A multi-core processor design space exploration algorithm based on in-context RL

Intro

The parsec-tests2 and m5out folders are used to put the sum of gem5&mcpat evaluation functions.

Since some of the previous reinforcement learning methods for processor design space exploration are one-time, that is, each time the constraint parameters are set, and then start training a new model suitable for the current scene, whenever the design requirements under the new scene (new constraints) are needed, the design needs to be re-designed from scratch, so the time cost is very high, this project is committed to applying In-Context RL to DSE, so as to realize the new constraint parameters that can modify the design framework and quickly apply it to the new scene.

Quick Start

1. Prepare for simulation envirmoent

docker pull lucifercn22/dse-cpu:v2.0.0
docker run -it --net=host --name=dse lucifercn22/dse-cpu:v2.0.0 bash

2. Prepare python envirmoent

git clone https://github.com/xue-yun-liang/icrl.git
conda create -n icrl python==3.8
conda activate icrl
cd icrl & pip install requirement.txt

3. Install environment

The environment part of the repository itself is separated, and the installation of the environment needs to refer to this repository Then, if you see the picture below, it indicates that the environment for evaluating the function is not a problem and you can start other development.