Skip to content

Instruction Pointer Classifier and Dynamic Degree Stream based Hardware Cache Prefetching

Notifications You must be signed in to change notification settings

aniketp/ipcp-dsp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

32 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Hardware assisted Cache Prefetching

Hardware prefetchers play an important role in hiding long data access latency in various multiprocessor systems. Significant number of prefetchers have been proposed using state-of-the-art techniques which independently augment the existing instruction pointer based and cache address based prefetching.

In this paper, we attempt to categorically evaluate the submissions of the Third Data Prefetching Championship (DPC3). Additionally, we also propose IPCP++, an enhanced version of Instruction Pointer Classifier based prefetcher (IPCP) reinforced with IP-Delta based sequence predictor and IP-based stride prefetcher (Sangam++). The confluence of these IP based prefetching techniques achieves a speedup of 9.39% over no prefetching when averaged over 20 single-thread 6XX SPEC CPU 2017 traces.

We also present a dynamic degree stream prefetcher. The dynamic stream prefetcher (DSP) takes into account factors like prefetch accuracy, cache pollution and strength of the stream in a given window of 1000 cycles to determine the prefetch degree dynamically.

This research work is done as a part of the Course Project for Advanced Computer Architecture (CS622A), Fall Semester - 2019, instructed by Prof. Mainak Chaudhury.

The simulation results for all the prefetchers are compiled here. And the project paper can be found here.

Group Members (G16)

Name Email Roll
Aditya Rohan raditya@iitk.ac.in 160053
Aniket Pandey aniketp@iitk.ac.in 160113

Brief Build Instructions

  • Install champsim repository
 > cd scripts
 > ./install-champsim.sh
  • Install DPC3 traces (SPEC 2017)

Either install the traces from within champsim repository in the project or move already existing traces to champsim/dpc3_traces.

  • Build CPU executables and run simulation on all prefetchers for all available traces.
 > go run main.go

NOTE If you already have built the CPUs using above command and introduced new traces for which you want to re-run the simulation, use

 > go run main.go skip-cpu

This skips building the CPU executables again.

Acknowledgement

  • The Third Data Prefetching Championship (online)

  • Sangam: A multi-component core cache prefetcher (paper)

  • Bouquet of instruction pointers: Instruction pointer classifier based hardware prefetching (paper)

About

Instruction Pointer Classifier and Dynamic Degree Stream based Hardware Cache Prefetching

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages