We are trying to use the 5.2 ixp linux IPP libraries on the PXA270 xscale processor for JPEG decompression/scaling.
Our gcc toolchain does not support VFP floating point so the IPP libs would never link. By reading the forum, we figured out that if we modified the ELF headers in the IPP libs, we could then link. We have done this and been able to run our application using IPP.
We are using the CJPEGDecoder sample class to do jpeg decompression and have run into a problem where the DCT functions do not seem to work correctly. The output data is incorrect for DCT values of 1_2 and 1_4.
We have run our same application on windows using the 5.2 IA32 IPP LIBS and verified that our code and the DCT functions work correctly. The do not return an error code.
These calls are used in the codec for the DCT values of 1_2 and 1_4:
ippsMul_16s
ippiDCT8x8To2x2Inv_16s_C1I
ippiDCT8x8To4x4Inv_16s_C1I
ippiAdd128_JPEG_16s8u_C1R
It seems like the most likely culprits are the DCT calls
Can anyone tell me if the VFP hack we have made could be the cause of this problem? What does VFP actually do? Since ixp should run on the xscale (from intel and this forum), does that mean that the libraries for ixp have a problem? Can anyone shed any light on what might be our problem and a possible solution. Help?
Thanks,
Michael
