I am new to the MLIR Community

MLIR
1

Hi,

I am an ML Compiler Engineer and experienced MLIR user. I love LLVM, I am passionate about Compilers and ML.
I work weekdays and weekends, always eager to learn and find new challenges. I am just a lambda programmer, but I really hope one day I will be able to make bigger impacts in the field of ML Compilers.
I am still figuring out how (all advices are welcome btw), but some recent thinking about myself and my work lead me to the realization that despite all the work I’ve been doing, I never contributed (or even engaged) with the LLVM / MLIR Community.

And that’s why I am posting today, to get some advice on how to become a more active member ? Where to start ?
My original idea was to contribute to MLIR Codebase, and I’ll try and find some time to do that. But things will be frantic for the next 3-4 months and I don’t know if I’ll actually get much time for that.
I was hoping instead to learn more from other fellow MLIR passionates. I tried looking at the Office Hours, but there doesn’t seen to be anything really specific to MLIR / ML Compilers. Are there any other online meetings ? Places to meet with the community ? (I live in a place in Europe far from technological hubs so conferences / networking events isn’t really an option for me).

I would also love to have some online chat with other members of this community.
To give a few more details about myself, I work as an ML Compiler Engineer. The team is quite small, but with a large scope, which is really cool because I can touch to many different things: TF / PyTorch Backends, XLA, HLO, torch-mlir, triton.
I love doing graph transformations / optimizations. Those can be high-level: fusion, buffers saving, layout conversion, dispatch to multiple IPs with cost model, reordering / fusing quant operations to try and keep floats as intermediates only. But also Lower level: Tiling, Vectorization, Memory / Compute reordering, Device To Shared memory promoting.
And these days I am more focused on backends (emitting LLVM IR, tweaking the backend).
And on my spare time I love reading fantasy novels (though after reading the most 2 infamous unfinished series I am not sure if I can get over it), Learning Languages (I am actually terrible at it but enjoy discovering cultures), and Eating Chinese Hot-Pot (How I wish I could go back to Shenzhen or Beijing).
If any of these sounds cool don’t hesitate to contact me.

Thanks Everyone, I can’t wait to get more Involved.
NS.

2 Likes
2

Hi there,

the best way to contribute is to fix a bug or implement a useful feature. There’s a bunch of bug reports here: Labels · llvm/llvm-project · GitHub, and an order of magnitude more unimplemented “TODOs” in the codebase. If you are already familiar with MLIR, it shouldn’t be very difficult to pick up one in the area that looks most interesting and directly contact people who are involved in the area (git blame is usually a good start to identify those).

As for contact points, there is a discord server LLVM that has MLIR-specific and location-specific channels. Most of the live discussion happens during ODMs, which I expect will resume after the summer hiatus. Face-to-face communication is usually happening in (Euro)LLVM developer meetings and affiliated events. Personally, these feel worthwhile to attend even if some travel is required. Some locations have infrequent LLVM social meetings, you can usually find announcements in Community - LLVM Discussion Forums.