How to create a Simple Lock Framework for C++ Synchronization

Daywei — Tue, 21 Aug 2012 06:37:00 GMT

Daywei 2012-08-21 14:37 发表评论

Daywei — Fri, 23 Dec 2011 09:45:00 GMT


BOOL CD3DCameraViewWnd::GetImageInfo(DWORD* pPixelBuffer, int& width,int& height)
{
	BOOL result = FALSE;

	IDirect3DSurface9 *_surfaceback = NULL; 
	if (FAILED(m_pDevice->GetBackBuffer(0, 0, D3DBACKBUFFER_TYPE_MONO, &_surfaceback)))
	{
		result = FALSE;
	}
	  

	D3DSURFACE_DESC surfaceDesc; 
	if (!FAILED(_surfaceback->GetDesc(&surfaceDesc)))
	{
		width = surfaceDesc.Width;
		height = surfaceDesc.Height;
	}

	D3DLOCKED_RECT lockedRect; 
	HRESULT hr = _surfaceback->LockRect( &lockedRect,0, 0); /* pointer to receive locked data指向甌���到的内存区域*/ /* lock entire surface*/// no lock flags specified 

	if (!FAILED(hr))
	{
		pPixelBuffer = (DWORD*)lockedRect.pBits;
		result = TRUE;
	}

	_surfaceback->UnlockRect();
	_surfaceback->Release();

	return result;
}

但是�q�样存在着性能的问题，�l�测试获取一帧需�?00ms左右�Q��ؓ什么会存在�q�样的问题？因�ؓ�q�边有资源锁定操作�?

现在关键的问题就在于,资源锁定的操作速度��L��非常之慢.当然,你会跟我�? OpenGL��g��能够快捷地完成这��Q�?但是,在Direct3D�?资源锁定操作��实是很慢的.�q�里面一个主要的原因�?API,驱动,以及��g要处理一些不可回避的后台操作.那就是GPU与CPU是�ƈ行运行的,若不加�Q何措�?��引��L��似多�U�程�E�序同步时的竞态条件的问题.

如果你试囑֎�修改的资源正同时被一个位于GPU处理序列中的指��o使用,那么整个渲染��程��׃��因�ؓ你的资源锁定而停��或强制��h��(stalls and flushes).停顿(stall)会一直持�l�到你完成了对资源的修改�q�调用Unlock().而强制刷�?flush)则会要求GPU在你得到�q�个资源的访问权之前完成目前所有的��d��.

如何去解册��个问题？下面的参考资料中有一些解��x��案，我没有经�q�认真测试，我试了其中一个GetRenderTargetData �q�样的一�U�方�?�Q�感觉不太好用，why��L��Reference3

下面�l�出我的解决�Ҏ(gu��)��

BOOL CD3DCameraViewWnd::GetImageInfo(DWORD* pPixelBuffer, int& width,int& height)
{
	BOOL result = FALSE;

	IDirect3DSurface9 *_surfaceback = NULL; 
	if (FAILED(m_pDevice->GetBackBuffer(0, 0, D3DBACKBUFFER_TYPE_MONO, &_surfaceback)))
	{
		result = FALSE;
	}
	  

	D3DSURFACE_DESC surfaceDesc; 
	if (!FAILED(_surfaceback->GetDesc(&surfaceDesc)))
	{
		width = surfaceDesc.Width;
		height = surfaceDesc.Height;
	}

	LPDIREC3DSURFACE9 surf;

	if(FAILED(m_pDevice->CreateOffscreenPlainSurface(width, height,D3DFMT_A8R8G8B8,D3DPOOL_SYSTEMMEM, &surf,NULL)))

		result = FALSE;

	D3DXLoadSurfaceFromSurface(surf, NULL,NULL,_surfaceback,NULL,NULL,D3DX_FILTER_NONE,0);

	D3DLOCKED_RECT lockedRect; 
	HRESULT hr = surf->LockRect( &lockedRect,0, 0); /* pointer to receive locked data指向甌���到的内存区域*/ /* lock entire surface*/// no lock flags specified 

	if (!FAILED(hr))
	{
		pPixelBuffer = (DWORD*)lockedRect.pBits;
		result = TRUE;
	}

	surf->UnlockRect();

	surf->Release();
	_surfaceback->Release();

	return result;
}

参考资料：Reference1.http://www.cnblogs.com/mixiyou/archive/2010/02/25/1673060.html

Reference 2.http://www.cnblogs.com/mixiyou/archive/2010/02/25/1673425.html

Reference 3.http://blog.csdn.net/Nightmare/article/details/1707362

Reference 4.http://www.cnblogs.com/lancidie/archive/2011/3/14.html

Daywei 2011-12-23 17:45 发表评论

SQLite C++ Class

Daywei — Thu, 17 Nov 2011 07:11:00 GMT

摘要: Code highlighting produced by Actipro CodeHighlighter (freeware)http://www.CodeHighlighter.com/-->////////////////////////////////////////////////////////////////////////////////// CppSQLite3&nbs... 阅读全文

Daywei 2011-11-17 15:11 发表评论

欧美午夜精品电影,亚洲第一网站免费视频,亚洲欧美一级二级三级

How to create a Simple Lock Framework for C++ Synchronization

SQLite C++ Class