HRESULT CPushPinDesktop::FillBuffer(IMediaSample *pSample)
{
	LocalOutput("video frame requested");

	__int64 startThisRound = StartCounter();
	BYTE *pData;

    CheckPointer(pSample, E_POINTER);
	if(m_bReReadRegistry) {
	  reReadCurrentStartXY(1);
	}

	
	if(!ever_started) {
		// allow it to startup until Run is called...so StreamTime can work see http://stackoverflow.com/questions/2469855/how-to-get-imediacontrol-run-to-start-a-file-playing-with-no-delay/2470548#2470548
		// since StreamTime anticipates that the graph's start time has already been set
		FILTER_STATE myState;
		CSourceStream::m_pFilter->GetState(INFINITE, &myState);
		while(myState != State_Running) {
		  // TODO accomodate for pausing better, we're single run only currently [does VLC do pausing even?]
		  Sleep(1);
		  LocalOutput("sleeping till graph running for audio...");
		  m_pParent->GetState(INFINITE, &myState);	  
		}
		ever_started = true;
	}


    // Access the sample's data buffer
    pSample->GetPointer(&pData);

    // Make sure that we're still using video format
    ASSERT_RETURN(m_mt.formattype == FORMAT_VideoInfo);

    VIDEOINFOHEADER *pVih = (VIDEOINFOHEADER*) m_mt.pbFormat;

	boolean gotNew = false; // dedupe stuff
	while(!gotNew) {

      CopyScreenToDataBlock(hScrDc, pData, (BITMAPINFO *) &(pVih->bmiHeader), pSample);
	
	  if(m_bDeDupe) {
			if(memcmp(pData, pOldData, pSample->GetSize())==0) { // took desktop:  10ms for 640x1152, still 100 fps uh guess...
			  Sleep(m_millisToSleepBeforePollForChanges);
			} else {
			  gotNew = true;
			  memcpy( /* dest */ pOldData, pData, pSample->GetSize()); // took 4ms for 640x1152, but it's worth it LOL.
			  // LODO memcmp and memcpy in the same loop LOL.
			}
	  } else {
		// it's always new for everyone else (the typical case)
	    gotNew = true;
	  }
	}

	// capture some debug stats (how long it took) before we add in our own arbitrary delay to enforce fps...
	long double millisThisRoundTook = GetCounterSinceStartMillis(startThisRound);
	fastestRoundMillis = min(millisThisRoundTook, fastestRoundMillis); // keep stats :)
	sumMillisTook += millisThisRoundTook;

	CRefTime now;
	CRefTime endFrame;
	now = 0;
	CSourceStream::m_pFilter->StreamTime(now);
	if((now > 0) && (now < previousFrameEndTime)) { // now > 0 to accomodate for if there is no reference graph clock at all...also at boot strap time to ignore it XXXX can negatives even ever happen anymore though?
		while(now < previousFrameEndTime) { // guarantees monotonicity too :P
		  LocalOutput("sleeping because %llu < %llu", now, previousFrameEndTime);
		  Sleep(1);
          CSourceStream::m_pFilter->StreamTime(now);
		}
		// avoid a tidge of creep since we sleep until [typically] just past the previous end.
		endFrame = previousFrameEndTime + m_rtFrameLength;
	    previousFrameEndTime = endFrame;
	    
	} else {
		// if there's no reference clock, it will "always" think it missed a frame
	  if(show_performance) {
		  if(now == 0) 
			  LocalOutput("probable none reference clock, streaming fastly");
		  else
	          LocalOutput("it missed a frame--can't keep up %d %llu %llu", countMissed++, now, previousFrameEndTime); // we don't miss time typically I don't think, unless de-dupe is turned on, or aero, or slow computer, buffering problems downstream, etc.
	  }
	  // have to add a bit here, or it will always be "it missed a frame" for the next round...forever!
	  endFrame = now + m_rtFrameLength;
	  // most of this stuff I just made up because it "sounded right"
	  //LocalOutput("checking to see if I can catch up again now: %llu previous end: %llu subtr: %llu %i", now, previousFrameEndTime, previousFrameEndTime - m_rtFrameLength, previousFrameEndTime - m_rtFrameLength);
	  if(now > (previousFrameEndTime - (long long) m_rtFrameLength)) { // do I even need a long long cast?
		// let it pretend and try to catch up, it's not quite a frame behind
        previousFrameEndTime = previousFrameEndTime + m_rtFrameLength;
	  } else {
		endFrame = now + m_rtFrameLength/2; // ?? seems to not hurt, at least...I guess
		previousFrameEndTime = endFrame;
	  }
	    
