CA1309198C - Parallel rendering of smoothly shaded color triangles with anti-aliased edges for a three dimensional color display - Google Patents

Abstract

PARALLEL RENDERING OF SMOOTHLY SHADED COLOR
TRIANGLES WITH ANTI-ALIASED EDGES FOR A
THREE DIMENSIONAL COLOR DISPLAY

Abstract of the Disclosure The present invention comprises a method for utiliz-ing an SIMD computer architecture in conjunction with a host processor and coordinate processor to render quality, three-dimensional, anti-aliased shaded color images into the frame buffer of a video display system. The method includes a parallel algorithm for rendering an important graphic primi-tive for accomplishing the production of a smoothly shaded color three-dimensional triangle with anti-aliased edges. By taking advantage of the SIMD
architecture and said parallel algorithm, the very time consuming pixel by pixel computations are broken down for parallel execution. A single coordinate processor computes and transmits an overall triangle record which is essentially the same for all blocks of pixels within a given bound-ing box which box in turn surrounds each triangle.
The individual pixel data is produced by a group of M x N pixel processors and stored in the frame buffer in a series of repetitive steps wherein each step corresponds to the processing of an M x N block of pixels within the bounding box of the triangle.
Thus, each pixel processor performs the same opera-tion, modifying its computations in accordance with triangle data received from the coordinate processor and positional data unique to its own sequential connectivity to the frame buffer, thus allowing parallel access to the frame buffer.

Description

YO9~6-065 1 3 ~
~, 1 j,~
PAR~LLEL RENDERI~G OF SMOOT~LY SHAD~D COLOR
TRIANGLES WïT~I ~NTI-ALIASED EDGES FOR A
THREE DIMENSIONAL COLOR DISPLA~
.

~ FIELD OF THE INVENTION

The present invention relates generally to the field of video displays. It relates more particularly to methodc for making video data computations whereby a significantly improved display of three~dimensional color images projected upon a two-dimensional display screen is achieved. It relates still more particularly to such methods whereby shading and anti-aliased are achieved in a highl~ efficient and improved manner.

BACXGROUND OF THE INVENTION

A primary consideration in the field of video data processing applications is the time involved in executing operations on the extremely large quanti-ties of data involved. As modern displays become ever more sophisticated, larger volumes of more complex data are involved. For e~ample, typical low-end displays are becoming available having a dimension of 1000 x 1000 pixels or 1 million pixels values. More particularly in the field of color graphics, each pixel is represented by a significant plurality of bits ~often 24 bit1 representing red, green and blue intensity values which must be processed in complex video data processing applica-tions. When a single host processor mus~ handle all graphic c~mputations, the result typically is an unacceptably lons execution time.

2 l 3 q~ 9 D' It is relatively well known in the computer art that where a given prcblem requires a great many computa-tions that a certain amount of time can be saved by performing certain computations in parallel. In ~ video data processing this can be done by employing a plurality o~ processors to simultaneously operate on different parts of the image. One solution is to have a plurality of simple processors with a proces-sor for every pixel in the image. A special purpose control processor issues data and special instruc-tions to each of the plurality of pixel processors.
It has similarly been known to arrange such pixel processors variously in an array, or in a line wherein the whole raster scan image can be processed concurrently, a line at a time, or as subsets of the array, i.e., sub arrays.

Although such systems have been known in the prior art, they have not been widely used due to the system overhead of supplying data and commands to the plurality of pixel processors and subsequently storing the data produced in the frame buffers.
Also, in such applications the frame buffers them-selves become very complex. Accordingly only very simple pixel operations have been performed by such prior art multiprocessor systems as exemplified by the follow~ng publications.
.

i) Fuchs, II. Poulton, I., Paeth, A and Bell A., "Developing Pixel Planes, A Smart Memory Based Roster Graphics System," 1982 conferenced on Ad vanced Research in VLSI, at MIT, Jan 17, 1982 pp.
137-146.

ii) Clark, J. H., and Hannah, M. R. "Distributed . ,. , . , , ., . ~ . . ~. . ~

'~ Y0986~0fi5 3 l 30ql 9 8 Processing in a High Performance S~art Image ~emo-ry", Lambda, 4th ~uarter 1980 ~p. 40-45.

As indicated above, the types of operations which have heretofore been avâilable with such proce~sor per pixel video architectures have been extremely limited in nature and, more particularly, limited in - the complexity of the operations performed. In the '~ field of color graphics, due to the amount of data involved per pixel, the manipulative operations become qui-te complex and frequently require rela-tively involved computations as compared to black/white or grayscale operations. The types of applications that have been envisioned for color graphics are complex, and have been limited due to the large amount of time necessary to perform the operations on a pixel by pixel basis. Further, due to the complexity of the operations, it has never been considered feasible to perform such operations as shading of three-dimensional lmages and anti-aliasing on anything approaching a processor per pixel type of display architecture which might be constructed as an SIMD pixel processor architec-ture.

In video displays, when projecting a three-dimen-sional image onto a two-dimensional surface, the attaining of uniformly shaded areas in the polygons making up the image and anti-aliasing of the inter-secting edges of these polygons to provide an enhanced image has always been a complex problem.
The basic geometry for providing the shading and anti-aliasing has long been known as will be set forth more fully subse~uently. However, providing such enhanced images via an SIMD architecture has 1 30q 1 ~8 never been considered feasible due to the complexity of the computations which must be performed. Hence, such image upgrading i.5 conventional]y done serial-ly, pixel by pixel in a host or image processor.

In such color images r the triangle is the primitive ~ shape for modeling three-dimensional surfaces.
Although rectangles have also been ~sed, trian~les have the important advantage that they retain ~heir planarity under perspective transformation. To render a three-dimensional ~urface, the triangles are projected onto a two-dimensional surface, the image plane. The triangles are drawn from back to front to remove hidden surfaces (by known algo-rithms). In order to provide optimized displays, it is desired that each triangle be smoothly shaded and have anti-aliased edges. This is well known. The smooth shading creates the effect of a continuous surface, and anti-aliasing of edges results iD
smooth ra~her than jagged boundaries which ~urther contribute to the effect of continuity of the displayed image. Conventional serial systems have not been able to generate such images at acceptable speeds. The herein disclosed invention greatly improves this situation by exploiting parallelism in an economical SIMD pixel processing architecture realizing a novel program which is run identically on each processor and contains the requisite data modification constructs whereby each processor is - ~ capahle o~ modifying its own particular pixel in an array of pixels to produce a smoothly shaded, anti-aliased, three-dimensional image on the display screen.

` YO986-065 .~
5 1 3 0 q ~

DESCRIPTION CF THE PRI()R ART

The following references are cumulative to the references disclosed and discussed in the following ` "Description of the Embodiment" section of the specifications and were found pursuant to a prior art search in the United States Patent and ~rademark - Office for the subject matter of the present inven-tion. Their relevancy is explained below.

An article ~y H. Gourand entitled "Continuous Shading of Curved Surfaces", IEEE Transactions on Computers, Vol. 20, No. 6, June '71 pages 623-629 discloses well known methods of shading, but does not disclose nor suggest the present method of performing such operation in paraliel with anti-allasing on an SIMD architecture.

The book "Computer Image Generation'l edited by B. J.
Schachter, published by J. Wiley & Sons, N.Y. (1983) describes shading as performed by Gourand. The 2~ present invention performs linear interpolation but not for a point on a scanline. It is done for a point on a triangle with inrormation about the triangle givQn to a parallel pixel processor archi-tecture. The prQsent invention performs a fast and simple m~thod of calculating the smoothing to take advantage of parallel processing. Schachter de-scribes the problem of anti~aliasing but does not disclose a method far removing the effects using the parallelism in an SIMD pixel processor architecture.

3~ U. S. Patent No. 3,602,702 of Warnock discloses a method for rendering 3D shaded polygons on a 2D
surface with hidden surface elimination. The , ~ YO986-065 6 ~ 3091 q~

present inventlon co~prises a novel method 'or calcuilating shading and anti-aliasing by providing parallel processors which use information pro~tided once per -triangle. The disclosed operations are performed on 2D triangles in which 3D infsrmation has been captured in rate data as will be described later.

U. S. Patent No. 3,665,408 of Erclahl, et al de-scribes a system for displaying bounded areas on a screen with an intensi~y _unction for that area.
The present in~ention comprises a fast method for providing anti-aliasing for such an imaqe after it has been rendered to 2D with rate data. The present method for smoothing is a simple calculation with small cost, performed contemporaneously with anti-aliasing is done.

U. S. Patent No. 4,343,037 of Bolton shows a pipe-line processor to perform perspective transfor-mation. In the present system, the 3D image has already been rendered as a 2D image in the frame buffer. Further, it provides parallel processing to anti-alias the edges in the frame buffer with smoothing done at small e~tra cost. This is not suggested in Bolton.

U~ S. Patents No. 4,156,~37 of Okada et al, and 4,208,719 of Lotz et al, broadly disclose parallel or pipe-lined architectures for use with particular video displays but otherwise have little to do with an SIMD computer architecture organized to perform simultaneous shadinq and anti-aliasing operations on 3D color triangles projected onto a 2D color dis-play.

