Zip compression predictor in 32 bit mode

Hello

I'm writing plugin to import psd files and I have problem with layers in 32-bit mode. Layers are compressed in zip with prediction.

I can decompress without any problems but I don't have any idea about predictor... I tried adopt predictor used in 16-bit mode (previous pixel difference) and extend it to 4 bytes, but not to avail. I also tried some other methods like previous byte diference (exactly like in EXR format) with integer math- no succes. I get only color noise.

So I have question: what is prediction algorithm in 32-bit mode? Is it previos pixel difference? Is it integer or float math??

Any advise would be very helpful for me. I hope this is not military secret...

Thanks