这一章讲述了Direct3D的设备对象，它提供了渲染流水线的一个抽象层。流水线的操作是通过设备的属性方法配置的。渲染在设备上的图片是要么2D拷贝操作，要么通过三维场景渲染。

    设备的属性和方法控制了整个图形流水线，属性一般是通过Get和Set方法管理。每个设备都有一组它自己的capabilities。设备也是resource对象的容器，Direct3D通过IDirect3DResource9接口管理资源。图形流水线的很多行为都是通过Render state控制的。应用程序可能需要频繁的设置render states，但是使用state block来保存一组设备属性（包括render state），使应用程序的设置更加方便和容易。

 

IDirect3DDevice9

   

    IDrect3DDevice9控制流水线的行为，它包含大量的属性和方法。下面我们将探讨它的每个属性和方法。device 对象代表了渲染流水线。我们也可以从流水线里取到图片，提供流水线场景数据，然后将场景渲染到图片上。我们能够把流水线分成几个大的section：顶点数据Assembly,顶点处理，primitive Assembly以光栅化，像素处理，frame buffer和Video scan out。IDirect3DDevice9的接口如下：

 

Read-Only Properties
GetAvailableTextureMem	可用的纹理内存，单位是KB
GetBackBuffer	back buffer image surface 接口
GetCreationParameters	设备的创建参数
GetDirect3D	creating IDirect3D9接口
GetDisplayMode	视频显示模式
GetFrontBufferData	front buffer 的拷贝
GetNumberOfSwapChains	Number of swap chain
GetRasterStatus	Raster display status
GetRenderTargetData	surface data in a render target
GetSwapChain	swap chain on the device
Write-only Properties
SetCursorPosition	cursor position
SetCursorProperties	cursor image and hot spot
SetDialogBoxMode	GDI compatability flag
Properties
GetClipPlane	user-defined clipping plane
SetClipPlane
GetClipStatus	clip status of rendered primitives
SetClipStatus
GetCurrentTexturePalette	current texture palette index
SetCurrentTexturePalette
GetDepthStencilSurface	Depth surface 接口
SetDepthStencilSurface
GetFVF	Flexible vertex format
SetFVF
GetGammaRamp	Gamma correction lookup table
SetGammaRamp
GetIndices	current index buffer
SetIndices
GetLight	light definition
SetLight
GetLightEnable	light enabled flag
LightEnable
GetMaterial	Primitive material properties
SetMaterial
GetNPatchMode	N-Patch tessellation mode
SetNPatchMode
GetPaletteEntries	Texture palette entries
SetPaletteEntries
GetPixelShader	pixel shader program handle
SetPixelShader
GetPixelShaderConstantB	Boolean constant registers
SetPixelShaderConstantB
GetPixelShaderConstantF	floating-pont constant registers
SetPixelShaderConstantF
GetPixelShaderConstantI	integer constant registers
SetPixelShaderConstantI
GerRenderState	Pipeline control values
SetRenderState
GetRenderTarget	Target surface for rendered pixels
SetRenderTarget
GetSamplerState	Texture sampler control values
SetSamplerState
GetScissorRect	Scissor test rectangle
SetScissorRect
GetSoftwareVertexProcessing	Vertex processing control
SetSoftwareVertexProcessing
GetStreamSource	source vertex data buffers
SetStreamSource
GetStreamSourceFreq	stream source sampling frequency
SetStreamSourceFreq
GetTexture	Texture resources used by each stage
SetTexture
GetTextureStageState	Texture stage control values
SetTextureStageState
GetTransform	tranformation matrix
SetTransform
GetVertexDeclaration	Vertex component declaration
SetVertexDeclaration
GetVertexShader	vertex shader program handle
SetVertexShader
GetVertexShaderConstantB	boolean constant registers
SetVertexShaderConstantB
GetVertexShaderConstantF	floating-point constant registers
SetVertexShaderConstantF
GetVertexShaderConstantI	Integer constant registers
SetVertexShaderConstantI
GetViewport	Rendering viewport extent
SetViewport
Methods
BeginScene	Make the start of a scene
BeginStateBlock	Make device state for capture
Clear	Clear the current viewport on the device
ColorFill	Fills a rectangular area with a color
CreateAdditionalSwapChain	Create an additional swap chain on a windowed device
CreateCubeTexture	Create a cube map texture sources
CreateDepthStencilSurface	Create a depth/stencil surface resource
CreateIndexBuffer	Create an index buffer resource
CreateOffScreenPlainSurface	Create an off-screen surface resource
CreatePixelShader	Create a pixel shader
CreateQuery	Create a device query
CreateRenderTarget	Create a render target resource
CreateStateBlock	Create a state block
CreateTexture	Create a texture resource
CreateVertexBuffer	Create a vertex buffer resource
CreateVertexDeclaration	Create a vertex declaration
CreateVertexShader	Create a vertex shader
CreateVolumeTexture	Create a volume texture resource
DeletePatch	Destroy a cached patch tesselation
DrawIndexedPrimitive	Draw indexed primitives from the current streams.
DrawIndexedPrimitiveUP	Draw indexed primitives from user data.
DrawPrimitiveUP	Draw primitives from usr data.
DrawRectData	Rectangular higher order surface patch
DrawTriPatch	Triangular higher order surface patch.
EndScene	Make the end of a scene
EndStateBlock	Capture device state changes
EvictManagedResources	Flush managed resources from the device
MultiplyTransform	Post-multiply onto a device transformation matrix
Present	Presents a rendered image for display
ProcessVertices	Process vertices in software
Reset	Resets the device's display characteristcs
ShowCursor	Displays or hides the cursor
StretchRect	Device memory blit with stretching
TestCooperativeLevel	Checks exclusive ownership of the device
UpdateSurface	system to device memory surfaces
UpdateTexture	system to device memory texture transfer.
ValidateDevice	Validate current device state.