	}

	// accomodate for 0 to avoid startup negatives, which would kill our math on the next loop...
	previousFrameEndTime = max(0, previousFrameEndTime); 

    pSample->SetTime((REFERENCE_TIME *) &now, (REFERENCE_TIME *) &endFrame);
	//pSample->SetMediaTime((REFERENCE_TIME *)&now, (REFERENCE_TIME *) &endFrame); 
    LocalOutput("timestamping video packet as %lld -> %lld", now, endFrame);

    m_iFrameNumber++;

	// Set TRUE on every sample for uncompressed frames http://msdn.microsoft.com/en-us/library/windows/desktop/dd407021%28v=vs.85%29.aspx
    pSample->SetSyncPoint(TRUE);

	// only set discontinuous for the first...I think...
	pSample->SetDiscontinuity(m_iFrameNumber <= 1);

#ifdef _DEBUG
    // the swprintf costs like 0.04ms (25000 fps LOL)
	double m_fFpsSinceBeginningOfTime = ((double) m_iFrameNumber)/(GetTickCount() - globalStart)*1000;
	swprintf(out, L"done video frame! total frames: %d this one %dx%d -> (%dx%d) took: %.02Lfms, %.02f ave fps (%.02f is the theoretical max fps based on this round, ave. possible fps %.02f, fastest round fps %.02f, negotiated fps %.06f), frame missed %d", 
		m_iFrameNumber, m_iCaptureConfigHeight, m_iCaptureConfigWidth, getNegotiatedFinalWidth(), getNegotiatedFinalHeight(), millisThisRoundTook, m_fFpsSinceBeginningOfTime, 1.0*1000/millisThisRoundTook,   
		/* average */ 1.0*1000*m_iFrameNumber/sumMillisTook, 1.0*1000/fastestRoundMillis, GetFps(), countMissed);
	LocalOutput(out);
	set_config_string_setting(L"frame_stats", out);
#endif
    return S_OK;
}
HRESULT CPushPinDesktop::FillBuffer(IMediaSample *pSample)
{
	__int64 startThisRound = StartCounter();
	BYTE *pData;

    CheckPointer(pSample, E_POINTER);
	if(m_bReReadRegistry) {
	  reReadCurrentPosition(1);
	}

    // Access the sample's data buffer
    pSample->GetPointer(&pData);

    // Make sure that we're still using video format
    ASSERT(m_mt.formattype == FORMAT_VideoInfo);

    VIDEOINFOHEADER *pVih = (VIDEOINFOHEADER*) m_mt.pbFormat;

	// for some reason the timings are messed up initially, as there's no start time at all for the first frame (?) we don't start in State_Running ?
	// race condition?
	// so don't do some calculations unless we're in State_Running
	FILTER_STATE myState;
	CSourceStream::m_pFilter->GetState(INFINITE, &myState);
	bool fullyStarted = myState == State_Running;
	
	boolean gotNew = false;
	while(!gotNew) {

      CopyScreenToDataBlock(hScrDc, pData, (BITMAPINFO *) &(pVih->bmiHeader), pSample);
	
	  if(m_bDeDupe) {
			if(memcmp(pData, pOldData, pSample->GetSize())==0) { // took desktop:  10ms for 640x1152, still 100 fps uh guess...
			  Sleep(m_millisToSleepBeforePollForChanges);
			} else {
			  gotNew = true;
			  memcpy( /* dest */ pOldData, pData, pSample->GetSize()); // took 4ms for 640x1152, but it's worth it LOL.
			  // LODO memcmp and memcpy in the same loop LOL.
			}
	  } else {
		// it's always new for everyone else!
	    gotNew = true;
	  }
	}
	// capture how long it took before we add in our own arbitrary delay to enforce fps...
	long double millisThisRoundTook = GetCounterSinceStartMillis(startThisRound);
	fastestRoundMillis = min(millisThisRoundTook, fastestRoundMillis); // keep stats :)
	sumMillisTook += millisThisRoundTook;

