D3D三层Texture纹理经像素着色器实现渲染YUV420P

    简单记录一下这两天用Texture实现渲染YUV420P的一些要点。

    在视频播放的过程中，有的时候解码出来的数据是YUV420P的。表面（surface）通过设置参数是可以渲染YUV420P的，但Texture纹理似乎不支持直接渲染YUV420P。表面（surface）用法似乎比较单一，通常用来显示数据，用Texture的话就可以用上D3D的许多其他功能，做出一些效果。当然，这看个人需求，通常而言显示视频数据用表面就够了。

1.利用像素着色器把YUV420P数据转为RGB32

   视频播放过程中帧与帧之间是有固定时间间隔的。若解码解出来的是YUV420P的数据，要用Texture渲染的话，就需要把数据转为RGB32的（应该是要转成RGB32的，没做过详细调查，看到的例子中纹理所用的数据都是RGB32的），如果这个过程交给CPU去做的话，会提高CPU的占用率，用GPU来做这项工作则就释放了CPU的一部分压力。

    本文考虑的方式是用三层纹理分别存放YUV420P的Y、U、V分量（这个词不知道用对没有），然后像素着色器把三个分量的数据计算成RGB32的数据然后显示。这是本文的核心内容。

像素着色器的HLSL代码如下：

sampler YTex;
sampler UTex;
sampler VTex;


struct PS_INPUT
{
    float2 y    : TEXCOORD0;
    float2 u    : TEXCOORD1;
    float2 v    : TEXCOORD2;
};


float4 Main(PS_INPUT input):COLOR0
{      
    float y = tex2D(YTex,input.y).r;
    float u = tex2D(UTex, input.u.xy / 2).r  - 0.5f;
    float v = tex2D(VTex,input.v.xy / 2).r  - 0.5f;            
                
    float r = y + 1.14f * v;
    float g = y - 0.394f * u - 0.581f * v;
    float b = y + 2.03f * u;
                
    return float4(r,g,b, 1);
}

     HLSL代码可以直接写在txt文件中，sampler可视作标识纹理层和采样级的对象，Direct3D将把每一个sampler对象唯一地与某一纹理层关联起来。具体的HLSL语法请自行查资料，我也是粗略知道是怎么回事，就不误人子弟了。在代码中通过调用D3DXCompileShaderFromFile函数可以从文件编译像素着色器。但实际上，我个人不是很喜欢这种把代码放在一个单独文件里面的做法，这种代码应该尽可能的编进exe里面。但是我还只是初步了解D3D，不知道怎么把它编进exe里面，如果有人知道，还望指教。

ID3DXBuffer* shader      = 0;
ID3DXBuffer* errorBuffer = 0;

hr = D3DXCompileShaderFromFile(
    "ps_multitex.txt",
    0,
    0,
    "Main", // entry point function name
    "ps_2_0",
    D3DXSHADER_DEBUG, 
    &shader,
    &errorBuffer,
    &MultiTexCT);

// output any error messages
if( errorBuffer )
{
    ::MessageBox(0, (char*)errorBuffer->GetBufferPointer(), 0, 0);
    d3d::Release<ID3DXBuffer*>(errorBuffer);
}

if(FAILED(hr))
{
    ::MessageBox(0, "D3DXCompileShaderFromFile() - FAILED", 0, 0);
    return false;
}

//
// Create Pixel Shader
//
hr = Device->CreatePixelShader(
    (DWORD*)shader->GetBufferPointer(),
    &MultiTexPS);

if(FAILED(hr))
{
    ::MessageBox(0, "CreateVertexShader - FAILED", 0, 0);
    return false;
}

d3d::Release<ID3DXBuffer*>(shader);

以上代码中，D3DXCompileShaderFromFile函数从文件ps_multitex.txt编译HLSL代码；参数Main是HLSL代码的入口函数，如上一点代码中所见。这个入口函数可以是自定义的其他的，但要注意保持一致；ps_2_0表示像素着色器的版本。CreatePixelShader函数创建像素着色器。

