nppGrayErode vs IppGrayErode

I am trying to use
I took the boxfilter npp sample and modified it
I also added ippiGrayErodeBorder_8u_C1R processing using the same data.

the ippiGrayErodeBorder_8u_C1R works as expected, the npp acts as a simple erode.

clearly I’m am doing something wrong. Anyone have any idea?

The project was attached, to see the difference go to the executable and type
grayErodeNPP.exe input sourcePixelMask.png

It takes about 15 seconds to finish the ipp call, 400 millisecond for the npp. (2.8 MB)