7 1 3 ~C3 SUMMARY AND OBJECTS OF THE INVENTION

It is a primary object of the present invention to provide a novel metho~ for rapidly producing smooth-ly shaded color triangles with anti-aliased ecges for use in a three-dimensional video display which utilizes the parallel execution capabilities of an SIM~ parallel processing architect~re.

It is another object of the invention to provide such a method wherein an identical parallel algo-rithm is utilized in each processor of the SIMD
architecture but with unique data for each processor to produce necessary pixel modifications in the display.

It is a further object of the invention to provide such a me~hod wherein all of the processors oE the SIMD architecture are organized into an M x N arrav, each processor modifying a different pixel related to a physically associated M x N array in the frame buffer and/or the display screen thus allowing parallel update to the frame buffer.

It is yet another object of the invention to provide such a method wherein a single coordinate processor computes general image data concerning a portion of an image being processed and wherein the lndi~idual pixel processors utilize this data to produce the actual pixel image data which is stored in the frame bufer and displayed.

Other objects, features and advantages of the invention will be apparent from the following description o- the herein disclosed preferred , YO98~-065 ~30~q'~

embodiment of the invention.

The objec~s of the present invention may be accom-plished by an SIMD computer architec~ure utilized in conjunction with a host processor and coordinate processor to render quali-ty, anti~aliased, shaded color images into the frame buffer of a video dispiay system. The method inclucles a parallel algorithm for rendering an important graphic primi-tive for accomplishing the production of a smoothly shaded color-three-dimensional triangle with anti-aliased edges for displa~ on a two-dimensional display screen. By taking advantage of the SIMD
architecture and said parallel algorithm, the verv time-consuming computations are bro~en down for parallel execution. A single coordinate processor computes and transmits an overall bounding box record which is essentially the same for all blocks -of pixels within a given bounding box which box in turn surrounds each triangle. The individual pixel data is produced by a group of M x N pixel proces-sors and stored in the frame buffer in a series of steps wherein each step corresponds to the process-ing of an M x N block of pixels within the bounding box.

The method allo~7s all of the processors to appear to be performing different functions while, in reality, they are all performing the same operation on a common instruction sequence. According to the méthod, a host processor provides the physical x,y,z data defining the location of each triangle in the rame burfer and on the display screen as well as its color. A coordinate processor computes for each triangle the instantaneous rate of change of the ... .. , . . . .~ . .

13~ql9~

-distance to any of the three sidcs of the triangle along hoth the .Y an v a~is and ,~le rate o' change of intensity of each or the primary colors, red, blue and green, along both the ; and y axes as a point ~ mo~es across the triangle.

The coordinate processor also computes the coordi-- nates of the origin of each of a plurality of M x N
pixel constituent blocks which constitute a bounding box which bounding bo~ completely encloses the triangle and wherein there are at least M x N
separate processors in the SI~ID processor svstem.
When all of the required rate of change data is computed for each triangle, the system controls cause the M x N array of processors (pixel proces-sors) to sequentiall~ compute new values for each constituent block of related M x N pi~els in the frame buffer properly modified for shading and anti-aliasing. This is accomplished by processing, ~ in paralle], all O r the pixels in a particular block and repeating the operations until all of the blocks constituting a bounding box for a particular trian~
gle have been processed and continuina the process until all triangles making up the display are processed.
' All computations on all blocks concerned with a particular triangle are provided with the same computed distance and color rate of change vaiues, the utilization of the rate of change values are ~eighted by the location of a particular pixel as 1 totally outside of a triangle, 2) totally interior of the triangle, or 3) within an anti-aliasing boundin7 region adjacent to each of the three sides of a triangle.

10 130qlq~

The method includes per~.,ittirlg the current color intensity value of the pixel in the frame buf._r to remain unchanged if condit:ion 1) above is found to exist; changillg the value of the pixel based only on ~ shading requirements if condition ~) above is found to exist; and changing the pixel value based on both shading and anti-aliasing reauirements, if condition 3) above is found to exist.

The method further includes a procedure operable in each pixel processors for computing a pixel location variable (a) whose value~ once computed, indicates which of condi.tions 1), 2) or 3) applies 'or a particular pixel.

The procedure begins with the triangle in the display which represents an object furthest from the viewer in the simulated 3D di.splay.

~'0986-065 11 1 309 1 q~

BRIEF DESCR:tPTIO~I OF THE DRA~`7ING~;

FIG. ] comprises a high-level -~unctional block diagram illustrating the use of an SIMD architecture to practice the inv2ntion.

FIG. 2 comprises a ~eometrical drawing explaining - how a particular computation is broken up and routed '~ to the various processor- to sequentially compute the necessary data.

FIG. 3 comprises a geometrical drawing illustrating how the rate of change of color data used for shading calculations is determined.

FIG. 4 comprises a geometrical drawing illustrating how anti-aliasing data is generated.

FIG. 5 is a flow diagram illustrating data flow ln 15 , the system which results in the recomputed shading and anti-aliasing video information.

FIG. 6 comprises a high-level functional ~]ock diagram of the overall system in somewhat more detail ~han FIG. l showing all of the major func-tional components in the system and their relation-ship to each other.
.
~IG. 7 comprises a ~iagram illustrating the func-tional relationship between individual pixel proces-sors and the assigned sections of the frame buffer as utilized in practicing the present invention.

FIG. 8 is a diagram illustrating two triangles on the display screen and their locating coordinates as YOg86-065 1 30q t ~3 they would appear on the display screen and also the data format of 'he coordinates which define each triangle making up the display which must be trans-ferred from the host to the coordinate processor during the prac-tice of the present invention.

FI5. 9 comprises of the data format and a related diagrammatic illustration of the coordinates of the individùal blocks within a bounding box which are p~riodically sent to each pixel processor and are used within the coordinate prccessor to control the repetitive operations within the method.

FIG. 10 comprises the data format for a single "triangle record" and related diagrammatic illustra-tion of set of data periodically transmitted from the coordinate processor to each pixel processor which is identical for a complete bounding box of a triangle.

FIG. 11 comprises a data format for a "block record"
and accompanying diagrammatic representation of data sent from the coordinate processor to each OL the pixel processors which data is revised ror each constituent block within the bounding box.

FIG. 12 comprises a flow chart of the sequence o~
operations which occur in the coordinate processor in carrying out the method of the present invention.

FIG. 13 comprises a flow chart of the steps involved in the "calculate bevel width" block of FIG. 12.

.. ... " . . . . . . .

` 13 ~ 30q 1 ~

FIG. 14 comprises a hiah ' evel flow chart of the operations which occur ~ithi~ one of the pi~i processors of the S~MD ~rocessor in carr~ing out the sequence of operations necessary in performing the ~ethod of the present invention.

FIG. 15 comprises a sub routine for computing the anti~aliasing Cactor as shown in Figure 14.

FIG. 16 comprises a Clow chart illustratin~ the operations necessary to computing the "shaded rgb"
values as shown in FIG. 14.

FIG. 17 comprises a sub routine illustrating the calculat~ons necessarv to produce the new red, blue and green values ~rn,gn,bn).

FIG~ 18 comprises a flow chart of the sub routine lS utilized to calculate the function "determa" which is utilized within the sub routine for compu.ing the anti-aliasing factor as shown in FIG. 15;

FIGS. l9A and l9B are diagrammatic illustrations of the geometric calculations performed by the coordi-nate processor and referenced in FIG. 12.

FIG. 20 comprises a data forma-t and an illustrative diagram showing a single "triangle data record" sent from the coordinate processor to each pixel proces-sor for a second embodiment of the invention.
FIG. 21 comprises a data format and a related diagram illustrating the "block record'l sent from the coordinate processor to each of the pixel processors when processing subsequent blocks within Yoss6-06s 1 3 nJ q 1 9 8 a bounding box ~or the second embodiment of the invention.

FIG. 2 comprises an overall flow chart similar to FIG. 12 of the operations which occur within the coordinate processor when practicing the method of tlle second embodiment of the invention.

FIG. 23 comprises a fiow chart similar to FIG. 14 of the operations which occur ln each pixel processor in carrying out the method of the present invention in accordance with the second embodiment.

" , , . , . , , , ~ . , YO9-86-065 15 l 30q 1 9~

DESCRIPTION OF THE PR~FERRED EMBODIMENT

The herein disclosed invention comprises a procedure by means of which an SIMD computer architecture can be used as an M x N pixel processor array to render quality three-dimensional color images into the frame buffer of an associated video display architecture to produce a two-dimensional display. l'he overall system includes a host processor including an attached video display unit wherein the host processor supplies the basic video data including display information concerning the individual three-dimensional triangles conventionally associated therewith. A coordinate processor is also included which can either be a stand alone processor or its functions can be performed by the host as a functional module. The coordinate processor performs initializing operations on the video data and controls the operation of and data supplied to the M x N pixel processor matri~ organized as an SIMD architecture. The method includes a parallel algorithm executable within the M x N pixel processor matrix for rendering an important graphic primitive -- a smoothly shaded color three-dimensional triangle with anti-aliased edges.

The overall system is shown in FIG. 1. This system comprises an SIMD architecture with the coordinate processor controlling a multiplicity of pixel processors.
The organization and operation of SIMD architectures is well known in the art. A particular architecture is set forth in copending Canadian Patent Application No.
502,805 filed February 26, 1986, of L. Lumelsky. An important feature of the present organization is its ability to access and process ` "0986-065 ~ 3n-q l q ~

multiple, multi-bit ~ixels in parallel. ~. second feature is the abi~ ty of a pi~el pro-essor to perform multiple ir.depenclent operations on d fferent segments of a pixel. Altl-ough the method to be described subsequently could be implemented on a system without SIMD architectural fcatures, tne lack of pi~el and operational parallelism ~ould substan-~ tially reduce the system performance.

