D3D学习笔记整理（五）

六、D3D与CUDA的互操作

CUDA与D3D可以进行互操作，也就是说可以将D3D的资源映射到CUDA地址空间，使在CUDA中可以对D3D读取和写入数据。CUDA实现这个功能的API有两组，这里以运行时API为例。

实现D3D与CUDA的互操作，从总体上看，主要多了D3D设备指定和资源注册及相应释放操作，以及在创建D3D设备时使用D3DCREATE_HARDWARE_VERTEXPROCESSING标识，见下例。

1、 初始化D3D设备

这个过程与一般的D3D设备创建过程并没有太多变化，注意用颜色表示的部分。另外，用于指定D3D设备的函数cudaD3D9SetDirect3DDevice必须在调用其它CUDA运行时API之前调用。

 HRESULT Cdig_D3DDlg::InitD3D9( UINT hWnd )

{

    m_hwndRender =GetDlgItem(hWnd)->GetSafeHwnd();//当前显示窗口

    LPDIRECT3D9 g_pD3D = NULL;     //Direct3D对象指针(不是设备指针)LPDIRECT3D9 g_pD3D = NULL;     //Direct3D对象指针(不是设备指针)

    // Create the D3D object.

    if( NULL == ( g_pD3D = Direct3DCreate9( D3D_SDK_VERSION ) ) )

        return E_FAIL;

 

    // Find the first CUDA capable device

    for(g_iAdapter = 0; g_iAdapter < g_pD3D->GetAdapterCount(); g_iAdapter++)

    {

        D3DCAPS9 caps;

        if (FAILED(g_pD3D->GetDeviceCaps(g_iAdapter, D3DDEVTYPE_HAL, &caps)))

            // Adapter doesn't support Direct3D

            continue;

        D3DADAPTER_IDENTIFIER9 ident;

        int device;

        g_pD3D->GetAdapterIdentifier(g_iAdapter, 0, &ident);

        cudaD3D9GetDevice(&device, ident.DeviceName);

        if (cudaSuccess == cudaGetLastError() )

            break;

    }

 

    // we check to make sure we have found a cuda-compatible D3D device to work on

    if(g_iAdapter == g_pD3D->GetAdapterCount() ) {

        printf("No CUDA-compatible Direct3D9 device available/n");

        printf("Test PASSED/n");

        // destroy the D3D device

        g_pD3D->Release();

        exit(0);

        return E_FAIL;

    }

    D3DPRESENT_PARAMETERS    g_d3dpp;

    RECT rc;      

    GetDlgItem(hWnd)->GetClientRect(&rc);

    g_pD3D->GetAdapterDisplayMode(g_iAdapter, &g_d3ddm);

 

    // Set up the structure used to create the D3DDevice

    //    D3DPRESENT_PARAMETERS d3dpp;

    ZeroMemory( &g_d3dpp, sizeof(g_d3dpp) );

    g_d3dpp.Windowed                = true;

    g_d3dpp.BackBufferCount         = 1;

    g_d3dpp.hDeviceWindow           = m_hwndRender;

    g_d3dpp.SwapEffect              = D3DSWAPEFFECT_DISCARD;

    g_d3dpp.BackBufferFormat        = g_d3ddm.Format;

    g_d3dpp.FullScreen_RefreshRateInHz = 0; // set to 60 for fullscreen, and also don't forg_et to set Windowed to FALSE

    g_d3dpp.PresentationInterval   = D3DPRESENT_INTERVAL_ONE; // D3DPRESENT_DONOTWAIT;

 

    //g_d3dpp.BackBufferWidth        = g_bQAReadback?g_WindowWidth:(rc.right  - rc.left);

    //g_d3dpp.BackBufferHeight       = g_bQAReadback?g_WindowHeight:(rc.bottom - rc.top);

    g_d3dpp.BackBufferWidth      = 512;

    g_d3dpp.BackBufferHeight       = 512;

 

    // Create the D3DDevice

    if ( FAILED( g_pD3D->CreateDevice( D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, m_hwndRender,D3DCREATE_HARDWARE_VERTEXPROCESSING,&g_d3dpp, &g_pd3dDevice ) ) )

        return E_FAIL;

 

 

    // Turn off culling, so we see the front and back of the triangle

    g_pd3dDevice->SetRenderState(D3DRS_CULLMODE, D3DCULL_NONE);

 

    // Turn off lighting, since we are providing our own vertex colors

    g_pd3dDevice->SetRenderState(D3DRS_LIGHTING, FALSE);

 

// Now we need to bind a CUDA context to the DX9 device

    // This is the CUDA 2.0 DX9 interface (required for Windows XP and Vista)

    cudaD3D9SetDirect3DDevice(g_pd3dDevice);

 

    return S_OK;

