DirectX-11 实例缓冲区返回 E_INVALID_ARGS

Question

伙计们。我正处于开始向我的游戏引擎添加实例化的地步。这是我制作的通用函数，因此我可以传入实例缓冲区指针和一些矩阵，以及要绘制的实例列表。但是，在它的第一个循环中，缓冲区为 NULL，因此它尝试对其进行初始化，但失败并显示 HR = E_INVALID_ARGS。我不确定参数的哪一部分是无效的。

目标是每个实例以 XMFLOAT4X4 的形式传递两个 4X4 矩阵。

“sizeofXMFLOAT4X4”变量只是一个健全性检查，以查看它读取的字节数为 (64, 4bytes * 4floats/vec * 4 vec = 64)。

如果有人能发现这里可能出了什么问题并指出来，我将不胜感激。如果您需要更多信息，请告诉我。谢谢。

void StaticSceneryListClass::UpdateInstanceBuffer(ID3D11Buffer *pInstanceBuffer, XMFLOAT4X4 VPMat, XMFLOAT4X4 tileMat,  vector<StaticSceneryStruct> culledInstanceList)
{
//Pass this the buffer you want to update, the VPMat for the frame, and a list of instances to draw and it will fill the buffer with the stuff and things. 
//If the buffer is not initialized yet, it will be initialized too.

//first, create the data needed.
vector<XMFLOAT4X4> instanceMats;

for (int i = 0; i < culledInstanceList.size(); i++)
{
    XMFLOAT4X4 WorldMat = xmCon->MatTranspose(xmCon->MatMultiply(culledInstanceList[i].worldMat, tileMat));
    XMFLOAT4X4 WVP = xmCon->MatTranspose(xmCon->MatMultiply(WorldMat, VPMat));
    instanceMats.push_back(WVP);
    instanceMats.push_back(WorldMat);
}

if (pInstanceBuffer == NULL)
{
    //buffer not initialized yet, initialize it.
    D3D11_BUFFER_DESC instBuffDesc;
    ZeroMemory(&instBuffDesc, sizeof(instBuffDesc));
    sizeofXMFLOAT4X4 = sizeof(XMFLOAT4X4);
    instBuffDesc.Usage = D3D11_USAGE_DYNAMIC;
    instBuffDesc.ByteWidth = sizeofXMFLOAT4X4 * 2 * culledInstanceList.size();
    instBuffDesc.BindFlags = D3D11_BIND_VERTEX_BUFFER;
    instBuffDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
    instBuffDesc.MiscFlags = 0;

    D3D11_SUBRESOURCE_DATA instData;
    ZeroMemory(&instData, sizeof(instData));

    instData.pSysMem = &instanceMats[0];
    HRESULT hr = graphicsPointers.pdev->CreateBuffer(&instBuffDesc, &instData, &pInstanceBuffer);

    if (FAILED(hr))
    {
        //Fails here with E_INVALID_ARG
        MessageBox(NULL, "Failed to create instance buffer!", "StaticSceneryList", MB_OK);
    }
}
else
{
    //buffer already initialized, need to use resource may and unmap.
    D3D11_MAPPED_SUBRESOURCE mappedResource;
    ZeroMemory(&mappedResource, sizeof(D3D11_MAPPED_SUBRESOURCE));

    //  Disable GPU access to the vertex buffer data.
    graphicsPointers.pdevcon->Map(pInstanceBuffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &mappedResource);
    //  Update the vertex buffer here.
    memcpy(mappedResource.pData, &instanceMats[0], sizeof(XMFLOAT4X4) * 2 * culledInstanceList.size());
    //  Reenable GPU access to the vertex buffer data.
    graphicsPointers.pdevcon->Unmap(pInstanceBuffer, 0);
}

更新 1：2017 年 12 月 2 日 2:02 CST 今天早上我一直在运行各种调试测试，结果很有趣。我尝试隔离问题，发现为数据传入 null 或为缓冲区传入一个新的本地指针什么也没做。我打开了调试层，它没有给我创建缓冲区语句的输出。我尝试使用缓冲区描述的参数来查看它会做什么，这些给了我一个错误。如果剔除列表的大小为 0 并且即使它不是 16 的倍数，我添加了一个捕获。这让它偶尔通过，但它实际上从未产生过工作缓冲区。事实上，它通常仍然失败... S_OK?? FAILED(hr) 以 S_OK 的 HR 下降，这没有多大意义。当我查看缓冲区指针的内存值时，它从 null 变为有地址，但它没有被“如果不是 null，则释放并设置为 null”块捕获。

这个有趣的副作用是它会导致各种巨大的内存泄漏。进程内存在大约 10 秒内以接近线性的速度攀升至超过 3gb（我此时关闭程序，它仍然非常灵敏）。

当然，我似乎也偶尔获得通过，然后似乎永远不会坚持，因为缓冲区是否已初始化的代码尚未触发。

这是当前更新的代码：

void StaticSceneryListClass::UpdateInstanceBuffer(ID3D11Buffer *pInstanceBuffer, XMFLOAT4X4 VPMat, XMFLOAT4X4 tileMat, vector<StaticSceneryStruct> culledInstanceList)
{
//Pass this the buffer you want to update, the VPMat for the frame, and a list of instances to draw and it will fill the buffer with the stuff and things. 
//If the buffer is not initialized yet, it will be initialized too.

