Loop on gpu

Hi there, for multiple cores on CPU you can use parfor. But for a loop on gpu how can you speed it up?

Any help?

Answer 1

[Edric Ellis]

The nearest equivalent on the GPU is to use arrayfun. This implicitly loops over each element of your input gpuArray, so there's no need for an explicit parfor loop.

Have a look at this demo which shows using a while loop inside a call to arrayfun on the GPU. In this case, the arrayfun call parallelises operations on each element of a large array, and the function running on the GPU includes a for loop.

[Maniprakash Reddy Gunturu]

Answer is so helpful for me but my problem is quite different from the above problem. can you please help me to solve the below problem.

 function [s,q]=initial(p,cn)

 if cn<5

 s=100./cn;

 q=s+p;

 cn(2,1)=s+2; 

 else

 s=500./cn;

 q=s+p;

 cn(2,1)=s+4;

 end

p=gpuArray([4;5;7;8;7;8;7;2]);

cn=gpuArray([15]);

[s,q]=arrayfun(@initial,p,cn)

my problem is similar to above code. i want to run the above code through "parallel computations using GPU".

my questions are

1. can use "ARRAYFUN" (my inputs (cn,p) have different size) to solve the above problem? i need to get "cn(i+1)" value from s(i).

2. if not how can i solve above problem through "parallel computations using GPU".

[AVINABA BISWAS]

Yes. I have also the similar situation. Means I have created a function and I want to call this function by using inputs which are already in gpuarray.