	CRefTime now;
	CRefTime endFrame;
    CSourceStream::m_pFilter->StreamTime(now);

    // wait until we "should" send this frame out...
	if((now > 0) && (now < previousFrameEndTime)) { // now > 0 to accomodate for if there is no reference graph clock at all...also boot strap time ignore it :P
		while(now < previousFrameEndTime) { // guarantees monotonicity too :P
		  Sleep(1);
          CSourceStream::m_pFilter->StreamTime(now);
		}
		// avoid a tidge of creep since we sleep until [typically] just past the previous end.
		endFrame = previousFrameEndTime + m_rtFrameLength;
	    previousFrameEndTime = endFrame;
	    
	} else {
	  if(show_performance)
	    LocalOutput("it missed a frame--can't keep up %d", countMissed++); // we don't miss time typically I don't think, unless de-dupe is turned on, or aero, or slow computer, buffering problems downstream, etc.
	  // have to add a bit here, or it will always be "it missed some time" for the next round...forever!
	  endFrame = now + m_rtFrameLength;
	  // most of this stuff I just made up because it "sounded right"
	  //LocalOutput("checking to see if I can catch up again now: %llu previous end: %llu subtr: %llu %i", now, previousFrameEndTime, previousFrameEndTime - m_rtFrameLength, previousFrameEndTime - m_rtFrameLength);
	  if(now > (previousFrameEndTime - (long long) m_rtFrameLength)) { // do I need a long long cast?
		// let it pretend and try to catch up, it's not quite a frame behind
        previousFrameEndTime = previousFrameEndTime + m_rtFrameLength;
	  } else {
		endFrame = now + m_rtFrameLength/2; // ?? seems to work...I guess...
		previousFrameEndTime = endFrame;
	  }
	    
	}
	previousFrameEndTime = max(0, previousFrameEndTime);// avoid startup negatives, which would kill our math on the next loop...
    
	// LocalOutput("marking frame with timestamps: %llu %llu", now, endFrame);
    pSample->SetTime((REFERENCE_TIME *) &now, (REFERENCE_TIME *) &endFrame);
	//pSample->SetMediaTime((REFERENCE_TIME *)&now, (REFERENCE_TIME *) &endFrame); //useless seemingly

	if(fullyStarted) {
      m_iFrameNumber++;
	}

	// Set TRUE on every sample for uncompressed frames http://msdn.microsoft.com/en-us/library/windows/desktop/dd407021%28v=vs.85%29.aspx
    pSample->SetSyncPoint(TRUE);

	// only set discontinuous for the first...I think...
	pSample->SetDiscontinuity(m_iFrameNumber <= 1);

    // the swprintf costs like 0.04ms (25000 fps LOL)
	m_fFpsSinceBeginningOfTime = ((double) m_iFrameNumber)/(GetTickCount() - globalStart)*1000;
	swprintf(out, L"done frame! total frames: %d this one %dx%d -> (%dx%d) took: %.02Lfms, %.02f ave fps (%.02f is the theoretical max fps based on this round, ave. possible fps %.02f, fastest round fps %.02f, negotiated fps %.06f), frame missed %d", 
		m_iFrameNumber, m_iCaptureConfigHeight, m_iCaptureConfigWidth, getNegotiatedFinalWidth(), getNegotiatedFinalHeight(), millisThisRoundTook, m_fFpsSinceBeginningOfTime, 1.0*1000/millisThisRoundTook,   
		/* average */ 1.0*1000*m_iFrameNumber/sumMillisTook, 1.0*1000/fastestRoundMillis, GetFps(), countMissed);
//#ifdef _DEBUG // probably not worth it but we do hit this a lot...hmm...
	LocalOutput(out);
	set_config_string_setting(L"frame_stats", out);
//#endif
    return S_OK;
}