 

Device methods by pipeline section

Vertex Assembly
	BeginScene
	CreateIndexBuffer
	CreateVertexBuffer
	DeletePatch
	DrawIndexedPrimitive
	DrawIndexedPrimitiveUP
	DrawPrimitive
	DrawPrimitiveUP
	DrawRectPatch
	DrawTriPatch
	EndScene
	SetIndices
	SetNPatchMode
	SetRenderState
	SetStreamSource
	SetStreamSourceFreq
	SetVertexDeclaration
Vertex Processing	CreateVertexDeclaration
	CreateVertexShader
	LightEnable
	MultiplyTransform
	ProcessVertices
	SetClipPlane
	SetClipStatus
	SetFVF
	SetLight
	SetMaterial
	SetRenderState
	SetSoftwareVertexProcessing
	SetTransform
	SetVertexShader
	SetVertexShaderConstantB
	SetVertexShaderConstantF
	SetVertexShaderConstantI
	SetViewPort
Primitive Rasterization	SetRenderState
Pixel  Processing	CreatePixelShader
	MultiplyTransform
	SetCurrentTexturePalette
	SetPaletteEntries
	SetPixelShader
	SetPixelShaderConstantB
	SetPixelShaderConstantF
	SetPixelShaderConstantI
	SetRenderState
	SetSamplerState
	SetTexture
	SetTextureStageState
	SetTransform
Frame Buffer	Clear
	ColorFill
	CreateAdditionalSwapChain
	CreateCubeTexture
	CreateDepthStencilSurface
	CreateOffscreenPlainSurface
	CreateRenderTarget
	CreateTexture
	CreateVolumeTexture
	EvictManagedResources
	GetAvailableTextureMem
	GetBackBuffer
	GetFrontBufferData
	GetNumberOfSwapChains
	GetRenderTargetData
	GetSwapChain
	Reset
	SetDepthStencilSurface
	SetDialogBoxMode
	SetRenderState
	SetRenderTarget
	SettScissorRect
	SetViewPort
	StretchRect
	UpdateSurface
	UpdateTexture
Video Scan out	GetDisplayMode
	GetRasterStatus
	Present
	SetCursorPosition
	SetCursorProperties
	SetGammaRamp
	ShowCursor

 

Capabilities

 

   一旦创建了设备,某些capabilities允许轻微的变化。 在某个设备上调用GetDeviceCaps可以返回一个真实设备对象的capabilities，但是如果在Direct3D对象上调用GetDeviceCaps则返回设备通用的一些capabilities。

    HRESULT GetDeviceCaps(D3DCAPS9 * value);

typedef struct _D3DCAPS9

{

D3DDEVTYPE   DeviceType;

UINT　AdapterOrdinal;

DWORD Caps;

DWORD Caps2;

DWORD  Caps3;

DWORD   PresentationIntervals;

DWORD  CursorCaps;

DWORD  DevCaps;

DWORD  PrimitiveMiscCaps;

DWORD  RasterCaps;

DWORD  ZCmpCaps;

DWORD  SrcBlendCaps;

DWORD  DestBlendCaps;

DWORD   AlphaCmpCaps;

DWORD  ShadeCaps;

DWORD  TextureCaps;

DWORD  TextureFilterCaps;

DWORD  CubeTextureFilterCaps;

DWORD  VolumeTextureFilterCaps;

DWORD  TextureAddressCaps;

DWORD  VolumeTextureAddressCaps;

DWORD LineCaps;

DWORD  MaxTextureWidth;

DWORD MaxTextureHeight;

DWORD MaxVolumeExtent;

DWORD MaxTextureRepeat;

DWORD MaxTextureAspectRatio;

DWORD MaxAnisotropy;

float   MaxVertexW;

float   GuardBandLeft;

float   GuardBandTop;

float    GuardBandRight;

float   GuardBandBottom;

float   ExtentAdjust;

DWORD StencilCaps;

DWORD  FVFCaps;

DWORD  TextureOpCaps;

DWORD  MaxTextureBlendStages;

DWORD   MaxSimultaneousTextures;

DWORD  VertexProcessingCaps;

DWORD  MaxActiveLights;

DWORD  MaxUserClipPlanes;

DWORD MaxVertexBlendMatrices;

DWORD  MaxVertexBlendMatrixIndex;

float MaxPointSize;

DWORD MaxPrimitiveCount;

DWORD MaxVertexIndex;

DWORD MaxStreams;

DWORD MaxStreamStride;

DWORD VertexShaderVersion;

DWORD MaxVertexShaderConst;

DWORD PixelShaderVersion;

float  PixelShader1XMaxValue;

DWORD DevCaps2;

float MaxNpatchTessellationLevel;

DWORD Reserved5;

UINT  MasterAdapterOrdinal;

UINT  AdapterOrdinalInGroup;

UINT  NumberOfAdaptersInGroup;

UINT DeclTypes;

DWORD NumSimultaneousRTs;

DWORD StretchRectFilterCaps;

D3DVSHADERCAPS2_0 VS20Caps;

D3DPSSHADERCAPS2_0 PS20Caps;

DWORD VertexTextureFilterCaps;

DWORD  MaxVShaderInstructionsExecuted;

DWORD  MaxPShaderInstructionsExecuted;

DWORD  MaxVertexShader30InstructionSlots;

DWORD MaxPixelShader30InstructionsSlots;

}

 

