Advice for approaching data locality improvements in legacy code

[aledinola]

I read the part on Numerical Recipes (Fortran 90 edition) about array syntax, but I think it is a bit outdated. I tried coding up a simple dynamic programming example in the post linked below and it turned out that the versions using array syntax instead of loops are slower. The code of the example below is also available on github

Performance of vectorized code in ifort and ifx

I created a small Fortran example of discrete dynamic programming that occurs quite often in macroeconomics. It is an optimal savings problem for a consumer who received idiosyncratic shocks to their income. I was interested in comparing loop-based Fortran code vs vectorized code, both in ifort and in ifx. Just to avoid misunderstandings, the definition of “vectorizing” that I am using is the following: I vectorize if I replace this code do i=1,n yvec(i) = exp(xvec(i)) enddo with the followin…