Parallel convolution realization (task 2)

.github/workflows/dotnet.yml

@@ -0,0 +1,30 @@
+# This workflow will build a .NET project
+# For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-net
+
+name: .NET
+
+on:
+  push:
+    branches: [ "main" ]
+  pull_request:
+    branches: [ "main" ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+    - name: Setup .NET
+      uses: actions/setup-dotnet@v4
+      with:
+        dotnet-version: 8.0.x
+    - name: Install OpenCV
+      run: sudo apt-get update && sudo apt-get install -y libjpeg-dev libpng-dev libtiff-dev libavcodec-dev libavformat-dev libswscale-dev libv4l-dev libxvidcore-dev libx264-dev libopencv-dev
+    - name: Restore dependencies
+      run: dotnet restore
+    - name: Build
+      run: dotnet build --no-restore
+    - name: Test
+      run: dotnet test --no-build --verbosity normal

.gitignore

@@ -17,6 +17,7 @@ bld/
 # .NET Core
 project.lock.json
 project.fragment.lock.json
+BenchmarkDotNet.Artifacts
 
 # ASP.NET Scaffolding
 ScaffoldingReadMe.txt

Convolver.sln

@@ -7,6 +7,16 @@ Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "src", "src", "{E7170ACE-9AA
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "BMPConvolver.Core", "src\BMPConvolver.Core\BMPConvolver.Core.csproj", "{523C0934-7353-45C5-AB89-DFB54C98071F}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "BMPConvolver.Cli", "src\BMPConvolver.Cli\BMPConvolver.Cli.csproj", "{A99E9047-15E4-492A-8C77-B2C82CAF0B0B}"
+EndProject
+Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "test", "test", "{5F4B7EAD-0AB8-4FBF-85F1-EA0BEE09832F}"
+EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "BMPConvolver.Tests", "test\BMPConvolver.Tests\BMPConvolver.Tests.csproj", "{69A51A1D-CE2E-4C75-AD2E-64ADC5F8E593}"
+EndProject
+Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "bench", "bench", "{2507B97A-9F2C-4A9C-A7A3-58541DDFA976}"
+EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "BMPConvolver.Benchmarks", "bench\BMPConvolver.Benchmarks\BMPConvolver.Benchmarks.csproj", "{4E6BFCD2-D954-4FB9-AC30-1024FA7A9A3C}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Any CPU = Debug|Any CPU
@@ -35,5 +45,8 @@ Global
 	EndGlobalSection
 	GlobalSection(NestedProjects) = preSolution
 		{523C0934-7353-45C5-AB89-DFB54C98071F} = {E7170ACE-9AAD-4689-ADC6-22E048D93B80}
+		{A99E9047-15E4-492A-8C77-B2C82CAF0B0B} = {E7170ACE-9AAD-4689-ADC6-22E048D93B80}
+		{69A51A1D-CE2E-4C75-AD2E-64ADC5F8E593} = {5F4B7EAD-0AB8-4FBF-85F1-EA0BEE09832F}
+		{4E6BFCD2-D954-4FB9-AC30-1024FA7A9A3C} = {2507B97A-9F2C-4A9C-A7A3-58541DDFA976}
 	EndGlobalSection
 EndGlobal

