vectorizing or speeding looped code

i am trying to speed up some code with multiple functions and have found the one that takes the most time (it is all a converted fortran code). it runs a nested for loop 4 times (each with modified input). i've had some success elsewhere in the code eliminating slow points but this one just cant get right. any ideas? (in 2016b for compatability reasons)

for II= 1:N
    XP(1)= 1.0D0;
    for JJ=2:IORD1
        XP(JJ)=XP(JJ-1)*double(A1(II));
    end
    for JJ= 1:IORD1
        for KK= 1:IORD1
            B(JJ,KK)=B(JJ,KK)+XP(JJ)*XP(KK);
        end
        C(JJ)=C(JJ)+XP(JJ)*A2(II);
    end
end

Answer 1

[Turlough Hughes]

This should help, though I've already made assumptions about the sizes of arrays. How many rows/columns are in each variable?

XP(1) = 1;
for II= 1:N
    XP(2:end) = XP(1:end-1)*double(A1(II));
    B = B + XP(:).*XP(:).';
    C = C + XP*A2(II);
end

[Paul Schenk]

the dimensions change (get smaller) in each of the 4 runs. they will have to be declared each time.

[Turlough Hughes]

Did the code work?

Some questions: is XP a column vector/row vector? Does A2 have as many elements as XP? Also, does B contain IORD1 rows and columns? Similarly does C have IORD1 elements?

Answer 2

[Paul Schenk]

not as entered. tripled cpu time.

here are preallo's to variables:

ORDER_MAX = 20;

C = zeros(ORDER_MAX+1,1);

IORD=MAXOR;

IORD1=IORD+1;

B = single(zeros(ORDER_MAX+1,ORDER_MAX+1)); % Solution matrix

XP = double(zeros(ORDER_MAX+1,1)); % Array of solutions

[Paul Schenk]

a1 and a2 are linear (1,2000) arrays that are usually only 10% filled in.

[Turlough Hughes]

Vectorisation is usually faster but not always. Probably best to just provide a minimum working example (google it) of code that I (or others) can test and profile. Otherwise it really is just guess work.

Answer 3

[Paul Schenk]

hope this works for you. it is essentially a customized LS fit solution.

[Turlough Hughes]

Can you give some typical inputs aswell?

[Paul Schenk]

example:

eqor=1, n=222 and a1 and a2 are linear vectors of numbers N long (they change each time and can be random for this test)

[Turlough Hughes]

eqor=1 ? I assume that's MAXOR?

[Paul Schenk]

yes, my mistake!