Closed
Description
Follow-up of #102. Currently we state that complex numbers will be supported in the next version (v2) of the standard. However, there is nothing forbidding the complaint libraries from implementing them, if they choose so (ex: CuPy has full support already).
In today's call, people seem to be happy about the idea that we specify the expected behaviors of APIs that could accept (or generate) complex numbers, so that we can provide a guideline for those libraries choosing to support it, and also earn a long enough time to potential changes or iterations before v2 is finalized.
Therefore, off top of my head a few linalg PRs will need some changes to set up the expectations, including
- Eigensolvers (WIP: Add specifications for computing eigenvalues and eigenvectors #113)
- SVD (Add SVD function specification #114)
- QR (Add specification for computing the qr factorization #126)
In addition, this would also unblock FFT (#159).