README.md

@@ -1 +1,172 @@
+# Запуск
 
+Для запуска приложения выполните следующую команду:
+
+```bash
+dotnet run --project src/BMPConvolver.Cli/BMPConvolver.Cli.csproj <input.bmp> <output.bmp> [опции]
+```
+
+Опции включают:
+- `--mode seq|par`: режим выполнения (последовательный или параллельный)
+- `--partition pixels|rows|cols|grid`: способ разделения работы
+- `--grid <block width>x<block height>`: размер сетки для grid partition
+- `--border zero|clamp`: режим обработки границ
+- `--kernel box3|sharpen|identity`: предустановленный kernel
+- `--kernel-text "..."`: kernel в текстовом формате
+- `--kernel-file path.txt`: kernel из файла
+
+## Тесты
+
+Для запуска тестов:
+
+```bash
+dotnet test
+```
+
+## Бенчмарки
+
+Для запуска бенчмарков:
+
+```bash
+dotnet run --project bench/BMPConvolver.Benchmarks/BMPConvolver.Benchmarks.csproj
+```
+
+# Анализ производительнсти
+- **Зависимость от ширины и высоты**: При фиксированной высоте увеличение ширины приводит к росту времени в 4-5 раз, тогда как при фиксированной ширине увеличение высоты дает рост в 2-3 раза. Такая разнца связана с особенностями кэширования памяти и последовательным доступом к данным в массивах.
+- **Зависимость от способа параллелизма**: 
+  - **Sequential**: Базовый последовательный метод, без параллелизма. Примерно в 5 раз медленнее самого эффективного параллельного метода
+  - **Parallel.Pixels**: Параллелизм по каждому пикселю создает огромное количество задач, что приводит к высоким накладным расходам на синхронизацию. Самый медленный метод с высокой дисперсией.
+  - **Parallel.Rows**: Параллелизм по строкам эффективен для изображений с большим количеством строк. Разделяет изображение на горизонтальные полосы, минимизируя конфликты памяти. Часто быстрее Parallel.Columns.
+  - **Parallel.Columns**: Параллелизм по столбцам показывает себя лучше на изображениях с большим количеством столбцов. Однако он может страдать от неравномерности нагрузки при широких изображениях.
+  - **Parallel.Grid**: Наиболее эффективный метод, разделяющий изображение на прямоугольные блоки (сетку). Балансирует нагрузку между потоками, минимизирует накладные расходы и обеспечивает лучшую масштабируемость. Часто в 2-3 раза быстрее Sequential и в 10+ раз быстрее Parallel.Pixels.
+
+## Border mode: Zero
+### Диаграмма
+![Столбчатая_диаграмма_Zero](public/img/Border_mode_Zero_columnar.png)
+
+### График
+![График_Zero](public/img/Border_mode_Zero_Graphics.png)
+
+### Таблица
+| Method           | Width | Height | Mean       | Error      |
+|----------------- |------ |------- |-----------:|-----------:|
+| Sequential       | 1024  | 1024   |  11.347 ms |  0.1680 ms |
+| Parallel.Pixels  | 1024  | 1024   |  24.677 ms |  1.3195 ms |
+| Parallel.Rows    | 1024  | 1024   |   2.510 ms |  0.0188 ms |
+| Parallel.Columns | 1024  | 1024   |   2.313 ms |  0.0259 ms |
+| Parallel.Grid    | 1024  | 1024   |   2.247 ms |  0.0162 ms |
+|                  |       |        |            |            |
+| Sequential       | 1024  | 2048   |  22.219 ms |  0.0945 ms |
+| Parallel.Pixels  | 1024  | 2048   |  51.281 ms |  5.7572 ms |
