The present invention relates generally to graphics processing, and more particularly to triangle interpolation processing in homogeneous space.
Polygon (triangle) interpolation is one of the computationally intensive tasks in graphics hardware because this interpolation is done for each pixel of each object in a scene. Each triangle may have several attributes like depth, colors, numerous textures and the quality of the image generated is greatly dependent on the accuracy of interpolation. The overall frame generation rate also depends on interpolation speed. The hardware implementation of these functions takes significant part of the gate budget of a modern graphics chip and is one the most critical parts of the chip.
Several solutions to the triangle interpolation problem have been implemented in different graphics architectures. Most of the current approaches fall into the following three categories: (i) fixed point interpolation in screen Cartesian coordinates with perspective correction; (ii) fixed point interpolation in screen barycentric coordinates with perspective correction; and (iii) fixed point interpolation in homogeneous coordinates.
Approaches in the first two categories require projecting of all parameters to screen space (i.e., division by W) for further interpolation and later perspective correction (i.e., division by 1/W) per pixel. These approaches are shown in
The third approach avoids the redundant projection of parameters to screen space with further correction and calculates the same homogeneous barycentrics, as shown in
More particularly, disadvantages of the first approach, fixed point interpolation in screen Cartesian coordinates with perspective correction, include: (a) redundant calculation with projection (division by W) of all parameters to screen space to make it linear; (b) the steps of project, interpolate, and correct to recover the true value; (c) redundant parameter delta setup calculation for interpolation including 1/W (accuracy problems arise); (d) redundant true parameter value recovery in each pixel by dividing to interpolated 1/W_pix value; and (e) a significant amount of dedicated hardware.
Disadvantages of the second approach, fixed point interpolation in screen barycentric coordinates with perspective correction, include: (a) same redundant calculation with projection (division by W) of all parameters to screen space to make it linear; (b) the steps of project, interpolate, and correct to recover the true value; (c) same redundant true parameter value recovery in each pixel by dividing to interpolated 1/W_pix value; and (d) a significant amount of dedicated hardware, which cannot be used for any other tasks.
Disadvantages of the third approach, fixed point interpolation in homogeneous coordinates, include: (a) calculations for pixel barycentrics must all have been done at the pixel level and, in the case of multi-pixel triangles, the number of calculations grows multiplicatively; and (b) a significant amount of hardware dedicated to the task.
Thus, there is a need for a method and apparatus pertaining to polygon interpolation in graphics hardware that significantly reduces the number of calculations needed and does not require a significant amount of dedicated hardware.
The present invention is directed towards the above-mentioned need. The present invention addresses the problem of fast, accurate and efficient polygon interpolation in graphics hardware to find correct value of each parameter (Z, color, multiple texture coordinates) of every pixel of triangle. For algorithm implementation, the present invention uses a programmable single instruction multiple data (SIMD) scalar unit, which can be also used for further pixel processing according to Microsoft D×9,10 and OpenGL API requirements for programmable graphics machines. The triangle interpolation algorithm of the present invention can be executed on this programmable SIMD scalar unit with efficiency that is close to the efficiency of dedicated hardware. In fact, all interpolation operations are executed in floating point arithmetic with the highest possible accuracy.
In accordance with a purpose of the invention as described herein, a method is provided for obtaining an attribute within a triangle. The method steps include (1) obtaining the vertices of a triangle, where each vertex is represented by a set of coordinates in a world coordinate space and has at least one attribute, (2) for each vertex, transforming the world space coordinates and the attribute of the vertex to coordinates and an attribute in viewer space to create viewer space coordinates and a viewer space attribute, where said viewer space coordinates are homogeneous coordinates, computing a set of homogeneous coefficients of the triangle based on the viewer space vertex homogeneous coordinates, where said homogenous triangle coefficients include perspective data, and projecting the viewer space coordinates of the vertex to coordinates in 2D screen space, (3) determining, in the 2D screen space, pixels that are affected by the triangle based on the 2D screen space coordinates, and (4) for each pixel affected by the triangle, computing, based on the homogeneous triangle coefficients, a set of barycentric coefficients in viewer space, and performing a linear interpolation based on the set of viewer space barycentric coefficients and the viewer space attributes of the triangle vertices to obtain the attribute of the pixel affected by the triangle.
One advantage of the present invention is the invention uses fewer arithmetic operations compared to prior art implementations and, thus, has better performance compared to triangle interpolation in screen space and later perspective correction.
Another advantage of the present invention is that it avoids parameter projection for interpolation in screen space and further perspective correction computations to recover true value in the viewpoint space.
