topic Windows/Linux Differences when using ippiResizeCubic_ or ippiResizeLanczos_ in Intel® Integrated Performance Primitives

Windows/Linux Differences when using ippiResizeCubic_ or ippiResizeLanczos_

Greg2 — Tue, 04 Mar 2014 15:43:15 GMT

I am seeing differences when running ippiResizeCubic_8u_C1R and ippiResizeLanczos_8u_C1R between Windows and Linux. There are pixel value differences that are off by one. They tend to be closer to edges, but differences show up in the middle of the image as well. I am using version 8.0 update 1. Is this a known issue? Is there any way to get them to match?

Thanks,
Greg

Hi Greg,

Valentin_K_Intel — Thu, 06 Mar 2014 16:45:43 GMT

Hi Greg,

For different operation systems a compiler can make different optimizations, which can have a different order of arithmetic operations that can lead to small deviations of computation results. If intermediate calculations are in floating point values and output values are integer, the small deviations can lead to rounding errors (±1). I believe that this is the reason of the differences. So there is no any way to get them to match on Windows and Linux.

Best regards,

Valentin