Device Capabilities By pipleline section

Vertex Assembly	Dev Caps:
	Draw Prim TL Vertex
	Quintic RT Patches
	RT Patches
	RT Patch Handle Zero
	TL Vertex System Memory
	TL Vertex Video Memory
	Dev Caps 2:
	Adaptive Tess N Patch
	Adaptive Tess RT Patch
	D Map N Patch
	Stream Offset
	Vertex Elements Can share Stream offset
	Decl Types
	FVF Caps
	Max Primitive Count
	Max Streams
	Max Stream Stride
	Max Vertex Index
	Max N Patch Tessellation Level
Vertex Processing	Dev Caps:
	HW Transform And Light
	Pure Device
	Max Active Lights
	Max User Clip Planes
	Max Vertex Blend Matrices
	Max Vertex Blend Matrices Index
	Max Vertex Shader 30 Instruction Slots
	Max Vertex Shader Const
	Max V Shader Instruction Executed
	Primitive Misc Caps:
	Clip Plane Scaled Points
	Clip TL Verts
	Fog Vertex Clamped
	Raster Caps
	Fog Range
	Fog Vertex
	Vertex Processing Caps
	Vertex Shader Version
	Vertex Texture Filter Caps
	VS 20 Caps
Primitive Processing	Max User Clip Planes
	Primitive Misc Caps:
	Clip Plane Scaled Points
	Clip TL Verts
	Cull CW
	Cull CCW
	Cull None
Primitive  Rasterization	Dev Caps:
	HW Rasterization
	Pure Hardware
	Extents Adjust
	Guard Band Left
	Guard Band Top
	Guard Band Right
	Guard Band Bottom
	Line Caps:
	Antialias
	Texture
	Max Point Size
	Max Vertex W
	Raster Caps:
	Color Perspective
	Shade Caps
	Texture Caps:
	Projected
Pixel Processing	Caps 2:
	Can Auto Gen Mipmap
	Cube Texture Filter Caps
	Dev Caps:
	Separate Texture Memories
	Texture Non Local Vid Mem
	Texture System Memory
	Texture Video Memory
	Max Anisotropy
	Max Pixel Shader 30 Instruction Slots
	Max P Shader Instructions Executed
	Max Simultaneous Textures
	Max Texture Aspect Ratio
	Max Texture Blend Stages
	Max Texture Height
	Max Texture Repeat
	Max Texture Width
	Max Volume Extent
	Pixel Shader Version
	Pixel Shader 1* max Value
	Primitive Misc Caps:
	Per Stage Constant
	TSS Arg Temp
	PS 20 Caps
	Raster Caps
	Anisotropy
	Mipmap LOD Bias
Frame Buffer	Alpha Cmp Caps
	Caps 3:
	Alpha FullScreen Flip or Discard
	Copy To System Mem
	Copy To Vid Mem
	Dest Blend Caps
	Line Caps:
	Alpha Cmp
	Blend
	Fog
	Z    Test
	Num Simultaneous RTS
	Primitive Misc Caps
	Blend Op
	Color Write Enable
	Independent Write Masks
	Mask Z
	MRT Independent Bit Depths
	MRT Post Pixel Shader Blending
	Separate Alpha Blend
Vide Scan out	Adapter Ordinal In Group
	Caps:
	Read Scan line
	Caps 2
	Can Calibrate Gamma
	Full Screen Gamma
	Caps 3
	Linear To sRGB Presentation
	Cursor Caps
	Dev Caps
	Can Render after Flip
	Master Adapter Ordinal
	Number of Adapters in Group
	Presentation Intervals

 

Resources

 

      设备与资源对象协同工作来执行场景渲染。 资源对象通常是场景数据的容器。两维和三维的纹理分别使用surface和volume来存放他们的内容。 back buffer,depth/stencil buffer 和设备的render target属性也是使用surfaces来存放数据。

        纹理接口的继承层次图。

　　基类接口包括IDirect3DVolume9和IDirect3DResource9。Volume对象并不是继承IDirect3DResource9的，所以它并不参与资源管理。剩下的资源都是继承于IDirect3DResource9。另外所有的Texture对象都是继承于IDirect3DBaseTexture9。

 

IDirect3DResource9

　　　　

　　IDirect3DResource9是所有资源对象的基本接口，它给可管理的资源提供了资源内存管理功能。它只能管理创建在D3DPOOL_MANAGED池里面的资源。Managed的资源都有一个整数优先级，数字越大优先级越高。优先级越低的资源就可能先被discarded。 Non-Managed 资源优先级总是0。在同样的优先级，Direct3D总是优先discard那些最近使用最少的资源。在渲染的时候如果应用程序使用的资源内存超过了显卡的资源，Direct3D将使用最近使用最多的策略来释放资源。应用程序也可以显示的调用资源管理器来释放资源，EvictManageResource可以完成这个操作。释放的设备内存不必是连续的，分配一个对象可能会由于设备内存不足而失败，即使资源管理器已经释放了足够的空间。

     创建在D3DPOOL_DEFAULT的资源不能进行管理，并且不能被资源管理器释放。应用程序只能手工的分配，拷贝和释放default pool里面的内存。

    当纹理加载到设备里面是，你可以先使用GetAvailableTextureMem来查询可用纹理内存空间的估计值，然后根据值来调节纹理的detail。这个是值只是一个粗略的估计值不是一个准确的值。

 

interface IDirect3DResource9 : IUnknown

{

  //read -only properties

 

HRESULT GetDevice(IDirect3DDevic9 ** value);

D3DRESOURCETYPE GetType();

 

//read /write properties

DWORD GetPriority();

DWORD  SetPriority();

HRESULT  GetPrivateData(REFGUID　data_guid, void* value, DWORD * size);

HRESULT  SetPrivateData(REFGUID   data_guid, const void * value, DWORD size, DWORD flags);

//methods

HRESULT FreePrivateData(RFEGUID data_guid);

void PreLoad();

}

 

GetDevice方法返回拥有这个资源的设备。资源不能跨设备共享。资源类型可以通过GetType获取。根据这个类型我们就可以知道这个资源IID,然后使用QueryInterface就可以得到具体的资源接口。GetPrivateDat和SetPrivateData可以将任意的数据快绑定到任何Direct3D资源上。

 

