使用 ILGPU 进行 GPU 加速运行 C# 代码时出现问题

问题描述 投票:0回答:1

我正在尝试使用 ILGPU 进行 GPU 处理来加速我的 C# 代码。但是,我遇到了以下错误:

错误 CS0117:“加速器”不包含“创建”的定义。

Accelerator accelerator = Accelerator.Create(...);

错误 CS1729:“Context”不包含采用 0 个参数的构造函数。

Context context = new Context();

我的目标是使用 ILGPU 将油画效果应用于位图图像,并且我遵循了 ILGPU 文档。但是,我似乎遗漏了一些东西或遇到兼容性问题。

这是我的代码的简化版本:

int valu = int.Parse(str[1]);
Bitmap resultImage = new Bitmap(cc);
BitmapData inputImageDa = resultImage.LockBits(new Rectangle(0, 0, width, height), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
BitmapData resultImageData = cc.LockBits(new Rectangle(0, 0, resultImage.Width, resultImage.Height), ImageLockMode.WriteOnly, PixelFormat.Format32bppArgb);
unsafe
{
    Parallel.For(0, height, y =>
    {
        byte* inputRow = (byte*)inputImageDa.Scan0 + (y * inputImageDa.Stride);
        byte* resultRow = (byte*)resultImageData.Scan0 + (y * resultImageData.Stride);
        Parallel.For(0, width, x =>
        {
            Color dominantColor = GetDominantColor(inputImageDa, x, y, valu);
            byte* resultPixel = resultRow + x * 4;
            resultPixel[0] = dominantColor.B;
            resultPixel[1] = dominantColor.G;
            resultPixel[2] = dominantColor.R;
            resultPixel[3] = 255;
        });
    });
}
cc.UnlockBits(resultImageData);

如果您能提供有关如何解决这些问题并使用 ILGPU 成功运行我的 C# 代码以进行 GPU 加速的指导或建议,我将不胜感激。

谢谢!

编辑:

抱歉,我忘了告诉大家,这是我尝试过的代码:

using System;
using System.Drawing;
using System.Drawing.Imaging;
using System.Linq;
using ILGPU;
using ILGPU.Runtime;

class Program
{
    static void Main()
    {
        // Replace with your image loading logic
        Bitmap originalImage = new Bitmap("your_image_path.jpg");
        int brushSize = 5; // Adjust the brush size as needed

        Bitmap resultImage = ApplyOilPaintingEffect(originalImage, brushSize);

        // Save or display the result image
        resultImage.Save("result_image.jpg");
    }

    static Bitmap ApplyOilPaintingEffect(Bitmap originalImage, int brushSize)
    {
        int width = originalImage.Width;
        int height = originalImage.Height;

        // Convert Bitmap to ILGPU memory buffer
        using (var context = new Context())
        {
            using (var accelerator = new Accelerator(context, AcceleratorType.Cuda)) // Adjust the AcceleratorType as needed
            {
                var inputBuffer = originalImage.ToBuffer2D(context, accelerator);
                var resultBuffer = accelerator.Allocate<Argb32>(width, height);

                // Process pixels on GPU
                using (var kernel = accelerator.LoadAutoGroupedStreamKernel<int, ArrayView<Argb32>, Index2>())
                {
                    kernel.DefaultNumIterations = inputBuffer.Length;

                    kernel.Body = (index, inputPixels, outputIndex) =>
                    {
                        var value = inputPixels[index];
                        var hist = new int[256 * 3];

                        // Collect histogram in the local neighborhood
                        for (int i = -brushSize; i <= brushSize; i++)
                        {
                            for (int j = -brushSize; j <= brushSize; j++)
                            {
                                int x = outputIndex.X + i;
                                int y = outputIndex.Y + j;

                                if (x >= 0 && x < width && y >= 0 && y < height)
                                {
                                    var neighborPixel = inputPixels[x, y];
                                    hist[neighborPixel.R]++;
                                    hist[256 + neighborPixel.G]++;
                                    hist[512 + neighborPixel.B]++;
                                }
                            }
                        }

                        // Find the dominant color
                        int maxIndex = hist.ToList().IndexOf(hist.Max());
                        int channel = maxIndex % 256;
                        int color = maxIndex / 256;

                        outputIndex = index / width;

                        // Set the dominant color to the output pixel
                        outputIndex.X = index - outputIndex.X * width;
                        var dominantColor = new Argb32((color == 0) ? channel : 0, (color == 1) ? channel : 0, (color == 2) ? channel : 0, 255);
                        inputPixels[index] = dominantColor;
                    };

                    kernel(inputBuffer.Length, inputBuffer, resultBuffer);
                }

                // Convert ILGPU memory buffer back to Bitmap
                Bitmap resultImage = resultBuffer.ToBitmap2D(context, accelerator);

                return resultImage;
            }
        }
    }
}
c# parallel-processing gpu gdi
1个回答
0
投票

对于您的问题,我发现ilgpu文档的这一页很有帮助。

简而言之,您可以通过命令

Context.Create(...)
Context.CreateDefault()
创建上下文,然后选择一个设备
d
,然后通过命令
d.CreateAccelerator(context)
创建加速器。

顺便说一句:由于文档仅涵盖选定的主题,因此我一般建议也浏览 ilgpu 项目的samples 文件夹

© www.soinside.com 2019 - 2024. All rights reserved.