RudolfWeeber

WIP: async GPU force/torque back-transfer, launch kernels earlier

20

This uses the thrust pinned memory allocator for the vectors receiving gpu forces/torques on the host. Then, the back-transfer can be asynchroneous. The host force/torque vector have to remain globals,...

Replace Stokes sphere test

5

The force on the sphere should be determined by a surface integral involving the surface normal and the stress around the sphere. The consistency of that would be sufficient to...

Walberla provides only one stress getter per node. It is not entirely clear, what that returns. Likely the hydrostatic part is not included, but the diagonals are not zero for...

Requirements for reworked collision detection

8

This is my understanding of a dicsussion during the Espresso offline meeting The feature creates bonds between colliding particles. It can additionally creates "non-sliding contacts" by creating virtual sites related...

Pull out integration kenerls for rotational dof

1

and place together with positional ones. This is already the case for steepest descent, but not for the rest

Non-blocking particle data collection for gpu methods

1

Gpu methods like LB_GPU or GPU-based electro/magnetostatics collect particle data on the head node at the beginning of the force calculation. This is blocking, which potentially lets nodes idle. Assuming...