Miscellaneous Properties

  

     有几个设备属性与图形流水线没有直接的联系。 GetCreationParameters方法使用D3DDEVICE_CREATION_PARAMETERS结构体返回设备创建的参数。

HRESULT GetCreationParameters(D3DDEVICE_CREATION_PARAMETERS * value);

HRESULT GetDirect3D(IDirect3D** value);

typedef struct _D3DDEVICE_CREATION_PARAMETERS

{

UINT  AdapterOrdinal;

D3DDEVTYPE DeviceType;

HWND  hFocusWindow;

DWORD BehaviorFlags;
}D3DDEVICE_CREATION_PARAMETERS;

创建设备的IDirect3D实例可以通过GetDirect3D获取。

 

设备查询

 

     设备查询就是从设备的驱动层获取信息。主要用途是渲染统计和设备的时间通知。只要设备查询对象和设备对象存在，你就可以向设备查询最新的信息。可以通过CreateQuery创建查询对象,D3DQUERYTPE是查询信息的类别。

 

typedef enum _D3DQUERYTYPE

{

D3DQUERYTYPE_BANDWIDTHTIMINGS = 17,

D3DQUERYTYPE_CACHEUTILIZATION = 18,

D3DQUERYTYPE_EVENT = 8,

D3DQUERYTYPE_INTERFACETIMINGS = 14,

D3DQUERYTYPE_OCCLUSION = 9,

D3DQUERYTYPE_PIPELINETIMINGS = 13,

D3DQUERYTYPE_PIXELTIMINGS = 16,

D3DQUERYTYPE_RESOURCEMANAGER = 5,

D3DQUERYTYPE_TIMESTAMP = 10,

D3DQUERYTYPE_TIMESTAMPFDISJOINT = 11,

D3DQUERYTYPE_TIMESTAMPFREQ = 12,

D3DQUERYTYPE_VCACHE = 4,

D3DQUERYTYPE_VERTEXSTATS = 6,

D3DQUERYTYPE_VERTEXTIMING = 15

} D3DQUERYTYPE;

HRESULT CreateQuery(D3DQUREYTYPE kind, IDirect3DQuery9  ** result);

interface IDirect3DQuery9 : IUnknown

{

//read-only properties

HRESULT GetDevice(IDirect3DDevice9  ** value);

D3DQUERYTYPE GetType();

HRESULT GetData(void * data, DWORD size,DWORD flags);

DWORD GetDataSize();

          //methods

          HRESULT ISSUE(DWORD flags);

}

一个查询对象通常有三种状态：signaled, building, 或者issued。Issue 方法用于将一个状态转换标记为有信号。当设备驱动返回了查询信息时，它也可以改变一个查询对象的状态。查询信息将通过渲染命令流发送给设备。Issue的flags参数将用于标识查询的开始和结束。

#define D3DISSUE_END (1<<0)

#define  D3DISSUE_BEGIN (1<<1)

在查询开始和结束之间，查询对象可以汇报被发布的primitive的情况。顶点cache，资源管理统计和事件查询都有一个隐式的开始，应用程序issue查询对象结束。隐式查询开始如下表。统计查询自从present调用后收集的数据，然而时间和顶点cache查询返回自设备创建后的数据。对于occlusion, disjoint timestamp和timing 查询，应用程序只能显示的指定查询开始和结束。

Query Type	Beginning
Vertex Cache	CreateDevice
Resource Manager Statistics	Present
Vertex Statistics	Present
Event	CreateDevice

     为了取得查询的结果，GetDataSize和GetData方法可以获得查询信息的拷贝。每种查询结果都对应一个数据结构。如下表，

Query Type	Data Type
Bandwidth Timings	D3DDEVINFO_D3D9BANDWIDTHTIMINGS
Cache Utilization	D3DDEVINFO_D3D9CACHEUTILIZATION
Event	BOOL
Interface Timings	D3DDEVINFO_D3D9INTERFACETIMINGS
Occlusion	DWORD
Pipeline Timings	D3DDEVINFO_D3D9INTERFACETIMINGS
Pixel Shader Timings	D3DDEVINFO_D3D9STAGETIMINGS
Resource Manager Statistics	D3DDEVINFO_RESOURCEMANAGER
Timestamp	UINT64
Timestamp Disjoint	BOOL
Timestamp Frequency	UINT64
Vertex Cache	D3DDEVINFO_CACHE
Vertex Shader Timings	D3DDEVINFO_D3D9STAGETIMING
Vertex Statistics	D3DDEVINFO_D3DVERTEXSTATS

GetData的flags参数可以让你同步刷新设备的命令队列。

事件查询可以用来当命令流里的某个特定点被驱动处理的时候通知应用程序，Occlusion查询返回传递到depth测试的像素数目。如果值为0，标识这组primitive全部被obscured。应用程序使用这个信息来cull occlueded物体。

顶点统计和资源管理查询只在debug时支持，retail runtime不支持这两个查询。

 

• 资源管理统计查询

   资源管理统计查询返回一个D3DDEVINFO_RESOURCEMANAGER结构体，它包含了一个D3DRESOURCESTATS结构体数组，每个资源类型对应一个D3DRESOURCESTATS。D3DRTYPECOUNT定义了资源类型数目。

#define D3DRTYPECOUNT (D3DRTYPE_INDEXBUFFER + 1)

typedef struct _D3DDEVINFO_RESOURCEMANAGER

{

D3DRESOURCESTATS  stats[D3DRTYPECOUNT];

} D3DDEVINFO_RESOURCEMANAGER;

   D3DRESOURCESTAS不仅包括自从上次调用Present后的统计数据外，还包含自从设备Reset后保存的统计数据。每个资源的统计数据暗示着D3DPOOL_MANAGED池里的资源的performance。 其他池里面的资源不在这个这个统计数组里面。

typedef struct _D3DRESOURCESTATS

{

// Data collected since last Present()

