About 18 months ago,I attended an Intel webinar where they presented a new and exciting version of one of their library. The key point was the following: previously, when applying multiple image processing filters to an image, the 1st filter was done by looping over the whole image, the 2nd was also another loop over the whole image, ... The advantage of the library was that filter were now designed to perform on rows, so the trash would not be thrashed as much. And Intel presented a lib that would handle all that.
I am now looking for info on this lib, but I now cannot remember which library it is, or where to find the doc for this feature. Is this IPP? Can someone point me in the right direction please?
This webinar was about Deferred Mode Image Processing Layer (DMIP), a library we built on top of IPP, so it use IPP low-level optimization and in addition to that it also tries to optimize computational pipeline by subdividing the data to keep processed chunk hot in processor's cache and run several such chunks in parallel through set of processing operations
It is released as a part of IPP sample package. We would be interested to hear your feedback on thislayer.