在 computeShader 代码运行后将数据存储在 GraphicsBuffer 中总是 (0,0,0)

Question

工作环境：Unity3D 2021.2.7.f1 Direct3D11

我尝试让拉普拉斯平滑在 GPU 上运行对于这种情况，我在其他 VertexBuffer（输入图形缓冲区）和 outVertexBuffer（输出 Grpahics 缓冲区）中设置了 Compute Shader，不幸的是，我在将数据存储到 GraphicsBuffer（存储顶点 Vector3）时遇到了一个奇怪的问题，我将其用作计算着色器的“输出” .

ComputeShader 组件的分配：

public void RunSmoothingShader(int kernel,int verticesNo)
{
    ComputeBuffer connectionsBuffer = new 
ComputeBuffer(_builder._triangleBudget*6,sizeof(int));

    ComputeBuffer offsetsBuffer = new ComputeBuffer(_builder._triangleBudget, sizeof(int));
    ComputeBuffer _counterBuffer = new ComputeBuffer(1, 4, ComputeBufferType.Counter);
    connectionsBuffer.SetData(VertexConnection.connectionsTable);
    offsetsBuffer.SetData(VertexConnection.offsetsTable);
    smoothingShader.SetBuffer(0,"ConnectionsBuffer",connectionsBuffer);
    smoothingShader.SetBuffer(0,"OffsetsBuffer",offsetsBuffer);
    smoothingShader.SetBuffer(0, "Counter", _counterBuffer);
    smoothingShader.SetBuffer(0,"OutVertexBuffer",outVertexBuffer);
    smoothingShader.SetInt("MaxVerticesNumber", verticesNo);
    smoothingShader.SetBuffer(0,"VertexBuffer", _builder._vertexBuffer);
    
 
    smoothingShader.Dispatch(kernel,64,64,64);
}

outVertexBuffer 的初始化：（stride 等于 12，因为类型为 float3）

 outVertexBuffer =
            new GraphicsBuffer(GraphicsBuffer.Target.Vertex, 5000000, 12);

在 LaplacianSmoothing ComputeShader 中，我想重新计算顶点的位置并使用 Store3 函数将它们存储到 outVertexBuffer 中。下面是 ComputeShader 中 LaplacianSmoothing 的代码（HLSL 风格）：

#pragma kernel LaplacianSmooth

RWTexture2D<float4> Result;
RWByteAddressBuffer VertexBuffer;
RWStructuredBuffer<uint> Counter; // used only for counting
RWByteAddressBuffer ConnectionsBuffer;
RWByteAddressBuffer OffsetsBuffer;
RWByteAddressBuffer OutVertexBuffer;
int MaxVerticesNumber;

#define SIZEOF_UINT 4
#define SIZEOF_FLOAT3 12

[numthreads(8,8,8)]
void LaplacianSmooth (uint3 id : SV_DispatchThreadID)
{
    uint currentIndex = 512*512*id.z + 512*id.y+id.x;
    if (currentIndex < MaxVerticesNumber)
    {
        float3 currentVertex = asfloat(VertexBuffer.Load3(2*currentIndex*SIZEOF_FLOAT3));
        float3 sumNeighVertex = 0;
        uint neighborAdrr=0;
        int currentOffset = OffsetsBuffer.Load(currentIndex*SIZEOF_UINT);
        int nextOffset = OffsetsBuffer.Load((currentIndex+1)*SIZEOF_UINT);
        int connectionCount = nextOffset-currentOffset;
        for (int i=0; i<connectionCount;i++)
        
        {
            neighborAdrr = ConnectionsBuffer.Load((currentOffset+i)*SIZEOF_UINT);// Adress of neighbor;
        
            sumNeighVertex += asfloat(VertexBuffer.Load3(neighborAdrr*2*SIZEOF_FLOAT3));
        }

        sumNeighVertex = sumNeighVertex / connectionCount;

        float3 newVert = sumNeighVertex+currentVertex;
        OutVertexBuffer.Store3(currentIndex*SIZEOF_FLOAT3,asuint(currentVertex));
    }
}

下一步我想从 outVertexBuffer 获取数据并将它们赋给 vertexArray GraphicsBuffer.GetData(Vector3[] array) 函数。

public void OnSmoothMeshLaplacian()
{

PrepareSmoothing();
var mesh = GetComponent<MeshFilter>().mesh;
Mesh testMesh = Instantiate(mesh);
GameObject.Find("Laplacian").GetComponent<MeshFilter>().sharedMesh = testMesh;
//testMesh = LaplacianFilter(testMesh, 5,1f);

Vector3[] vertexArray;
vertexArray = new Vector3[2*testMesh.vertices.Length];
var network = VertexConnection.BuildNetwork(testMesh.triangles);
RunSmoothingShader(smoothingShader.FindKernel("LaplacianSmooth"), testMesh.vertexCount);
outVertexBuffer.GetData(vertexArray);
testMesh.SetVertexBufferData(vertexArray, 0, 0, vertAndPosList.Length);
testMesh.RecalculateBounds();
Debug.Log("Mesh:" + mesh.bounds.size);
Debug.Log("Test Mesh:" + testMesh.bounds.size);
Debug.Log($"Shrinkage: {(mesh.bounds.size - testMesh.bounds.size).magnitude / mesh.bounds.size.magnitude * 100f}%");
    }

我在getData函数后设置了断点，查看数据，都是(0,0,0)。我使用 RenderDoc 检查缓冲区的工作方式，用谷歌搜索着色器和主程序中使用的所有函数的文档，但没有任何解释缺少分配数据。我想补充一点，当我也使用 inputBuffer 作为输出时，我会得到数据（逻辑上不正确，但仍然不是 (0,0,0)。

Answer 1

试试这个：

outVertexBuffer = new GraphicsBuffer(GraphicsBuffer.Target.Structured, 5000000, 12);

请注意更改后的目标。图形缓冲区需要至少一个计算目标标志才能绑定到计算内核。 Target.Vertex 不是这样的标志。您仍然可以添加它，但在您的情况下不需要。

事实上 outVertexBuffer 可能只是一个 ComputeBuffer。无需使用 GraphicsBuffer，因为您只需将其复制到 CPU 侧的网格。

编辑---

另一件事： SetVertexBufferData() 要求您首先配置所有顶点属性，您传递的数据应该基本上是原始顶点数据：位置、法线、uvs（如果有）等。

如果您只想设置顶点，也许使用 Mesh.SetVertices().

会更容易

在 computeShader 代码运行后将数据存储在 GraphicsBuffer 中总是 (0,0,0)

Storing data inside GraphicsBuffer after computeShader code runs always are (0,0,0)

c#

gpu

hlsl

unity3d

compute-shader