search

NVIDIA / MatX

An efficient C++17 GPU numerical computing library with Python-like syntax
https://nvidia.github.io/MatX
BSD 3-Clause "New" or "Revised" License
1.21k stars 82 forks source link

[BUG] Increase MatX TensorViewToNumpy and NumpyToTensorView to support rank = 5 #425

Closed nvjonwong closed 1 year ago

nvjonwong commented 1 year ago

Current TensorViewToNumpy and NumpyToTensorViews only work on rank < 5 matrices.

Proposing to increase to add rank=5 support and add a static_assert so users will know about the implementation limitation.

cliffburdick commented 1 year ago

Ideally TensorViewToNumpy should work with any rank and the DLPack support we added. The other way is it as easy since it's going from dynamic ranks to static, and you have to bound it to a point.

nvjonwong commented 1 year ago

This is my attempt at being able to intake more arbitrary numpy data. https://github.com/NVIDIA/MatX/pull/428