OIDN wgpu interoperability

A helper library to create shared buffers between OIDN and wgpu.

Getting started

Creating the device

Simply replace the adapter.request_device call with oidn_wgpu_interop::Device::new (using adapter.request_device only if oidn_wgpu_interop::Device::new fails). You are then able to call device.wgpu_device to get the created wgpu device and device.oidn_device to get the OIDN device.

Creating shared buffers

To create a shared buffer call device.allocate_shared_buffers. The shared buffer may be used with usages BufferUsages::COPY_SRC | BufferUsages::COPY_DST. To get the wgpu buffer call buffer.wgpu_buffer and to get the OIDN buffer call buffer.oidn_buffer. It is recommended to minimise the number of shared buffers that exist at a given time due to them each requiring a separate allocation.

Synchronisation

There is no synchronisation between OIDN and wgpu currently possible. This lack of synchronisation means that after using any SharedBuffer in wgpu, but before using it with OIDN, all wgpu command buffers that use the shared buffer must finish. The same must happen in the opposite direction, any OIDN functions that use this buffer must have finished.

Platform Support

Currently, this only supports DirectX12 and Vulkan (on Windows using VK_KHR_external_memory_win32). This code could be expanded to Vulkan (on Linux using VK_KHR_external_memory_fd) and Metal. Due to some devices being unsupported by OIDN it is recommended to support a mode that copies to the cpu and then into an OIDN buffer anyway.