BOOL　bThrashing;

DWORD ApproxBytesDownloaded;

DWORD NumEvicts;

DWORD NumVidCreates;

DWORD LastPri;

DWORD NumUsed;

DWORD NumUsedVidMem;

 

// Persistent data

DWORD  Workingset;

DWORD  WorkingSetBytes;

DWORD  TotalManaged;

DWORD  TotalBytes;

} D3DRESOURCESTATS;

    Persistent数据设备内存working set和一些汇总数据。Working set成员是video内存资源对象的数目。WorkingSetBytes是working set的大小。TotalManaged 成员是可管理资源对象的数目，TotalBytes是他们总的大小。

  bThrashing成员设置是否资源管理器thrashing。当资源管理器不断的把资源流化到video内存，thrashing 就发生了。例如，假定场景里面有很多物体，每个都使用纹理集合中的一个纹理。如果物体按照纹理排序顺序绘制，极有可能是必需的纹理将会装载到设备内存里面。如果物体绘制的顺序不根据纹理顺序，资源管理器将会花很多时间在working set的不同纹理之间来回thrashing。

  ApproxBytesDownloaded ，NumEvicts和NumVidCreates成员描述了自从上次调用Present后资源管理器的资源消耗行为。LastPri给出了上一次从设备内存里面回收的资源对象的优先级。NumUsed和NumUsedInVidMem分别表示已经设置到设备的资源对象的数目和已经present到设备内存上资源对象的数目。

 

• 顶点统计查询

设备也能统计自上次调用Present以来的顶点处理，如三角形的数目以及clipping后增加的三角形数目。

typedef struct _D3DDEVINFO_D3DVERTEXSTATS

{

DWORD NumRenderedTriangles;

DWORD NumberExtraClippingTriangles;

}D3DDEVINFO_D3DVERTEXSTATS;

• 顶点cache查询

顶点cache查询返回设备上顶点cache的大小等信息。顶点cache是一块临近GPU的内存cache，它避免访问一些最近使用的顶点内存。这个查询返回的是D3DDEVINFO_VCACHE结构体。

typdef struct _D3DDEVINFO_VCACHE

{

DWORD Pattern  //必须是四字符代码CACH

DWORD OptMethod;  //当它的值是0的时候，应用程序必须使用尽可能长的三角形步长来最大化顶点cache。 如果值是1表示应用程序将使用CacheSize大小的顶点Cache来优化。

DWORD CacheSize;

DWORD  MagicNumber;// 当OptMethod 是的时候，它决定了重启三角形步长的时间。
} D3DDEVINFO_VCACHE, *LPD3DDEVINFO_VCACHE;

• PIX 相关查询

   PIX是DirectX应用程序的性能测量工具。余下的查询类型返回性能测量的数据。一个驱动可能不支持所有查询。如果值是0，它将忽略测量的结果。cache利用查询，返回texel和顶点cache的命中率。

  typedef struct _D3DDEVINFO_D3D9CACHEUTILIZATION

{

     float TextureCacheHitRate;

     float PostTransformVertexCacheHitRate;

}D3DDEVINFO_D3D9CACHEUTILIZATION;

 

   所有的时间查询返回都是流水线各个部分的查询时间消耗情况。这个测量反应查询开始和结束之间的百分比。

typedef struct _D3DDEVINFO_D3D9BANDWIDTHTIMINGS

{

float MaxBandwidthUtilized;

float  FrontEndUploadMemoryUtilizedPercent;

float VertexRateUtilizedPercent;

float TriangleSetupRateUtilizedPercent;

float FillRateUtilizedPercent;

}D3DDEVINFO_D3D9BANDWIDTHTIMINGS;

 

typedef struct _D3DDEVINFO_D3D9INTERFACETIMINGS

{

float  WaitingForGPUToUseApplicationResourceTimePercent;

float  WaitingForGPUToAcceptMoreCommandsTimePercent;

float  WaitingForGPUToStayWithinLatencyTimePercent;

float  WaitingForGPUExclusiveResourceTimePercent;

float  WaitingForGPUOtherTimePercent;

} D3DDEVINFO_D3D9INTERFACETIMINGS;

 

typedef struct _D3DDEVINFO_D3D9PIPELINETIMINGS

{

float VertexProcessingTimePercent;

float  PixelProcessingTimePercent;

float  OtherGPUProcessingTimePercent;

float  GPUIdleTimePercent;

} D3DDEVINFO_D3D9PIPELINETIMINGS;

   顶点shader timing和pixel shader timing查询都返回了D3DDEVINFO_D3D9STAGETIMINGS结构体里面的信息。

typdef struct _D3DDEVINFO_D3D9STAGETIMINGS

{

  float  MemoryProcessingPercent;

  float ComputationProcessingPercent;
}D3DDEVINFO_D3D9STAGETIMINGS;

 

设备状态

 

   设备对象的属性控制渲染流水线的行为，它的方法给流水线供应数据。设备的渲染，采样器，纹理阶段的状态属性控制了很多流水线行为，总结如下表：

 

