GPGPU-Sim Project Setup Guide

This README provides detailed instructions on setting up and running simulations using GPGPU-Sim, focusing on L2 cache management in GPU architectures. We have kept our code in Github at this link https://github.com/GPU-Project/GPU-L2-Cache-Management , If you want to go through this README.md that is well formatted, you can preview at https://github.com/GPU-Project/GPU-L2-Cache-Management , the instruction setup will be very easy as the commands to execute are well highlighted in README.md (this same file)

Prerequisites

• Ensure that you have atleast 3.5 GB of free space before running this project in CIMS CLUSTER.
• Using this README, we have successfully ran our code on CIMS cluster.
• All the Instructions should be executed in CIMS cluster

Setup and Execution Steps

Step 1: Build GPGPU-Sim Sandbox

Create a GPGPU-Sim sandbox environment using Apptainer (already available readily in CIMS). This step sets up an isolated environment for your GPGPU-Sim project.

apptainer build --sandbox gpgpu-sim-sandbox docker://socalucr/gpgpu-sim    

Expected time 10 mins and ignore all the warnings

Step 2: Enter the Sandbox Environment

Enter the sandbox environment using Apptainer:

apptainer exec --fakeroot --writable gpgpu-sim-sandbox /bin/bash  

Step 3: Navigate to Root Directory inside the sandbox container

Once inside the container:

cd /root/  

Step 4: Verify CUDA Installation Path

Check that the CUDA_INSTALL_PATH environment variable is correctly set to /usr/local/cuda:

echo $CUDA_INSTALL_PATH

Step 5: Set NVIDIA Compute SDK Location

Set the NVIDIA Compute SDK location:

export NVIDIA_COMPUTE_SDK_LOCATION=/root/NVIDIA_GPU_Computing_SDK/

Step 6: Verify NVIDIA Compute SDK Location

Confirm the setting of the NVIDIA_COMPUTE_SDK_LOCATION variable:

echo $NVIDIA_COMPUTE_SDK_LOCATION

Step 7: Clone GPU L2 Cache Management Repository

Clone the repository containing L2 cache management which contains our code:

git clone https://github.com/GPU-Project/GPU-L2-Cache-Management

Step 8: Copy Modified Source to GPGPU-Sim

Clone the repository containing L2 cache management which contains our code:

cp -r GPU-L2-Cache-Management/src gpgpu-sim_distribution/

Step 9: Copy Modified Benchmark SH file to /root

Clone the repository containing L2 cache management which contains our code:

cp -r GPU-L2-Cache-Management/run_benchmark.sh /root

Step 10: Navigate to GPGPU-Sim Distribution

Change directory to the GPGPU-Sim distribution:

cd gpgpu-sim_distribution/

Step 11: Setup Environment and Compile

Set up the environment and compile the simulator:

source setup_environment
make clean
make

Compilation takes approximately 10 minutes. Ignore warnings if they appear.

Step 12: Prepare Benchmark Environment

Prepare the environment for running benchmarks:

cd ..
mkdir test
cp -a ./gpgpu-sim_distribution/configs/GTX480/* ./test/
cd ispass2009-benchmarks/
make -f Makefile.ispass-2009
./setup_config.sh GTX480
cd ..
cd test/

The make process for benchmarks may take around 5 minutes and ignore warnings

Step 13: Test the Installation

Run a specific benchmark (e.g., RAY) and extract relevant metrics like cache miss rate:

../ispass2009-benchmarks/bin/release/RAY 4 4 | grep miss_rate

Now all the environment and benchmark for the project is setup and you should see Total cache miss rate

Step 14: Change the permissions for the run_benchmark file

Change the permissions for the run_benchmark file

cd /root
chmod +x run_benchmark.sh

Step 15: RUN Benchmark for a specific cache policy and benchmark

Change the permissions for the run_benchmark file

./run_benchmark.sh Q RAY

The Possible values for cache policies are L(Least Recently Used), F(First-In, First-Out), M(Most Recently Used), Q(Least Frequently Used), R(Random Replacement), S (Second Chance (or Clock) Page Replacement Policy), P(Last-In, First-Out)

The Possible values for Benchmark algorithms are CP (COULOMBIC POTENTIAL), MUM (MUMmerGPU), RAY(RAY), NQU(NQU), BFS(BFS)