Ying-Jer Kao

Added tests for BlockUniTensor and Bond

1

Bump up test coverage

Tensor indices are `int`

1

In the process of making clang compiling works, I notice that the Tensor indices are defined as `int` to allow for -1, but people regard the indices as `unsigned int`...

Now that CuSolver has implemented LU decomposition https://github.com/NVIDIA/CUDALibrarySamples/tree/master/cuSOLVER/Xgetrf I propose to replace `magma_zgetrf_gpu` in `cuDet_internal.cu` with `cusolverDn` https://docs.nvidia.com/cuda/cusolver/index.html#cusolverDnXgeqrf With this update, we can drop the dependency on MAGMA.

Torch FX as code generator for AD

1

PyTorch 2.0 released the FX module, which consists of three main components: a symbolic tracer, an intermediate representation, and Python code generation. https://pytorch.org/docs/stable/fx.html Example code: https://github.com/pytorch/examples/tree/1bef748fab064e2fc3beddcbda60fd51cb9612d2/fx This allows for manipulation...

Refactor internal functions using std::variant and visitor pattern

10

In commit https://github.com/Cytnx-dev/Cytnx/commit/710c98c73d24dba02bdbecad54dcfc8dcdf5b6fc , multiple dispatch using `std::variant` together with the visitor pattern. @ianmccul already provided an example implementation in `Kron_internal.hpp` and a simpler example is available here https://en.cppreference.com/w/cpp/utility/variant/visit2 We...

I noticed that when UNI_ICPC flag is turned on, no overloading of complex arithmetics are provided in `complex_arithmetic.hpp` and `complex_arithmetic.cpp`. For example, https://github.com/Cytnx-dev/Cytnx/blob/8474bbfd4c856be7b13721006cf084dd9dfeb427/src/utils/complex_arithmetic.cpp#L5-L15 @kaihsin I wonder what is the reason...

Bug-prone using unsigned long long to handle vector/storage size

1

I see a potential bug in handling vector/storage size using `unsigned long long`. In several internal functions, the length is treated as an `unsigned long long` which corresponds to `uint64_t`....