2.sampler与纹理关联

    创建纹理层。本文实现YUV420P渲染的方法采用了三层纹理，每层纹理分别存放Y、U、V数据。

Device->CreateTexture ( Width, Height, 1, D3DUSAGE_DYNAMIC, D3DFMT_L8, D3DPOOL_DEFAULT, &YTex, NULL ) ;
Device->CreateTexture ( Width / 2, Height / 2, 1, D3DUSAGE_DYNAMIC, D3DFMT_L8, D3DPOOL_DEFAULT, &UTex, NULL ) ;
Device->CreateTexture ( Width / 2, Height / 2, 1, D3DUSAGE_DYNAMIC, D3DFMT_L8, D3DPOOL_DEFAULT, &VTex, NULL ) ;

    sampler与纹理的关联

// 
// Get Handles
//

YTexHandle      = MultiTexCT->GetConstantByName(0, "YTex");
UTexHandle        = MultiTexCT->GetConstantByName(0, "UTex");
VTexHandle        = MultiTexCT->GetConstantByName(0, "VTex");

//
// Set constant descriptions:
//

UINT count;

MultiTexCT->GetConstantDesc(YTexHandle,      &YTexDesc, &count);
MultiTexCT->GetConstantDesc(UTexHandle, &UTexDesc, &count);
MultiTexCT->GetConstantDesc(VTexHandle,    &VTexDesc, &count);

MultiTexCT->SetDefaults(Device);

    设置纹理/sampler的状态，这一部分我是在渲染的时候做的，也可以直接写在HLSL代码中。在后面渲染部分还会见到这些代码，其实是同一段代码，我只是为了表述纹理与sampler关联的一个整体过程，把它预先从渲染部分截了出来，希望不会造成误解。

// Y tex
Device->SetTexture(     YTexDesc.RegisterIndex, YTex);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

// U tex
Device->SetTexture(     UTexDesc.RegisterIndex, UTex);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

// string tex
Device->SetTexture(     VTexDesc.RegisterIndex, VTex);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

3.渲染YUV420P

    获得YUV420P数据。本文直接读取的YUV420P数据。

打开文件代码：

if((infile=fopen("test_yuv420p_320x180.yuv", "rb"))==NULL){
    printf("cannot open this file
");
    return false;
}

读取数据并将数据copy到纹理中：

if (fread(buf, 1, Width*Height*3/2, infile) != Width*Height*3/2){
        // Loop
        fseek(infile, 0, SEEK_SET);
        fread(buf, 1, Width*Height*3/2, infile);
    }


//
// Render
//

Device->Clear(0, 0, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, 0xffffffff, 1.0f, 0);

plane[0] = buf;
plane[1] = plane[0] + Width*Height;
plane[2] = plane[1] + Width*Height/4;

D3DLOCKED_RECT d3d_rect;
byte *pSrc = buf;
//Locks a rectangle on a texture resource.
//And then we can manipulate pixel data in it.
LRESULT lRet = YTex->LockRect(0, &d3d_rect, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest = (byte *)d3d_rect.pBits;
int stride = d3d_rect.Pitch; 
for(int i = 0;i < Height;i ++){
    memcpy(pDest + i * stride,plane[0] + i * Width, Width);
}

YTex->UnlockRect(0);

D3DLOCKED_RECT d3d_rect1;
lRet = UTex->LockRect(0, &d3d_rect1, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest1 = (byte *)d3d_rect1.pBits;
int stride1 = d3d_rect1.Pitch; 
for(int i = 0;i < Height/2;i ++){
    memcpy(pDest1 + i * stride1 / 2,plane[1] + i * Width / 2, Width / 2);
}

UTex->UnlockRect(0);

D3DLOCKED_RECT d3d_rect2;
lRet =  VTex->LockRect(0, &d3d_rect2, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest2 = (byte *)d3d_rect2.pBits;
int stride2 = d3d_rect2.Pitch; 
for(int i = 0;i < Height/2;i ++){
    memcpy(pDest2 + i * stride2 / 2,plane[2] + i * Width / 2, Width / 2);
}

VTex->UnlockRect(0);

渲染：

Device->BeginScene();

Device->SetPixelShader(MultiTexPS);
Device->SetFVF(MultiTexVertex::FVF);
Device->SetStreamSource(0, QuadVB, 0, sizeof(MultiTexVertex));