+| Parallel.Rows    | 1024  | 2048   |   4.961 ms |  0.0466 ms |
+| Parallel.Columns | 1024  | 2048   |   5.252 ms |  0.0444 ms |
+| Parallel.Grid    | 1024  | 2048   |   4.517 ms |  0.0601 ms |
+|                  |       |        |            |            |
+| Sequential       | 1024  | 4096   |  48.451 ms |  0.1076 ms |
+| Parallel.Pixels  | 1024  | 4096   | 110.977 ms | 19.1888 ms |
+| Parallel.Rows    | 1024  | 4096   |   9.724 ms |  0.0335 ms |
+| Parallel.Columns | 1024  | 4096   |  10.589 ms |  0.0641 ms |
+| Parallel.Grid    | 1024  | 4096   |   8.937 ms |  0.0491 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 1024   |  24.379 ms |  0.1265 ms |
+| Parallel.Pixels  | 2048  | 1024   |  49.390 ms |  6.5355 ms |
+| Parallel.Rows    | 2048  | 1024   |   4.982 ms |  0.0440 ms |
+| Parallel.Columns | 2048  | 1024   |   6.388 ms |  0.1713 ms |
+| Parallel.Grid    | 2048  | 1024   |   5.264 ms |  0.1405 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 2048   |  48.880 ms |  0.2759 ms |
+| Parallel.Pixels  | 2048  | 2048   | 102.091 ms |  8.1502 ms |
+| Parallel.Rows    | 2048  | 2048   |   9.650 ms |  0.0547 ms |
+| Parallel.Columns | 2048  | 2048   |   9.829 ms |  0.1386 ms |
+| Parallel.Grid    | 2048  | 2048   |   9.884 ms |  0.0655 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 4096   |  98.503 ms |  0.8038 ms |
+| Parallel.Pixels  | 2048  | 4096   | 236.901 ms |  4.6947 ms |
+| Parallel.Rows    | 2048  | 4096   |  18.314 ms |  0.2121 ms |
+| Parallel.Columns | 2048  | 4096   |  20.790 ms |  0.1510 ms |
+| Parallel.Grid    | 2048  | 4096   |  19.169 ms |  0.1776 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 1024   |  48.863 ms |  0.2360 ms |
+| Parallel.Pixels  | 4096  | 1024   | 103.113 ms | 11.9344 ms |
+| Parallel.Rows    | 4096  | 1024   |   8.819 ms |  0.0642 ms |
+| Parallel.Columns | 4096  | 1024   |   9.795 ms |  0.0708 ms |
+| Parallel.Grid    | 4096  | 1024   |   9.115 ms |  0.2993 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 2048   |  96.191 ms |  3.7277 ms |
+| Parallel.Pixels  | 4096  | 2048   | 223.957 ms | 24.3798 ms |
+| Parallel.Rows    | 4096  | 2048   |  16.587 ms |  0.0625 ms |
+| Parallel.Columns | 4096  | 2048   |  22.102 ms |  0.0916 ms |
+| Parallel.Grid    | 4096  | 2048   |  20.067 ms |  0.2590 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 4096   | 203.417 ms |  3.6494 ms |
+| Parallel.Pixels  | 4096  | 4096   | 483.104 ms |  3.5202 ms |
+| Parallel.Rows    | 4096  | 4096   |  37.907 ms |  0.3498 ms |
+| Parallel.Columns | 4096  | 4096   |  45.954 ms |  0.4596 ms |
+| Parallel.Grid    | 4096  | 4096   |  38.499 ms |  0.2745 ms |
+
+
+## Border mode: Clamp
+
+### Диаграмма
+![Столбчатая_диаграмма_Clamp](public/img/Border_mode_Clamp_columnar.png)
+
+### График
+![График_Clamp](public/img/Border_mode_Clamp_Graphics.png)
+
+### Таблица
+
+| Method           | Width | Height | Mean       | Error      |
+|----------------- |------ |------- |-----------:|-----------:|
+| Sequential       | 1024  | 1024   |  10.899 ms |  0.0522 ms |
+| Parallel.Pixels  | 1024  | 1024   |  25.133 ms |  1.8560 ms |
