Parallelizing Quadtree Image Decomposition with mixed approaches

Authors: Gabriele Masina and Andrea Stedile

Project required for the fulfillment of the Advanced Computing Architectures course at University of Trento, taught by professor R. Passerone in the 2021/2022 A.Y. We implemented and evaluated various approaches to the parallelization of the Quadtree image decomposition algorithm, such as CPU-based, CUDA-based and mixed approaches.

Compile

Locally:

mkdir build && cd build
conan install --build=missing ..
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build .

On the University of Trento's HPC cluster, do this instead:

module load gcc91
module load cuda-11.1
mkdir build && cd build
conan install --build=missing ..
/apps/cmake-3.20.3/bin -DCMAKE_TOOLCHAIN_FILE=../cmake/toolchain-hpc.cmake -DCMAKE_BUILD_TYPE=Release ..
/apps/cmake-3.20.3/bin --build .

Usage

Once the project is built, four apps are produced: qt-rec-td, qt-rec-bu, qt-it-gpu, qt-it-cpu.

To print the usage of an application, do, for example, ./qt-rec-td -h.

Usage: qt-rec [options] 

Optional arguments:
-h --help               shows help message and exits [default: false]
-v --version            prints version information and exits [default: false]
--input                 specify the input file [required]
--detail-threshold      specify the detail threshold [default: 13]
--no-output-file        suppress the production of the resulting image [default: false]
--csv                   writes timing information to csv file [default: false]