Request for CUDA features (virtual memory management and graph) in SYCL

[mgrabban]

While trying to migrate tiny-cuda-nn (https://github.com/nvlabs/tiny-cuda-nn) into SYCL, I realized that it uses -

• CUDA driver API for virtual memory management (https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__VA.html#group__CUDA__VA). My understanding is this helps with dynamic memory management. There is an NVIDIA blog (https://developer.nvidia.com/blog/introducing-low-level-gpu-virtual-memory-management/) on this. There is also cuda-sample on this (https://github.com/NVIDIA-developer-blog/code-samples/tree/master/posts/cuda-vmm).
• CUDA Graph API (https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#cuda-graphs). A blog on CUDA graph can be found here: https://developer.nvidia.com/blog/cuda-graphs/.

Per dpct (or SYCLomatic), these are not available in SYCL, so is there any plan to support these in near future? Or, maybe these are already supported in some way in SYCL, and in that case, I would appreciate if someone could point me to the right docs.

[mabraham]

There's a proposed extension to SYCL standard at #5626 and draft implementation at #7627

[reble]

This is for the graph part. A draft for SYCL virtual memory extension can be found here:
#8954

[npmiller]

Closing this, as we now have both virtual memory and SYCL graph extensions, merged and working.

Feel free to re-open or submit a new ticket if there are any issues remaining.