ashvardanian/less_slow.cpp

Add cuGraph examples

Opened this issue · 10 comments

ashvardanian commented

We currently have a pretty extensive list of graph storage and processing approaches on the CPU, but lack GPU analogs. This can be a great opportunity to showcase both the more advanced and rarely used parts of the Nvidia libraries, as well as explore atomic variables and concurrent data-structure implementations on massively parallel chip architectures.

This is also worth covering on the Python side with the available Python cuGraph bindings:

syedshazli commented

I can get started on this, can you elaborate on what you'd like to see as well as where in the project structure this should be placed?

  • ❤️1
ashvardanian commented

The cuGraph invocations should go into less_slow.cu, while the benchmark logic should go after the CPU-side Graph-related benchmarks in less_slow.cpp.

  • 👍1
syedshazli commented

Nice, I think you might have to assign the issue to me and I'll get started. I'll ask any questions in here along the way

  • ❤️1
ashvardanian commented

Done! Looking forward to your PR @syedshazli!

  • 🎉1
syedshazli commented

Hi Ash, thanks for assigning me the issue. I've worked with Google benchmark in the past, but I was trying to use cmake as described in the README. Was wondering if you had anything like this message pop up before? I think my current version of Cmake (3.22) won't allow me to run the commands which require Cmake 3.25. When I try to upgrade versions, I get this.

Image

ashvardanian commented

@syedshazli , you can fetch a newer CMake from PyPi 🤗

  • 👍1
syedshazli commented

Hi Ash, I'm currently a University student having finals and projects due this week and next week, so I won't be able to set aside much time until next Friday. If you want to assign this PR to someone else or work on it yourself, feel free,if not then I can get started next week. Sorry for the inconvenience

ashvardanian commented

No rush! Best of luck with exams! Submit a PR whenever ready, @syedshazli 🤗

  • 😄1
syedshazli commented

Thank you!

syedshazli commented

Hey Ash, I really like the way the repo is right now, but to be honest I've never used CUDA which is why I'm learning CUDA C right now. Sicne it might take some time to get the basics of CUDA down, I'm not sure how much help I can be when referencing Cugraph. This repo seeks to get the most efficient C++/Cuda code out there, and I don't know if I can writie it given I don't have the basics of it down yet.

So far reading the Cuda_by_example pdf out there by Nvidia.

  • 👍1