Vertex Assembly	RS Patch Edge Style	RS Adaptive Tess X, Y,Z,W
	RS Position Degree	RS Enable Adaptive Tessellation
	RS Normal Degree	RS Max Tessellation Level
	RS Color Vertex	RS Min  Tessellation Level
	RS Point Size	SS D Map offset
Vertex Processing	RS Tween Factor	RS Clipping
	RS Vertex Blend	RS Indexed Vertex Blend Enable
	TSS Tex Coord Index	RS Ambient Material Source
	RS FOG Vertex Mode	RS Diffuse Material Source
	RS Range Fog Enable	RS Specular Material Source
	RS Ambient	RS Emissive Material Source
	RS Local Viewer	RS Lighting
	RS Specular Enable	RS Normailze Normals
	RS Shade Mode	TSS Texture Transform Flags
	RS Cull Mode	RS Clip Plane Enable
Pixel Processing		SS Border Color
	RS Fill Mode	SS Mag Filter
	RS Last Pixel	SS Min Filter
		SS Mip Filter
	RS Specular Enable	SS Mip Map LOD Bias
	RS Texture Factor	SS Max Mip level
	RS Wrap 0——15	SS Max Anisotrophy
	SS Address U, V, W	TSS Color Arg 0-2
	TSS Color Op	TSS Constant
	TSS Alpha Arg 0-2	RS Fog Table Mode
	TSS Alpha Op	RS Fog Density
	TSS Result Arg	RS Fog End
	TSS Bump Env Mat 00-11	RS Fog Start
	TSS Bump Env L Scale	RS Fog Color
	TSS Bump Env L Offset	RS Fog Enable
	RS Depth Bias	RS Slope Scale Depth Bias
	SS sRGB Texture	RS Antialiased Line Enable
Frame Buffer	RS Alpha Ref	RS Alpha Test Enable
	RS Alpha Func	RS Z Enable
	RS Z  Func
	RS Stencil Enable	RS Stencil Func
	RS Stencil Fail	RS Stencil Z Fail
	RS Stencil Pass	RS Alpha Blend Enable
	RS Src Blend	RS Dest Blend
	RS Blend Op	RS Dither Enable
	RS Color Write Enable	RS Stencil Write Mask
	RS Z Write Enable	RS Multi Sample Antialias
	RS Multi Sample Mask	RS Scissor Test Enable
	RS Src Blend Alpha	RS Separate Alpha Blend Enable
	RS Dest Blend Alpha	RS Blend Op Alpha
	RS Blend Factor	RS CCW Stencil Fail
	RS CCW Stencil Z Fail	RS CCW Stencil Pass
	RS sRGB Write Enable	RS CCW Stencil Func
	SS Element Index	RS Color Write Enable 1-3
		RS Two Sided Stencil Mode

  

注意：RS,SS和TSS 分别表示render, sampler和texture stage states。

Render state, sampler state和纹理阶段state都是32位的数量，并且携带一个名字和值。名字取值于D3DRENDERSTATETYPE, D3DSAMPLERSTATETYPE和D3DTEXTURESTAGESTATETYPE 枚举值。这些属性是通过GetRenderState， SetRenderState, GetSamplerState，SetSamplerState,GetTextureStageState,SetTextureStageState方法进行管理。

HRESULT GetRenderState(D3DRENDERSTATETYPE kind, DWORD * value);

HRESULT SetRenderState(D3DRENDERSTATETYPE kind, DWORD * value);

HRESULT  GetSamplerState(DWORD stage, D3DSAMPLERSTATETPE kind, DWORD * value);

HRESULT SetSamplerState(DWORD stage, D3DSAMPLERSTATETYPE kind, DWORD * value);

HRESULT GetTextureStageState(DWORD stage, D3DTEXTURESTAGESTATETYPE kind, DWORD* value);

HRESUL　SetTextureStageState(DWORD stage, D3DTEXTURESTAGESTATETYPE kind, DWORD value);

 