+| Parallel.Rows    | 1024  | 1024   |   2.423 ms |  0.0177 ms |
+| Parallel.Columns | 1024  | 1024   |   2.210 ms |  0.0153 ms |
+| Parallel.Grid    | 1024  | 1024   |   2.339 ms |  0.0334 ms |
+|                  |       |        |            |            |
+| Sequential       | 1024  | 2048   |  23.711 ms |  0.1069 ms |
+| Parallel.Pixels  | 1024  | 2048   |  49.684 ms |  5.6650 ms |
+| Parallel.Rows    | 1024  | 2048   |   4.385 ms |  0.0375 ms |
+| Parallel.Columns | 1024  | 2048   |   4.687 ms |  0.0334 ms |
+| Parallel.Grid    | 1024  | 2048   |   4.933 ms |  0.0552 ms |
+|                  |       |        |            |            |
+| Sequential       | 1024  | 4096   |  43.248 ms |  0.2635 ms |
+| Parallel.Pixels  | 1024  | 4096   | 102.942 ms | 13.7667 ms |
+| Parallel.Rows    | 1024  | 4096   |   9.432 ms |  0.0324 ms |
+| Parallel.Columns | 1024  | 4096   |  10.499 ms |  0.1717 ms |
+| Parallel.Grid    | 1024  | 4096   |   9.198 ms |  0.0589 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 1024   |  22.027 ms |  0.1667 ms |
+| Parallel.Pixels  | 2048  | 1024   |  47.339 ms |  5.4978 ms |
+| Parallel.Rows    | 2048  | 1024   |   4.699 ms |  0.1312 ms |
+| Parallel.Columns | 2048  | 1024   |   6.117 ms |  0.1464 ms |
+| Parallel.Grid    | 2048  | 1024   |   5.095 ms |  0.0555 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 2048   |  47.986 ms |  0.2603 ms |
+| Parallel.Pixels  | 2048  | 2048   |  97.947 ms |  9.1101 ms |
+| Parallel.Rows    | 2048  | 2048   |   8.399 ms |  0.1387 ms |
+| Parallel.Columns | 2048  | 2048   |   9.672 ms |  0.1301 ms |
+| Parallel.Grid    | 2048  | 2048   |   9.903 ms |  0.2243 ms |
+|                  |       |        |            |            |
+| Sequential       | 2048  | 4096   |  90.954 ms |  6.1812 ms |
+| Parallel.Pixels  | 2048  | 4096   | 233.673 ms | 26.1762 ms |
+| Parallel.Rows    | 2048  | 4096   |  20.059 ms |  0.3383 ms |
+| Parallel.Columns | 2048  | 4096   |  21.500 ms |  0.2026 ms |
+| Parallel.Grid    | 2048  | 4096   |  20.564 ms |  0.3094 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 1024   |  43.379 ms |  0.2542 ms |
+| Parallel.Pixels  | 4096  | 1024   | 111.441 ms | 20.5576 ms |
+| Parallel.Rows    | 4096  | 1024   |   8.383 ms |  0.0320 ms |
+| Parallel.Columns | 4096  | 1024   |   9.556 ms |  0.0487 ms |
+| Parallel.Grid    | 4096  | 1024   |  10.015 ms |  0.0773 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 2048   |  98.364 ms |  1.2503 ms |
+| Parallel.Pixels  | 4096  | 2048   | 213.736 ms | 31.6862 ms |
+| Parallel.Rows    | 4096  | 2048   |  17.360 ms |  0.0749 ms |
+| Parallel.Columns | 4096  | 2048   |  22.656 ms |  0.1534 ms |
+| Parallel.Grid    | 4096  | 2048   |  18.816 ms |  0.2907 ms |
+|                  |       |        |            |            |
+| Sequential       | 4096  | 4096   | 190.988 ms | 16.7715 ms |
+| Parallel.Pixels  | 4096  | 4096   | 470.957 ms |  4.9517 ms |
+| Parallel.Rows    | 4096  | 4096   |  35.611 ms |  0.3819 ms |
+| Parallel.Columns | 4096  | 4096   |  47.020 ms |  0.2583 ms |
+| Parallel.Grid    | 4096  | 4096   |  35.363 ms |  0.5478 ms |

