Direct3D对加载到内存中的网格模型提供了优化功能，通过网格模型优化可以明显提高三维模型的渲染速度，这对渲染速度要求较高的三维图形程序和游戏具有非常重要的现实意义。

对于程序员而言，对网格模型进行优化是比较简单的，只需调用接口ID3DXMesh的方法Optimize()，该函数的声明如下：

Generates a new mesh with reordered faces and vertices to optimize drawing performance.

HRESULT Optimize(
  DWORD Flags,
  CONST DWORD * pAdjacencyIn,
  DWORD * pAdjacencyOut,
  DWORD * pFaceRemap,
  LPD3DXBUFFER * ppVertexRemap,
  LPD3DXMESH * ppOptMesh
);

Parameters

Flags

[in] Specifies the type of optimization to perform. This parameter can be set to a combination of one or more flags from D3DXMESHOPT and D3DXMESH (except D3DXMESH_32BIT, D3DXMESH_IB_WRITEONLY, and D3DXMESH_WRITEONLY).

pAdjacencyIn

[in] Pointer to an array of three DWORDs per face that specifies the three neighbors for each face in the source mesh. If the edge has no adjacent faces, the value is 0xffffffff. See Remarks.

pAdjacencyOut

[in, out] Pointer to an array of three DWORDs per face that specifies the three neighbors for each face in the optimized mesh. If the edge has no adjacent faces, the value is 0xffffffff.

pFaceRemap

[in, out] An array of DWORDs, one per face, that identifies the original mesh face that corresponds to each face in the optimized mesh. If the value supplied for this argument is NULL, face remap data is not returned.

ppVertexRemap

[out] Address of a pointer to an ID3DXBuffer interface, which contains a DWORD for each vertex that specifies how the new vertices map to the old vertices. This remap is useful if you need to alter external data based on the new vertex mapping.

ppOptMesh

[out] Address of a pointer to an ID3DXMesh interface, representing the optimized mesh.

Return Values

If the method succeeds, the return value is D3D_OK. If the method fails, the return value can be one of the following: D3DERR_INVALIDCALL, E_OUTOFMEMORY.

Remarks

This method generates a new mesh. Before running Optimize, an application must generate an adjacency buffer by calling ID3DXBaseMesh::GenerateAdjacency. The adjacency buffer contains adjacency data, such as a list of edges and the faces that are adjacent to each other.

This method is very similar to the ID3DXBaseMesh::CloneMesh method, except that it can perform optimization while generating the new clone of the mesh. The output mesh inherits all of the creation parameters of the input mesh.

Direct3D提供了7种网格模型优化方式，由枚举常量D3DXMESHOPT定义：

Specifies the type of mesh optimization to be performed.

typedef enum D3DXMESHOPT
{
    D3DXMESHOPT_COMPACT = 0x01000000,
    D3DXMESHOPT_ATTRSORT = 0x02000000,
    D3DXMESHOPT_VERTEXCACHE = 0x04000000,
    D3DXMESHOPT_STRIPREORDER = 0x08000000,
    D3DXMESHOPT_IGNOREVERTS = 0x10000000,
    D3DXMESHOPT_DONOTSPLIT = 0x20000000,
    D3DXMESHOPT_DEVICEINDEPENDENT = 0x40000000,
} D3DXMESHOPT, *LPD3DXMESHOPT;

Constants

D3DXMESHOPT_COMPACT

Reorders faces to remove unused vertices and faces.

D3DXMESHOPT_ATTRSORT

Reorders faces to optimize for fewer attribute bundle state changes and enhanced ID3DXBaseMesh::DrawSubset performance.

D3DXMESHOPT_VERTEXCACHE

Reorders faces to increase the cache hit rate of vertex caches.

D3DXMESHOPT_STRIPREORDER

Reorders faces to maximize length of adjacent triangles.

D3DXMESHOPT_IGNOREVERTS

Optimize the faces only; do not optimize the vertices.

D3DXMESHOPT_DONOTSPLIT

While attribute sorting, do not split vertices that are shared between attribute groups.

D3DXMESHOPT_DEVICEINDEPENDENT

Affects the vertex cache size. Using this flag specifies a default vertex cache size that works well on legacy hardware.

Remarks

The D3DXMESHOPT_STRIPREORDER and D3DXMESHOPT_VERTEXCACHE optimization flags are mutually exclusive.

