See You at the 2026 US LLVM Developers' Meeting

Next week I will be at the 2026 US LLVM Developers' Meeting (October 26–28, Santa Clara, CA). It is my first time there, and I am really looking forward to finally meeting in person many of the people I have been talking to on Discourse, Discord, and in code reviews.

The poster. I will present a poster on A Multi-Dimensional, Per-Pass Empirical Study of the LLVM Optimization Pipeline (arXiv). In short, we ran every cumulative prefix of the -O3 pipeline on 30 PolyBench/C kernels and looked at runtime, binary size, hardware counters, and energy, pass by pass. The pipeline turns out to be non-monotone and back-loaded, and most of the speedup comes from a small core of passes. If you work on the pass pipeline, on performance tracking, or you simply disagree with the methodology, please come by. I would love to hear your feedback.

MLIR round table. I am also looking forward to participating in the MLIR round table. Most of my day-to-day work is on MLIR (see for example this deep dive into a linalg.matmul and this post on lowering MLIR to AArch64 with SME), so I am happy to chat about linalg, vectorization, and code generation for CPUs.

MLIR Meetup. On Thursday, October 29, right after the conference, I will also attend the first half of the morning of the MLIR Meetup 2026 at the Intel Customer Experience Center in Santa Clara.

Thanks. This trip would not have happened without a few people. A big thank you to Andrzej (GitHub, LinkedIn) for the mentorship, and to both Ege (GitHub, LinkedIn) and Andrzej for the support. Thanks also to the LLVM Foundation for fully funding my participation, and in particular to Tanya and everyone working with her for all the effort behind the scenes to make it possible.

Get in touch. If you are around and want to talk, just stop me in the hallway, or reach out beforehand on LLVM Discourse, by email, on X, LinkedIn, or Telegram.

See you in Santa Clara!