Any other DL/HPC related algorithm that CUTLASS could adapted in?

[Peter9606]

Just wondering if there is any other algorithm in deep learning or HPC domains that CUTLASS could fit in.

[hwu36]

There are a lot. Here are those that are on top of my head.

Block sparse: https://github.com/openai/blocksparse

Mos of things here: http://clmathlibraries.github.io/clBLAS/group__BLAS3.html

Semi ring gemm: https://github.com/hpcgarage/cuASR

Others can add more.

[Peter9606]

Personally I'd like to have "Plan your project", as I'm curious about the road map.

[hwu36]

I created a blank project. Feel free to play with it you want.

As to the roadmap, CUTLASS is tightly connected to NVIDIA hardware development. I will see how much that I can say.

[Peter9606]

I'd like to add some open issues as "To Do" in Kanban, but I cannot find "Add Cards" in the project.

[hwu36]

I can see add cards in the upper right corner.

[Peter9606]

Following is my view. Do we have the same options?

[hwu36]

if you close menu, you can see "add cards" left to the Menu button.

[Peter9606]

I'm afraid I still cannot find it.

[hwu36]

I don't know much about how to use project feature. I guess only the maintainer can add cards. Actually a card was created automatically when I created this project, but I don't see it in your screenshot.

[Peter9606]

That's alright.
Actually I'm thinking about adding a couple of "TODO" items to track open issues, some of them are listed here:

• Profiling parallel split-K gemm
• Resurrect GEMV
• and more...

[hwu36]

Thank you very much. There are couple more that contributor outside of Nvidia can help us.

• fix all the bugs when building for maxwell and pascal.
• fully support clang.
• clean up the code.
• add "-Werror -Wall"

We are encouraging the community help us ;)

[Peter9606]

Hope these tasks can be manged well by utilizing Project tool