Optimize run time of octave/matlab function - multithreading

Question

I am using the Google Earth Toolbox, for Octave, which is extremely similar to the same library made for MATLAB. See this link, for the MATLAB doc for the tool.
Unfortunately I am having a bit of trouble with a certain function: ge_quiver. This function creates a quiver plot and translates the result into KML code. My problem with it
is that it is a bit too slow for my liking.
For a data variable of 30x30 size, the function takes around 10 seconds to complete. This is fine, except for the fact that I am trying to make an animation quiver plot, with 23 other same sized data variables (24 total). Since Octave (and MATLAB for that matter) normally only runs on one thread, the whole thing is run serially, and takes 10*24 = 240 s.
I have tried parfor loops, but they have only removed 10 seconds from the total run time. I have a computer with 16 cores. Considering, this is embarrassingly parallel (no dependence between variables), this should ideally take:
10 seconds for the first 16 variables
+ 10 seconds for the 8 remaining variables
------------------------------------------------------
= 20 seconds total.
I haven't even tested arrayfun, because I don't know how to adapt my function (p_ge_quiver). Nonetheless, many have said that arrayfun will not make it faster.
The following code is approximately what I am trying to do. Remember that the variable
data_u is of size 30x30x24. Same thing for data_v, lonand lat.
...
...
[YYYY,MM,DD,HH,mm,ss] = ncdate(NCFILE);
date_s.year = YYYY;
date_s.month = MM;
date_s.day = DD;
date_s.hour = HH;
date_s.minute = mm;
date_s.second = ss;
TIMESTEP = 60;
parfor i = 1:size(data_u,3)
dt = TIMESTEP*i;
kml(i) = p_ge_quiver(lon,lat,data_u,data_v,dt,TIMESTEP,data_s);
endparfor
...
...
function kml = p_ge_quiver(lon,lat,u,v,data_time,step,date_s)
% V_GE_QUIVER - Writes the quiver plot into KML.
% Get date variables.
YYYY = date_s.year;
MM = date_s.month;
DD = date_s.day;
HH = date_s.hour;
mm = date_s.minute;
ss = date_s.second;
% Date format. Use Google's.
F = 'yyyy-mm-ddTHH:MM:SSZ';
% Start and end dates of data.
tStart = datestr(datenum(YYYY,MM,DD,HH,mm+data_time,ss),F);
tEnd = datestr(datenum(YYYY,MM,DD,HH,mm+data_time+step,ss),F);
% Quiver plot.
kml = ge_quiver(lon,lat,u,v,'timeSpanStart', tStart, ...
'timeSpanStop' , tEnd , ...
'visibility' , 0 , ...
'lineColor' , 'AAFFFFFF');
end

Answer 1

Seems to me that the problem is on what ge_quiver does internally which you don't show. Aside that, you should note that this language is designed to write vectorized code (and not only because of the difference in speed).
Looking at your code, you can simplify it lot (and this usually does have an effect in speed as well) by making use of a vectorized syntax:
step = 60;
dt = step:step:(step * size (data_u, 3));
## Get date variables.
YYYY = date_s.year;
MM = date_s.month;
DD = date_s.day;
HH = date_s.hour;
mm = date_s.minute;
ss = date_s.second;
## Date format. Use Google's.
F = "yyyy-mm-ddTHH:MM:SSZ";
tStart = datestr (datenum (YYYY, MM, DD, HH, mm + dt , ss), F);
tEnd = datestr (datenum (YYYY, MM, DD, HH, mm + dt + step, ss), F);
At this point, tStart and tEnd have all the dates that you need, one time per row. The only required loop is the following:
for i=1:rows(tStart)
kml = ge_quiver (lon, lat, data_u, data_v, "timeSpanStart", tStart(i,:), ...
"timeSpanStop" , tEnd(i,:) , ...
"visibility" , 0 , ...
"lineColor" , "AAFFFFFF");
endfor
At this point is all dependent on what ge_quiver does. The mathworks site you linked appears as under maintenance. I looked into the Google Earth toolbox website and couldn't find that function (they have a ge_quiver3 function though).
I would argue that a function for this type of languages should allow for vector inputs, so go and confirm it. Maybe you don't need a loop at all.