The OpenCL Working Group has published a draft extension (cl_khr_cooperative_matrix) that brings cooperative matrix operations—a key technology for accelerating ML inference—to OpenCL, developed in collaboration with Arm, Intel, and Qualcomm. A companion extension to expose these capabilities directly in the OpenCL C language is also in progress, and the community is invited to review both drafts and provide feedback before they are finalized

The Khronos Group is a consortium of technology companies and industry experts that develops open standards for graphics, compute, and vision processing. Through initiatives like Vulkan, OpenGL, and WebGL, Khronos Group defines cross-platform APIs for accelerating graphics rendering and computational tasks across diverse hardware platforms. Developers can learn about Khronos Group's standards, participate in working groups, and leverage open APIs for building high-performance graphics and compute applications.

Khronos

The OpenCL Working Group has published a draft extension (cl_khr_cooperative_matrix) that brings cooperative matrix operations to OpenCL, developed with Arm, Intel, and Qualcomm. Cooperative matrix ops allow sub-groups of work-items to jointly load, multiply, and accumulate matrix tiles, mapping directly onto hardware matrix acceleration units for ML inference. The extension enables OpenCL implementations to accept SPIR-V modules using SPV_KHR_cooperative_matrix, with a new device query API for supported matrix sizes and types. A companion extension to expose these capabilities in OpenCL C (without hand-authored SPIR-V) is also in progress, with a clang frontend RFC published on LLVM Discourse. Both drafts are open for community feedback before finalization.

OpenCL Cooperative Matrix Extensions Are Here