Speeding up stream2 with parallel computing?

I've got a set of data that I'm looping through almost 4 million times, and it's iterative, so I can't just parfor it. The whole script takes a couple of days to run. I ran it through the built in profiler and 83-87% of the computation time is spent in stream2. I'm leaning towards GPU, since the lab I'm working for has a computer with a Nvidia Tesla K40, but my parallel computing experience is nonexistent.

The inputs to stream2 are all arrays except of the step size and max vertices, which I need to be able to control. I've tried putting it in to arrayfun and putting all the inputs in to gpuArrays, but stream2 throws an error:

   Error using stream2>parseargs (line 100)
   Wrong number of input arguments.

   Error in stream2 (line 29)
   [x, y, u, v, sx, sy, step, maxvert] = parseargs(nargin,varargin);

Here's the code I'm working with:

    OPTIONS = [0.05 , 2];

    EP = arrayfun(@stream2,C.x/10,C.y/10,C.u/10,C.v/10,Xin,Yin);

Where C is a structure array. As I loop through, the output of stream2 from the n-1 loop will be the Xin and Yin for the nth loop.

Any thoughts on speeding this up?

Edit: Added the @ in front of my function, turns out I had forgotten my inputs to stream2 weren't the same size and shape, so arrayfun probably isn't the solution.

Answer 1

[Matt J]

Not

EP = arrayfun(stream2,....)

You need the @,

EP = arrayfun(@stream2,...)

[Stuart]

Thanks. Looks like I'd forgotten, not all my inputs are the same size and shape, so arrayfun won't take them.