These and other features, aspects and advantages of the present invention will become better understood with regard to the following description, appended claims, and accompanying drawings where:
In a graphics system embodying the present invention, primitives are decomposed into triangular polygons that are further processed to provide a better representation of the images. To illustrate such processing,
To allow better understanding of the method in
Thus, in the triangle level, the triangle data set is first characterized in the world coordinate space by the three vertices with texture coordinates as follows:
The world coordinate space (three-dimensional space) serves to define objects prior to any geometric transformations. Given this triangle data set in the world coordinate space, the triangle data set is then transformed into viewer coordinate space and clipped to a view volume. Essentially, the triangle first specified in world space is treated with respect to a defined viewpoint. In the viewer space, the eye or viewpoint is the origin for coordinates and the view ray is along the Z-axis. The resulting triangle data set in viewer coordinate space, i.e., the triangle 120′ with coordinates in viewer (homogenous) space as shown in
A triangle setup phase follows the transformation to viewer coordinate space, in which, three (3) sets of the homogenous coefficients (ai, bi, ci) are derived from the triangle data set in the homogenous coordinate space, where, for example, a1=Y2h*W3−Y3h*W2. More broadly, the coefficients are derived by:
ai=Yjh·Wk−Ykh·Wj;
bi=Xjh·Wk−Xkh·Wj; and
ci=Xjh·Ykh−Xkh·Yjh,
and where i, j & k=1, 2 & 3.
Concurrently, as is preferred, the triangle viewer coordinate data set is projected to the screen coordinate space, to produce {[X1s, Y1s], [X2s, Y2s], . . . [X3s, Y3s]}.
Screen coordinate space involves further transformations where the object geometry in screen coordinate system is projected and transformed to give depth to the displayed object. Accordingly, as shown in
As further shown in
Lastly, in the interpolation phase, the homogeneous coordinates and attributes for each pixel are linearly interpolated using the barycentric coordinates previously calculated. This includes linear interpolation of the Z-coordinate in homogenous space, Zh, the perspective correction parameter, W, and the attributes, Pah, Pbh, . . . in the homogenous space. For each pixel, the resulting Z-coordinate, Z_pix, is:
Z—pix=α·Z1h+β·Z2h+γ·Z3h;
The resulting perspective correction parameter, W_pix, is:
W—pix=α·W1+β·W2+γ·W3; and
for each of its attributes, after interpolation the resulting attributes, Phi, for each pixel, is
Phi—pix=αP1hi+β·P2hi+γ·P3hi.
After the linear interpolation, the resulting pixels data with depth is described as:
More particularly, together,
X=αx1+βx2+γx3, and
Y=αy1+βy2+γy3.
It is noted that the point P is defined by a unique set of barycentric coordinates and further that the coordinates satisfy the requirement that α+β+γ=1.
Thus, based on the barycentric coefficients, the coordinates, X,Y, of each pixel affected by the triangle can be derived. This means that the point P can be calculated as a linear combination of the vertices (P1, P2, P3). As the barycentric coefficients are applicable to interpolation of any attribute of the pixel, including Z-coordinate, color, UV coordinates, etc., once they are calculated the attributes can be calculated as well. In particular, α, β, and γ are calculated as shown in Equation (1) of
When the three triangle vertices are enumerated with i=1, 2, 3, j=i mod3+1, and k=j mode3+1, and the substitutions into Equation (1) of ai=yi−yk, bi=xk−xj, ci=xj*yk−xk*yj, are made, as shown in Equation (2), the barycentric coefficients can be restated as shown in Equation (3).
As shown in Equation (4), w, the perspective correction parameter of the point P, is a function of the barycentric coordinates, α, β, and γ, and the perspective correction parameters, w1, w2, w3, of the three vertices. It is noted that the perspective correction parameter, w, is preserved in order to preserve the information on the distance to the viewpoint. However, before re-calculation of the barycentric coordinates in the homogenous (viewpoint) space can be done, the vertices, xi,yi, are converted to coordinates in the homogenous space, {tilde over (x)}i,{tilde over (y)}i, using the corresponding parameters, wi. The coordinates ({tilde over (x)},{tilde over (y)}) of P in the homogenous space are then derived using w.
