How to check if OptiX is available

[maxfrei750]

I ran into a similar error as mentioned in #167, but for libnvoptix.so.1.

When loading Mitsuba, I receive:

jit_optix_init(): libnvoptix.so.1 could not be loaded -- disabling OptiX backend! Set the DRJIT_LIBOPTIX_PATH environment variable to specify its path.

However, if I insist on running with a CUDA variant, I receive RuntimeError('Could not initialize OptiX!') (rightly so). Since there is no solution yet (waiting for NVIDIA, see here), I'd like to catch the error by checking if OptiX is available and setting the Mitsuba variant to LLVM, if it is not.

Is there a way to check if OptiX is available, using drjit or mitsuba from Python?

Thanks!

EDIT: I think I found the relevant part of the code (jitc_optix_check) in https://github.com/mitsuba-renderer/drjit-core/blob/044f04b54fce4af81fd841352fed594889e6b608/src/optix_api.cpp, but is there any way to call it from Python?

[njroussel]

The mechanism to dynamically load shared libraries is mostly handled by jitc_find_library in drjit-core (code). Unfortunately, this function is not directly bound to Python, and I can't think of a clever way to manipulate any other existing code to re-use this.

You would need to mimic it's behaviour, which is pretty straightforward. All we're doing is dlopen("libnvoptix.so.1"), which you should be able to reproduce with ctypes. In Dr.Jit a failed load of this library deactivates the OptiX backend.

[maxfrei750]

Thanks for the quick and extremely helpful reply. I created an answer with an example code.

[maxfrei750]

Thanks to the input of @njroussel, I came up with the following function that checks if OptiX is available:

import ctypes

def is_optix_available() -> bool:
    """Check, if OptiX is available."""
    try:
        ctypes.cdll.LoadLibrary("libnvoptix.so.1")
    except OSError:
        return False

    return True