Q. regarding DMIP based design

Q. regarding DMIP based design

Hi all,

I have been recently looking at and experimenting with DMIP, and I have a basic question:

Suppose my task is to extract 2 columns from an Image, and add them. This I would like to do in a loop, for a different

image each time, but same column index.

It is easy to write a graph for this.

However, if using DMIP, after the source node that stores the image,

1) what node type should I use for the columns?

2) Can I overload the "add" to operate on columns?

Further, in the "classical IPP" based thinking, I would have used the GetRow/Columns for the line extraction, and ippsAdd variants for the

additions. Do I still use these, or I can use the loop based C-style summation and colums extraction, and rely on DMIP to substitute with an appropriate code?

Thanks,

Sid.

3 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hi Sid,

By "columns" do you mean real pixel-wide column or sub-picture of specific width and whole image height ? DMIP is  generally oriented on processing images by horizontal slices. This kind of operations better uses CPU cache.

Regards,
Sergey 

Regards,
Sergey

>>...Suppose my task is to extract 2 columns from an Image, and add them...

You could transpose the source image and addition of two rows ( formely columns ) will be significantly easier and faster.

Leave a Comment

Please sign in to add a comment. Not a member? Join today