Vulkan/OpenCL Interop Extension In Development, OpenCL-Next Continues
Neil Trevett posted on GitHub today some brief comments following a user inquiring about the state of Vulkan OpenCL interoperability. Most of what he said isn't particularly new news and was in fact covered last summer when it comes to OpenCL-Next, Vulkan and OpenCL continuing to co-exist, etc, but here is the summary of his post today:
- The Khronos OpenCL Working Group will soon be starting on a Vulkan/OpenCL interoperability extension. NVIDIA already has been working on such but now it's up to the working group to come to a multi-vendor approved extension. This is akin to OpenGL and Vulkan interoperability.
- The OpenCL road-map will continue to be "very actively" developed. Contrary to some still saying OpenCL is being canned or merged with Vulkan, that is not the case, as we've been reporting. Both Vulkan and OpenCL will continue to co-exist though some overlap in compute functionality. Vulkan will continue seeing GPU compute capabilities but principally is a graphics API focused on GPUs while OpenCL also has to worry about FPGAs and other types of accelerators really outside of the cope of Vulkan.
- Neil reaffirmed that OpenCL Next will have "significantly more deployment flexibility" so OpenCL platforms can offer more tailored functionality. (As reported previously, for example, SVM is expected to be made "optional" and this greater flexibility is what should allow the NVIDIA proprietary driver to move on to supporting this yet-to-be-released OpenCL spec compared to currently being at v1.2 without full v2.0+ support.)
His comments in full can be read on GitHub.
5 Comments
