HotGauge

A framework for characterizing hotspots in next-generation processors

[[TOC]]

Docker Setup

HotGauge can be tested and run inside a Docker container. Simply run ./docker_build.sh followed by ./docker_run.sh on a machine with Docker installed. Then follow the instructions below for "Subsequent Use" below.

Local Machine Setup

This codebase requires python 3 and was developed using python 3.4

Initial Setup

1. Clone this repository
2. Set up a virtual-environment
   1. Create the virtual-environment: python -m venv env
   2. Activate the virtual-environment
      • source env/bin/activate if using bash
      • source env/bin/activate.csh if using csh
      • source env/bin/activate.fish if using fish
   3. Update pip pip install --update pip
   4. Install required modules: pip install -r requirements.txt
3. Set up Sniper (the performance simulator)
   1. Clone the Sniper (the performance simulator) git repository by following the instructions on https://snipersim.org/w/Getting_Started
   2. Use cd sniper && git log to checkout the commit for version 6.1
   3. Apply the patch using patch_Sniper.sh
4. Set up McPAT (the power simulator)
   1. Download and patch McPAT (the power simulator) using get_and_patch_McPAT.sh
5. Set up 3d-ice (the thermal simulator)
   1. Download and patch 3d-ice (the thermal simulator) using get_and_patch_3DICE.sh
   2. Compile and test 3d-ice
      1. make sure compatible version of bison, flex, and superLU are installed. See compilation hints below for more details
      2. Compile the heatsink_plugin using make in ./3d-ice/heatsink_plugin/
      3. Compile 3D-ICE executables using make in ./3d-ice/
      4. Test 3D-ICE executables using make test
6. Try out the scripts in the ./examples/ directory. Running python simulation_with_warmup.py in the examples/ directory is a good way to test HotGauge. (hint: make sure you have your python virtual-environment activated!)
7. Utilize HotGauge as you see fit!

Subsequent Use

1. Activate the environment: source env/bin/activate

Sniper Compilation Hints

Requirments

• Python 2

3D-ICE Compilation Hints

Requirments

As per the documentation, 3D-ICE was developed using the following versions of its dependancies

• gcc 7.4.0
• bison 3.0.4
• flex 2.6.4
• blas 3.7.1
• SuperLU 4.3

The pluggable heat-sink interface---which is used by HotGauge---has its own set of dependancies:

• OpenModelica 1.16.0
• Pugixml 1.8.1
• Python 3 header files
• pkg-config

HotGauge Development Environment (RHEL 6.10)

HotGauge was developed and tested on a server running Red Hat Enterprise Linux Server release 6.10 (RHEL 6). The development machine had the following versions of 3D-ICE's dependancies:

• gcc 4.8.0
• bison 2.4.1
• flex 2.5.35
• blas 3.2.1
• SuperLU 4.3
• pkg-config 0.23

HotGauge Example Patch

This repository also comes with a second patch file that enables compilation on the development server. This patch must be applied after the base HotGauge patch file While this patch file may not be exactly the changes you will need to make to get 3D-ICE installed on your system, it is included with the hope that it will substantially help you overcome the compilation issues we encountered, should you also encounter them. The details of the changes made by this patch file are given below.

Local copy of OpenModelica

A script was created to download and compile a local copy of OpenModelica (omc). In order to get omc to compile, a few changes needed to be made, including the removal of calls to asciidoc, which was an incompatible version.

Download source code for pugixml

A script was created to download the source files for pugixml and add their compilation and inclusion to the required Makefile.

Disable installation of SuperLU

The development machine already has a compatible version of SuperLU, so the patch changes the SuperLU installation file to disable installation.

Disable heatsink-plugin python loader

The python loader for the heatsink-plugin also had to be disabled due to a python version issue.

Repository Contents

The following directories contain the code developed for use with this project. Each directory has a different purpose, as described below.

examples

This directory contains some example scripts that do various tasks such as generating floorplans as well as running thermal-simulations and doing subsequent analyis.

3d-ice

This is where the thermal simulator, 3d-ice, is compiled. HotGauge uses a custom version of 3d-ice, which is generated by apply the supplied patch file.

HotGauge

The python package for HotGauge. This includes scripts to run simulations and process the outputs.