At this point, it should be clearly understood that no claim is made that the geometry of shading and anti-aliasing is novel per se as the methods of obtaining this type of shading ard anti-aliasing are well known in the field or computer image generation as will be apparent rom the sub.sequent description lS in which references are made to prior art articles describing certain of the geometricai operations involved. What is deemed to be novel is ~he pres-ently disclosed method of anal~zing a particular trianyle and surrounding it with a boundir.g box which is known to completely enclose the triangle and which is, in turn, composed of a plurality of constituent blocks having a dimension of ~ ~ N
pixels. Assuming that the SIMD architecture is composed of M x N processors, each of said pro-cessors is uniquely connectable to the frame buffer in that it is able to access and analyze a particu-lar pixel Ln each of said constituent bloc~s. It should further be understood that the entire frame buffer is organized into a plurality Oc said con-stituent blocks although only a certain contiguous subset will be involved in any particular bounding ~ox .

Utilizing these basic organ1zational concepts of the . .

1 3091 9~

ar~ay o ~l x N pixel processors relati~e to the frame buffer, a unique parailel algorithm has been devised. The coordinate processor precomputes certain data rela~ive to a particular triangle being ~ processed and transmits this informatlon in parallcl to all of the pixel processors which uniquely use this data in accordance with their own positional ~ inter relationship within a constituent bo.c to recompute or upgrade the color information original-ly stored in the frame buffer so that proper shading and anti-aliasing is accomplished.

According to one further aspect of the invention, a novel method of computing the anti-aliasing factors 50 that they may be utili~ed in the parallel algo-rithm is set forth which uniquely enables 'he processors to individuallv determine what effect an anti-aliasing factor will have on their computation and, iE ~uch an effect is necessary, to automatical-ly include this in the computation.
The specific details of the invention will no~l be set forth. The overal7 process is illustrated diagrammatically in FIG. 1. The host processor starts the process by performing an initial set O r calculations ~hich are known and passes the results to the coordinate processcr. Some of these results are passed on to the pixel processors hy the coordi-nate processor. Then a series of iterations begin under the control of the coordinate processor. At the start of each iteration, the coordinate proces-sor supplies a set a values to the pixel processors.
The pixel processors use these values to calculate the tria~gle pixels and to combine the triangle pixels ~lith the pixels currently stored in the frame ~'09~6-065 1~ 1 30q 1 ~

buf er. At the hegir.ninc3 of the operation, the pixels currently stored in the frame buffer repre-sent those placed there by the host processor without consideration given to shading and anti-aliasing. The itera-tion is repeated within the pixel pro'cessors until the entire triangle has been rendered a constituent block at a time.
-If there are M N pixel processors, then one M x N
block of pi:-els is processed during each iteration as described above. A simple way of covering the triangle is to process all the constituent blocks within its bounding box. This is illustrated in FIG. 2. A more efficient alternative would be to process only the constituent blocks which are wholly or partly within the triangle. In the presently disclosed embodiment, this is not done e.g. all of the constituent blocks are processed; however, it would be comparatively trivial for the coordinate . processor to make the requisite calculations before performing the iteration which would evaluate a constituent block which laid wholly outside of the subject triangle being processed. In any case, the coordinate processor determines the order and selection of the blocks to be processed.
The following is a very brief description of the processing performed by each pixel processor. It will be elaborated upon subsequent~y. During each iteration, each pixel processor evaluates six variables using the linear relation:
v = vb + i*vx + j*vy -where vb is the value of the variable at the lower-left-hand pixel of a particular constituent block, vx is the variation of v wlth respect to x, - ~'09$6-OG5 19 1 30~ 1 ~8 vy is the variation o~ v ~ith respect to y, and (i,j) are the illdices deiining the relati~-e position of a particular pixel ~ithin the constitucnt blcck.
The six variables are:
Distances from the triangle lines to the pixel, adl, dd2, dd3 and the colors, r,g,b.
The rates of change with respect to x and y are supplied only once berore the iterations begin.
However, to maintain precision, the values at the lower-left-hand pix~l of each constituent block could be supplied at the start of each lteration.

The distance variables are illustrated in FIG. 3.
ddl, dd2, and dd3 are the distances from the three sides of the triangle to a particular pixel. A
negative distance means the pixel is "outside" a line. A positive distance means the pixel is "inside" a line. A pixel is inside the triangle if it is inside all three triangle lines. The defini-tions dx and dy are illustrated in the insert to FIG. 3.

Thus to ~ecapitulate, the following data is supplied to the pixel processor by the coordinate processor:

Upon each iteration of the pixel Processors (per const~tuent b_ock) r,g,b, dl, d2, and d3.

Once for each triangle (per boundino box) Rx, gx, bx, dlx, d2x, d3x;
ry, gy, by, dly, d2y, d3y~
, ," . .. : ,. , . ~ ~. . . .

` YO986-065 20 1 309 1 ~8 The ~a~ this data is used in the over~ll process will be set forth more full~ subsequently.

To understand how the anti-aliasing computation is performed according to the present invention refer-ence should be made to FIG. 4. The three distances, ddl, dd2, and dd3 are used to calculate the - anti-aliasing factor (a). Ia) takes care of anti-aliasing at triangle edges and helps determine how to combine the triangle pixel with the contents of the frame buffer. There are many ways of calcu-lating (a). A simple technique will be described here. The idea is to treat the triangle as if it were a three-dimensional object with beveled edges.
The bevel is defined by the three lighter lines within the primary heavier lines defining the triangle as shown in the figure. (a) is defined to be the height of the triangle at the center of the piY.el. The width, wk, of the beveled region is a 20 function of dxk and dyk. In this case, let wk = max~¦dxk¦, ¦dyk¦) and define the height of the triangle in the interi-or region to be one. Thus, the height outside the triangle is zero and the height within the bevel is some value between zero and one.

For each edge, k(l,2,3,) each pixel processor uses the distance, ddk, from the line containing the edge to get the value ak. The ak's are defined as fOllGWS:
if ddk < 0 then ak = 0, if 0 < ddk < wk then ak = dk/wk, and if ddk > wk then ak = lo When all three edges are taken into account, a = min~a;,a2,a3,) FIG. 4 shows several examples of " ~

~0986-06S

21 ~ 3 ~1 q 8 pi~els relative to their locatlon outside of the triangle, inside of the triangle, and within one or more of the beveled ed~es.
Examples:
L is totally within the triangle, ~a) - 1 M is totally outside the triangle, (a) = 0 N is on one edge of the triangle, 0 < (a) ~ 1 P is on two edges o~ the triangle, 0 ~ (a) < 1 The last step is to combine the triangle pixel values with the contents of the frame buffer. This ls diagrammed in a very high-level fashion in FIG.

5. The cld pi~el values are depicted as being within the block entitled "old pixel" and comprise values rc, go, bo. The new values computed by the present system are deplcted as being within the box marked "new pixel" and comprise rr, gg, bb. The way that these values are specifically computed will also be set forth in detail subsequently. These values are combined by the pixel processor as illustrated and the combining function is controlled by the anti-aliasing factor computed for that particular pixel by the present procedure. The new results of this computation are returned to the frame buffer and indicated by the block "new result"
and comprise rn, gn, bn.

- Having thus generally described the overall proce-dure of the present invention, the specific details of the operation will now be set forth with respect to FIGS. 6 through 23. It is pointed out that the following description refers specifically to the preferred embodiment and may differ in certain minor respects from the general description immediately preceding.

.. . .. . .. .

22 1 ~ql ~ ~

As is well known in a color display, a pixel is represented by the three primary colors, red, green and blue. Further, in conventional frame buffer storage systems, a particular pixel value is a composite of the ~hree primary colors wherein a digital code represents the intensity value.
Accordingly throughout this write-up, a particular pixel value consisting of ~he three primary colors is simply referred to as rgb.

1~') A number of assumptions are made about the indi-vidual triangles making up an overall display in the host before they are sent to the coordinate proces-sor. It is assumed that an rgb value has been assigned to each vertex of each triangle in the l$~ manner described in [1] and on page 30 of [5].
These shading values are the intensities of r,g,b at each ver~ex. The shading values will be linearly interpolated for points in the triangles by the coordinate processor and the individual pixel 20n processors.

It is assumed that the three-dimensional surface has been projected onto a two-dimensional surface as well known in the art and described in Chapter 8 of reference [3]. It is also assumed that the trian-25j gles are depth sorted (by z value) as described by ~4] and page 558 in [3~. The triangles closest to the background are sent to the coordinate processor for processing first. The triangles next to the background will be anti-aliased with the background 300 by the pixel processors. Later triangles may cover parts of the previously processed edges. This will of course result in a modified rgb intensity value being stored in those pixel locations within the YOg86-065 23 l 3~9 1 q8 frarne buf~er as will be descrlbed subsequentl~.

