Reproduce “Multi-Dimensional Vector ISA Extension for Mobile In-Cache Computing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alireza Khadem, Daichi Fujiki, Hilbert Chen, Yufeng Gu, Nishil Talati, Scott Mahlke, Reetuparna Das
This paper introduces the Multi-dimensional Vector Extension (MVE), a new vector ISA tailored for mobile in-cache computing that efficiently handles multi-dimensional data patterns. MVE achieves an average 2.9x performance speedup and 8.8x energy reduction compared to Arm Neon SIMD extensions, and outperforms a mobile GPU for certain workloads.
Key insights
- Mobile workloads frequently exhibit multi-dimensional data access patterns that existing one-dimensional vector ISAs fail to efficiently exploit.
- Tailoring a vector ISA to multi-dimensional data patterns can drastically improve SIMD lane utilization and reduce data movement in in-cache computing.
- Integrating computation directly within the L2 cache (in-cache computing) significantly reduces energy consumption by mitigating the 'memory wall'.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.