Skip to content

Latest commit

 

History

History
92 lines (69 loc) · 3.58 KB

README.md

File metadata and controls

92 lines (69 loc) · 3.58 KB
Raw

gem5 skylake config

This repo contains experiments to study performance of gem5 against a real hardware. The experiments are based on gem5art framework.

Repo structure

  • gem5-configs: contains python scripts that configure for skylake micro-architecture in gem5.
    • system/core.py: contains the below CPU classes.
      • UnCalibCPU: configured based on the documentation available for the micro-architecture.
      • CalibCPU: considering the absence of certain features in gem5, for example, absence micro-op fusion in gem5, this configuration offsets such deficiencies by improving some other parameters.
      • MaxCPU: models a CPU with maximum pipeline widths, minimum pipeline and instruction latencies.
    • system/caches.py: classic cache model.
    • system/MESI_Two_Level.py: Ruby cache system for MESI protocol.
    • system/se.py: sets a system to run experiments in SE mode.
    • system/fs.py: sets a system to run experiments in FS mode.
    • run-se.py: SE mode run script.
    • run-spec-fs.py: FS mode run script.
  • results: contains the experiment results.
  • launch_microbm_experiment.py: script to launch the microbenchmark experiments.
  • launch_spec2006_experiment.py: script to launch the spec2006 experiments.

Running experiments

Note: The following steps describe how to setup and run experiments with microbenchmarks, refer this tutorial for running experiments with spec2006.

Step 1 - Clone this repo

git clone https://github.com/darchr/gem5-skylake-config.git
cd gem5-skylake-config

Step 2 - Clone and build gem5

git clone https://gem5.googlesource.com/public/gem5
cd gem5
scons build/X86/gem5.opt -j8

Step 3 - Clone and build microbenchmarks

git clone https://github.com/darchr/microbench.git
# Checkout adjust-ters branch, this branch contains benchmarks with different iteration to capture complete behavior of sim-objects in gem5.
cd microbench
git fetch --all
git checkout adjust-iters
# Build the benchmarks
make
cd ..

Step 4 - Setup the environment

# Create a virtual python3 environment before using gem5art
virtualenv -p python3 venv
source venv/bin/activate

# gem5art can be installed using pip
pip install gem5art-artifact gem5art-run gem5art-tasks

Step 5 - Create a database

docker run -p 27017:27017 -v <absolute path to the created directory>:/data/db --name mongo-<some tag> -d mongo

Step 6 - launching the experiment

# To launch experiment for the Calib config and control flow benchmarks
python launch_experiment.py Calib ctrl
# To launch experiment for the all configs and all benchmarks
python launch_experiment.py all all

References

These are primary references for skylake micro-architecture.

  1. Agner Fog, Microarchitecture of Intel and CPUs
  2. Agner Fog, Instruction tables
  3. Wikichip, Skylake Microarchitecture(server)
  4. Intel, Architecture Optimization Manual

These are for configuring a micro-architecture in gem5

  1. gem5 haswell config
  2. A Study on the Impact of Instruction Set Architectures on Processor's Performance