[drjit] Automatic backend selection

[maxfrei750]

Is there a way to automatically adapt the backend that is used for drjit functions, e.g. dr.zeros?

Motivation: When writing custom functions, I'd like to use drjit data types that match the mitsuba variant that the user selected. When creating variables, I can resort to things like mi.Float, to automatically get the right data type. Is there something similar for functions like dr.zeros?

Or should I create the array of zeros using numpy and convert to a vector/tensor using mitsuba?

[maxfrei750]

The solution was rather simple. Functions like dr.zeros expect a data type as first argument. So one can simply pass a dynamic data type like mi.Float to them, e.g.:

dr.zeros(mi.Float, 3)

[njroussel]

Hi @maxfrei750

Indeed, most mitsuba types are just aliases for drjit types. Passing directly the mitsuba type to the few Dr.Jit operations that require it should almost always work 😄