CN101697226B

CN101697226B - Novel method for asynchronous pipeline topographic drawing

Info

Publication number: CN101697226B
Application number: CN200910236281XA
Authority: CN
Inventors: 李胜; 唐晨; 汪国平; 陈毅松
Original assignee: Peking University
Current assignee: Beijing Tong Jian Telit Intelligent Systems Engineering Technology Co ltd
Priority date: 2009-10-23
Filing date: 2009-10-23
Publication date: 2012-01-18
Anticipated expiration: 2029-10-23
Also published as: CN101697226A

Abstract

The invention discloses a novel asynchronous assembly line terrain rendering method, which belongs to the technical field of digital terrain rendering. The drawing pipeline of the method of the present invention includes three steps of data preparation, CPU operation and GPU operation in sequence, and the method includes three parallel drawing pipelines at the same time, and the three pipelines do not perform the same process at the same time. the steps. The step of preparing data includes requesting data from a local disk or network, receiving data and organizing data. The CPU operation steps include tree building and geometric meshing operations. The invention can be used for terrain rendering, and has the advantages of fast speed, solving delay, stable frame rate and controllability.

Description

Novel method for asynchronous pipeline topographic drawing

Technical field

The present invention relates to terrain rendering, relate in particular to a kind of new method for asynchronous pipeline topographic drawing, belong to digital terrain rendering technique field.

Background technology

Traditional terrain rendering algorithm (such as ROAM or other algorithms) is pipeline synchronization terrain rendering method normally, its rendering pipeline, the drawing process of the one whole of a frame in other words, experience following several steps sequentially:

Prepare data → achievement → gridding → drafting

For the above-mentioned first step-preparation data, the terrain rendering system of general FX size need not carry out, during still to the massive terrains data; Because the landform zone of drawing is not fixed; Therefore should obtain new data to data source (this locality or network) inquiry, but, wait for that the time of DSR may be long owing to reasons such as magnetic disc i/o delay and network delays; Will drag slow frame per second like this, and follow-up step can be carried out just after must waiting for first step completion.

The conventional synchronization algorithm all is that the disposable internal memory that is written into is then contribute and adopted adaptive strategy to data often, and speed can be very fast.But for the huge massive terrains data and network application widely of can't disposable loading going into internal memory; Current data of asking (altitude figures or data texturing) possibly temporarily can not obtain; Make and before contributing, to wait for that speed can lower naturally.Though can do some tactful treatment for property, do not have to have lost image quality such as data to replace, do like this to have improved performance with regard to value with an acquiescence, are level lands because a lot of places possibly occur, the such rough sledding in mountain peak has appearred but have suddenly.

In addition; Contribute time-dependent with geometric grid consumption in the size of the landform of being created, recursive hierarchical and number of triangles, existing landform algorithm all is difficult to control the number of triangles of every frame; Say nothing of the achievement time and the frame per second of the every frame of control; Even as ROAM, adopt the algorithm of adaptive mode, it adopts dynamic division and merges controls number of triangles, and it is stable completely also to be difficult to let frame per second accomplish.For landform in large scale and the complicated grid of high precision, may let achievement and gridding process very very long, increase the stand-by period of subsequent step and follow-up rendering pipeline on foot.If but the reduction of grid complexity, perhaps the landform scale reduces, and can lose display quality again.

Therefore; The state of the art to the terrain rendering technical field; Need a kind of new terrain rendering method; Can reduce because of plot step consuming time (such as preparing data step, or achievement and gridding step) to the stand-by period that subsequent step or follow-up rendering pipeline cause, improve total volume drawing efficient.

Summary of the invention

The objective of the invention is to overcome the problem that exists in the prior art, propose a kind of new terrain rendering method.

Different with existing pipeline synchronization terrain rendering method, terrain rendering method of the present invention is a kind of method for asynchronous pipeline topographic drawing.

The rendering pipeline of method for asynchronous pipeline topographic drawing according to the invention comprises the preparation data sequentially; CPU operation and three steps of GPU operation; It is characterized in that; Said method comprises three parallel said rendering pipelines at synchronization, and said three streamlines do not carry out identical said step at synchronization.