typedef enum _D3DRENDERSTATETYPE

{

D3DRS_ALPHABLENDENABLE = 27,

D3DRS_ALPHAFUNC = 25,

D3DRS_ALPHAREF = 24,

D3DRS_ALPHATESTENABLE = 15,

D3DRS_AMBIENT  = 139,

D3DRS_AMBIENTMATERIALSOURCE = 147,

D3DRS_BLENDOP = 171,

D3DRS_CLIPPING = 136,

D3DRS_CLIPPLANEENABLE = 152,

D3DRS_COLORWRITEENABLE = 168,

D3DRS_COLORVERTEX = 141,

D3DRS_CULLMODE = 22,

D3DRS_DEBUGMONITORTOKEN= 165，

D3DRS_DESTBLEND=20,

D3DRS_DIFFUSEMATERIALSOURCE = 145,

D3DRS_DITHERENABLE = 26,

D3DRS_EDGEANTIALIAS = 40,

D3DRS_EMISSIVEMATERIALSOURCE = 148,

D3DRS_FILLMODE = 8,

D3DRS_FOGCOLOR = 34,

D3DRS_FOGDENSITY = 38,

D3DRS_FOGENABLE =28,

D3DRS_FOGEND = 37,

D3DRS_FOGSTART = 36,

D3DRS_FOGTABLEMODE = 35,

D3DRS_FOGVERTEXMOE = 140,

D3DRS_INDEXEDVERTEXBLENDENABLE = 167,

D3DRS_LASTPIXEL= 16，

D3DRS_LIGHTING = 137,

D3DRS_LINEPATTERN = 10,

D3DRS_LOCALVIEWER = 142,

D3DRS_NORMALIZENORMALS = 143,

D3DRS_MULTISAMPLEANTIALIAS =  161,

D3DRS_NORMALORDER = 173,

D3DRS_PATCHEDGESTYLE = 163,

D3DRS_PATCHSEGMENTS = 164,

D3DRS_POINTSCALE_A = 158,

D3DRS_POINTSCALE_B = 159,

D3DRS_POINTSCALE_C = 160,

D3DRS_POINTSCALEENABLE= 157,

D3DRS_POINTSIZE = 154,

D3DRS_POINTSIZE_MAX = 166,

D3DRS_POINTSIZE_MIN = 155,

D3DRS_POINTSPRITEENABLE = 156,

D3DRS_POSITIONNOORDER = 172,

D3DRS_RANGEFOGENABLE = 48,

D3DRS_SHADEMODE = 9,

D3DRS_SOFTWAREVERTEXPROCESSING = 153,

D3DRS_SPECULARENABLE = 29,

D3DRS_SPECULARMATERIALSOURCE = 146,

D3DRS_SRCBLEND = 19,

D3DRS_STENCILENABLE = 52,

D3DRS_STENCILFAIL = 53,

D3DRS_STENCILFUNC = 56,

D3DRS_STENCILMASK = 58,

D3DRS_STENCILPASS = 55,

D3DRS_STENCILREF = 57,

D3DRS_STENCILWRITEMASK = 59,

D3DRS_STENCILZFAIL = 54,

D3DRS_TEXTUREFACTOR = 60,

D3DRS_TWEENFACTOR = 170,

D3DRS_VERTEXBLEND = 151,

D3DRS_WRAP0 = 128,

D3DRS_WRAP1 =  129,

D3DRS_WRAP2 = 130,

D3DRS_WRAP3 = 131,

D3DRS_WRAP4 = 132,

D3DRS_WRAP5 = 133,

D3DRS_WRAP6 = 134,

D3DRS_WRAP7 = 136,

D3DRS_ZBIAS = 47,
D3DRS_ZENABLE = 7,

D3DRS_ZFUNC = 23,

D3DRS_ZVISIBLE = 30,

D3DRS_ZWRITEENABLE = 14

} D3DRENDERSTATETYPE;

 

typedef enum _D3DSAMPLERSTATETYPE

{

D3DSAMP_ADDRESSU = 1,

D3DSAMP_ADDRESSV = 2,

D3DSAMP_ADDRESSW = 3,

D3DSAMP_BORDERCOLOR = 4,

D3DSAMP_DMAPOFFSET = 13,

D3DSAMP_ELEMENTINDEX = 12,

D3DSAMP_MAGFILTER = 5,

D3DSAMP_MAXANISOTROPY = 10,

D3DSAMP_MAXMIPLEVEL = 9,

D3DSAMP_MINFILTER = 6,

D3DSAMP_MIPFILTER = 7,

D3DSAMP_MIPMAPLODBIAS = 8,

D3DSAMP_SRGBTEXTURE = 11

} D3DSAMPLERSTATETYPE;

 

typedef enum _D3DTEXTURESTAGESTATETYPE

{

D3DTSS_ADDRESSU = 13,

D3DTSS_ADDRESSV = 14,

D3DTSS_ADDRESSW = 25,

D3DTSS_ALPHAARG0 = 27,

D3DTSS_ALPHAARG1 = 5,

D3DTSS_ALPHAARG2 = 6,

D3DTSS_ALPHAOP = 4,

D3DTSS_BORDERCOLOR = 15,

D3DTSS_BUMPENVLOFFSET = 23,

D3DTSS_BMPENVLSCALE = 22,

D3DTSS_BUMPENVMAT00 = 7,

D3DTSS_BUMPENVMAT01 = 8,

D3DTSS_BUMPENVMAT10 = 9,

D3DTSS_BUMPENVMAT11 = 10,

D3DTSS_COLORARG0 = 26,

D3DTSS_COLORARG1 = 2,

D3DTSS_COLORARG2 = 3,

D3DTSS_COLOROP = 1,

D3DTSS_MAGFILTER = 16,

D3DTSS_MAXANISOTROPY = 21,

D3DTSS_MAXMIPLEVEL = 20,

D3DTSS_MINFILTER = 17,

D3DTSS_MIPFILTER = 18,

D3DTSS_MIPMAPLODBIAS = 19,

D3DTSS_RESULTARG = 28,

D3DTSS_TEXCOORDINDEX = 11,

D3DTSS_TEXTURETRANSFORMFLAGS = 24

}

 

  所有的状态值都是32位，API把他们当做成DWORD。但是每个渲染，采样和纹理阶段状态值将绑定到一个32为的数据类型，但是它不一定是DWORD。这些类型的大部分都是枚举值，但是有些状态值是浮点数。因为Direct3D使用了IEEE的单精度浮点值，它也是32位的。但是当调用属性方法的时候，他必须转换到DWORD。一般使用reinterpret_cast方法来进行转换。

