Parallel Pixelation on CPU and GPU

由衷地感谢滕老师的指导与帮助。

This is a project for learning parallel programming, which realizes pixelation of .wkt data on CPU and GPU platform.

1. On the CPU platform, this project realizes the parallel programming of pThread and OpenMP.
2. On the CPU platform,this project realizes the parallel programming of CUDA.

Requirements

• CMake>=3.22
• Eigen3
• CUDA Toolkit

Installation

1. Download from repo

$ git clone --recursive https://github.com/KMushaL/ParallelPixelization.git

2. Build

---

Configure Options

• USE_OPENMP
  • This option is enabled by default. It enables OpenMP instead of pThread.
• USE_CUDA
  • This option is enabled by default. If enables CUDA and to further generate the corresponding executable file.

---

Then run the follow shell to complete build.

$ cmake -S . -B [-D <option>=<value>]
$ cd build
$ make

3. Run

There are some params can be passed when run the executable file.

CPU

---

Params

• in

  • Specify the path to the .wkt file, If no value is passed, the bigpolygon.wkt file in the res directory will be used.
  • e.g. --in ${path to your file}

• ps

  • A double value specifying the pixel size, the default value is 10.
  • e.g. --ps 1

• th

  • An integer value specifying the number of threads to run, the default value is 16.
  • e.g. --th 8

---

Then run the follow shell.

$ ./CPU_PIXEL.out [--<param> <value>]

GPU

---

Params

• in

  • Specify the path to the .wkt file, If no value is passed, the bigpolygon.wkt file in the res directory will be used.
  • e.g. --in ${path to your file}

• ps

  • A double value specifying the pixel size, the default value is 1.
  • e.g. --ps 1

---

Then run the follow shell.

$ ./GPU_PIXEL.out [--<param> <value>]

Test Results

There are some test results on my laptop, the configuration is as follows:

CPU: Intel 11th Gen Core i7-11800H

GPU: Nvidia Geforce RTX 3060(Laptop)

---

1. pThread vs OpenMP (pixel_size=1)

---

2. OpenMP vs CUDA (pixel_size=0.05)

   OpenMP(16 threads): 17715.5s

   CUDA: 19.9403s

---

3. Different size of block on CUDA

   1. pixel_size=0.05

      

   2. pixel_size=0.01

      

---