Generally, the inventive method is followed two principle, and the one, in streamline, carry out successively before and after above-mentioned three steps, promptly prepare data step formerly, secondly CPU operates, and the GPU operation is last; Two is to carry out successively before and after the step identical in three streamlines, such as with reference to figure 1, and streamline i-1, the preparation data of i and i+1, CPU operation and GPU operate and all carry out successively.For another example, with reference to figure 1, after the preparation data step of streamline i is accomplished, because the completion as yet of the CPU operation steps of streamline i-1, so the CPU operation steps of streamline i entering wait just begins to carry out after the CPU of streamline i-1 operation steps is accomplished; Similarly, after the CPU of streamline i+1 operation steps is accomplished, because the completion as yet of the GPU operation steps of streamline i, so the GPU operation steps of streamline i+1 entering wait just begins to carry out after the GPU of streamline i operation steps is accomplished.

Fig. 1 is a synoptic diagram of describing the inventive method practical operation flow process, has only shown streamline i-1 among the figure, three streamlines of i and i+1.It is pointed out that streamline is dynamic change, after accomplishing such as the GPU operation steps of streamline i-1; Mean that streamline i-1 finishes, this moment, the preparation data step of streamline i+2 began to carry out (if the preparation data step of streamline i+1 is not accomplished as yet certainly, then waiting for after it is accomplished beginning) always; Mean that streamline i+2 begins, at this moment, still and deposit three streamlines; That is, i, i+1 and i+2.

It is thus clear that above-mentioned second principle is intended to draw mode with the language description of summarizing streamlined asynchronous parallel shown in Figure 1.

Adopt this mode to reduce the stand-by period effectively, improve the whole efficient of drawing.

Preparation data step according to the invention comprises to local disk or network requests data.

CPU operation steps according to the invention comprises at least contributes and the geometric grid operation.

GPU operation steps according to the invention comprises at least draws operation.

To sum up, the present invention proposes a kind of novel method for asynchronous pipeline topographic drawing, and through encapsulating a kind of general synchronized algorithm, no matter the ROAM algorithm of traditional quaternary tree or adaptivity algorithm can be done asynchronous process.Method for asynchronous pipeline topographic drawing of the present invention mainly contains following three advantages:

1. speed is fast

Any landform algorithm; Basically the main standard of passing judgment on its quality is exactly its achievement speed; Asynchronous algorithm is assigned to several frames to the achievement time and goes to accomplish, so speed can improve tens times moment because be asynchronous achievement; Asynchronous mode can reach the drafting of millions number of triangles equally, and its performance index is considerably beyond similar other algorithm speeds.Because between several frames, the viewpoint mobile range is very little, it is not very greatly that topographic details changes, so jump phenomena clearly can not occur simultaneously.

2. solve and postpone

Asynchronous pipeline method of the present invention can come maximized reduction to postpone through pipelining, and asynchronous algorithm can come smooth delay through a kind of delay update strategy in addition, avoids occurring false data loss display quality.

3. frame rate stable and controllability

Asynchronous pipeline method of the present invention is completely fixed through setting its update mode can accomplish fully in each frame to contribute time of each operation, and this is very important for some to the harsh especially efficient software of time requirement.Simultaneously, the present invention also can according to present procedure the task number that will carry out, come dynamic achievement update time of distributing each frame, be that other any existing synchronized algorithms all can't be accomplished in this.

Description of drawings

Fig. 1 is a synoptic diagram of describing the inventive method practical operation flow process;

Fig. 2 is a synoptic diagram of describing embodiment of the invention AM/BAM operation strategy.

Embodiment

Combine accompanying drawing that the present invention is further described through specific embodiment below.

Present embodiment adopts the asynchronous method for drafting of three pipeline parallel methods as shown in Figure 1 to carry out terrain rendering; Asynchronous technique and pipelining are unified; Existing terrain rendering flow process is divided into 3 different parallel modules, prepares data from being I/O operation (perhaps wait for network operation, essence is the same) in essence; Achievement is the CPU operation with geometric gridization; Drafting is GPU operation, and these three can walk abreast and carry out, i.e. the preparation data manipulation three of the achievement of the drafting of i-1 bar streamline operation, i bar streamline and gridding operation, i+1 bar streamline can executed in parallel.

