- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
I have a large complex image processing application that uses the commercial IPP 4.1 static libraries. When I try to upgrade the application to use IPP 5.1 commercial libraries I notice that it performs poorly overall. Average CPU usage increases by 4.2 % (also each processor seems to have more utilization) and in particular one function that we benchmarked, ippiWarpPerspective_8u_C3R was about 4.7 ms slower overall. I tried using the dynamic version of the new library to see how that fared, with little success. I tried using a variety of threads (2, 4, 8) with no success. I even tried setting the CPU manually, as I saw this helped one poster but this also didn't make any difference.
The major change in getting our app to build with IPP 5.1 was several functions (template matching mostly) didn't require a working buffer. So I removed these buffers in our app - some of these were globals so it was fairly efficient. I assume that these functions now handle the buffers internally (?).
I was expecting a performance gain, not a loss overall in using this new version of the library. I know that there were supposed to be significant benchmarking improvements in the 5.x series. Is there something I'm missing in using this new version to get performance gains?
Thanks,
-David Weiss
I have a large complex image processing application that uses the commercial IPP 4.1 static libraries. When I try to upgrade the application to use IPP 5.1 commercial libraries I notice that it performs poorly overall. Average CPU usage increases by 4.2 % (also each processor seems to have more utilization) and in particular one function that we benchmarked, ippiWarpPerspective_8u_C3R was about 4.7 ms slower overall. I tried using the dynamic version of the new library to see how that fared, with little success. I tried using a variety of threads (2, 4, 8) with no success. I even tried setting the CPU manually, as I saw this helped one poster but this also didn't make any difference.
The major change in getting our app to build with IPP 5.1 was several functions (template matching mostly) didn't require a working buffer. So I removed these buffers in our app - some of these were globals so it was fairly efficient. I assume that these functions now handle the buffers internally (?).
I was expecting a performance gain, not a loss overall in using this new version of the library. I know that there were supposed to be significant benchmarking improvements in the 5.x series. Is there something I'm missing in using this new version to get performance gains?
Thanks,
-David Weiss
Link Copied
0 Replies
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page