at each level of the software stack. The SYCL SC working group has included members from several key industry players, including AMD, Mercedes-Benz,* and Qualcomm*.
→ Check out the complete recording and presentation slides for the ‘SYCL SC State of the Union’ talk.
→ More information about the SYCL SC Working Group is available here.
Before SYCL SC, the Khronos Group has introduced safety-critical extensions for several of its other open standards over the past 20 years. For example:
-
OpenGL* SC™ extension of the OpenGL ES (OpenGL for Embedded Systems) graphics API
-
Vulkan* SC™ extension of the Vulkan graphics and compute API
-
OpenVX™ (Open Vision Acceleration) specification, including guidance for safety-critical use cases in accelerated machine learning and computer vision
→ Check out the SYCL SC session recording from [00:01:25], in which the presenter explains the evolution of Khronos’ safety-critical standards in detail.
‘SAFETY-CRITICAL’ Systems: An Important Concept
A safety-critical system is one whose failure could lead to severe consequences, such as loss of life, environmental harm, significant property damage, or major financial loss. These systems are integral to industries like aerospace, automotive, healthcare, nuclear energy, defense, and railways. Examples include flight control systems, automotive emergency braking systems, nuclear reactor controls, train signaling systems, secure communication, and radar systems.
To ensure reliability, software in these systems is typically certified against industry-specific safety standards, such as ISO 26262 in the automotive sector. These standards define safety integrity levels (SILs) that reflect the system's criticality and the rigor required for its development.
While absolute security is ideally unattainable, certification ensures that documented, systematic measures have been taken to mitigate risks. Importantly, functional safety must be addressed at the system level, not just in isolated components. Modern safety-critical applications often leverage high-performance hardware, such as the Intel® A760A discrete GPU, to meet demanding computational requirements.
Given the maturity of the C++ ecosystem, including robust development tools, performance analyzers, sanitizers, optimized libraries, and established coding guidelines, C++ and C++-based languages like SYCL are widely adopted in safety-critical domains.
→ Watch the session video from [00:02:55] to learn about safety critical systems in more detail.
The Khronos* SYCL SC™: An Overview
Fig.1: Offerings of the SYCL SC extension
The SYCL SC initiative aims to define a safety-critical extension of the SYCL 2020 specification, focusing on improving compatibility and compliance through targeted API modifications. These changes include alignment with industry standards such as the MISRA (Motor Industry Software Reliability Association)’s coding guidelines.
As a foundational component of safety-critical systems, SYCL SC is guided by the following core principles:
-
Robustness: Ensuring reliable and fault-tolerant behavior.
-
Determinism: Guaranteeing predictable execution.
-
Simplification: Reducing complexity to facilitate safety assurance.
Simplification is particularly critical, as a less complex API significantly eases the process of formal verification and safety certification. To achieve these goals, the SYCL SC working group systematically removes, modifies, and adds features to the SYCL 2020 specification, ensuring all changes are aligned with the group’s safety-oriented design philosophy.
While SYCL SC provides a foundation for developing safety-critical applications, it is important to clarify its scope and limitations:
-
Not a Safety Implementation Guide: SYCL SC does not prescribe how to implement a safe application or runtime. It defines an API that facilitates safety, but the responsibility for safe implementation lies with the developer.
-
No Safety Guarantees: SYCL SC alone cannot guarantee system safety. Functional safety is a system-level property that depends on hardware, software, and process integration. Running SYCL SC on unsafe hardware, for example, cannot yield a safe system.
-
Not a Certification: SYCL SC itself is not certifiable as a safety-critical product. However, implementations of SYCL SC and systems built using it can be certified, and the specification is designed to support such efforts.
-
Not a Process or Standards Manual: SYCL SC does not dictate how to apply industry-specific safety processes or standards. It is up to the implementer to select and apply the appropriate guidelines based on their use case.
-
No Hardware Assurance: SYCL SC does not address or mitigate hardware-level safety issues. It cannot compensate for deficiencies in the underlying platform.
In summary, SYCL SC supports the development of safety-critical systems, but it is not a complete solution. Its effectiveness depends on its use within a broader safety engineering context.
→ Learn more about what SYCL SC IS and what it IS NOT from [00:12:15] in the session recording.
Key Focus Areas of SYCL SC
Here’s what the SYCL SC working group has been up to in the past few years as its efforts for advancing the SYCL specification to meet the demands of safety-critical systems:
-
Deterministic error handling: Replacing C++ exception-based error handling with a deterministic model suitable for safety-critical environments.
-
Object and memory model: Introducing a predictable memory model to eliminate non-determinism caused by dynamic memory allocation, such as container reallocations.
Apart from the above two major features, other aspects of focus include:
-
Standard container adaptation: Modifying C++ standard containers to ensure safe and deterministic behavior.
-
Device selection: Defining mechanisms for deterministic and certifiable device targeting.
-
Host-side thread safety: Clarifying thread safety guarantees and providing mechanisms to query implementation support.
-
Fallback queue behavior: Establishing predictable behavior for fallback execution paths.
-
SYCL alignment: Balancing compatibility with SYCL 2020 while introducing necessary modifications for safety certification.
-
Specification management: Utilizing Git-based workflows for transparent and traceable specification development.
An example of a potential real-world application of SYCL SC is AUTOSAR (AUTomtive Open System ARchitecture), a global partnership of key leaders in the automotive and software industry. It is a global standardization initiative that defines a common software architecture for automotive electronic control units (ECUs). It aims to improve software modularity, scalability, and interoperability across different vehicle platforms and manufacturers. AUTOSAR is divided into two main platforms: the Classic Platform, optimized for real-time and resource-constrained systems, and the Adaptive Platform, designed for high-performance computing environments such as those required for autonomous driving and advanced driver assistance systems (ADAS). By promoting standardized interfaces, hardware abstraction, and compliance with safety and cybersecurity standards, AUTOSAR enables more efficient development, integration, and maintenance of complex automotive software systems. The Khronos Group collaborates with AUTOSAR and contributes to it in domains requiring parallel computing and hardware acceleration (e.g., vision processing, AI inference, sensor fusion).
Accelerate and Safeguard Your Applications with SYCL!
Explore the SYCL standard in greater depth. Get started with SYCL SC and ensure functional security of safety-critical applications.
We encourage you to check out our oneAPI and AI developer tools, which allow you to leverage SYCL for high-performance, cross-vendor, heterogeneous computing. These tools can be included in our developer toolkits or downloaded in stand-alone versions.
Additional Resources