代码之家 › 专栏 › 技术社区 › Elad Maimoni

无法从Direct3D 11中的内存正确初始化2D纹理

direct3d11 direct3d directx-11 textures directx

Elad Maimoni · 技术社区 · 8 年前

我试图在系统内存中生成一个简单的数组,表示R8G8B8A8纹理,然后将该纹理传输到GPU内存。

首先,我分配一个数组并用所需的绿色数据填充它:

    frame.width = 3;
    frame.height = 1;
    auto components = 4;
    auto length = components * frame.width * frame.height;
    frame.data = new uint8_t[length];

    frame.data[0 + 0 * frame.width] = 0;  frame.data[1 + 0 * frame.width] = 255;  frame.data[2 + 0 * frame.width] = 0;  frame.data[3 + 0 * frame.width] = 255;
    frame.data[0 + 1 * frame.width] = 0;  frame.data[1 + 1 * frame.width] = 255;  frame.data[2 + 1 * frame.width] = 0;  frame.data[3 + 1 * frame.width] = 255;
    frame.data[0 + 2 * frame.width] = 0;  frame.data[1 + 2 * frame.width] = 255;  frame.data[2 + 2 * frame.width] = 0;  frame.data[3 + 2 * frame.width] = 255;

然后,我创建纹理对象并将其设置为像素着色器资源:

    D3D11_TEXTURE2D_DESC textureDescription;

    textureDescription.Width = frame.width;
    textureDescription.Height = frame.height;

    textureDescription.MipLevels = textureDescription.ArraySize = 1;
    textureDescription.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
    textureDescription.SampleDesc.Count = 1;
    textureDescription.SampleDesc.Quality = 0; 
    textureDescription.Usage = D3D11_USAGE_DYNAMIC;
    textureDescription.BindFlags = D3D11_BIND_SHADER_RESOURCE;
    textureDescription.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
    textureDescription.MiscFlags = 0;


    D3D11_SUBRESOURCE_DATA initialTextureData;
    initialTextureData.pSysMem = frame.data;
    initialTextureData.SysMemPitch = frame.width * components;
    initialTextureData.SysMemSlicePitch = 0;

    DX_CHECK(m_device->CreateTexture2D(&textureDescription, &initialTextureData, &m_texture));
    DX_CHECK(m_device->CreateShaderResourceView(m_texture, NULL, &m_textureView));
    m_context->PSSetShaderResources(0, 1, &m_textureView);

我的期望是GPU内存将包含3x1绿色纹理,每个纹理在alpha通道中将有1.0f。但是,这不是通过Visual Studio图形调试器检查加载的纹理对象可以看到的情况:

有人能解释一下发生了什么事吗?我怎样才能解决这个问题?

1 回复 | 直到 8 年前

Nico Schertler 8 年前

让我们看看您的阵列寻址方案(索引随着您提供的维数而减少):

frame.data[0] = 0;  frame.data[1] = 255;  frame.data[2] = 0;  frame.data[3] = 255;
frame.data[3] = 0;  frame.data[4] = 255;  frame.data[5] = 0;  frame.data[6] = 255;
frame.data[6] = 0;  frame.data[7] = 255;  frame.data[8] = 0;  frame.data[9] = 255;

重新订购,我们得到

data[ 0] = 0                  B pixel 1
data[ 1] = 255                G pixel 1
data[ 2] = 0                  R pixel 1
data[ 3] = 0  (overwritten)   A pixel 1
data[ 4] = 255                  pixel 2
data[ 5] = 0
data[ 6] = 0
data[ 7] = 255
data[ 8] = 0                    pixel 3
data[ 9] = 255
data[10] = undefined
data[11] = undefined

如您所见,这正是调试器向您显示的数据。

所以你只需要修改你的地址方案。正确的公式是:

index = component + x * components + y * pitch,

pitch = width * components

为了推导这个公式,您只需要考虑当您增加一个变量时,您必须跳过多少个索引。E、 g.当您增加当前组件时,只需进一步增加一个条目(因为所有组件彼此相邻)。另一方面,如果增加y坐标,则需要跳过与行中相同数量的条目(称为间距,即图像宽度乘以密集堆积的组件数)。