It is furthcr assumed that ~he triangles ha~e been received by the cGordinate processor enlarged slightly~ This allows the "beveled" areas to overlap. As stated previously, the beve7ed areas are the thin strip inside a triangle where pi~els are anti-aliased and the use of these bevels in determining an anti-aliasing vall~e ~a) will be described later. For the present implementation since the triangles are enlarged first, first the edge of a triangle is anti-allased with an edge in the ba~kground. Later that edge is anti-aliased with an edge of a triangle in the foreground. In this manner, an edge of a triangle close to the viewer has been anti-aliased with two shaded pixels.

The amount of enlarging performed before the host transmits the triangles to the CPU uses the same , geometry shown in the embodiment. The enlarging on each triangle is by its corresponding bevel width.
It is quite possible that other implementations of the basic invention disclosed herein may not have the beveled areas overlap. Whether they do or not does not effect the basic structure of running the pixel processors in parallel.
~5 FIG. 6 illustrates a system diagram with the host processar passing a description of triangles from its memory to the control processor. The control processor stores data tables in its RAM which it computes from triangle data recei~ed from the host.
The control processor will in turn pass data to the pixel processors which run as a "single instruction . , . . . ~ , .

. ~4 ~ 309~ ~8 multiple data" (SIMD) arc~itected s~stem such as disclosed in reference [2]. As will be apparent from the previous discussion in the following description, all of the pi:sel processors execu.e precisely the same program although they will of course be processing different data values due to their own unicue position in the particular M x N
constituent block.

It should be noted at this point that in the ?re.sent embodiment, it is assumed that there are sixteen pixel processors (e.g. 4 x 4) in the SIMD architec-ture; however, it is obvious that this is merely chosen ~or convenience of representation. Any value of M x N could be used depending on the desired system performance.

Each of the sixteen illustrated pixel processors has access to one sixteenth of the frame buffer. Stated differently, each processor can access seLectively every sixteenth pixel stored in the frame buffer.
The organization of this access may be designed into the system. For the present invention~ it is assumed that the frame buffer is organized into a set of blocks which are referred to herein as constituent blocks each block being four pixels by four pixels. Thus each pixel processor according to the design of the system has access to a particular pixel in each of the constituent blocks making up the entire frame buffer which pixel of course will be in the same relative position in each block.

The output of the frame buffer drives the display at whatever refresh rate is required thereby and does nct form a part o the present invention.

1 30q 1 ~8 FIG. 7 illustrates graphicctllty ~he organization o~
the pi.~el processors and their rel~ionship to a predeterminèd relative position of a pixel within each of the constituent blocks making up the frame buffer. Thus in each cons~ituent block, the pixel processor 1 is cor.nected to the upper-left-hand pi.~el in the four by four array and pixel processor 16 is connected to the lower-right-hand pixel~
Selection means, not shown, would of course assure that the pi.Yel processors are only connected to an r,g,b triplet of values in a single consti,uent block during any one processing cycle as will be well understoodu FIG. 8 shows the triangle geometry records which lS describe each of the trianqles ma];ing up the total display as they would be stored in the host. As will be understood, each triangle is represented by a set of three point vertices. Each point can be , identified by an xy address coordinate and by a color intensity value rgb. These records are transmitted by the host to the coordinate processor.
As stated previously, each corner of the triangle is defined by its xy coordinates and its color which color is used for the Gourard shading as described in reference rl]. Also, the triangles, at this stage~ are assumed to have been depth sorted. These triangle geometry records are processed by the coordinate processor to produce data records which are in turn sent to the individual plxel processors.

FIG. 9 illustrates a triangle placed in the frame buffer (diagrammatically) with its bounding box and a number of constituent blocXs making up the bound-ing box. The diagram at the top of the fiqure .

.

YO986-C6~

130~1q8 illustrates the data record generated b~ the coordi-nate processor and ke?t or the ~urpose of identif~--ng the progress of t'le operation as a given bound-ing box and its constituent blocks are processed.
- It should be clearly ~Inderstood that this merely represents the record ~eeping of the coordinate processor and not the triangle geometrv calcula-tions. It should also be understood that the bounding ~o~ is shown ~s exactly lying on or being congruent with the lower-left-hand corner of the triangle and being substantially outside and above the upper-right-hand corner of the triangle. In all probability, the bounding box would be totally outcide of all the points of the trianale. The requirement is that it must totally enclose or surround the triangle and must be made up of com-plete four by four constituent blocks that are addressable in a predetermlned fashion in the frame buffer as will be understood.

Referring again to the diagram at the top of FIG. 9, it will ~e noted that the coordinates ax, ay defines the origin of the first constituent block within the bounding box and the coordinates bx, by comprise the coordinates of the origin of the last constituent block within the bounding bo~. The field marked column ~Icol~ and "row" re~er to the number of rows and columns of constituent blocks within the bound-ing box. As will be appreciated, this data is used together with counters to control the number of iterat~ans of the present computations in order to completely evaluate a given triangle.

FIG. 10 illustrates a triangle record which is passed from the coordinate processor to all of the , .

~0986-065 ~7 l 3 O~l q ~

pixel processors. This data will be used when each constituent block n the trian~le's bounding box is processed. All the data is rate data except wl, w2, and w3 which refers to the bevel widths whose use will be described later.

As illustrated in the diagram at the top of the figure, the rate of change of the three distances with respect to both the x and y coordinate are shown. These are dlx, dly; d2x, d2y; and d3x, d3y.
These represent the rate of change of distance to the three sides of the triangle as a point moves a unit distance in the x and the y direction. This record also contains the three items of rate of change color data e.g. rx, ry; gx, gv; and bx, by.
These latter six values represent the rate of change of intensity of the three primar~y colors a~ain as a pixel moves a unit distance along either the x or y axis. The following example illustrates the compu tation of several of these variables reCerenced to ~0 the Figure which computations are well known in the art and are adequately described in the references cited.

Example:

dlx=b-a (The change in distance relative to edge l for moving pixel in the x direction) d2y=d-c ~The change in distance relative to edge 2 for moving 1 pixel in the y direction ~
. , ,, . . . ,, , ~ . ~ ~.

, 2~ l 30qt 9 ~

rx=rb-ra(The change in intensity of .he color red for movin~
pixel in the x direction) gy=gc-ya(The change in intensity of the color green for moving 1 pixel in the y direction3 FIG. 11 illustrates a constituent bloc~ record and the meaning of its elements. A triangle and a constituent block ~ithin a bounding box are shown in the lower part of the figure. It is also noted that the bounding box is shown as ideally touching the three corners o' the triangle although in most situations this would not be the case. Shown in the upper part of the figure are the calculated rgb intensities and the distance to the triangle edges for the corner pixel of a constituent block. The sixteen pixel processors will each process cne of the pixels in the four bv four block. The pixel processors processing pixels which are not in a corner use the triangle rate data to calculate the distance and rgb values to calculate their particu-lar distance and rgb values from their knowledge of their relationship to the origin pixel and thus compute shaded, anti-aliase~ values which are to ~e updated in the frame buffer.

FIG. 12 comprises a flow chart of the coordinate processor routine for the first and preferred embodiment of the invention. It is believed that the flow chart is self explanatory in as far as the particuiar functions required. Reference is appro-priately made in the figure to subsequent figures which show details of the particular operations ~0986-065 1 3~ 1 q~
2~

specified in the various hlocks o this flow chart.

Block 10 implies the receipt of successive triangle data from the host processor wherein a new triangle will be accessed from the host when a pre~ious one has been completeLy evaluated until all triangles have been processed. Triangle record data is computed in blocks 12 through 20 and sent to all Oc the individual pixel processors. It is noted that this data is the same for all iterations of the pixel processors concerning a single triangle or bounding box.

Block 22 controls the consecutive iterations of the pixel processors-as they compute data for each successive constituent block. The data illustrated in FIG. 9 is of course utilized for this purpose.
Blocks ,4 through 30 compute the data which is unique to each constituent block and must be updated on each iteration of the pixel processors. This data is illustrated in FIG. 11.

Thus to recapitulate, for each triangle a plurality of constltuent blocks within the triangle bounding box are processed. In this manner, the data is sent from the coordinate processor to the individual pixel processors. General data is sent about the triangle once via block 20; and upon each iteration of the pi~el processors, the new constituent block data is sent as shown in block 30 of the figure.
Thus, all of the required data is sent to the pixel processors. In the pixel processors, the frame buffer will be read and written and anti-aliased and shaded rgb -~alues will replace or be combined with values previousl~ in the frame buffer.

. . . , , ~ . . ~

30 l 30q 1 q~

The distances and rgb values as illustra'ed in FIG.
lC wi-th respect to the or-gin of the frame buf~er are calculated. These are used in the inner loop of the coordinate processor procedure to calculate the distances and rgb's with respect to each constituent block in the bounding box. As will be seen latex, formulas for calculating values with respect to a constituent block use the distance and rgb values from the origin of the frame buffer as a reference.
Similarly, the formulas in a pixel processor for calculating with respect to a given pixel use values known wi-th respect to the origin of a partlcular constituent block.

Thus, the rate changes in distance and rgb intensity are calculated once for a triar.gle (bounding box~
and used for each constituent block in the coordi-nate accessor's computaticns and later are used for each pixel's computations within the individual , pixel processors.

