Improving performance of interp1 function inside for loop

Hello,

I'm looking for some insight on troubleshooting a matlab fcn bottleneck in a Simulink project. According to the code profiler this Matlab in-line function is (understandably) hogging the runtime:

function [preview,preview_a] = fcn(dp,X_matrix,rho_matrix)

    %preallocate
    rho_matrix_t = rho_matrix'
    preview_a=zeros(1,185);

    for i=1:dp
    preview(i)=X+i;
    preview_a(i)=interp1(X_matrix,rho_matrix_t,preview(i));
    end

I understand the interp1 will inevitably consume some CPU time, but could there be some approach to improve the above code?

In the meantime I have these following in mind:

1. Vectorization: building preview(1:dp) then do interp1 operation in batch (but I suspect interp1 doesn't support this)
2. Using other built-in Simulink block that does equivalent jobs? (I've been looking at interp block, will see if it does the job).

Any suggestions are welcomed. :)

Regards,

Arya

Answer 1

[Walter Roberson]

interp1() is fine with accepting a vector of points to interpolate at. Building preview ahead of time should work well.

[Arya]

Indeed, it turns out interp1() is happy to accept vector points, this is what I did:

function [preview,preview_a] = fcn(dp,X_matrix,rho_matrix)

%preallocate
  rho_matrix_t = rho_matrix'
  preview_a=zeros(1,185);

   i=1:dp 
   preview(i) = X+i
   preview_a(i)=interp1(X_matrix,rho_matrix_t,preview(i));

Much simpler than I thought, and seem to accelerate the execution time a bit.

Cheers!