Is the MPM solver differentiable?

[kavenio]

I found some related discussions on differentiability in the MuJoCo-backend/MJWarp solver (#1483, google-deepmind/mujoco_warp#1226) and Newton's collision pipeline (#2164), but those seem focused on the rigid-body/contact solver rather than material point method (MPM).

Is the MPM solver in Newton currently differentiable (e.g. via Warp's autodiff/Tape), or is that not yet supported? I'm interested in gradient-based optimisation of material parameters or initial conditions through an MPM simulation.

Any info on the current status or known limitations would be great. Thank you!