Device State	Data Type
RS Adaptive Tess X, Y,Z, W	float
RS Alpha Blend Enable	BOOL
RS ALPHA Func	D3DCMPFUNC
RS ALPHA Ref	[0,255]
RS ALPHA Test Enable	BOOL
RS Ambient	D3DCOLOR
RS Ambient Material Source	D3DMATERIALCOLORSOURCE
RS Antialiased Line Enable	BOOL
RS Blend Factor	float
RS Blend Op	D3DBLENDOP
RS Blend Op Alpha	D3DBLENDOP
RS CCW Stencil Fail	D3DSTENCILOP
RS CCW Stencil Func	D3DCMPFUNC
RS CCW Stencil Pass	D3DSTENCILOP
RS CCW Stencil Z Fail	D3DSTENCILOP
RS Clipping	BOOL
RS Clip Plane Enable	BOOL
RS Color Write Enable	D3DCOLORWRITEENABLE
RS Color Write Enable 1-3	D3DCOLORWRITEENABLE
RS Color Vertex	BOOL
RS Cull Mode	D3DCULL
RS Debug  Monitor Token	D3DDEBUGMONITORTOKENS
RS Depth Bias	float
RS Dest Blend	D3DBLEND
RS Dest Blend Alpha	D3DBlEND
RS Diffuse material Source	D3DMATERIALCOLORSOURCE
RS Dither Enable	BOOL
RS Emissive Material Source	D3DMATERIALCOLORSOURCE
RS Enable Adpative Tessellation	BOOL
RS Fill Mode	D3DFILLMODE
RS Fog Color	D3DCOLOR
RS Fog Density	float
RS Fog Enable	BOOL
RS FOG END	float
RS FOG Start	float
RS FOG Table Mode	D3DFOGMODE
RS FOG Vertex Mode	D3DFOGMODE
RS Indexed Vertex Blend Enable	BOOL
RS Last Pixel	BOOL
RS Lighting	BOOL
RS Local Viewer	BOOL
RS Max Tessellation Level	DWORD
RS Min Tessellation Level	DWORD
RS Normal Degree	D3DDEGREETYPE
RS Normalize Normals	BOOL
RS Multi Sample Antialias	BOOL
RS Multi Sample Mask	[0,2~m-1]
RS Patch Edge Style	D3DPATCHEDGESTYLE
RS Point Scale A, B, C	float
RS Point Scale Enable	BOOL
RS Point Size	float
RS Point Size Max	float
RS Point Size min	float
RS Point Sprite Enable	BOOL
RS Position Degree	D3DDEGREETYPE
RS Range Fog Enable	BOOL
RS Scissor Test Enable	BOOL
RS Separate Alpha Blend Enable	BOOL
RS Shade Mode	D3DSHADEMODE
RS Slope Scale Depth  Bias	float
RS Specular Enable	BOOL
RS Specular Material Source	D3DMATERIALCOLORSOURCE
RS Src Blend	D3DBLEND
RS Src Blend Alpha	D3DBLEND
RS sRGB Write Enable	BOOL
RS Stencil Enable	BOOL
RS Stencil Fail	D3DSTENCILOP
RS  Stencil Mask	[0,2~s -1]
RS Stencil Pass	D3DSTENCILOP
RS Stencil Ref	[0,2~s -1]
RS Stencil Write Mask	[0,2~s-1]
RS Stencil Z Fail	D3DSTENCILOP
RS Texture Factor	D3DCOLOR
RS Tween Factor	float
RS Two Sided Stencil Mode	BOOL
RS Vertex Blend	D3DVERTEXBLENDFLAGS
RS Wrap 0 - 15	DEDWRAPCOORD
RS Z Enable	D3DZBUFFERTYPE
RS Z Func	D3DCMPFUNC
RS Z Write Enable	BOOL
SS Address U, V,W	D3DTEXTUREADDRESS
SS Border Color	D3DCOLOR
SS D Map Offset	DWORD
SS ELEMENT Index	DWORD
SS Mag Filter	D3DTEXTUREFILTERTYPE
SS Max Anisotropy	DWORD
SS Max Mip Level	[0,l-1]
SS Min Filter	D3DTEXTUREFILTERTYPE
SS Mip Filter	D3DTEXTUREFILTERTYPE
SS Mip Map LOD Bias	float
SS sRGB Texture	BOOL
TSS Alpha Arg 0 -2	D3DTEXTUREARG
TSS ALPHA Op	D3DTEXTUREOP
TSS BUMP Env L Offset	float
TSS Bump Env L Scale	float
TSS Color Arg 0-2	D3DTEXTUREARG
TSS Color OP	D3DTEXTUREOP
TSS Constant	float
TSS Result Arg	D3DTextureARG
TSS Tex Coord Index	[0,15] D3DTSS_TCI
TSS Texture Transform Flags	D3DTEXTURETRANSFORMFLAGS

 

• Miscellaneous Render state

   RS Debug Monitor Token 并不能控制渲染，也并不与流水线的某个具体环节有任何关系。RS Debug Monitor Token 控制了当使用debug runtime时从流水线返回的调试信息。Debug token是通过D3DDEBUGMONITORTOKENS枚举定义的。

typedef enum _D3DDEBUGMONITORTOKENS

{

D3DDMT_ENABLE = 0,

D3DDMT_DISABLE = 1

} D3DDEBUGMONITORTOKENS;

 

• Setting group of State

   通常在Direct3D应用程序里面，有一组渲染和纹理阶段状态需要设置。如果写每个SetRenderState和SetTextureStageState太冗长。 我们能使用group of state减少单个的调用。

 

Device State Blocks

   状态block都是com对象， 为你的应用程序提供了一种方法cache一组设备属性为以后使用。比如， 两个状态block可以分用于编辑器里未选中和已选中物体的外表。IDirect3DStateBlock9用于管理状态block。每个状态block都关联一个设备。一旦一个状态block被创建，Apply方法将设备属性应用到设备。Capture捕捉state block的当前值。

interface IDirect3DStateBlock9: IUnknown

{

HRESULT GetDevice(IDirect3DDevice ** value);

 

HRESULT Apply();

HRESULT Capture();
}

   有两种方法创建一个状态block， 然后使用特定的属性值来填充它。创建State block的第一种方法是调用CreateStateBlock。

HRESULT CreateStateBlock(D3DSTATEBLOCKTYPE kind, IDirect3DStateBlock9 ** result);

typedef enum _D3DSTATEBLOCKTYPE

{

D3DSBT_ALL = 1,

D3DSBT_PIXELSTATE = 2,

D3DSBT_VERTEXSTATE = 3

}D3DSTATEBLOCKTYPE;

   获取一个状态block对象的第二种方法是调用BeginStateBlock，设置设备属性，然后调用EndStateBlock。一旦BeginStateBlock调用以后，下面的方法可以被它捕捉他们的状态。

   LightEnable, SetSamplerState, SetClipPlane, SetStreamSource, SetIndices, SetTexture, SetLight, SetTextureStageState, SetMaterial, SetTransform, SetPixelShader, SetVertexShader, SetPixelShaderConstantB, SetVertexShaderConstantB, SetPixelShaderConstantF, SetVertexShaderConstantF, SetPixelShaderConstantI, SetVertexShaderConstantI, SetRenderState, SetViewPort。

  当设备重新获得的时候，状态block需要销毁然后重建。

 

Pure Devices

 

   当创建设备的时候，我们可以请求一个pure的硬件设备，它执行更少的设备状态管理。Pure设备将会有更好的性能优势，因为runtime和driver并不比保存一份non-queryable状态。如果D3DCAPS::DevCaps的D3DDEVCAPS_PUREDEVICE设置，它将支持一个pure设备。