LLVM21 is a stinker and has worse codegen in general, and a bug forced disabled loop auto-vectorization. Turned out to be a blessing: we clearly identified areas we were over-reliant on implicit compiler optimizations and were able to explicitly restructure our code to what we really wanted. The result is that we now have clean generic SIMD subroutines in places we previously relied on auto-vectorization. And we did it better and as a result produced faster code (see below). Pure ASCII throughput improved more than 20% (30% on Linux) which is insane and is purely because we wrote better SIMD than an auto-vectorizer can. Lit. We also found it stopped inlining automatically in certain places which destroyed some benchmarks based on real world corpuses. As a result, we now explicitly inline those backed by benchmarks. Wonderful. Some things slowed down more than can be attributed to noise. I'm looking into that now but they're minor benchmarks. The important ones are parity or better.

Congratulations to @clattner_llvm, @iamtimdavis, and the entire @Modular team on today's announcement that Qualcomm has agreed to acquire the company for ~$4 billion. We at @GreylockVC have been fortunate to back Chris, Tim, and the team since their seed round, and it's been a privilege to partner with them from the start. Chris and Tim are an incredibly rare combination of world-class systems thinking and product taste, that can take deep technical complexity and make it highly usable for millions of developers. When they started Modular a little over four years ago, they had a prescient thesis that AI needed an open platform to runs models at state-of-the-art performance across heterogeneous hardware, without requiring developers to rewrite anything. They delivered it with Mojo and MAX, and now they'll accelerate that mission by pairing their software with Qualcomm's silicon leadership. We're grateful to have invested alongside @davemuni @GVteam, @deepnishar @generalcatalyst, and Chris Ré, and we're excited for this important milestone for Modular.