bench/BMPConvolver.Benchmarks/BMPConvolver.Benchmarks.csproj

@@ -0,0 +1,18 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <ItemGroup>
+    <ProjectReference Include="..\..\src\BMPConvolver.Core\BMPConvolver.Core.csproj" />
+  </ItemGroup>
+
+  <ItemGroup>
+    <PackageReference Include="BenchmarkDotNet" Version="0.15.8" />
+  </ItemGroup>
+
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <TargetFramework>net8.0</TargetFramework>
+    <ImplicitUsings>enable</ImplicitUsings>
+    <Nullable>enable</Nullable>
+  </PropertyGroup>
+
+</Project>

bench/BMPConvolver.Benchmarks/ConvolverBench.cs

@@ -0,0 +1,87 @@
+using BenchmarkDotNet.Attributes;
+using BenchmarkDotNet.Columns;
+using BenchmarkDotNet.Configs;
+using BenchmarkDotNet.Jobs;
+using BMPConvolver.Core;
+using BMPConvolver.Core.WorkPartitioning;
+
+namespace BMPConvolver.Benchmarks;
+
+[Config(typeof(Config))]
+[HideColumns(Column.RatioSD, Column.AllocRatio, Column.Gen0, Column.Gen1, Column.Gen2)]
+public class ConvolutionBench
+{
+    private sealed class Config : ManualConfig
+    {
+        public Config()
+        {
+            AddJob(Job.Default.WithWarmupCount(3).WithIterationCount(15));
+        }
+    }
+    private const int Seed = 1;
+
+    [Params(1024, 2048, 4096)]
+    public int Width { get; set; }
+
+    [Params(1024, 2048, 4096)]
+    public int Height { get; set; }
+
+    [Params(BorderMode.Zero, BorderMode.Clamp)]
+    public BorderMode Border { get; set; }
+
+    public int KernelSize { get; set; } = 3;
+
+    public int Grid { get; set; } = 8;
+
+    private GrayImage _input = null!;
+    private Kernel _kernel = null!;
+
+    [GlobalSetup]
+    public void Setup()
+    {
+        _input = RandomImage(Width, Height);
+        _kernel = RandomKernelOddFrom(KernelSize, KernelSize);
+    }
+
+    [Benchmark(Baseline = true, Description = "Sequential")]
+    public GrayImage Sequential()
+        => Convolver.ConvolveSequential(_input, _kernel, Border);
+
+    [Benchmark(Description = "Parallel.Pixels")]
+    public GrayImage ParallelByPixels()
+        => Convolver.ConvolveParallel(_input, _kernel, Border, PartitioningMode.Pixels, gridX: Grid, gridY: Grid);
+
+    [Benchmark(Description = "Parallel.Rows")]
+    public GrayImage ParallelByRows()
+        => Convolver.ConvolveParallel(_input, _kernel, Border, PartitioningMode.Rows, gridX: Grid, gridY: Grid);
+
+    [Benchmark(Description = "Parallel.Columns")]
+    public GrayImage ParallelByColumns()
+        => Convolver.ConvolveParallel(_input, _kernel, Border, PartitioningMode.Columns, gridX: Grid, gridY: Grid);
+
+    [Benchmark(Description = "Parallel.Grid")]
+    public GrayImage ParallelByGrid()
+        => Convolver.ConvolveParallel(_input, _kernel, Border, PartitioningMode.Grid, gridX: Grid, gridY: Grid);
+
+    private static GrayImage RandomImage(int width, int height)
+    {
+        var rnd = new Random(Seed);
+        var pixels = new float[width * height];
+        for (var i = 0; i < pixels.Length; i++)
+            pixels[i] = (float)rnd.NextDouble();
+        return new GrayImage(width, height, pixels);
+    }
+
+    private static Kernel RandomKernelOddFrom(int widthRequested, int heightRequested)
+    {
+        var width = widthRequested <= 1 ? 1 : (widthRequested % 2 == 1 ? widthRequested : widthRequested + 1);
+        var height = heightRequested <= 1 ? 1 : (heightRequested % 2 == 1 ? heightRequested : heightRequested + 1);
+
+        var rnd = new Random(Seed);
+        var weights = new float[width * height];
+        for (var i = 0; i < weights.Length; i++)
+            weights[i] = ((float)rnd.NextDouble() - 0.5f) * 0.5f;
+        return new Kernel(width, height, width / 2, height / 2, weights);
+    }
+}
+

bench/BMPConvolver.Benchmarks/Program.cs

@@ -0,0 +1,12 @@
+using BenchmarkDotNet.Running;
+
+namespace BMPConvolver.Benchmarks;
+
+public static class Program
+{
+    public static int Main(string[] args)
+    {
+        BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args);
+        return 0;
+    }
+}

src/BMPConvolver.Cli/BMPConvolver.Cli.csproj

@@ -0,0 +1,18 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <ItemGroup>
+    <ProjectReference Include="..\BMPConvolver.Core\BMPConvolver.Core.csproj" />
+  </ItemGroup>
+
+  <ItemGroup>
+    <PackageReference Include="SixLabors.ImageSharp" Version="3.1.12" />
+  </ItemGroup>
+
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <TargetFramework>net8.0</TargetFramework>
+    <ImplicitUsings>enable</ImplicitUsings>
+    <Nullable>enable</Nullable>
+  </PropertyGroup>
+
+</Project>