Directshow源码分析之推模式

在windowsSDK中有一个推模式的Source Filter例子，位于SDK安装目录samples\C++\Directshow\Ball下。

下面做一下简要分析：

功能：Live Source不断产生视频帧，演示在封闭围墙内碰撞的弹球的运动轨迹。使用该filter构建Filter Graph及其输出结果如下图：

类的继承结构关系为：

CBouncingBall继承自CSource

CBallStream继承自CSourceStream

可以知道，从CSource派生的Filter，它的输出Pin会使用一个数据线程，将Sample不断的推出去，具体实现如下：

//// DoBufferProcessingLoop//// Grabs a buffer and calls the users processing function.// Overridable, so that different delivery styles can be catered for.HRESULT CSourceStream::DoBufferProcessingLoop(void) {    Command com;    OnThreadStartPlay();//开始数据传送前调用，子类可以重新实现以进行一些初始化    do {while (!CheckRequest(&com)) {    IMediaSample *pSample;<span style="white-space:pre"></span>//输出Pin上得到一个空的Pin    HRESULT hr = GetDeliveryBuffer(&pSample,NULL,NULL,0);    if (FAILED(hr)) {                Sleep(1);continue;// go round again. Perhaps the error will go away    // or the allocator is decommited & we will be asked to    // exit soon.    }    // 子类必须实现的一个函数，填写Sample的实际内容    hr = FillBuffer(pSample);    if (hr == S_OK) {hr = Deliver(pSample);//往下一个filter发送数据                pSample->Release();                // downstream filter returns S_FALSE if it wants us to                // stop or an error if it's reporting an error.                if(hr != S_OK)                {                  DbgLog((LOG_TRACE, 2, TEXT("Deliver() returned %08x; stopping"), hr));                  return S_OK;                }    } else if (hr == S_FALSE) {                // derived class wants us to stop pushing datapSample->Release();DeliverEndOfStream();return S_OK;    } else {                // derived class encountered an error                pSample->Release();DbgLog((LOG_ERROR, 1, TEXT("Error %08lX from FillBuffer!!!"), hr));                DeliverEndOfStream();                m_pFilter->NotifyEvent(EC_ERRORABORT, hr, 0);                return hr;    }            // all paths release the sample}        // For all commands sent to us there must be a Reply call!if (com == CMD_RUN || com == CMD_PAUSE) {    Reply(NOERROR);} else if (com != CMD_STOP) {    Reply((DWORD) E_UNEXPECTED);    DbgLog((LOG_ERROR, 1, TEXT("Unexpected command!!!")));}    } while (com != CMD_STOP);    return S_FALSE;}

CBaseFilter的FillBuffer的实现：

<pre name="code" class="cpp">HRESULT RtpStream::FillBuffer(IMediaSample *pms){CheckPointer(pms, E_POINTER);ASSERT(m_Ball);BYTE *pData;long lDataLen;pms->GetPointer(&pData);lDataLen = pms->GetSize();ZeroMemory(pData, lDataLen);//请屏，将图像帧设置为黑色{CAutoLock cAutoLockShared(&m_cSharedState);// If we haven't just cleared the buffer delete the old// ball and move the ball onm_Ball->MoveBall(m_rtSampleTime - (LONG)m_iRepeatTime);//在图像帧上画出弹球当前的位置m_Ball->PlotBall(pData, m_BallPixel, m_iPixelSize);// The current time is the sample's startCRefTime rtStart = m_rtSampleTime;// Increment to find the finish timem_rtSampleTime += (LONG)m_iRepeatTime;//为Sample打上时间戳pms->SetTime((REFERENCE_TIME *)&rtStart, (REFERENCE_TIME *)&m_rtSampleTime);}pms->SetSyncPoint(TRUE);return NOERROR;}

可以看到，m_Ball是一个CBall类，主要完成弹球的位置计算以及绘制，当要在新位置上绘制弹球时，一般首先调用MoveBall

进行新位置的计算，然后调用PlotBall在计算得到的位置上绘制一个球。

下面是理解该例子的几个要点：

1.如何创建输出Pin。因为FIlter-Pin采用的是CSource-CSourceStream类结构，所以实际输出Pin的创建一般在Filter的构造函数中。

<pre name="code" class="cpp">RtpStreamFilter::RtpStreamFilter(LPUNKNOWN lpunk, HRESULT *phr) :CSource(NAME("Rtp Source Filter"), lpunk, CLSID_BouncingBall){ASSERT(phr);CAutoLock cAutoLock(&m_cStateLock);//首先进行内存分配，m_paStreams是一个指针数组，因为只有一个Pin，所以分配一个单元m_paStreams = (CSourceStream **) new RtpStream*[1];if (m_paStreams == NULL){if (phr)*phr = E_OUTOFMEMORY;return;} //新生成一个CSourceStream实例m_paStreams[0] = new RtpStream(phr, this, L"A Rtp Source Filter!");if (m_paStreams[0] == NULL){if (phr)*phr = E_OUTOFMEMORY;return;}} 

