The USERMESO package of LAMMPS is a fully GPU-accelerated package for Dissipative Particle Dynamics. Instead of being merely a translation of the conventional molecular dynamics, the package integrates several innovations that specifically targets CUDA devices
- atomics-free warp-synchronous neighbor list construction algorithm;
- 2-level particle reordering which aligns with the cell list lattice boundaries for generating strictly monotonic neighbor list
- locally transposed neighbor list;
- redesigned non-branching transcendental functions (sin, cos, pow, log, exp, etc.);
- overlapped pairwise force evaluation and halo exchange using CUDA streams for hiding communication and kernel launch latency;
- radix sort with GPU stream support;
- pairwise random number generation based on per-timestep binary particle signatures and the Tiny Encryption Algorithm;
It can achieve tens of times speedup on a single CUDA GPU over 8-16 CPU cores. The work is featured by a NVIDIA Parallel Forall blog article Accelerating Dissipative Particle Dynamics Simulation on Tesla GPUs.
The package can be freely used and redistributed under the GPL v3 license. However we would greatly appreciate if you could cite the following paper:
Tang, Yu-Hang, and George Em Karniadakis. "Accelerating dissipative particle dynamics simulations on GPUs: Algorithms, numerics and applications." Computer Physics Communications 185.11 (2014): 2809-2822.
NVCC and a MPI implementation is required for compilation of the code.
cd <working_copy>/src
make yes-molecule
make yes-user-meso
make meso ARCH=[sm_30|sm_35|sm_52|sm_60|...]
To run a simple example of DPD fluid
cd <working_copy>/example/simple
../../src/lmp_meso -in [sp.run|dp.run] -var case [25|48|64]
sp.run: single-precision
dp.run: double-precison
case: simulation box size (number density = 4)
| Amphiphilic polymers in microfluidic channel |
|---|
 |
| Membrane wrapping into a vesicle | Multi-layer vesicle |
|---|---|
 |
 |