It will be noted that the majority of the computa-tional blocks in FIG. 12 indicate references to other figures which detail the particular function being performed in the indicated blocks. It will also be noted that in a number of the blocks of the flow chart, reference is made to performing certain operations within a particular "block". This of course refers to the constituent blocks which are determined to lie within a particular bounding box.

FIG. 13 illustrates the mathematical operations involved in performing the calculation of the beveled ~idth wk~ In this figure, blocl; 1 indicates the entry into the routine. Block 2 indicates the , . .. . . .

~0986-065 31 l 3 09l q ~

three calculations for the three bevels and block 3 indicates an exit. As will be noted, the signif -cance of the bevel is that it is a function of the rate of change of distance. Thus, the rates of change along both the x and y coordinates are compared and the larger is chosen.

- FIG. 1~ is a flow chart for the pixel processor routine. It is this routine which is run in paral-lel in all of the sixteen processors (M x ~J) of the SIMD architecture and, together with the preprocess-ing done by the coordinate processor as set forth in FIG. l, comprises the essence of the present invention. It is this routine which is run in each of the pi~el processors rrom beginning to end on each iteration as the block of sixteen pi~rel proces-sors computes the data for a new constituent block.
Even though the anti-aliasing factor computed earl~
in the routine is determined to be zero which . indicates that a pixel is outside of the triangle and will result in no change in the stored value within the frame buffer; the method is still run through completely in order to maintain the full parallelism of the program which is of course necessary with SIMD architectures.

Referring now to the figure, block 1 indicates the initiation of the procedureu Block 2 indicates that a test is made to determine if all the triangles making up the complete image have yet been pro-cessed. If not, the system continues to block 3.

In block 3, the pixel processor obtains the data from the coordinate processor which is constant for a complete triangle or bounding box. These are the 13~9~q8 rates of change of rcrb, tlle rate of change o~
distan^es and the bevel w dths as shown in rIG. 10.
The procedure then con-tinues to block 4 where a test is made to determine if there are any more constitu-ent blocks within a bollnding bo~ for a particular triangle. If the last constituent block has been processad, the procedure branches bac~ to block ~, otherwise it proceeds to block 5.

In block 5, the data is read from the coordinate processor that is speci~ic to the particular con-stituent block being processed. These inslude the three distances from the origin of the constituent block to the triangle edges and the color intensity vallles rgk. This is the constituent block record which was described previously ~ith respect to FIG.
11. The procedure then moves to block 6 wherein the anti-aliasing factor (a) is computed. The details of this computation are e:iplained in FIG. 15.
..
In block 7, the shaded color intensity values rgb which are indicated in the Ligures as rr, gg, bb are computed as detailed in FIG. 16.

At this point, it should be noted that what is being done by the system is that a determination is made as to whether a computed shaded value or the old value from the frame buffer or a fraction of the two is written back into the frame buffer. To do this, blocks 8, 9 and 10 are invoked.

Block 8 reads the old color intensity values from the frame buffer, ro, go, bo. Block 9 calculates the new -olor intensity values rn, gn, bn using the antl-aliasing factor as aetailed in FIG. 17.

~ Y0986-065 ~3~qlq8 Block lC causes the new color intersit~r value to be stored in the frame buffer at the appropriate pixel location.

~ This ends the pixel ~rocessor routine and the procedure returns to block 4 where tests are made to see if there are any remaining constituent blocks to be processed for the particular triangle~

FIG. l5 shows the computation of the antl-aliasing factor ~a). The i and j values shown in the compu-tation block 2 are unique to each particular pixel processor and determine that the pixel p~ocessor is working on its particular pixel in the block. Thus, the pixel processor computing the pixel in the lower-left-hand corner or origin of a constituent lS block would have values i = 0, j = 0. Conversely, the piYel processor processing the pixel in the upper-right-hand corner of the constituent block , would have the values i = 3, j = 3 etc.

The function "determa" is in turn detailed Eurther in FIG. 18. This function is invoked three times to determi~e the three anti-aliasing factors al, a2, a3. In each instance the appropriate value of ddk and wk are utilized as will be understood.

Finall~, the anti-aliasing factor (a) for that particular pixel is determined as the minimum of al, a2, a3. This value is used in the computation block 9 of FIG. 14 which will also be well understood.

The following are brief definitions of the variables shown in block 2 of FIG. 15.

~ . , ~ YO986-065 1 3'J~ 1 qg dl,d2,d3 - are distarces, read from the coordinate ~rocessor, frcm the origin of a constitucnt block to the three edges of the triangle i,j - are indices unique to each pixel processor specifying the location of the pixel in the constituent block - they vary 4rom 0-3 as explained above ddl,dd2,dd3 - are computed distances from a pixel to the triangle edges "determa" i9 a function which returns (a) which is, in turn, based on a distance and bevel width for each edge (a~ - is the anti-aliasing factor .'., wl,w2,w3 - are the bevel widths (wk)-read once from the coordinate processor for each triangle.

FIG. 16 illustrates the computation of the shaded color intensity values of block 7 of FIG. 1~. As described previously with respect to FIG. 15, i and j are indices which are unique to the particular pixel processor and allow thé particular pixel 25 processor to adjust for its own particular pixel location in the constituent block. As will be apparent, the values r, g, b, utilized in the computation block 2 of the figure, were transmitted to the pixel processor in block 5 from the coordi-nate processor. The values rx, ry, gx, gy, bx, by, were transmitted to the pixel processors and the ~ YO986-065 1 3 ~

coordinate processor in block 3.

The following are ~rief definitions of the variables shown in the computation block 2 of FIG. 16.
~, r,g,b - are color intensity -~alues read from the coordinate processor for each constituent block.

i,j - are indices identical to those of FIG~ 15 and vary from 0-3.

rx,ry - are rate of change of eolor values along the gx,gy the x and y axes and are read once from hx,by the eoordinate processor for each triangle (bounding box).

FIG. 17 illustrates the anti-aliasing factor (a) being utilized to generate a new eolor intensity value rn, gn, bn to be returned to the frame buffer at the partieular pixel loeation at the end of the pixel proeessor procedure. All of the variables in the eomputational bloek 2 of FIGo 17 have been defined. Anti-aliasiny factor (a) was computed in the proeedure outlined in FIG. 15 and the shaded eolor intensity values rr, gg, bb were eomputed in the proeedure detailed in FIG. lÇ and the old eolor intensity values ro, go, bo are aeeessed from the frame buffer at the appropriate pixel location.
As one further note to the proeedure of FIG. 17, as stated previously, the anti-aliasing faetor (a) may be a 1, a 0, or a fraetion, the new eolor is a new (n) eomputed shaded valu~, the old ~o~ value in the . , ~ , . .

36 l 3 oql ~ ~

frame bu fer or a fractiorl of each. If (a) is a 1, the new pixel value for the pixel ~ust computed is placed unaltered in the frame buffer, if it is a 0, it is known that ~he pixel is outside the triangle and the old value in the frame buffer will be left unaltered; or ~inally if ~a) is a fraction, this frâction s used to accomplish the anti-aliasing in accordance with the procedure shown in the figure.

FIG. 18 specif~es the computations involved in determining whether the anti-aliasing factor (a) for the particular triangle edge relative to the pixel being processed is a 1, 0 or a fraction. As will be apparent, this procedure is followed to aetermine three anti-aliasing factGrs7 al, a2, a3 for each of the three edges~ of the triangle relative to a given pixel being processed. In block 2, the rate of change of distance ddk is compared with 0. If it is less than 0, the system proceeds to block 3 and the procedure returns a value of 0 for that particular anti-aliasing factor. If not, the procedure contin-ues to block 4 where the value ddk is compared with the width wk. If ddk is greater than wk, the system proceeds to block 5 and the procedure returns a value of 1 for the particular value of ddk. If ddk is less than wk, the system branches to block 6 where the specified ratio ddk/wk is returned as the value of the anti-aliasing factor for that edge.
Regardless of which path is taken, the function will take the same time to be executed. This will allow the pixel processors to all execute in parallel, as required of such an SIMD architecture~

FIGS. l9A and l9B are the diagrammatic representa-tions of the geometry and trigonometry involved in , ,~ , . . ; . .

~0986-065 37 l 3 ~ql 93 calculating the rate of change of distarces to the three edges of the triangle and the rate of change of colors as a pixel is moved unit directions in the x and y directions as descri~ed previously with 5 ~ respect to FIG. 10. More particularl~, FIG. 19A
describes the computation with the distances dkx, dky and FIG. 19B describes the computation of the three rates of change of color intensity values e.g.
rx, ry. In the top half of each figure, a diagram-1~ matic representation of the coordinate processor's computation is shown and in the bottom half of each figure, the evaluation of the indices i,j which is done in each pixel processor is ill~strated diagram-maticallyO Also a set of straight for~ard geometric formulas is set forth, by way of example, for each of these figures subsequentlv.

