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] cuTENSOR 2.0 support

lucifer1004 opened this issue · comments

Now that cuTENSOR 2.0 has been released for some time, I think it is natural for MatX to follow this update and enjoy the performance gains brought about by this new version.

I just realized that since cuTensorNet is used in MatX, this update is blocked until cuTensorNet adopts cuTENSOR 2.0.

cuQuantum 24.03 will move to use cuTENSOR 2.

cuQuantum 24.03 will move to use cuTENSOR 2.

Thanks for the information.

@lucifer1004 you just have read my mind because we were just talking about this. I'll try to get to it next week. Do you use einsum right now?

I need to use einsum. See #597.

@lucifer1004 we will wait until the new cuquantum is released so that it's compatible with cutensor 2.x.

@lucifer1004 the existing (older) cutensor + einsum is useable now and we will integrate the new cutensornet once ready

FYI cuQuantum 24.03 is out.

Updated to cuQuantum 24.03 and verified fix.