Support derivative related metrics. Details can be found in our DRBSD8 paper.
Support Lorenzo+Quantization prototype function (for both CPU and GPU).
The prototype includes the prediction and quantization (for both CPU and GPU).
The input is the original raw dataset (stored in binary format - single precision) and error bound.
The output will be the set of integers which need to be compressed losslessly.
You can use automake to build the code.
To enable GPU, you need to run the command like this:
./configure --with-cuda=.... --prefix=....
the -with-cuda will switch on the GPU function and specify the installation path of the CUDA (the root directory that contains bin, lib64, ...).
Then, make; make install