NVIDIA / MatX

An efficient C++17 GPU numerical computing library with Python-like syntax

Home Page:https://nvidia.github.io/MatX

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[FEA] Add operator input support to FFT/solver

cliffburdick opened this issue · comments

The functions that use backend libraries currently cannot take operators as inputs, with the exception of matmul. The reason was because these functions didn't allow iterators as inputs or outputs. However, to make the API more uniform we can detect if the input is an operator and do a copy to an async-allocated tensor instead. This would save the user from creating temporary tensors and should have no performance impact.

@luitjens already fixed this for FFT it looks like. Solver is the only interface missing.