A Programming Model for Efficient Inter-Kernel Control-Flow on Memory-Mapped Near-Data(…)

on YouTube ↗10m

" summary

Arachne is a programming model that moves control flow onto a memory‑mapped NDP device by introducing a device_main routine and C++ NDP task classes, extending RISC‑V M2NDP and Google’s highway library for vector SPMD kernels. It reduces host‑device round trips, is evaluated on an M2NDP simulator, and reports up to 8.3x speedup on single‑source shortest path vs host‑driven launches while improving programmability compared with manual CUDA graph construction and exposing device_main and Arachne APIs.

AI-written summary · about

" chapters (8)

1. 0:00Introduction
2. 1:30CXL memory access challenges
3. 2:57Existing NDP model limits
4. 3:40Arachnne programming model
5. 5:25Unified NDP task abstraction
6. 6:28Highway CD vector programming
7. 7:30Performance evaluation
8. 8:15Q&A

" related talks

• 17mThe Real Challenge for RISC-V Vector ProcessorsJohn Min · RISC-V Forum: Vector & Machine Learning 2021
• 01h 21mPanel+ClosingUmit Ogras, William J. Dally, Axel Jantsch, Michael Kishinevsky · NOCS 2022
• 27mConsistency and Coherence of the NVIDIA Grace-Hopper SuperchipReese Levine, Tyler Sorensen, Vijay Nagarajan, Ryan Stutsman, Soham Bagchi, Sanya Srivastava · ISMM 2026
• 41mUsing Microarchitectural Design to Break KASLR and MoreAnders Fogh · RuhrSec 2017
• 19mNuclei System Technology Releases UX1030H with Full Support for RVA23Dr. Peng Chen · RISC-V Summit North America 2025
• 19mHow NOT To Program an Out-of-order Vector ProcessorDongjie Xie, Chip Kerchner · RISC-V Summit North America 2025
• 13mMemPool Flavors: Between Versatility and Specialization in a RISC-V Manycore ClusterLuca Benini, Sergio Mazzola, Yichao Zhang, Marco Bertuletti, Diyou Shen · RISC-V Summit Europe 2025
• 17mRISC-V is Ready for Powering the Era of Intelligent General ComputingCharlie Su · RISC-V Summit North America 2025