FIG.19A illustrates the geometric constructions used to compute the requisite data in the coordinate ~ processor and in the pixel processors. In the coordinate processor, the distance from the origin of a constituent block to an edge is calculated. In the pixel processor, the distance rom a pixel ~o an edge is calculatPd. In both cases, the rate of change valuss need to be known. These are calculat-ed by the coordinate processor and are shown as the values dxk and dyk (where k = 1-3). In both cases, a distance from an origin needs to be known. The coordinate processor determines these. The coordi-nate processor uses the formula for ddl to calculate the distance from the origin of a constituent block to edge 1 etc. A pixel processor uses the same formula to find the distance from its particular pixel to the edge utilizing the appropriate indices 38 1 309 1 q~ -It should be noted that if a point is in a triangle, then all distances from the lines to the point will all be positive or negative depending on whether points are chosen in clockwise or counterclockwise ~ direction. Therefore, if a point is moved from point 1 to point 2 and the first dis~ance i5 nega-tive, then reversing the direction and moving from point 1 to point 3 will produce a positive distance.
Three positive distances will then indicate that a point is in the triangle~ There,ore in a pixel processor if the ddl is r.egative, then ddl, dd2 and dd3 are multiplied by a -1. This insures that for a point iIl a triangle all the distances are positive.

For purposes of the present embodiment the index variables i,j are two bits wide. It is assumed that there are no multipliers in the pixel processor; but the multiplications can be rapidly carried out by addition. This simplifies the hardware required for , the pixel processors. Also in the embodiment i and j account for the "low order" effects on a distance or color calculation. The registers for calculating i*dx, far example, do not have to be wide. The dl value contains the high order effect of the final value for ddl.
.
In the second embodiment of-the invention, ~riefly described with respect to FIGS. 20 through 23, it is assumed that a pixel processor has a multiplier. In that case, i and j are si~nificantly wider quanti-ties since they will have to cover indices spanning the complete bounding box rather than just a four by foux constituent block. As will be shown, dl will be sent once for each triangle and a new i and j will be sent for each block. This second embodiment 39 1 3~ 1 q ~

reduces the total amount of data sent from the coordinate processor to the individual pixel proces-sors.
Calculated In Coordinate Processor:
.
2 Xl ~Y Y2 Yl L - ((Qx) 2 + (~y) 2) ~ = lenath of line segment dl = ~lY2-Yl~`2 = distance to an edge (1) from x,y L

dx = - ay = rate of change of distances per unit x L

dy = ~x = rate of change of distances per unit y L

Calculated_in Coordinate Processor (1) and Pixel ' Processor (2) ddl = dl + i dx + j dy = distance indexed to an edge from:

1) The origin of a constituent block in the coordinate processor, with i, j indexed from the origin of the bounding box.

2) A particular pixel within the constituent block with i, j indexed from the corner (origin) of the constituent block.

.. . . . .

YO9~6-065 40 13~q~q~

FIG. l9B shows the formuias for calculating ulti-mately a shaded red value (rr) proceeding from the coordinate processor through the operations in a pixel processor. As in the dis-tance calculations, the rate data and a red value r from an orisin is required. Identical calculations would, Oc course, be performed for blue (b) and green (g).

It should be noted that the coordinate processor calculations described above could alternatively be performed in the host ar.d transmitted to the coordi-nate processor and then to the pixel processors by the coordinate processor. They have been shown and described as bein~ performed in the coordinate processor for clarity.

Calculated in Coor~inate Processor D = (x2y3 - y~x3) + (Y1~3 lY3 (xlY2 - y2x2 ) r = (X2y3-y2x3)rl + (Y2~3~XlY3)r2 + (XlY2 Y2X2) 3 D
rx = (Y2-y3)r1 + (Y3-Yl)r2 + (~l-Y2)r3 D
ry = (x3-x2)r2 + (~1~X3)r2 +~X2-xl)r3 D

Calculated the Coordinate Processor (1) or in the Pixel Processor (2):
.
rr = r + rx . i + ry . y where it is the value of red in (1) the corner of . . .

'i0386-065 41 ~30~q~

the constituent block and (~ at a particu1ar pixel in the constitllent block.

A second embodiment of the invention will now be briefly described since it is ver~ similar to the prior embodiment with the exception of the fact that -the contents of the triangle and block recorcls are changed. In this embodiment, more data is p aced in the triangle record sent once to the pixel proces-sors and less d~ta is ~laced in the constituent block record which must be recomputed and transmit-ted each time a new constituent block is processed.

FIG. 0 is a figure quite similar to FIG. lO wherein the data contained in a triangle record is shown at the top of the figure and a diagrammatic representa-tion or the geometrical meaning of these records is shown in the bottom portion. Added to the iriangle record are the items OL data shown in the bottom line of the data record. These are dl, d~, d3, r, g, b. These represent, respectively, the distances to the edges from the corner of the bounding box and the color intensit~ of the colors ryb at the corner or origin of said bounding box.

FIG. 21 shows the ~ew constituent block record sent from the coordinate processor to the pixel proces-sors each time a new constituent block is being processed. The ;ndices i,c and j,c from the coordi-nate processor locate the origin of the particular constituent block being processed relative to the origin of the bounding box. In the pixel processors these will be added to i,p and j,p (p is for pixel) which are unique to the particular pixel within the constituent block being evaluated by the pixel . . ~ , .

~2 13~91~8 processor. As ~ill be understood, in the new pixel processor routine flow chart of figure 23, the pixel processor will be able o calculate the required distance dl, d2, d3 and color, rgb ~Jalues utilizin~
' x,i~ coordinate data calculated from these ne~
indices.

- It will be noted in FIG. 21 by way of example that a particular constituent block within a triangle is shown. ~5 indicated in the legend next to the figure, ic = 8 and ~c = 4 for the origin of this constituent block.

FIG. 22 ~omprises a 'low chart of the modified coordinate processor routine of the second embodi-ment. It differs from the flow chart of the first embodiment as set f~orth in FIG. 12 in those boxes marked by the symbol ~ , as descrihed previously for each triangle, the distance from the corner or origin of the bounding box is calculated once and also the color intensit~ value rgb of the corner or origin of the bounding box is similarly calculated by the coordinate processor. These values are sent to the pi~el processors in the "new" triangle record. This is indicated in block 3 of FIG. 2~.
Blocks 3 through 8 of the remainder of the flow chart remain identical to those of FIG. 12 with the slight e~ception of block 7; in this case, the additional information in the triangle record is sent to the pixel processors.

In block 9 of the procedure, ic and jc which locate the constituent block in the bounding box are computed by the coordinate processor and sent to the pixel processors on each iteration of this inner . .

!3 13~9t~'~

loop. As will be seen, olocks 26 and ~8 of FIG. 12, which calcul~ted the distances and colors, have been elimlnated from the coordinate processor routine.
The ne~ constituent block data containing the proper S ~ ic and jc are sent to the pixel processors in bLock 10 and the system returns to block 8 cf the inner loop.

FIG. 23 illustrates the new pixel processor routine and is accord,r.gly very cimilar to FIG. 14 described in detail previousl~. As stated with respect to FIG. ~2, all of the blocks of FIG. 23 remain identi-cal to those of FIG. 14 with the exceptions of those marked bv a symbol ~ . Thus in block 3 of the procedure of 23 the modified triangle data is read from the coordinate processor including the rgb and dl, d2, d3 values for the triangle referenced to the origin of the boundin~ bo:c.

The only other difference in the routine is included in bloc~ S in the pixel processor routine wherein only the values ic and jc need -to be read from the coordinate processor rather than the distances and colors for the origin of the specific constituent block. As was explained previously, these new indices will allow the proper evaluation of the values rgb and dl, d2 and d3 to be properly indexed - for purposes of computing the anti-aliasing factor in block 6. The same calculations are performed in block 6 through lO as were performed in the previ-ously described procedure of FIG. 14 with the exception that the values for ddl, dd2, dd3 are computed using dl, d2, d3 indexed now from the origin of the bounding box rather than from the origin of the indiv-dual constituent block. As will - YO9~6 065 44 1 3 o~l q 8 ..
be appreciated, the result will be identical.

This completes the description cf the present preferred embodiment of the invention with the modifications described relative to the ~low charts of FIGS. 22 and 23. There are many other changes and modifications which could be made in the details - of the procedure without departing from the spirit and scope of the present invention as set forth in the appended claims, namely of providing a complete-ly parallel procedure which mar be performed on all the processors o~ an SIMD architected pixel proces-sor system which allows such an SIMD architecture to be used Lor the complicated procedures necessary in rendering smoothly shaded, anti-aliased, three-di-mensional, co].or triangles onto a two-dimensional surface.

. , ~ Yos~6-06s ~ 45 l 39~

REFERENCES

[1] (Gou) Gourard H., "Continuous Shading of Curved Surfaces," IEEE Transac~ions or Computers, C-20(6), pp. 623-628. June 1971.

5- [2~ (Fly) Flynn, M. J., "Very High Speed Computing Systems," Proc. of the IEEE, 54:1901-9, 1966.

[3] (Fol) Foley, J. and A. Van Dam, "Fundamentals of Interactive Computer Graphics~"
Addison-Wesley, New York, 1982.

l~ [4] ¦N~W) Newell, M. E., R. G. Newell and T. L.
Sancha, "A New Approach to the Shaded Picture Problem," Proc. ACM Nat. Conf., pp. 443, 1972.

[5] (Sch) Schacter, Bruce J. (editor), "Computer ~ Image Generation," John Wiley and Sons, New York, 1983.

[6] (Tho) Thomas Jr., George B., "Calculus and Analytic Geometry," Addison-Wesley, Reading, MA, 1966.

