Re: Faster approach for total(data,dimension) possible? [message #66999] |
Fri, 26 June 2009 07:44  |
rogass
Messages: 200 Registered: April 2008
|
Senior Member |
|
|
Dear all,
thank you for your suggestions. I will pick up JD's approach - like
several times :). For a very large data set (eg. 5000x5000x2000) I
will generally broke it up into chunks. Then within the chunks, I will
generate a mask for the first channel where the pixels have zero
values assuming they are zero-channel representatives. This mask will
be used with the total approach and the ~mask with JD's For-Loop-
approach. After this I will merge both results - result1+result2 eq 2.
This may be faster for a very large data set. Additionally, the
presence of a working gpulib will be tested and used to increase the
computation speed.
All the best
CR
|
|
|