I am doing a feasiblity study of a project which is going to use IPP functions. I tried a VC1 transform function ippiTransform8x8Inv_VC1_16s_C1R() but it was giving wrong output(Seems to have some precision problems). Is there anybody who have already experienced this problem? I am using IPP version 5.3.3.082.
Please have a look at thecode snippet
int16 inp, out0,out1;
dstSizeNZ.height = 8;
dstSizeNZ.width = 8;
for(i = 64; i--;)
inp = i;
InverseDCT8x8_16bit(out0, inp, 8); // VC1 8x8 16bit transform C version
ippiTransform8x8Inv_VC1_16s_C1R( inp, 16, out1, 16,dstSizeNZ );
Can you report this into Intel IPP product in Intel Premier Support? So we can take further look on this issue. You may also include the bitstream you tested with we would try to replicate this precision issue.
In the meantime, you may also check with the latest version IPP 60 beta to see ifit has same issue or not.