WO1995027267A1

WO1995027267A1 - Z-buffering in 3-d computer graphics

Info

Publication number: WO1995027267A1
Application number: PCT/GB1995/000752
Authority: WO
Inventors: Peter Robert Warnes
Original assignee: Argonaut Technologies Limited
Priority date: 1994-03-31
Filing date: 1995-03-31
Publication date: 1995-10-12
Also published as: GB9406511D0

Abstract

A method of processing computer graphics information prior to rendering an image on a display, which includes comparison of the z (depth) coordinate of points having the same x, y coordinates on different surfaces of a 3-D scene so as to determine and store in a z-buffer those points which should appear on the screen, and to eliminate those which are hidden and should not appear, wherein the process includes two stages in the first of which all the z-comparisons on a scan line are made without interruption and the outcome is stored in the z-buffer, and in the second, only if the z-buffer content has changed, the changed values are used to update a frame buffer containing image rendering values.

Description

Z-Buffering in 3-D Computer Graphics
This invention concerns z-buffering in 3-D computer graphics.

In 3-D computer graphics, the z-coordinate of a point (x,y) on an object in the image space decides whether it will be visible when compared with the z-coordinate of the same point (x,y) of another object in the same image space. Thus, after the 3-D coordinates of a multitude of objects have been calculated, the rendering of the final 2-D image of a scene, showing some objects in front of others, involves this comparison step-by-step over the whole scene. The algorithm for such removal of hidden surfaces involves storage in a memory, the z-buffer, of z depth values of the nearest surface for every pixel in an image. As a new object is added to a scene, the z-buffer has to be updated with new z values of the new object.

This process is called z-buffering. Details can be found in "3-D Computer Animation" by Vince, Addison-Wesley (1992).

The current technique is as follows:
To plot each pixel, the z-buffer is read and compared against the z depth of the new pixel. If the new value is nearer than the z-buffer value, then the z-buffer value is updated. The pixel colour is then calculated and written to the frame buffer. This is repeated pixel-by-pixel across each scan line of the screen.

The sequence of memory accesses for every updated pixel is : read z-buffer'.

write z-buffer; write frame buffer;
Particularly if the z-buffer and the frame buffer are on the same memory bus, then this alternating pattern of accesses can be slow when using fast localised access memories (like fast page mode DRAM).

The problem is emphasised when only slower processing units and smaller memories are available, as in home microcomputers. This inhibits the quality of rendering when real-time animation is required, for example for computer games.

Accordingly, the invention proposes a method of processing computer graphics information prior to rendering an image on a display, which includes comparison of the z (depth) coordinate of points having the same x, y coordinates on different surfaces of a 3-D scene so as to determine and store in a z-buffer those points which should appear on the screen, and to eliminate those which are hidden and should not appear, characterised in that the process includes two stages in the first of which all the zcomparisons on a scan line are made without interruption and the outcome is stored in the z-buffer, and in the second, only if the z-buffer content has changed, the changed values are used to update a frame buffer containing image rendering values.

This method is used to optimise the z-buffering process by using two separate passes per scanline, the first of which reads and updates the z-buffer and the second of which updates the frame buffer accordingly. This reduces memory wait states, and in the event that no pixels require update, reduces the amount of setup for rendering.

The two separate scanline passes involve: (1) a z-pass, in which the z-buffer is read and updated if necessary, whilst keeping a record of which pixels are updated, (2) a render pass, in which pixels are rendered into the frame buffer if the pixel record says they need updating.

If during the z-pass no pixels were updated, then the render pass would not be necessary, thus saving time.

This new method keeps all the memory accesses in the first pass to the z-buffer, and in the second pass to the frame buffer.

The following example of the procedure illustrates the method in more detail as applied to some exemplary hardware.

Pass 1 .... Read The Underlying Z Values
The Z values of objects already drawn on the screen map where the new object is to be placed are read into a temporary RAM. So the DRAM screen buffer address is computed from X,Y, the screen base register. A list of these DRAM addresses can be placed in a FIFO for queuing burst fetches from the external DRAM.

Some time later, the DRAM will have responded with the data and this is placed directly into the pixel engine temporary RAM. The target address within this internal
RAM is taken from an X counter. As soon as the Z values are read in, they are compared with the new Z value (which is incrementing by dZ/dX as the values arrive from the DRAM). If the new Z object is visible, then the new pixel is marked as such by writing the current X value into the scheduler RAM, ready for the next traversal of the object.

Pass 2 .... Read The Texture Value, Perform The Intensity
Calculation ....

Since only the visible texels need to be fetched, using the schedule computed by the previous traversal of the object, the Texture address is computed from U and V, and the address is placed into the address queue for the DRAM.

Some time later, the DRAM will respond with texel data, which is expanded through a CLUT and then multiplied with an intensity value and placed into the pixel engine temporary RAM in a similar way to the Z values. As the results are read in, the X values are again written into the scheduler RAM, unless a zero value is encountered in the texture, which is interpreted as transparent, so the new pixel is not written.

The next stage depends on whether values have been set in the ALPHA (a) fields of the new object.

Pass 3a .... If Alpha Has Not Been Set....Write The New Z
Values
Using the X values remaining in the scheduler RAM, the new
Z values of the visible pixels are written.

Pass 3b ....Alpha Blending
If alpha blending is required, then the new Z values are not written, but the pixel values behind the new translucent object need to be read.

Some time later, the 'old' pixel values are read, combined using an blending multipler with the new values in the temporary RAM, and then written back to this RAM, which implies multi-ported RAM - or perhaps written back to the
Z value location, ready for the final pass.

Pass 4 .... Pixel Write.

The schedule list is traversed for the final time, and any pixels which need to be written are fed to the DRAM FIFO scheduler.

When the pixel span operation is complete, the calculations in Y can be made, and a new span operation can begin.

In practice, particularly in computer games programs which have to be designed for computers of limited processing capacity and memory, this technique can be advantageous.

It can be faster, especially when a new object is introduced to the scene which is in fact obscured by a nearer one.

Claims

1. A method of processing computer graphics information prior to rendering an image on a display, which includes comparison of the z (depth) coordinate of points having the same x, y coordinates on different surfaces of a 3-D scene so as to determine and store in a z-buffer those points which should appear on the screen, and to eliminate those which are hidden and should not appear, characterised in that the process includes two stages in the first of which all the z-comparisons on a scan line are made without interruption and the outcome is stored in the zbuffer, and in the second, only if the z-buffer content has changed, the changed values are used to update a frame buffer containing image rendering values.

2. A method as claimed in claim 1 wherein a record is kept in a store of those pixels for which the z-buffer content has changed.

3. A method as claimed in claim 1 or 2 wherein the two stages are performed in respect of each scanline of a scene before proceeding to the next scanline.

4. A method of processing computer graphics information substantially as herein described with reference to the accompanying drawings.