In the small timing test I just did, leaving the value inside the struct was slightly faster. However, it is entirely possible that I simply did not do enough calculation with the variable to make up for the cost of doing the assignment to the variable.
Basically... you should time both versions.
Be careful, though: I have multiple times found that if I use something like
result1 = timeit(first_test,0)
result2 = timeit(second_test,0)
that first test I do is notably slower, so exchanging first_test and second_test would give very different results about which was faster. It is not just the first timeit() call either: if I use a loop to call timeit repeatedly then the results for the entire first loop are odd.
str = struct('foo', rand(1,50), 'bar', 'hello there how are you?');
F1 = @() fun_direct_struct(str);
F2 = @() fun_tovar(str);
N = 100;
T1a = zeros(1, N);
T1b = zeros(1, N);
T2a = zeros(1, N);
for K = 1 : N
T1a(K) = timeit(F1, 0);
end
for K = 1 : N
T2a = timeit(F2, 0);
end
for K = 1 : N
T1b = timeit(F1, 0);
end
plot(1:N, T1a, 'k*-', 1:N, T1b, 'kv--', 1:N, T2a, 'b.:')
legend({'T1a', 'T1b', 'T2a'});
In the above, T1b and T2a are pretty much constant in the time they take each iteration, but T1a varies a lot, especially initially. If you were to only measure F1 the first time then you would get a very different impression about which function was slower.