}

2、 顶点创建与资源注册

这里首先创建一个顶点buffer，然后注册的CUDA上，其中可以通过设定函数cudaD3D9ResourceSetMapFlags的标识符来决定资源的读写属性。

HRESULT Cdig_D3DDlg::InitVB()//D3DXCreateTextureFromFile

{

if (FAILED(g_pd3dDevice->CreateVertexBuffer(65536*sizeof(CUSTOMVERTEX), 0, D3DFVF_CUSTOMVERTEX, D3DPOOL_DEFAULT,/

    &g_pVB, NULL)))

{

    return E_FAIL;

}

 

 

    // Initialize interoperability between CUDA and Direct3D9

    // Register vertex buffer with CUDA

    cudaD3D9RegisterResource(g_pVB, cudaD3D9RegisterFlagsNone);

 

// cudaD3D9MapFlagsWriteDiscard: Specifies that CUDA kernels which access this resource will not read from this resource

//and will write over the entire contents of the resource,

//so none of the data previously stored in the resource will be preserved.

    cudaD3D9ResourceSetMapFlags(g_pVB, cudaD3D9MapFlagsWriteDiscard);

    return S_OK;

};

3、CUDA写入顶点数据

首先调用cudaD3D9MapResources函数进行资源映射（第一个参数指定D3D资源数目），再由函数cudaD3D9ResourceGetMappedPointer得到显存地址，这里也可以使用CUDA Array和2D pitch空间，即cudaD3D9ResourceGetMappedArray和cudaD3D9ResourceGetMappedPitch.

函数cudaD3D9UnmapResources用于解除资源映射。

void Cdig_D3DDlg::runCuda()

{

    HRESULT hr = S_OK;

 

    // Map vertex buffer to Cuda

    float4* dptr;

 

    // CUDA Map call to the Vertex Buffer and return a pointer

    cudaD3D9MapResources(1, (IDirect3DResource9 **)&g_pVB);

    cudaD3D9ResourceGetMappedPointer( (void **)&dptr, g_pVB, 0, 0);

    const unsigned int g_MeshWidth    = 256;

    const unsigned int g_MeshHeight   = 256;

    float anim=0.1f;

    // Execute kernel

    simpleD3DKernel(dptr, g_MeshWidth, g_MeshHeight, anim);

    // CUDA Map Unmap vertex buffer

    cudaD3D9UnmapResources(1, (IDirect3DResource9 **)&g_pVB);

}

4、渲染

渲染的方式与一般的D3D渲染完全相同，这里给出一个参考示例。

HRESULT Cdig_D3DDlg::render()

{

    if( g_pd3dDevice )

    {

        //

        // Draw the scene.

        //

        g_pd3dDevice->Clear(0, 0,

            D3DCLEAR_TARGET ,

            D3DCOLOR_XRGB(0,0,0), 1.0f,0);

 

 

        // Run CUDA to update vertex positions

        runCuda();

 

        g_pd3dDevice->BeginScene();

 

        // Setup the world, view, and projection matrices

        SetupMatrices(IDC_RENDERVIEW);

        SetupLights();

    // 渲染顶点缓冲区内容

        /*g_pd3dDevice->SetStreamSource(0, g_pVB, 0, sizeof(CUSTOMVERTEX));

        g_pd3dDevice->SetFVF(D3DFVF_CUSTOMVERTEX);

        g_pd3dDevice->DrawPrimitive(D3DPT_TRIANGLESTRIP, 0, 2*50-2);*/

 

       

        // Render the vertex buffer contents

        g_pd3dDevice->SetStreamSource( 0, g_pVB, 0, sizeof(CUSTOMVERTEX) );

        g_pd3dDevice->SetFVF( D3DFVF_CUSTOMVERTEX );

        g_pd3dDevice->DrawPrimitive( D3DPT_POINTLIST, 0, 65536 );

 

 

        g_pd3dDevice->EndScene();

        //将渲染的结果存储到后台缓冲区

        //0表示更新所有的缓冲区

        g_pd3dDevice->Present(0, 0, 0, 0);

    }

    return S_OK;

}

5、资源释放

Never forget to release these resources after you don’t use them again.

HRESULT Cdig_D3DDlg::cleanup()

{

if( g_pVB != NULL ) {

        // Unregister vertex buffer

       cudaD3D9UnregisterResource(g_pVB);

       

        g_pVB->Release();

    }

 

// Uninitialize CUDA data and context

    cudaThreadExit();

 

    // Nothing to Destroy.

    if( g_pd3dDevice )

        g_pd3dDevice->Release();

   

 

   

    return S_OK;

}

 

到此为止，D3D的入门学习告一段落了，更高级的三维渲染技术需要在工程实践中得到锻炼和提升，路漫漫其修远兮，唯有上下求索耳！