The D3DXMESHOPT_SHAREVB flag has been removed from this enumeration. Use D3DXMESH_VB_SHARE instead, in D3DXMESH.

D3DXMESH

Flags used to specify creation options for a mesh.

typedef enum D3DXMESH
{
    D3DXMESH_32BIT = 0x001,
    D3DXMESH_DONOTCLIP = 0x002,
    D3DXMESH_POINTS = 0x004,
    D3DXMESH_RTPATCHES = 0x008,
    D3DXMESH_NPATCHES = 0x4000,
    D3DXMESH_VB_SYSTEMMEM = 0x010,
    D3DXMESH_VB_MANAGED = 0x020,
    D3DXMESH_VB_WRITEONLY = 0x040,
    D3DXMESH_VB_DYNAMIC = 0x080,
    D3DXMESH_VB_SOFTWAREPROCESSING = 0x8000,
    D3DXMESH_IB_SYSTEMMEM = 0x100,
    D3DXMESH_IB_MANAGED = 0x200,
    D3DXMESH_IB_WRITEONLY = 0x400,
    D3DXMESH_IB_DYNAMIC = 0x800,
    D3DXMESH_IB_SOFTWAREPROCESSING = 0x10000,
    D3DXMESH_VB_SHARE = 0x1000,
    D3DXMESH_USEHWONLY = 0x2000,
    D3DXMESH_SYSTEMMEM = 0x110,
    D3DXMESH_MANAGED = 0x220,
    D3DXMESH_WRITEONLY = 0x440,
    D3DXMESH_DYNAMIC = 0x880,
    D3DXMESH_SOFTWAREPROCESSING = 0x18000,
} D3DXMESH, *LPD3DXMESH;

Constants

D3DXMESH_32BIT

The mesh has 32-bit indices instead of 16-bit indices. See Remarks.

D3DXMESH_DONOTCLIP

Use the D3DUSAGE_DONOTCLIP usage flag for vertex and index buffers.

D3DXMESH_POINTS

Use the D3DUSAGE_POINTS usage flag for vertex and index buffers.

D3DXMESH_RTPATCHES

Use the D3DUSAGE_RTPATCHES usage flag for vertex and index buffers.

D3DXMESH_NPATCHES

Specifying this flag causes the vertex and index buffer of the mesh to be created with D3DUSAGE_NPATCHES flag. This is required if the mesh object is to be rendered using N-patch enhancement using Direct3D.

D3DXMESH_VB_SYSTEMMEM

Use the D3DPOOL_SYSTEMMEM usage flag for vertex buffers.

D3DXMESH_VB_MANAGED

Use the D3DPOOL_MANAGED usage flag for vertex buffers.

D3DXMESH_VB_WRITEONLY

Use the D3DUSAGE_WRITEONLY usage flag for vertex buffers.

D3DXMESH_VB_DYNAMIC

Use the D3DUSAGE_DYNAMIC usage flag for vertex buffers.

D3DXMESH_VB_SOFTWAREPROCESSING

Use the D3DUSAGE_SOFTWAREPROCESSING usage flag for vertex buffers.

D3DXMESH_IB_SYSTEMMEM

Use the D3DPOOL_SYSTEMMEM usage flag for index buffers.

D3DXMESH_IB_MANAGED

Use the D3DPOOL_MANAGED usage flag for index buffers.

D3DXMESH_IB_WRITEONLY

Use the D3DUSAGE_WRITEONLY usage flag for index buffers.

D3DXMESH_IB_DYNAMIC

Use the D3DUSAGE_DYNAMIC usage flag for index buffers.

D3DXMESH_IB_SOFTWAREPROCESSING

Use the D3DUSAGE_SOFTWAREPROCESSING usage flag for index buffers.

D3DXMESH_VB_SHARE

Forces the cloned meshes to share vertex buffers.

D3DXMESH_USEHWONLY

Use hardware processing only. For mixed-mode device, this flag will cause the system to use hardware (if supported in hardware) or will default to software processing.

D3DXMESH_SYSTEMMEM

Equivalent to specifying both D3DXMESH_VB_SYSTEMMEM and D3DXMESH_IB_SYSTEMMEM.

