Just thinking a-loud: it really seems we should add tests for all this selection logic. It shouldn't be too hard to implement some mock environment to verify automatic configuration is happening…
Hi Alaska, good catch, thanks and approved!
Excellent and thanks for the adoption of this version!
Why do we need this is_cpu_denoiser_device ? single_denoiser_device : cpu_fallback_device? Can as well use the cpu_fallback_device for simplicity.
Error reporting reasons - I suspect that the…
How OptiX denoiser is supposed to be working with CPU rendering? This funcitonality seems to be lost with the #118841 ?
In order to use OptiX denoiser (or OIDN denoiser) on GPU with CPU…
The logic about how the device is gathered and used to pass to a specific denoiser implementation is already quite fragile, and hard to follow.
If you have some particular examples, I could…
Hi @Sergey, thanks for your PR and I think it could be simplified a bit in order to keep it more easy to read.
I think approach of checking single_denoiser_device on each occasion when it is used (here and below, in return statement) just make it unnecessary compilation. @Sergey, what about just modifying find_best_device to ensure, that it wouldn't return a nullptr?
Ive run the tests locally on the similar configuration that was asserting before. it all works fine. Thanks!
The PR description needs to be tweaked before landing to reflect the updated…