Re: octave vs. matlab vectorization

[matt]

thanks for the insight on compiler issues.

the matlab JIT compiler is doing something clever in the simple- 
minded codes i was experimenting with but i realise now that a lot of  
the problem is the overhead in allocating and defining arrays.

i had a vector q of length 10. in the loops code, it was fast to  
check through it elementwise; in the (semi) vectorized code, time was  
being spent setting up q and also tmp and (tmp>50), both length 10.

replacing 10 by 1000 puts the two codes on a par as now the overhead  
to set up the vectors is relatively small.

in octave, of course, the vectorized version wins by miles.