Is it possible to run with cuda and ti.f64 precision? I changed all f32 in mpmsolver to f64 and set default_fp to f64. I can still run demo_2d.py sucessfully. However, running demo_3d.py will trigger the following error:
[cuda_driver.h:operator()@80] CUDA Error CUDA_ERROR_INVALID_CONTEXT: invalid device context while calling malloc_managed (cuMemAllocManaged)