GitLab

* Set C++17 for latest pytorch versions. Add flags for CUDA 12 and 11.8

* Update setup.py

* remove import subprocess

* more robust way to compare version

---------
Co-authored-by: Jinze Xue <jinzexue@ufl.edu>

* Fixed: symeig deprication

This causes:
RuntimeError: This function was deprecated since version 1.9 and is now removed. Please use the `torch.linalg.eigh` function instead.

* Fix: slow tensor creation

This raises:
Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor.

Co-authored-by: Raymond G Schireman <raymond.schireman@uvm.edu>

* Fix cuaev build on CUDA 11.5 and latest torch

* Update aev.cu
Co-authored-by: Jinze (Richard) Xue <yueyericardo@gmail.com>

* Fix cub to 1.11.0 for cuda 10.2

* quick save

* quick save

* Fix --cuaev-all-sms

* expecttest

* fix ci

* Fix AEV and CUAEV GPU Device Bug

* format

* apply review suggestion

* apply review suggestion

* add a small test for 2x, which was completely missing from tests

* remove unused code

* flake8

* fix

* rm

* Wrap cub with CUB_NS_PREFIX and remove dependency on Thrust

* update readme

* format

* Update install_cuda.sh
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* assume-unchanged

* fix

* fix cuda ci

* init

* separate to h cpp cu files

* works

* fix

* clean

* update

* fix

* only keep one copy of Result

* const result&

* move declaration to header

* fix jit

* clean

* fix

* fix jit

* fix ci test

* clean

* revert to autograd in charge of it's result

* update

* fix memory issue of result.aev_t

* Update torchani/cuaev/aev.h
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update torchani/cuaev/aev.h
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update torchani/cuaev/aev.h
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* save

* Update torchani/cuaev/aev.h
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* update

* Update torchani/cuaev/cuaev.cpp
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update torchani/cuaev/cuaev.cpp
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update torchani/cuaev/aev.h
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* remove aev_t after forward

* remove release function

* fix

* fix jit

* update
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update install_cuda.sh

* Update install_cuda.sh

* Update install_cuda.sh

* update

* update

* update

* update

* update

* update

* update

* update

* save

* Remove constant memory usage

* Update torchani/cuaev/aev.cu
Co-authored-by: Jinze Xue <yueyericardo@gmail.com>

* Update torchani/cuaev/aev.cu
Co-authored-by: Jinze Xue <yueyericardo@gmail.com>

* Update torchani/cuaev/aev.cu
Co-authored-by: Jinze Xue <yueyericardo@gmail.com>

* format
Co-authored-by: Jinze Xue <yueyericardo@gmail.com>

* init

* init

* double backward test

* fix doublebackward test

* add another test

* rm gaev

* radial done

* angular init

* angular done

* update

* force training benchmark

* format

* update

* benchmark

* update

* update

* clean redundancy codes

* update

* adapt review request

* update

* update

* update

* update

* update

* update

* fix

* fix

* cuAngularAEVs code deduplicate

* pairwise double backward

* cuRadialAEVs dedup

* pairwiseDistance dedup

* format

* readme build notes

* save

* update

* save

* save

* update

* fix

* save

* add equations on comments
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* improve documentation and fix buggy behavior of shuffle() and cache()

* remove warnings

* Fast Build CUAEV

* fix

* --cuaev for fastbuild, --cuaev-all-sms for all

* update readme

* fix

* fix

* fix

* fix

* fix

* format

* Update torchani/cuaev/aev.cu
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* update

* rm

* update

* update

* update

* ase test atolerent

* preparation

* radial preparation 30%

* radial backward kernel done

* reuse Gmr (exp part) result for gradient

* radial kernel every block run by column major, to avoid atomicAdd waiting

* apply code review

* static_cast

* implicit cast

* format

* angular preparation

* angular backward works, but slow, AtomicAdd should be avoided

* angular opti: use share memory to avoid AtomicAdd

* format

* equation optimization

* remove unnecessary shared mem for atomi

* remove a lot (warpsize * nbr) unnecessary shared mem for atomj

* format

* update

* clean

* fix

* fix

* test file

* fix
Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

Co-authored-by: Gao, Xiang <qasdfgtyuiop@gmail.com>

* Update docs.yml

* Update docs.yml

* Update docs.yml

* Update docs.yml

* Update docs.yml

* Update docs.yml

* Apply suggestions from code review
Co-authored-by: Richard Xue <yueyericardo@gmail.com>
Co-authored-by: Richard Xue <yueyericardo@gmail.com>

* line-limit 120

* CRLF to LF

* clang-format

* line-limit 120