Each of the three modules in the rendering engine maintains state information that is set by using the Direct3D API. After all the state information is set, the rendering engine is ready to process display lists, which are known as execute buffers. Your application works explicitly with execute buffers only in Immediate Mode; Retained-Mode applications work at a higher level than this.
Execute buffers are fully self-contained, independent packets of information. They contain a vertex list followed by an instruction stream. The instruction stream consists of operation codes, or opcodes, and the data that is operated on by those opcodes. Direct3D's opcodes are listed in the D3DOPCODE enumerated type. The D3DINSTRUCTION structure describes instructions in an execute buffer; it contains an opcode, the size of each instruction data unit, and a count of the relevant data units that follow.
The following illustration shows the format of execute buffers.
The instructions define how the vertex list should be lit and rendered. One of the most common instructions is a triangle list (D3DOP_TRIANGLE), which is simply a list of triangle primitives that reference vertices in the vertex list. Because all the primitives in the instruction stream reference vertices in the vertex list only, it is easy for the transformation module to reject a whole buffer of primitives if its vertices are outside the viewing frustum.
The hardware determines the size of the execute buffer. You can retrieve this size by calling the IDirect3DDevice::GetCaps method and examining the dwMaxBufferSize member of the D3DDEVICEDESC structure. Typically, 64K is a good size for execute buffers when you are using a software driver because this size makes the best use of the secondary cache. When your application can take advantage of hardware acceleration, however, it should use smaller execute buffers to take advantage of the primary cache.
You can disable the lighting module or both the lighting and transformation modules when you are working with execute buffers. This changes the way the vertex list is interpreted, allowing the user to supply pretransformed or prelit vertices only for the rasterization phase of the rendering pipeline. Note that only one vertex type can be used in each execute buffer.
In addition to execute buffers and state changes, Direct3D accepts a third calling mechanism. Either of the transformation or lighting modules can be called directly. This functionality is useful when rasterization is not required, such as when using the transformation module for bounding-box tests.