.. , , . ,. , , ~ ~ , ` '~0986-065 ~, ~6 1 30q 1 q~

Description oE a ~referred ~ m Embodiment The follo~ing material comprises a program capable of executlng the pixel processor operations in an SIMD environment. It is written in the Pascal programming language. The actual program statements are in the left-hand column and various descriptive ~ "comments" relating to particular instructions or subsets of instructions are in the right-hand column as is conventional.

The program is intended to run in each processor of the ~ x N array of SIMD pixel processors to perform the shading and anti-aliasing functions for each of the associated M x N pixels in a particular constit-uent block of a bounding box as described previous-ly. The program assumes that the coordinate proces-sor communicates with each of the M x N pixel processars through a set of registers made up of three parts:

1) An op register to tell the pixel processors what type of data is in the data registers and what operations is to be performed.

2) x, y, z, and zz data registers which store all data read from the coordinate processor. zz is the index into an M x N array of pixels in the frame buffer, e.g., it is the origin address of the particular bounding block whose pixels are currently being processed (first embodiment). Each pixel consists of three color values Ir, g, and b). There is a separate pixel storage array to which each pixei processor is connected. All of these latter arrays together make up the frame buffer.
. . , ~09~6-065 47 1 3 ~91 q ~

3) Cpflag and ppflag are each olle bit flags for handshaking between the coordir.ate and pixel proces-sor~ When the coordinate processor writes to the pixel processor, it sets cpflag and when it reads, S it clears ppflag. Complimentary rules apply to the pixel processor. Handshaking is a conventional procedure in any communication system. The opera-- tion is both straightforward and well known and need not be described further.

The program 'tri' is program for a pixel processor communicating with the coordinate processor ~ia the registers described. For the purposes of this program, the coordinate and pixel processors are assumed to be conventional processors in terms of 15 instructlons performed, etc... The subsequent description comprises the overall procedure and the main variables in the program 'tri'.

- The followin~ comprises a listing/definition of the roqram variables and main program on the left side with comments on the right side.

Program tri;

var op,x,y,z,zz, ~ (hardware registers and rlags) cpflag,ppflag ( " ~ ) rx,gx,bx, (incremental values passed initially) ry,gy,by, dlx,d2x,d3x, -dly,d2y,d3y, (the next 6 values are supplied each iteration) d],d2,d3, (distances between a point and the sides of the triangle) 48 ~ f~

r,g,~, (red, qreen, and blue trianyle color values read from coord processor) i,j, (known pi~.el processor indices) opcode (holds op from eoord processor~
:integer;
fb:array(02,0.... max) of integer;

10 begin (of main program) waitin;
(zz needs to be latched for frame buffer aecesses) 15 Opcode =op;
(cp=coordiante processor) (pp=pixel processor) ease opcode of (the following are done once per tr-angle) 1: readep(rx,gx,bv);
2: readcp(ry,gy,by);
3: readep(dlx,d2x,d3x);
4: readep(dly,d2y,d3y);
(the following are done eaeh iteration) 5: begin readep(dl,d,',d3);
6: begin readep(r,g,);
' eompute; end;
otherwise; (don't do anything) end; (of main program) ( 30 The followinq is a li.sting of the ma-n subroutines on the left with eomments on the riaht.
__ ____ *~ ,.
proeedure eompute;

35 var ddl,dd2,dd3, (temp variables) wl,w2,w3, al,a2,a3, rr,gg,bb, ( "
40 a, (anti-alias value ) ro,go,bo, lold fb eolor values) rn,gn,bn (new fb eolor values) :integer;

Y098~-065 49 1 3~9 1 ~Q

begin (calculate the anti-alias value (a) for or.e triangle) ~ ddl:= dl + i*dlx + j*dly; (calculate a distance dd2:= d2 + i*d2x + j*d2y: for this pixel) dd3:= d3 + i*d3x + j*d3y;
wl:= max(dlx,dly); (calculate the bevel ~Jidth w2:= max(d2x,d2y); ~wk~ inside the tri.) w3:= max(d3x,d3y);
al:= determa(ddl,wl); (determine anti-aLiasing a2:= determa(dd2,w2);
values) a3:= determa*dd3,w3);
a:= min(al,a2,a3); (calculate the (a) for the triangle) ______________________ (* calculate the new (using the coord colors, color values *) calculate a smoothed value. Will be used ]ater to calcul-ate an anti-aliased color value.) rr:- r -1 i*rx + j*ry;
gg:= g + i*gx + j*gy;
bb:= b ~ i*bx ~ j*by;
________.~__________ _____ readfb(ro,go,bo); (read from frame buffer) rn:= a*rr + (l-a)*ro; (calculated new colors gn:= a*gg + (l-a)*go; which are anti-aliased by the (a~ amount) bn:= a*bb ~ a)*bo; (the two colors mixed are from the frame buffer and from the smoothed pixel just calculated) writefb~rn,gn,bn); (write the frame buffer with 1) averaged value, 2) the old value from the frame buffer, or 3) the value from coordinate processor) end;
____ ______________~_____ Y09-86-065 50 1 3Q~

The followinq ~ sed in the proqram on the left with comments on the riqht.

procedure waitin; (pp waits for input from (*--------------------*) the cp) begin repeat until cpflag=l;

procedure readcp(a,b,c: integer; (reads from cp) begin a:=x; b:=y; c:=z;
cpflag:=0 end ____________________ __________________________________________ function determa(dk,wk: .integer) :integer;
__________________________ _______________ begin if dk<=0 (returns (a) as a value then determa := 0 of 0,1 or a fraction) else if dk<wk then determa := dk div wk else determa := l;
end;

procedure writecp(a,b,c: integer);(pp writes to cp) begin ppflag:=l;
x:=a; y:=b; z:=c;
end;

procedure readfb(a,b,c: integer);
~*__,~________ ) begin a:~fb(O,zz); b:=fb(l,zz); c:-fb(2,zz) end;

l 309 ~

r~rocedure writerh(a,b,c: intc~er); (uses zz t.o write the ~rair e buf fer) begin fb(O,zz) :=a; fb(l,zz) :=b; fb(2,zz) :=c end;

Claims (14)

The embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. In a video display system including a host processor, a coordinate processor and a plurality of pixel processors organized in an SIMD architecture, a frame buffer for storing video data to be displayed on a display monitor, an improved method for rendering smoothly shaded, anti-aliased, three-dimensional triangles into said frame buffer wherein there are M x N processors for rendering M x N contiguous pixels of video data into said frame buffer, said pixel processors being functionally organized as an M x N array, said frame buffer being organized into a plurality of rectangular arrays of addressable M x N constituent blocks of pixels, each processor of said M x N array of processors being selectively connectable to a particular spatially related pixel storage location in each of said array of M x N pixel constituent blocks in said frame buffer, said method comprising a parallel procedure operable simultaneously in all of said pixel processors which in turn comprises the steps of:
1) determining the coordinates of a bounding box which completely encloses a particular triangle to be rendered, said bounding box being composed of a c x r array of M x N pixel constituent blocks, 2) determining the origin of the first constituent block, and the number of rows (r) and columns (c) of constituent blocks, included in said bounding box, 3) evoking said M x N array of pixel processors to operate completely in parallel to render a complete M x N constituent block of pixels, 4) updating the frame buffer with the newly rendered M x N array of pixels computed by the pixel processors, 5) repeating the operation of said M x N
array of pixel processors to render consecutive M x N constituent blocks of pixels until all such constituent blocks included in a particular bounding box have been rendered, and 6) utilizing the rendered contents of the frame buffer to drive the display monitor.

2. In a video display system including a host processor, a coordinate processor and a plurality of pixel processors organized in an SIMD architecture, and a frame buffer for storing video data to be displayed on a display monitor, an improved method for rendering smoothly shaded, anti-aliased, three-dimensional triangles into said frame buffer which comprises a parallel procedure operable simultaneously in all of said pixel processors wherein there are M x N processors rendering M x N contiguous pixels of video data into said frame buffer including the steps of:

1) accessing from the coordinate processor data about a triangle being rendered including rate of change of distances in both the x and y direction relative to the three sides of the triangle as a reference pixel moves across the display, rate of change of color intensity values as the pixel moves across the display and bevel widths for the three sides of the triangle which define regions in which anti-aliasing is to be performed, 2) determining if all of the constituent blocks in a bounding box have been evaluated and, if not, 3) accessing from the coordinate processor computed distance values to each of the three sides of the triangle from the origin of said M x N array and the color intensity values at the origin of said array, 4) indexing said distance and color data accessed in Step (3), for the particular pixel in said M x N array being processed by a particular processor, 5) computing the anti-aliasing factor (a) for the particular pixel being rendered, 6) computing the shaded color values (rr,gg,bb) for the particular pixel being shaded, 7) accessing the current color value (ro,go,bo) of the pixel from the frame buffer, 8) calculating new color values, (rn,gn,bn) for the pixel being rendered using the computed anti-aliasing factor (a), the shaded color values (rr,gg,bb) and the current color values (ro,go,bo), 9) storing this new color value in the frame buffer for access by the display monitor, 10) returning to step (2), 11) if it is determined that all constituent blocks in a bounding block have been evaluated returning to step (1).

