Thesis: Using Deep Reinforcement Learning for Automatic Code Optimization in the MLIR Compiler

#Performance #Physics #QCD #MLIR

https://hgpu.org/?p=30054

2025 AsiaLLVM Developers' Meeting Talks
Videos: https://www.youtube.com/playlist?list=PL_R5A0lGi1ADKfJbzpA0rMDCb5T3QGe5k
Slides: https://llvm.org/devmtg/2025-06/#program
#LLVM #MLIR

https://blog.llvm.org/posts/2025-03-11-flang-new/

Seamless acceleration of Fortran intrinsics via AMD AI engines

#AMD #AI #MLIR #Performance #Package

https://hgpu.org/?p=29780

Some recent #LLVM goodies in #Gentoo:

1. Well, I've said that already but all packages are now in llvm-*/, so they're easier to find.
2. I've added llvm-runtimes/offload, which restores the offloading functions of #OpenMP (available in 19.1.6+).
3. I've added llvm-core/mlir, now that it provides a proper dylib to link to (available in the most recent 20.x snapshot, no plans to backport to 19.x).
4. I've started working on #Flang too, but no promises. At the very least, I'd like to make it support #MLIR dylib upstream.
5. I've sent patches for standalone build support in #ClangIR, so hopefully it'll be ready for integration when it lands in mainstream.
6. I've sent a new llvm-r2 eclass for review, now with much cleaner support for cross-compilation scenarios. https://public-inbox.gentoo.org/gentoo-dev/20241221160053.780079-1-mgorny@gentoo.org/T/#t
7. I'm also slowly working on improving cross-compilation support for LLVM packages themselves, but no ETA on this. At the very least, I'd like to remove the necessity of building another complete copy of LLVM+Clang for CBUILD to compile #Clang.

The #LLVM developer room is back for the 11th consecutive year at #FOSDEM, on February 1st!
We're looking for presentation proposals on all aspects of #LLVM and #MLIR.
See the CFP for details https://discourse.llvm.org/t/cfp-fosdem-2025-llvm-dev-room/82779.
If you've got ideas, but you're not fully sure if they would be a good fit for the room, I'm more than happy to help you refine your abstract. Please don't be shy to reach out here or through one of my office hours https://llvm.org/docs/GettingInvolved.html#office-hours.
I hope to see many of you at FOSDEM!

#MLIR is ________

Here’s my #introduction.

At AMD, I lead a small team focused on high-performance 3D Graphics and #machinelearning.

I’m interested in Neural Radiance Fields, real-time super-resolution, generative models like #stablediffusion and text->3D, #gamedev, etc.

We rely on open-source software like #julialang, #pytorch, #fastai, #mlir, IREE, etc.

I live near Toronto with my wife and 2 kids.