Stories by matt_d

The Thing We All Obviously Want

ActPlane: Programmable OS-Level Policy Enforcement for Agent Harnesses

Fenwick trees for products mod 2ⁿ

A Fake Shell for Pangenomics

Making Equality Saturation Usable for Developing Vectorized Compilers

Reading AI Model Compilation in MLIR Through the Lens of Formal Theories

Liveness Proofs in Veil, Part I: The First Step

ParallelKernelBench: Can LLMs write fast multi-GPU kernels?

LXM: Better Splittable Pseudorandom Number Generators (and Almost as Fast) [video]

Code as Agent Harness

PICO: Performance Insights for Collective Operations

An Overview of Petri Net Theory [video]

2026 EuroLLVM Developers' Meeting Talks

VoltanaLLM: Energy-Efficient LLM Serving

Why Software Requirements Get Easier in an AI Economy

Lifting E-Graphs: A Function Isn't a Constant

Inference Compute Shapes Frontier LLM Evaluation

Concordia: JIT-Compiled Persistent-Kernel Checkpt for Fault-Tolerant Inference

SIMT-Step Execution: A Flexible Operational Semantics for GPU Subgroup Behavior

NektarIR: A Domain-Specific Compiler for High-Order FE Ops on Heterogeneous HW

1 2 3 4 5 6 7 8 9 10