Striving for SDR Performance Portability in the Era of Heterogeneous SoCs
summary
Claims mapping GNU Radio to heterogeneous SoCs and achieving near-portable speed: GNU Radio mapped to NVIDIA AMD and ARM GPUs plus NVIDIA Xavier Qualcomm Snapdragon and Xilinx Zynq, using a directive-based pipeline with OpenACC directives and an intelligent runtime scheduler. The stack includes GNR-tools for static and dynamic profiling and Aspen performance models with the openArc compiler targeting OpenCL. Initial ports show promising results and the goal is ~80% of expert-tuned performance across 5–10 architectures.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:33Recent Trends in Computing
- 2:49Fundamental Limits
- 5:30Predictions
- 7:50Hot Chips
- 8:47Mobile Market
- 10:43AMD GPU
- 11:51Memory Hierarchies
- 12:39Cosmic
- 15:11Performance Modeling
- 17:00Applications
- 17:50Profiles
- 20:17Aspen
- 22:52Compiler
- 25:31Runtime System
related talks
40mApplication and Challenges of Using GenAVB/TSN Software Stack for Industrial...
34mZephyr and OpenAMP
38mRemote Processor Framework in Zephyr RTOS
18mDistributed Embedded Systems Using Zephyr
41mGreybus for Robotics with Zephyr RTOS
42mAMP Virtio: A New Virto Transport for AMP Systems, with Focus on Zephyr, Linux, and Xen
52mAn Introduction to Asymmetric Multiprocessing: When this Architecture can be a Game Changer
18mPower Savings & NFV: How to reconcile these 2 constraints ?