D3DXMESH_MANAGED

Equivalent to specifying both D3DXMESH_VB_MANAGED and D3DXMESH_IB_MANAGED.

D3DXMESH_WRITEONLY

Equivalent to specifying both D3DXMESH_VB_WRITEONLY and D3DXMESH_IB_WRITEONLY.

D3DXMESH_DYNAMIC

Equivalent to specifying both D3DXMESH_VB_DYNAMIC and D3DXMESH_IB_DYNAMIC.

D3DXMESH_SOFTWAREPROCESSING

Equivalent to specifying both D3DXMESH_VB_SOFTWAREPROCESSING and D3DXMESH_IB_SOFTWAREPROCESSING.

Remarks

A 32-bit mesh (D3DXMESH_32BIT) can theoretically support (2^32)-1 faces and vertices. However, allocating memory for a mesh that large on a 32-bit operating system is not practical.

因为调用ID3DXMesh::Optimize()函数对网格模型进行优化时，需要用到网格模型中每个面的三个邻接信息，该信息在加载网格模型时得到：

ID3DXBuffer* material_buffer;

V_RETURN(D3DXLoadMeshFromXW(L"Dwarf.x", D3DXMESH_MANAGED, pd3dDevice, &g_adj_buffer, &material_buffer, NULL,
							&g_num_materials, &g_mesh));

D3DXMATERIAL* xmaterials = (D3DXMATERIAL*) material_buffer->GetBufferPointer();
g_mesh_materials = new D3DMATERIAL9[g_num_materials];
g_mesh_textures  = new IDirect3DTexture9*[g_num_materials];

for(DWORD i = 0; i < g_num_materials; i++)
{
	g_mesh_materials[i] = xmaterials[i].MatD3D;
	g_mesh_materials[i].Ambient = g_mesh_materials[i].Diffuse;

	WCHAR wfilename[256];
	RemovePathFromFileName(xmaterials[i].pTextureFilename, wfilename);

	g_mesh_textures[i] = NULL;

	if(xmaterials[i].pTextureFilename != NULL && lstrlen(wfilename) > 0)
	{
		V_RETURN(D3DXCreateTextureFromFileW(pd3dDevice, wfilename, &g_mesh_textures[i]));
	}
}

material_buffer->Release();

接着我们以三种方式对原始网格模型进行了优化：

DWORD* adj_in = (DWORD*) g_adj_buffer->GetBufferPointer();

V_RETURN(g_mesh->Optimize(D3DXMESHOPT_ATTRSORT | D3DXMESH_MANAGED, adj_in, NULL, NULL, NULL,
				 &g_mesh_attr_sort));

V_RETURN(g_mesh->Optimize(D3DXMESHOPT_STRIPREORDER | D3DXMESH_MANAGED, adj_in, NULL, NULL, NULL,
				 &g_mesh_strip_reorder));
	

V_RETURN(g_mesh->Optimize(D3DXMESHOPT_VERTEXCACHE | D3DXMESH_MANAGED, adj_in, NULL, NULL, NULL,
				 &g_mesh_vertex_cache));

渲染优化后的网格模型与未优化的网格模型相比没有任何区别：

// Clear the render target and the zbuffer 
V( pd3dDevice->Clear(0, NULL, D3DCLEAR_TARGET | D3DCLEAR_ZBUFFER, D3DCOLOR_ARGB(0, 0, 0, 0), 1.0f, 0) );

// Render the scene
if( SUCCEEDED( pd3dDevice->BeginScene() ) )
{
	for(DWORD i = 0; i < g_num_materials; i++)
	{
		pd3dDevice->SetMaterial(&g_mesh_materials[i]);
		pd3dDevice->SetTexture(0, g_mesh_textures[i]);

		switch(g_render_flag)
		{
		case OPT_NONE:
			g_mesh->DrawSubset(i);
			break;

		case OPT_ATTR_SORT:
			g_mesh_attr_sort->DrawSubset(i);
			break;

		case OPT_STRIP_REORDER:
			g_mesh_strip_reorder->DrawSubset(i);
			break;

		case OPT_VERTEX_CACHE:
			g_mesh_vertex_cache->DrawSubset(i);
			break;
		}
	}		

	RenderText();

	V(g_button_dlg.OnRender(fElapsedTime));

    V( pd3dDevice->EndScene() );
}

运行效果图：

主程序：

下载示例工程