OpenCL 2.0 Memory Model
A scoped extension of the C11 model for heterogeneous compute, adding memory scopes (work-item, work-group, device, all-SVM-devices) on top of C11's atomic orderings. At all-devices scope it coincides with C11; narrower scopes permit additional weak behaviours. Formalised and repaired alongside C11 by Batty, Donaldson & Wickerson (POPL 2016).
Ordering relationships
- Strictly weaker than
- C11/C++11 Memory Model — OpenCL 2.0 extends C11 with memory scopes; at all-SVM-devices scope it coincides with C11, but narrower scopes (work-group, device) permit synchronisation-free weak behaviours that flat C11 forbids. See Batty, Donaldson & Wickerson (POPL 2016).
- Compiles correctly to
- NVIDIA PTX Memory Model — OpenCL kernels compile to PTX on NVIDIA GPUs.
- AMD GPU Memory Model — OpenCL kernels compile to AMD's GPU ISA via the HSA-based model.
- Incomparable with
- Vulkan Memory Model — Both are scoped GPU API models, but differ in scope sets and synchronisation primitives — Vulkan adds explicit availability/visibility operations that OpenCL lacks; neither contains the other.
References
- Mark Batty, Alastair F. Donaldson, John Wickerson. Overhauling SC Atomics in C11 and OpenCL. POPL 2016, 2016. doi:10.1145/2837614.2837637
- Jade Alglave, Mark Batty, Alastair F. Donaldson, Ganesh Gopalakrishnan, Jeroen Ketema, Daniel Poetzl, Tyler Sorensen, John Wickerson. GPU Concurrency: Weak Behaviours and Programming Assumptions. ASPLOS 2015, 2015. doi:10.1145/2694344.2694391