pc-second-assignment

name: Giovanni Santini

mat_id: 235441

Index

• About
• Dependencies
• Build and run on the cluster
• Building manually
• Running manually
• Additional Information
  • Analyze data

About

This repository contains the code for the first assignment of the course, as well as the source for the report in latex and all the necessary files to run the benchmarks on the cluster.

There is a high level view of the structure of this repo:

• include/: project's headers
• src/: algorithms implementation
• benchmarks/: all benchmarks
  • plotting/: python code to generate plots
• cluster/: pbs files to run on the cluster
• tests/: testing the algorithms
• fuzz/: fuzzing the algorithms
• latex/: report source code

Dependencies

The code is written in C++20 and it is tested with gcc-8.5.0, 9.1.0, 14.1.0 and 14.2.0 on a linux 6.10 and 6.12 and freeBSD 14.1 machine. Alternative/previous compilers or different operating systems are not guaranteed to work.

This project depends on two libraries: valfuzz and tenno.

• Valfuzz is used to run the benchmarks. The benchmarks are registered through a macro and are executed multile times, logging vaious information. To learn more please visit the library's github page.
• Tenno implements is a superset of the C++23 (and the upcoming 26) standard library and It provides useful data structures and functionalities to aid the developement of the benchmarks. An example of a feature used is compile time pseudo random number generation to initialize the input data.

Both dependencies are automatically downloaded by the build system of your choice, thou cmake is primarly supported and Its use is advised.

Build and run on the cluster

The complete .pbs file can be found in cluster/first-assignment/first-assignment.pbs. This file builds the binaries and executes them. To submit the script for execution, run the following on the cluster:

git clone https://github.com/San7o/parallel-computing-cpp.git &&
git checkout second-assignment &&
export PC_PBS_FILE=$HOME/parallel-computing-cpp/cluster/second-assignment/second-assignment.pbs &&
chmod +x $PC_PBS_FILE &&
qsub $PC_PBS_FILE

the script will generate reports in $HOME/parallel-computing-cpp/benchamrks/plotting/reports/ containing collected data in csv format.

Building manually

This project primarly supports building with cmake >= 3.15.4. If you are on the cluster, please load the necessary modules first:

module load gcc91 &&
module load cmake-3.15.4 &&
module load mpich-3.2.1--gcc-9.1.0

To build the full benchmark suite, run the following command:

git clone https://github.com/San7o/parallel-computing-cpp.git &&
cd parallel-computing-cpp &&
git checkout second-assignment &&
cmake -Bbuild \
	  -D PC_BUILD_OPTIMIZED_O1=ON \
	  -D PC_BUILD_OPTIMIZED_O2=ON \
	  -D PC_BUILD_OPTIMIZED_O3=ON &&
cmake --build build \
      -j $(nproc)

Description of the arguments:

• PC_BUILD_OPTIMIZED_O{1,2,3}=ON: build a new target with specified optimization level
• --build build: output in the "build" directry
• -j $(nproc): build with maximum number of threads. nproc is part of the gnu utils, if you don't have access to the program you can manually select a number or remove this argument completly.

Additional information:

Some options are enabled by default such as PC_BUILD_OMP for omp support. You can build with clang by enabling PC_USE_CLANG but the project is not fully compatible with clang yet.

Running manually

The project depends on valFuzz which provides testing, fuzzing and benchmarking functionalities. To run the benchmarks, use the --benchmark flag. To get more informations about possible flags, please consult the help message with --help. The benchmarks are composed of a master process and worker processes, which will be build by default. You can run a benchmark like this:

module load gcc91 &&
module load cmake-3.15.4 &&
module load mpich-3.21--gcc-9.1.0 &&
export NUM_WORKERS=3 &&
export NUM_ITERATIONS=10 &&
export BUILD_DIR=build &&
mpirun -np 1 \
      ./$BUILD_DIR/tests \
      --benchmark \
      --num-iterations $NUM_ITERATIONS \
      --no-multithread \
      --report $OUTPUT_DIR/base_np4.txt \
		--reporter csv \
      : -np $NUM_WORKERS ./$BUILD_DIR/worker &&
unset NUM_WORKERS NUM_ITERATIONS BUILD_DIR

You could also run the full benchmarks by running the .pbs script:

chmod +x cluster/second-assignment/second-assignment.pbs &&
./cluster/second-assignment/second-assigmnet.pbs

There is a test .bps in the same directory to test the correctness of the alogrithms.

To collect information about the system machine, you can use the get-info.sh script:

./get-info.sh

Additional Information

Analyze Data

All the graphs found in the report are generated using seaborn in python. All the code can be found in the benchmarks/plotting directory.