[Octopus-devel] [Octopus-notify] svn commit: r4417 - in trunk/src: grid h_sys states sternheimer by dstrubbe

[Xavier Andrade]

On Fri, 25 Jul 2008, octopus-notify at tddft.org wrote:
> *) Changed explicit loops over NP in calc_beta to vector expressions
> with 1:NP instead -- isn't this potentially more efficient, as well as
> more compact in the code?

In theory, yes. But in practice, no. Many compilers (intel for example) 
take an expression like

a(1:n) = b(1:n) + c(1:n) + d(1:n)

and calculate it like this

tmp(1:n) = c(1:n) + d(1:n)
a(1:n) = b(1:n) + tmp(1:n)

which is clearly less efficient and requires extra memory. That is why we 
use explicit loops in performance critical parts.

Cheers,

Xavier