// Y tex
Device->SetTexture(     YTexDesc.RegisterIndex, YTex);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(YTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

// U tex
Device->SetTexture(     UTexDesc.RegisterIndex, UTex);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(UTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

// string tex
Device->SetTexture(     VTexDesc.RegisterIndex, VTex);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MAGFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MINFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_MIPFILTER, D3DTEXF_LINEAR);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_ADDRESSU, D3DTADDRESS_BORDER);
Device->SetSamplerState(VTexDesc.RegisterIndex, D3DSAMP_ADDRESSV, D3DTADDRESS_BORDER);

Device->DrawPrimitive(D3DPT_TRIANGLELIST, 0, 2);

Device->EndScene();
Device->Present(0, 0, 0, 0);

D3D Texture纹理渲染YUV420P的主要就是以上一些内容。完整工程代码：http://download.csdn.net/download/qq_33892166/9702415

本文的HLSL代码写出来的像素着色器画面有些偏黄，知道如何优化的朋友还请指教。

--------------------------------------------------------------2016.12.13 更新---------------------------------------------

本次更新修正遗留的画质问题。

更新1.HLSL代码

    HLSL代码中并不需要对U、V做除以2。新HLSL代码：

sampler YTex;
sampler UTex;
sampler VTex;


struct PS_INPUT
{
    float2 y    : TEXCOORD0;
    float2 u    : TEXCOORD1;
    float2 v    : TEXCOORD2;
};


float4 Main(PS_INPUT input):COLOR0
{      
    float y = tex2D(YTex,input.y).r;
    //这里不需要除以2
    float u = tex2D(UTex,input.u.xy).r  - 0.5f;
    float v = tex2D(VTex,input.v.xy).r  - 0.5f;            
                
    float r = y + 1.14f * v;
    float g = y - 0.394f * u - 0.581f * v;
    float b = y + 2.03f * u;
                
    return float4(r,g,b, 1);
}

更新2.copy数据到texture

    从内存copy数据到显卡的时候，U、V的数据copy上出了问题。新的读取数据并将数据copy到纹理中：

if (fread(buf, 1, Width*Height*3/2, infile) != Width*Height*3/2){
        // Loop
        fseek(infile, 0, SEEK_SET);
        fread(buf, 1, Width*Height*3/2, infile);
    }


//
// Render
//

Device->Clear(0, 0, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, 0xffffffff, 1.0f, 0);

plane[0] = buf;
plane[1] = plane[0] + Width*Height;
plane[2] = plane[1] + Width*Height/4;

D3DLOCKED_RECT d3d_rect;
byte *pSrc = buf;
//Locks a rectangle on a texture resource.
//And then we can manipulate pixel data in it.
LRESULT lRet = YTex->LockRect(0, &d3d_rect, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest = (byte *)d3d_rect.pBits;
int stride = d3d_rect.Pitch; 
for(int i = 0;i < Height;i ++){
    memcpy(pDest + i * stride,plane[0] + i * Width, Width);
}

YTex->UnlockRect(0);

D3DLOCKED_RECT d3d_rect1;
lRet = UTex->LockRect(0, &d3d_rect1, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest1 = (byte *)d3d_rect1.pBits;
int stride1 = d3d_rect1.Pitch; 
for(int i = 0;i < Height/2;i ++){
    //修改1：这个地方不需要除以2
    memcpy(pDest1 + i * stride1,plane[1] + i * Width / 2, Width / 2);
}

UTex->UnlockRect(0);

D3DLOCKED_RECT d3d_rect2;
lRet =  VTex->LockRect(0, &d3d_rect2, 0, 0);
if (FAILED(lRet)){
    return false;
}
// Copy pixel data to texture
byte *pDest2 = (byte *)d3d_rect2.pBits;
int stride2 = d3d_rect2.Pitch; 
for(int i = 0;i < Height/2;i ++){
    //修改1：这个地方也不需要除以2
    memcpy(pDest2 + i * stride2,plane[2] + i * Width / 2, Width / 2);
}

VTex->UnlockRect(0);

完整工程代码：http://download.csdn.net/download/qq_33892166/9710622