37 views (last 30 days)

Hi, I am developing a SPH (smoothed particle hydrodynamic) code of solving fluid equations in Matlab. I have successfully vectorized and implemented the code on GPU. The speed up is astonishing, ~10 times faster than CPU code. In the process I got invaluable suggestions from Matt J and Joss Knight for speeding up the code. Thank them for their wonderful suggestions.

In SPH, one has to find the neighbor particles in a given radius for every particles in the domain. I use the matlab function knnsearch for this purpose. Now the other part of the code (except neighbor search but solving the fluid equations) is so fast that the limiting part now is the knnsearch, which uses kdtree algorithm runing on CPU. It takes 85% percent of the runtime (see the following code profiler results)

The function 'knnCPU_kdtree_func' uses the matlab built-in function knnsearch with kdtree algorithm runing on CPU. The other functions are solving the real fluid equations runing on GPU only consumes 10% of the total time.

I wonder is there any GPU implementation of k-nearest neighbor search that I can free download and using as a function call in my matlab code? Many thanks.

Joss Knight
on 13 Dec 2018

knnsearch is supported on the GPU, so just use it!

Matt J
on 14 Dec 2018

Sign in to comment.

Matt J
on 13 Dec 2018

Edited: Matt J
on 13 Dec 2018

In SPH, one has to find the neighbor particles in a given radius for every particles in the domain. I use the matlab function knnsearch for this purpose.

Perhaps instead, you should use rangesearch with a pre-constructed KDTreeSearcher object. You could then use parfor to process different chunks of query data in parallel.

Sign in to answer this question.

Opportunities for recent engineering grads.

Apply Today
## 0 Comments

Sign in to comment.