Paper Reading

This repository contains notes on research papers, technical documents, and learning materials, as well as a record of my personal research path. My general interests include Computer Systems, Machine Learning, Computer Architecture, and Networking.

Research Path

• 26.1.9: Surveyed approximate nearest neighbor (ANN) methods. Concluded that this direction may not be suitable; the main bottleneck appears to be memory bandwidth.

• 26.1.16: Surveyed RISC-V extensions. Became particularly interested in compute-in-memory (CIM).

• 26.1.23: Read the RISC-V Vector Extension (RVV) specification.

• 26.1.31: Professor advised to look into Near-Memory Computation (NMC), which might help mitigate cache problems caused by long vectors. Observed that many optimizations happen at runtime rather than compile time. Research task this week: Survey Xuantie C910 to determine: Whether vector instructions are executed in-order or out-of-order. At what granularity instructions are executed in-order vs. out-of-order (e.g., instruction, micro-op, vector element).

• 26.3.6: TODO: Look into the micro-implementation of Ara.

• 26.3.13: Reporst on the micro-implementation, TODOs: Look into the detail implmentation of Ara (Simulator), wrtie benchmarks of simulators.

• 26.3.20: This week, write tests to determine whether vector chaining occurs in two specific scenarios: (1)Two nonsequential access instructions chaining. (2) Mask-generation instructions chain with instructions that consume the generated masks.

Reading List

RISC-V Vector Extension

• Saturn Microarchitecture Manual
• RVV Manual Book
• Saturn Microarchitecture Manual
• RVV Manual Book
• Arrow: A RISC-V Vector Accelerator for Machine Learning Inference
• K3
• Xuantie-910

HPC

• Communication-Avoiding General Matrix Multiplication within a single GPU

ANN

• HNSW paper

ML and AI

• Attention is All You Need
• Online Detection of Ai-generated Image