Stories by matt_d

Inside VOLT: Designing an Open-Source GPU Compiler

An MLIR Pipeline for Offloading Fortran to FPGAs via OpenMP

Inside Nvidia GPU: Blackwell's Limitations & Future Rubin's Microarchitecture

Kitsune: Enabling Dataflow Execution on GPUs with Spatial Pipelines

DMA Collectives for Efficient ML Communication Offloads

10 Myths of Scalable Parallel Languages Part 8: Striving Toward Adoptability

Slicing Is All You Need: Towards a Universal One-Sided Distributed MatMul

Machine Scheduler in LLVM – Part II

The content-addressed storage (CAS) model of incremental build systems

Defeating the Training-Inference Mismatch via FP16

Opportunistically Parallel Lambda Calculus

Place Capability Graphs: A General-Purpose Model of Rust's Ownership & Borrowing

Linear effects, exceptions, resources: Curry-Howard destructors correspondence

Making the Clang AST Leaner and Faster

Draw high dimensional tensors as a matrix of matrices

Wafer-Scale AI Compute: A System Software Perspective

Towards Automated GPU Kernel Generation

Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs

Triton Developer Conference 2025 Talks [video]

OpenEstimate Evaluating LLMs on Reasoning Under Uncertainty with Real-World Data

8 9 10 11 12 13 14 15 16 17