Dear Blaise and Xinian,
We have looked at this and discovered the culprit.
There is a very inefficient call to a third-party matrix library
which is called far too often. The problem - as you've discovered
is one of repeated memory allocation and deallocation. We have
changed the code here and the new version will appear in the next
patch. It is a factor of around 600 times faster! :)
All the best,
Mark
Blaise Frederick wrote:
> Ok, I have to amend my previous post - this time when I ran it, it did complete after 6 hours of cpu
> time. I did a little profiling, and it seems to be spending an awful lot of its time allocating and
> freeing memory in the depths of calcFFT. The results are pretty much what I expected, so I think it's
> actually working, just incredibly slowly.
>
> Blaise
>
>
|