Perform Gauss-broadening on line spectrum

[Maxim Gawrilow]

Dear Octave users,

I have created a bunch of spectral lines and would like to simulate the 
Gaussian broadened spectrum of these. The spectral lines are a Nx2 
matrix where the first column contains the position and the second the 
intensity of each line. What I want is to place a Gaussian at each 
position, scale it with the appropriate intensity and calculate the sum. 
In theory, this code would do the job:

gauss = @(x,x0,Int,sigma) 
Int./(sigma.*sqrt(2*pi)).*exp(-0.5*((x-x0)./sigma).^2);
x = linspace (0, 4000, 20000);
outdata = sum(gauss(x,inputdata(:,1),inputdata(:,2),0.8),1);
exportdata = [x', outdata'];

However, inputdata contains quite a few lines: at least half a million, 
sometimes even several dozen million entries. Execution of said code 
results in "out of memory of dimension too large for Octave's index 
type". Therefore I'd need a different approach.
I am able to get the correct result in gnuplot with

plot "inputdata" smooth kdensity bandwidth 0.8

but this is extremely slow. Already for half a million entries it takes 
5 minutes for a single plot, and will obviously take hours to compute 
cases with dozens of millions of entries.

Does anyone have an idea of speeding up this process? I guess 
convolution with a Gaussian would be a correct approach as well, but I 
didn't find any function that suits my case (conv only takes equidistant 
vectors, and I'm a bit of a loss at understanding kernel_density from 
econometrics).

With kind regards, Maxim

--
Maxim Gawrilow
Georg-August-Universität Göttingen
Institut für Physikalische Chemie
AG Suhm

Tammannstr. 6
37077 Göttingen