Gray Scott in 3D using sparse grids on GPU (Optimized)

Table of Contents

• Optimizations

Solving a gray scott-system in 3D using Sparse grids on gpu

This example show how to solve a Gray-Scott system in 3D using sparse grids on gpu

In figure is the final solution of the problem

More or less this example is the adaptation of the dense example in 3D

See also

Gray Scott in 3D

Optimizations

Instead of using the default decomposition algorithm based on parmetis we use BoxDistribution. This decomposition divide the space equally across processors. The way to use a different algorithm for decomposing the sparse grid is given by changing the type of the Sparse grid

 
typedef CartDecomposition<3,float, CudaMemory, memory_traits_inte, BoxDistribution<3,float> > Dec;
 
typedef sgrid_dist_id_gpu<3,float,aggregate<float,float,float,float>,CudaMemory, Dec> SparseGridType;
 

Because the geometry is fixed we are also using the option SKIP_LABELLING. With this option active after a normal ghost_get we are able to activate certain optimization patterns in constructions of the sending buffers and merging data.