🚀 Feature

Support for complex-valued torch.cholesky.

Motivation

Pytorch 1.6 supports complex-valued tensors, however, most of the matrix operations cannot be written in terms of their real analogue (in contrast to e.g. multiplication, which currently also does not support complex-valued data). Cholesky decomposition is important in decorrelating e.g. Monte Carlo data. The MAGMA backend also supports Cholesky decompositions for complex-valued data.

Pitch

torch.cholesky supports complex-valued data.

Alternatives

Possible workarounds

cc @ezyang @gchanan @zou3519 @anjali411 @dylanbespalko @nikitaved