Added patch for VASP.6.6.0

[AsukaHanaT]

Hi, I've modified the patch for VASP.6.6.0 to support the newly added GPU offloading feature for Intel and AMD GPUs.
I tested all the results against the tutorial examples included in the b4vasp documentation, and they seem correct.
However, I do believe the computational parameters (OMP_NUM_THREADS, KPAR, ENCUT, etc.) need to be carefully adjusted, especially for larger systems, to avoid memory-related issues. I am not sure whether this was already an existing issue with the LKPROJ scheme, or whether it was introduced by my patch.

[MicheleReticcioli]

diff_Asuka-vs-MR.txt
vasp.6.6.0_U-patch.26_MR.sh

Hi! Thank you very much for adapting the patch to VASP 6.6.0. We are happy to see that the community is actively participating in the project.

I do not currently have time to inspect the updated code carefully, but I had a quick look and ran a few tests. I am uploading a slightly modified version of the patch. What do you think of these modifications compared with your version? Let's discuss these modifications before merging your pull request.

Regarding memory usage, this is also an issue in the basic implementation included in the default VASP releases. Our patch improves the performance, and the possibility to leverage GPU offloading in your version further facilitates the calculations. In my tests unfolding large unit cells on the LEONARDO-CINECA HPC system, I had to stop at 5 k-points on CPUs, while I reached 100 k-points on GPUs without problems.

In case of memory issues, our recommendation is to split the unfolding band-structure calculation into multiple calculations by dividing the original k-path into smaller intervals. This can be easily done using the bands4vasp suite, including both the pre- and post-processing of the sub-interval folders.

Thanks again for your contribution!