IPP in Pentium 4 is slower than IPL

[alexm]

I am working with evaluation copy of IPP downloaded from here:

http://www.intel.com/software/products/ipp/downloads/ippwin_ev.htm

I used file w_ipp_ia32_itanium_em64t_eval_4_1_ev05.exe

It should be optimized for Pentium 4 processor. OS is WinXP Professional.

To test IPP performance, I created sample program which makes 2D convolution of 1024*1024 float image. Program is written using Image-Processing-IPL sample (using existing IPL functions). Another program uses IPL to make the same convolution.

Test results on Pentium 3:

IPL 105 ms

IPP 76 MS

Test results on Pentium 4:

IPL43 ms

IPP48 MS

Results are surprising: IPP is optimized for Pentium 4 processor, but IPL gives better results. What is wrong, maybe I need another IPP version?

Thanks.