In the actual implementation of present embodiment technical scheme, though each module in the streamline is the operation that is Myself that ceaselessly repeats continuously, for the pipelined process of one whole; In some cases; A lot of attributes wherein (such as the zone of the landform that begins, viewpoint position, height or the like) are unalterable; Must before preparing data, just confirm and locking; Transmit along with the carrying out of process then,, and in whole process, can not be modified so stream line operation need provide a unique attribute copy for each complete achievement rendering operations.In order to reach this purpose; The terrain rendering attribute data that present embodiment will be confirmed before preparing data step is stored in the session (Session), and the data in the said session supply said preparation data step and CPU operation steps (contributing and geometric gridization) to use.

There is kinds of schemes in the concrete implementation of said session; Present embodiment then adopts following scheme: the said terrain rendering attribute data that alternately stores adjacent said rendering pipeline through two sessions; In two adjacent rendering pipelines, the CPU of last streamline operation begins to carry out with the preparation data manipulation of back one streamline simultaneously simultaneously.

Specifically, present embodiment employing backstage, foreground technology realizes preparing the data synchronization problems between data and the CPU operation in the streamline, and wherein achievement and geometric gridization are foregroundings, and preparing data is consistency operations.Foregrounding is corresponding to a session object so, and consistency operation is corresponding to another one session object (should be session next time).Each just exchange the session on foreground and backstage when foreground and backstage all after the complete operation, the backstage just becomes the foreground so, and the foreground becomes the session information that the backstage just is updated to next round afterwards, realizes the transmission and the synchronization mechanism of pipeline data thus.

For instance, as shown in Figure 2, at AM/BAM conversion time point, the Sb (Sa, Sb represent two mutually different sessions) that is positioned at the backstage switches to the foreground, and the Sa that is positioned at the foreground then switches to the backstage of another streamline (i+1).

Fig. 2 compares with Fig. 1 and also has a difference, promptly in Fig. 2 instance, in order to realize the switching of AM/BAM session; The CPU operation of last streamline must begin to carry out with the preparation data manipulation of back one streamline simultaneously, such as, with regard to streamline i+1; In Fig. 1, the preparation data manipulation of i+1 can be carried out after the preparation data manipulation of i is accomplished, and in Fig. 2; The preparation data manipulation of i+1 also need be waited for a period of time after the preparation data manipulation of i is accomplished, so that begin simultaneously to carry out with the CPU operation of i; So that when Sb got into the foreground by the backstage, Sa also can get into backstage (corresponding to streamline i+1) by foreground (corresponding to streamline i-1).

Synchronous processing for achievement and geometric gridization and drawing process is also similar.Present embodiment alternately stores the required grid data of GPU operation of adjacent rendering pipeline through two display lists.

Specifically; In achievement and geometric gridization, adopt the display list technology all geometric grid data importing GPU rendering pipelines, so for GPU; Can let it draw a foreground display list (the just last in fact geometric grid of contributing and accomplishing); And contribute to this moment and gridding generates is a backstage display list, at every turn when contribute with the gridding completion after, exchange foreground display list and backstage display list; Continue to upgrade the backstage display list after exchanging then, the backstage display list before the exchange then changes to the foreground and carries out real-time rendering.

Claims

1. A novel asynchronous pipeline terrain drawing method, its drawing pipeline comprises preparing data in sequence, CPU operation and GPU operation three steps, it is characterized in that,

The method includes three parallel rendering pipelines at the same time, the three pipelines do not perform the same steps at the same time, and the same steps in the three pipelines are performed sequentially;

In one of the drawing pipelines, the terrain drawing attribute data to be determined is stored in a session before the step of preparing data, and the data in the session is used for the step of preparing data and the step of CPU operation;

The terrain rendering attribute data of the adjacent rendering pipelines are stored alternately through two sessions, and at the same time, in the two adjacent rendering pipelines, the CPU operation of the previous pipeline and the data preparation operation of the subsequent pipeline start simultaneously.

2. The novel asynchronous pipeline terrain rendering method according to claim 1, wherein the step of preparing data includes requesting data from a local disk or a network, receiving data and organizing data.

3. The novel asynchronous pipeline terrain rendering method according to claim 1, wherein the CPU operation steps include tree building and geometric meshing operations.

4. The novel asynchronous pipeline terrain rendering method according to claim 1, wherein the GPU operation step comprises a rendering operation.

5. The novel asynchronous pipeline terrain rendering method according to any one of claims 1-4, wherein two display lists alternately store the geometric grids and textures required for the GPU operations of the adjacent rendering pipelines data.