v. 5.0.017 bug in p4 vs p1 optimised h263 encoder?

[knutinh1]

I am experiencing visual artefacts that could be related to rounding errors in DCT or something similar when running p4 optimised versions of the IPP v 5.0.017 h263 encoder against known "reference" decoders, but not against IPP decoder itself. However, when I switch to p1-only optimisation, this error disappears. Is this a known error or am I doing something worng?

regards

k

[knutinh]

I use a long I-frame distance, however, I have modified the MB coding so that the 132 limit is never crossed. The artefacts appear on a much shorter timeline than 132 frames anyways.

My explanation is that either there is a bug in the IPP 8x8DCT for P4-HT, or there is some strange behaviour in the IPP example h263 encoder that cause differences between p1 and p4 DCTs to be amplified.

Using px instead of w7 for the following functions solved my issue:

ippiDCT8x8Inv_16s_C1I

ippiDCT8x8Inv_16s8u_C1R

regards

k

[Vladimir_Dudnik]

By the way, the latest available version of IPP is 5.1.1. Could you please evaluate the latest version of IPP and media sample?

Regards,
Vladimir