For loop question.

When looping over indices of an array I use 1:length(x),

x = rand(1,10); for i = 1:length(x) ... end

is there a more efficient way?

Answer 1

[Jan]

There is no more efficient way.

As Walter mentioned already accessing the indices is faster, when you use integer types. But then you have to think twice, or better 8 times, if a saturation is possible. If you are in doubt, use at least UINT32 or stay at doubles. Example:

No advantage for integers:

x = rand(1, 1e4);
for i = 1:1e4
  y = x(i);
end

for i = uint32(1):uint32(1e4)
  y = x(i);  % <- same speed as DOUBLE index
end

An integer loop index is faster, if it is used for indexing inside the loop:

x = rand(1, 1e4);
for i = 1:1e4
  y = sum(x(1:i));
end

u1 = uint32(1);
for i = uint32(1):uint32(1e4)
  y = sum(x(u1:i));  % <- faster than DOUBLE index vector
end

But the integer saturation does not cause an error message:

for i = uint8(1):uint8(1e4)  % !!! uint8(1e4) is 255!

Creating the index vector explicitly is slower:

v = 1:1e4;
for i = v      % Slower
  y = x(i);
end

for i = 1:1e4  % Faster
  y = x(i);
end

Using an explicit data vector can be nicer, but the speed depends on the details:

v = rand(1,1e4);
for av = v
  y = av;
end

[Andrei Bobrov]

+1

Answer 2

[Walter Roberson]

Better practice would be 1:numel(x) but otherwise what you are doing is fine.

Jan would probably point out that using uint32() indices are faster when they are sufficient to hold the index values.

[Jan]

@Walter: This is my "answer of the week"! I'm amused, that you do not create just your own answers, but even take into consideration the usual attitudes of other contributors. Thanks :-)

I'd prefer INT8 when the indices are in the range of 1 to 10. On the other hand a beginner will be confused by such tricks and the total program time (design phase + programming + testing + debugging + run time) will increase.