2.如何在输出Pin上提供首选的媒体类型列表，包括RGB32,RGB24,RGB565,RGB555,RGB8等?如下代码描

//// GetMediaType//// I _prefer_ 5 formats - 8, 16 (*2), 24 or 32 bits per pixel and// I will suggest these with an image size of 320x240. However// I can accept any image size which gives me some space to bounce.//// A bit of fun://      8 bit displays get red balls//      16 bit displays get blue//      24 bit see green//      And 32 bit see yellow//// Prefered types should be ordered by quality, zero as highest quality// Therefore iPosition =// 0    return a 32bit mediatype// 1    return a 24bit mediatype// 2    return 16bit RGB565// 3    return a 16bit mediatype (rgb555)// 4    return 8 bit palettised format// (iPosition > 4 is invalid)//HRESULT RtpStream::GetMediaType(int iPosition, CMediaType *pmt){CheckPointer(pmt, E_POINTER);//如果pmt为NULL，直接returnCAutoLock cAutoLock(m_pFilter->pStateLock());if (iPosition < 0){return E_INVALIDARG;}// Have we run off the end of types?if (iPosition > 4){return VFW_S_NO_MORE_ITEMS;}VIDEOINFO *pvi = (VIDEOINFO *)pmt->AllocFormatBuffer(sizeof(VIDEOINFO));if (NULL == pvi)return(E_OUTOFMEMORY);ZeroMemory(pvi, sizeof(VIDEOINFO));switch (iPosition){case 0:{  // Return our highest quality 32bit format  // since we use RGB888 (the default for 32 bit), there is  // no reason to use BI_BITFIELDS to specify the RGB  // masks. Also, not everything supports BI_BITFIELDS  SetPaletteEntries(Yellow);  pvi->bmiHeader.biCompression = BI_RGB;  pvi->bmiHeader.biBitCount = 32;  break;}case 1:{   // Return our 24bit format  SetPaletteEntries(Green);  pvi->bmiHeader.biCompression = BI_RGB;  pvi->bmiHeader.biBitCount = 24;  break;}case 2:{  // 16 bit per pixel RGB565  // Place the RGB masks as the first 3 doublewords in the palette area  for (int i = 0; i < 3; i++)  pvi->TrueColorInfo.dwBitMasks[i] = bits565[i];  SetPaletteEntries(Blue);  pvi->bmiHeader.biCompression = BI_BITFIELDS;  pvi->bmiHeader.biBitCount = 16;  break;}case 3:{   // 16 bits per pixel RGB555  // Place the RGB masks as the first 3 doublewords in the palette area  for (int i = 0; i < 3; i++)  pvi->TrueColorInfo.dwBitMasks[i] = bits555[i];  SetPaletteEntries(Blue);  pvi->bmiHeader.biCompression = BI_BITFIELDS;  pvi->bmiHeader.biBitCount = 16;  break;}case 4:{   // 8 bit palettised  SetPaletteEntries(Red);  pvi->bmiHeader.biCompression = BI_RGB;  pvi->bmiHeader.biBitCount = 8;  pvi->bmiHeader.biClrUsed = iPALETTE_COLORS;  break;}}// (Adjust the parameters common to all formats...)// put the optimal palette in placefor (int i = 0; i < iPALETTE_COLORS; i++){pvi->TrueColorInfo.bmiColors[i].rgbRed = m_Palette[i].peRed;pvi->TrueColorInfo.bmiColors[i].rgbBlue = m_Palette[i].peBlue;pvi->TrueColorInfo.bmiColors[i].rgbGreen = m_Palette[i].peGreen;pvi->TrueColorInfo.bmiColors[i].rgbReserved = 0;}pvi->bmiHeader.biSize = sizeof(BITMAPINFOHEADER);pvi->bmiHeader.biWidth = m_iImageWidth;pvi->bmiHeader.biHeight = m_iImageHeight;pvi->bmiHeader.biPlanes = 1;pvi->bmiHeader.biSizeImage = GetBitmapSize(&pvi->bmiHeader);pvi->bmiHeader.biClrImportant = 0;SetRectEmpty(&(pvi->rcSource)); // we want the whole image area rendered.SetRectEmpty(&(pvi->rcTarget)); // no particular destination rectanglepmt->SetType(&MEDIATYPE_Video);pmt->SetFormatType(&FORMAT_VideoInfo);pmt->SetTemporalCompression(FALSE);// Work out the GUID for the subtype from the header info.const GUID SubTypeGUID = GetBitmapSubtype(&pvi->bmiHeader);pmt->SetSubtype(&SubTypeGUID);pmt->SetSampleSize(pvi->bmiHeader.biSizeImage);return NOERROR;} 