Then, a number of calculations are performed to allow for the calculation of the barycentric coordinates in the homogenous space. These include 1) calculating three sets of homogenous space coefficients, ai, bi, ci, (per triangle) which are shared for every pixel affected by the triangle, 2) converting these to coefficients in the homogenous space, ãi,{tilde over (b)}i,{tilde over (c)}i, using the perspective correction parameters, wi, and 3) calculating di (per pixel), as shown in Equation (6) of
The difference between the approach of the present invention and the method of
Yet another difference is that the present invention provides very accurate and fast interpolation of triangles using the same ALUs for both triangle and pixel processing, with arbitrary interleaving of triangle and pixel processing instructions. The hardware SIMD unit used for interpolation is scalable to increase the performance of barycentric interpolation. The hardware unit is very flexible in terms of accuracy and performance, and it can be used for other graphics processing tasks to improve overall performance balance. Indeed, the programmable SIMD scalar unit, used for the algorithm implementation, can be used for further pixel processing according to Microsoft D×9,10 and OpenGL API requirements for programmable graphics machines. Also, the triangle interpolation method, in accordance with the present invention, is more suitable for implementation on programmable graphics processing unit and gives an efficiency close to the efficiency of a dedicated hardware implementation.
An apparatus of the present invention is implemented in one universal programmable unit replacing separate depth, color and texture address interpolators. To illustrate,
This hardware unit 10 is implemented as a two-ALUs unit where the scalar ALUs, 16 and 18, are implemented with shifted processing cycles. The cycle shifting is accomplished with the bypass registers 30 and 32. The unit 10 can process, in each clock, one triangle or a number of pixels (e.g., 2–4 pixels) depending on the required accuracy, and with the arbitrary interleaving of triangle and pixel processing instructions. The Reciprocal unit 20, is used for processing the division calculations (as described with reference to the method of
Also included in the hardware unit 10, are the triangle rasterizer 22, and the blank pixel screen coordinates data it produces are stored in pixel memory 24. The vertex geometry processing unit 26, produces data of the triangle coordinates in the homogenous space and this data is stored in the triangle memory unit 28.
To see how these and other instructions are used, an example program is set forth below. In this program, the instruction XPRDL is used a number of times to perform the calculations required by Equation (6) for i=0, 1, and 2 (i.e., for a0, a1, a2, b0, b1, b2, c0, c1, c2). Note that the index ‘i’ in this program equals 0, 1, 2 rather than i=1, 2, 3, as described above, but the results are not affected by this. For pixel barycentric coefficients calculations (α,β,γ), as required by Equation (5), the instructions include MOV, FBL (‘Folded Blend Mode’), FWD, FSUBL (“Folded Long”), etc. For pixel attribute interpolation as shown in
More specifically, below is the listing of the required calculations and their implementation in the instruction set of the SIMD ALU, for the (i) triangle setup and (ii) pixel processing, and (iii) pixel attribute interpolation.
To implement the forgoing required calculations, the following instructions are executed (with 9 instructions, for SIMD 4 bits (nibble) mode):
To implement the forgoing required calculations, the following instructions are executed (with 7 instructions):
To implement the forgoing required calculations, the following instructions are executed (with 2 instructions, for SIMD 8 bits (byte) mode):
Finally,
Thus, the method and interpolation algorithm of the present invention are implemented in floating point arithmetic units with high efficiency and accuracy; a performance improvement is made in triangle interpolation for triangles with size more than 1 pixels compared to prior art methods; and simplification is made of pixel level interpolation hardware by splitting the homogeneous barycentric coordinate calculation to triangle level and pixel level.
Although the present invention has been described in considerable detail with reference to certain preferred versions thereof, other versions are possible. Therefore, the spirit and scope of the appended claims should not be limited to the description of the preferred versions contained herein.
This application claims the benefit of U.S. Provisional Application Ser. No. 60/419,881, filed Oct. 19, 2002, and entitled “METHOD AND PROGRAMMABLE DEVICE FOR TRIANGLE INTERPOLATION IN HOMOGENEOUS SPACE,” which application is incorporated herein by reference.
Number | Name | Date | Kind |
---|---|---|---|
5361368 | Herzl et al. | Nov 1994 | A |
5361386 | Watkins et al. | Nov 1994 | A |
5550960 | Shirman et al. | Aug 1996 | A |
5877773 | Rossin et al. | Mar 1999 | A |
6137497 | Strunk et al. | Oct 2000 | A |
6275235 | Morgan, III | Aug 2001 | B1 |
6380936 | Olynyk | Apr 2002 | B1 |
6414683 | Gueziec | Jul 2002 | B1 |
6646648 | Donham | Nov 2003 | B1 |
6664958 | Leather et al. | Dec 2003 | B1 |
6686924 | Mang et al. | Feb 2004 | B1 |
Number | Date | Country | |
---|---|---|---|
20040145589 A1 | Jul 2004 | US |
Number | Date | Country | |
---|---|---|---|
60419881 | Oct 2002 | US |