Why do you say that you must have a number of bins that is not an integer? How can you have anything but a positive integer for the number of bins?
You can approximate pretty closely a bin width of 1/2*std(data)
data = randn(1000,1);
binwidth = 1/2*std(x);
numbins = round(range(data)/binwidth);
% now construct a histogram just to check bin width
[~,binctrs] = hist(data,numbins);
% look at bin width
mean(diff(binctrs))
% compare to 1/2*std(data)
1/2*std(data)
You should see there is very little difference between the bin width you want and what you get.
Now you can use that number of bins in histfit()
histfit(data,numbins)