3.如何在输出Pin连接时进行媒体类型的检查?如下代码

//// CheckMediaType//// We will accept 8, 16, 24 or 32 bit video formats, in any// image size that gives room to bounce.// Returns E_INVALIDARG if the mediatype is not acceptable//HRESULT RtpStream::CheckMediaType(const CMediaType *pMediaType){CheckPointer(pMediaType, E_POINTER);if ((*(pMediaType->Type()) != MEDIATYPE_Video) ||   // we only output video!(pMediaType->IsFixedSize()))                   // in fixed size samples{return E_INVALIDARG;}// Check for the subtypes we supportconst GUID *SubType = pMediaType->Subtype();if (SubType == NULL)return E_INVALIDARG;if ((*SubType != MEDIASUBTYPE_RGB8)&& (*SubType != MEDIASUBTYPE_RGB565)&& (*SubType != MEDIASUBTYPE_RGB555)&& (*SubType != MEDIASUBTYPE_RGB24)&& (*SubType != MEDIASUBTYPE_RGB32)){return E_INVALIDARG;}// Get the format area of the media typeVIDEOINFO *pvi = (VIDEOINFO *)pMediaType->Format();if (pvi == NULL)return E_INVALIDARG;// Check the image size. As my default ball is 10 pixels big// look for at least a 20x20 image. This is an arbitary size constraint,// but it avoids balls that are bigger than the picture...if ((pvi->bmiHeader.biWidth < 20) || (abs(pvi->bmiHeader.biHeight) < 20)){return E_INVALIDARG;}// Check if the image width & height have changedif (pvi->bmiHeader.biWidth != m_Ball->GetImageWidth() ||abs(pvi->bmiHeader.biHeight) != m_Ball->GetImageHeight()){// If the image width/height is changed, fail CheckMediaType() to force// the renderer to resize the image.return E_INVALIDARG;}return S_OK;  // This format is acceptable.}

4.当输出Pin连接成功后，需要协商Pin上以后传送数据使用的Sample的一些属性，比如Sample使用的内存的大小等。

//// DecideBufferSize//// This will always be called after the format has been sucessfully// negotiated. So we have a look at m_mt to see what size image we agreed.// Then we can ask for buffers of the correct size to contain them.//HRESULT RtpStream::DecideBufferSize(IMemAllocator *pAlloc,ALLOCATOR_PROPERTIES *pProperties){CheckPointer(pAlloc, E_POINTER);CheckPointer(pProperties, E_POINTER);CAutoLock cAutoLock(m_pFilter->pStateLock());HRESULT hr = NOERROR;VIDEOINFO *pvi = (VIDEOINFO *)m_mt.Format();pProperties->cBuffers = 1;pProperties->cbBuffer = pvi->bmiHeader.biSizeImage;ASSERT(pProperties->cbBuffer);// Ask the allocator to reserve us some sample memory, NOTE the function// can succeed (that is return NOERROR) but still not have allocated the// memory that we requested, so we must check we got whatever we wantedALLOCATOR_PROPERTIES Actual;hr = pAlloc->SetProperties(pProperties, &Actual);if (FAILED(hr)){return hr;}// Is this allocator unsuitableif (Actual.cbBuffer < pProperties->cbBuffer){return E_FAIL;}// Make sure that we have only 1 buffer (we erase the ball in the// old buffer to save having to zero a 200k+ buffer every time// we draw a frame)ASSERT(Actual.cBuffers == 1);return NOERROR;} 

5.如何响应质量控制时事件。需要在输出Pin上实现Notify函数

STDMETHODIMP RtpStream::Notify(IBaseFilter * pSender, Quality q){// Adjust the repeat rate.if (q.Proportion <= 0){m_iRepeatTime = 1000;        // We don't go slower than 1 per second}else{m_iRepeatTime = m_iRepeatTime * 1000 / q.Proportion;if (m_iRepeatTime > 1000){m_iRepeatTime = 1000;    // We don't go slower than 1 per second}else if (m_iRepeatTime<10){m_iRepeatTime = 10;      // We don't go faster than 100/sec}}// skip forwardsif (q.Late > 0)m_rtSampleTime += q.Late;return NOERROR;} // Notify

先到这里，想起来再补充。