//Skip this process if there aren't any instances to draw
if (culledInstanceList.size() > 0)
{
    //first, create the data needed.
    vector<XMFLOAT4X4> instanceMats;
    for (int i = 0; i < culledInstanceList.size(); i++)
    {
        XMFLOAT4X4 WorldMat = xmCon->MatTranspose(xmCon->MatMultiply(culledInstanceList[i].worldMat, tileMat));
        XMFLOAT4X4 WVP = xmCon->MatTranspose(xmCon->MatMultiply(WorldMat, VPMat));
        instanceMats.push_back(WVP);
        instanceMats.push_back(WorldMat);
    }

    if (pInstanceBuffer == NULL)
    {
        //buffer not initialized yet, initialize it.
        D3D11_BUFFER_DESC instBuffDesc;
        ZeroMemory(&instBuffDesc, sizeof(instBuffDesc));
        bytewidth = sizeof(XMFLOAT4X4) * 2 * culledInstanceList.size();
        if (bytewidth % 16 != 0)
        {
            //hasn't fallen into here yet, which makes sense.
            MessageBox(NULL, "Instance Buffer Byte Width Not a multiple of 16", "StaticSceneryList", MB_OK);
        }

        instBuffDesc.Usage = D3D11_USAGE_DYNAMIC;
        instBuffDesc.ByteWidth =  bytewidth;
        instBuffDesc.BindFlags = D3D11_BIND_VERTEX_BUFFER;
        instBuffDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
        instBuffDesc.MiscFlags = 0;

        D3D11_SUBRESOURCE_DATA instData;
        ZeroMemory(&instData, sizeof(instData));

        instData.pSysMem = &instanceMats[0];

        HRESULT hr = graphicsPointers.pdev->CreateBuffer(&instBuffDesc, &instData, &pInstanceBuffer);
        //HRESULT hr = graphicsPointers.pdev->CreateBuffer(&instBuffDesc, &instData, &debugInstanceBuffer); //DEBUG
        if (FAILED(hr))
        {
            //Fails here with S_OK lulwut
            MessageBox(NULL, "Failed to create instance buffer!", "StaticSceneryList", MB_OK);
            if (pInstanceBuffer != NULL)
            {
               //This doesn't seem to trigger even though the address is not 0x0000etc
                pInstanceBuffer->Release();
                pInstanceBuffer = NULL;
            }
        }
        else
        {
            //occasionally gets here but it doesn't seem to stick
            MessageBox(NULL, "Successfully created instance buffer.", "StaticSceneryList", MB_OK);
        }
    }
    else
    {
        //never seems to get to here.
        //buffer already initialized, need to use resource may and unmap.
        D3D11_MAPPED_SUBRESOURCE mappedResource;
        ZeroMemory(&mappedResource, sizeof(D3D11_MAPPED_SUBRESOURCE));

        //  Disable GPU access to the vertex buffer data.
        graphicsPointers.pdevcon->Map(pInstanceBuffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &mappedResource);
        //  Update the vertex buffer here.
        memcpy(mappedResource.pData, &instanceMats[0], sizeof(XMFLOAT4X4) * 2 * culledInstanceList.size());
        //  Reenable GPU access to the vertex buffer data.
        graphicsPointers.pdevcon->Unmap(pInstanceBuffer, 0);
    }
}//if there is anything to draw

Answer 1

因此，在没有运气的情况下进行了更多测试后，我将代码移到调用此函数的位置，并直接而不是间接地根据对象编写它，并且我让它工作，至少使用默认缓冲区。我不知道为什么，但是指针的传递发生了一些奇怪的事情，这会导致它在退出函数调用时最终为 null。

另外，请注意，当您在输入布局中声明四个 R32B32G32A32_FLOAT 向量系列，然后在顶点着色器中将它们作为四个 float4 拉入，然后使用 float4x4(vec1, vec2, vec3, vec4) 组合它们时，您在将 XMFLOAT4X4 放入缓冲区数据之前不要转置它！