3. In a video display system including a host processor, a coordinate processor and a plurality of pixel processors organized in an SIMD architecture, and a frame buffer for storing video data to be displayed on a display monitor, an improved procedure for rendering smoothly shaded anti-aliased three-dimensional triangles into said frame buffer which comprises a parallel procedure operable in all of said pixel processors simultaneously wherein there are M x N
processors renderinq M x N contiguous pixels of video data into said frame buffer including a method for preprocessing data to be utilized in said rendering procedure operable in said coordinate processor which comprises 1) accessing from the host processor the three-dimensional coordinates and the color intensity values for three vertices of a color triangle to be rendered, 2) determining the coordinates of a bounding box which completely encloses a particular triangle to be rendered, composed of a c x r array of M x N pixel constituent blocks, 3) determining the origin of the first constituent block, and the number of rows (r) and columns (c) of constituent blocks, included in said bounding box, 4) calculating the distances to each of the three sides of the triangle with respect to the frame buffer origin, 5) calculating the color intensity values at the frame buffer origin, 6) calculating rate of change of distance to the three edges of the triangle when moving one pixel along the x and y axes, 7) calculating the bevel widths for each of the three sides of the triangle, 8) calculating the rate of change of colors when moving one pixel along both the x and y axes, 9) broadcasting the data computed in Steps (6), (73 and (8) to all of the pixel processors, 10) determining if all of constituent blocks in a bounding box surrounding said triangle have been evaluated and if not 11) calculating the distance from the origin of the constituent block to the three edges of the triangle, 12) calculating the color values of the pixel at the origin of the constituent block, 13) broadcasting the items calculated in Steps (11) and (12) to the pixel processors, 14) again determining if said all of the constituent blocks in a bounding box surrounding the triangle have been evaluated and if so 15) determining if all of the triangles forming a complete display have been evaluated and if so exiting from the procedure, and if not returning to Step (1), 16) if it is determined in Step (10) that all of the constituent blocks have not been evaluated, determining the origin of the next constituent block and repeating Steps (11), (12) and (13).

4. An improves method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 2 wherein the step of computing the anti-aliasing factor (a) comprises, the steps of, 1) determining the distance (ddk) of the pixel from each of the three edges of the triangle, 2) determining an anti-aliasing factor a1, a2, a3 relative to each edge of the triangle wherein (ak) = 0 if a pixel is outside of the triangle, (ak) = 1 if a pixel is totally within the triangle, and (ak) = ddk wk if the pixel lies within the bevel for a particular edge and 3) selecting as the anti-aliasing value (a) for a particular pixel, the minimum of a1, a2, a3.

5. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 2 wherein the step of calculating the new color values (rn, bn, gn) for a pixel comprises calculating for each color (r, g, b) the following functions:

rn = a * rr + (1-a) * ro gn = a * gg + (1-a) * go bn = a * bb + (1-a) * bo wherein: * implies multiplication rr, gg, bb are the calculated shaded values, ro, go, bo are the old color values stored in the frame buffer.

6. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 2 wherein the step of computing the shaded color values (rr, gg, bb) for a particular pixel comprises calculating for each color the following functions:

rr = r + i * rx + j * ry gg = g + i * gx + j * gy bb = b + i * bx + j * by wherein: * implies multiplication rx, ry, gx, gy, bx, by are rate of change of color, i and j are indices for the pixel relative to the origin of the constituent block, r, g, b, are computed intensity values from the coordinate processor at the origin of a constituent block.

7. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 3 wherein the step of calculating the bevel widths (w1, w2, w3) comprises the steps of evaluating the six rate of change of distance values for a particular triangle d1x, d1y, d2x, d2y, d3x, d3y according to the relationship:

w1 = max (¦d1x¦, ¦d1y¦) w2 = max (¦d2x¦, ¦d2yl) w3 - max (¦d3x¦, ¦d3y¦) and utilizing the three bevel widths selected for subsequent calculations for the triangle.

8. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 3 wherein the calculations in steps (11) and (12) of claim 3 are indexed to the origin of the constituent block being evaluated.

9. In a video display system including a host processor, a coordinate processor and a plurality of pixel processors organized in an SIMD architecture and a frame buffer for storing video data to be displayed on a display monitor, an improved method for rendering smoothly shaded, anti-aliased, three-dimensional triangles into said frame buffer which comprises a parallel procedure operable simultaneously in all of said pixel processors wherein there are M x N processors rendering M x M contiguous pixels of video data into said frame buffer including the steps or:

1) accessing from the coordiante processor data about a triangle being rendered including rate of change of distances in both the x and y direction relative to the three sides of the triangle as a reference pixel moves across the display, rate of change of color intensity values as the pixel moves across the display, the bevel widths for the three sides of the triangle which define regions in which anti-aliasing is to be performed, the distances to each of the three sides of the triangle from the bounding box origin and the color intensity values at the bounding box origin, 2) determining if all of the constituent blocks in a bounding box have been evaluated and if not, 3) accessing from the coordinate processor index values locating the origin of the constituent block with respect to the bounding box origin, 4) indexing distance and color data from data accessed in steps (1) and (3) for the particular pixel in said M x N array being processed by a particular processor, 5) computing the anti-aliasing factor (a) for the particular pixel being rendered, 6) computing the shaded color values (rr,gg,bb) for the particular pixel being shaded, 7) accessing the current color value (ro,go,bo) of the pixel from the frame buffer, 8) calculating new color values, (rn,gn,bn) for the pixel being rendered using the computed anti-aliasing factor (a), the shaded color values (rr,gg,bb) and the current color values (ro,go,bo), 9) storing this new color value in the frame buffer for access by the display monitor, 10) returning to step (2), 11) if it is determined that all of the constituent blocks in a bounding box have been evaluated, returning to step (1).

10. In a video display system including a host processor, a coordinate processor and a plurality of pixel processors organized in an SIMD architecture, and a frame buffer for storing video data to be displayed on a display monitor, an improved procedure for rendering smoothly shaded anti-aliased three-dimensional triangles into said frame buffer which comprises a parallel procedure operable in all of said pixel processors simultaneously wherein there are M x N
processors rendering M x N contiguous pixels of video data into said frame buffer including a method for processing data to he utilized in said rendering procedure operable in said coordinate processor which comprises 1) accessing from the host processor the three-dimensional coordinates and the color intensity values for three vertices of a color triangle to be rendered, 2) determining the coordinates of a bounding box which completely encloses a particular triangle to he rendered, composed-of a c x r array of M x N pixel constituent blocks, 3) determining the origin of the first constituent block, and the number of rows (r) and columns (c) of constituent blocks, included in said bounding box, 4) calculating the distances to each of the three sides of the triangle with respect to the bounding box origin, 5) calculating the color intensity values at the bounding box origin, 6) calculating rate of change of distance to the three edges of the triangle when moving one pixel along the x and y axes, 7) determining the bevel widths for each of the three sides of the triangle, 8) calculating the rate of change of colors when moving one pixel along both the x and y axes, 9) broadcasting the data computed in Steps (4), (5), (6), (7) and (8) to all of the pixel processors, 10) determining if all of constituent blocks in a bounding box surrounding said triangle have been evaluated and if not 11) calculating the indices ic and jc which locate the origin of the current constituent block relative to the origin of the bounding box, 12) broadcasting the items calculated in Step (11) to the pixel processors, 13) again determining if said all of the constituent blocks in a bounding box surrounding the triangle have been evaluated and if so 14) determining if all of the triangles forming a complete display have been evaluated and if so exiting from the procedure, and if not returning to Step (1), 15) if it is determined in Step 10 that all of the constituent blocks have not been evaluated, repeating Steps (11) and (12).

11. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 9 wherein the step of computing the anti-aliasing factor (a) comprises the steps of 1) determining the distance (ddk) of the pixel from each of the three edges of the triangle, 2) determining an anti-aliasing factor a1, a2, a3 relative to each edge of the triangle wherein (ak) = 0 if a pixel is outside of the triangle, (ak) = 1 if a pixel is totally within the triangle, and (ak) = if the pixel lies within the bevel for a particular edge and 3) selecting as the anti-aliasing value (a) for a particular pixel, the minimum of a1, a2, a3.

12. An improved method for rendering shaded claim 9 wherein the step of calculating the new color values (rn, bn, gn) for a pixel comprises calculating for each color (r,g,b) the following functions:

rn = a * rr + (1-a) * ro gn = a * gg + (1-a) * go bn = a * bb + (1-a) * bo wherein: * implies multiplication rr,gg,bb are the calculated shaded values, ro,go,bo are the old color values stored in the frame buffer.

13. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 9 wherein the step of computing the shaded color values (rr,gg,bb) for a particular pixel comprises calculating for each color the following functions:
rr = r + i * rx + j * ry gg = g + i * gx + j * gy bb = b + i * bx + j * by wherein: * implies multiplication rx,ry,gx,gy,bx,by are rate of change of color, i and j are indices for the pixel relative to the origin of the constituent block, r,g,b are computed intensity values from the coordinate process-or at the origin of a constituent block.

14. An improved method for rendering shaded anti-aliased three-dimensional color triangles into a frame buffer as set forth in claim 10 wherein the step of calculating the bevel widths (w1,w2,w3) comprises the steps of evaluating the six rate of change of distance values for a particular triangle d1x, d1y,d2x,d2y,d3x,d3y according to the relationship:

w1 = max (¦d1x¦,¦d1y¦) w2 = max (¦d2x¦,¦d2y¦) w3 = max (¦d3x¦,¦d3y¦) and utilizing the three-bevel widths selected for subsequent calculations for the triangle.