JP2002008059A - Data processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform a perspective conversion at a high speed by approximation of division by a lower cost arithmetic. SOLUTION: As the perspective projection converting processing for projecting aggregation of plural vertexes (Xi, Yi, Zi) of a three dimensional shape onto an XY perspective plane surface, a first processing for setting reference points (Xc, Yc, Zc) of the aggregation of the vertexes, a second processing for computing a difference (dZi) between a Z-coordinate (Zc) of the reference point and a Z-coordinate (Zi) of the vertex, and a third processing for computing a perspective x-coordinate (SXi) and a perspective Y-coordinate (SYi) corresponding to the X-coordinate and the Y-coordinate of the vertex with an approximate formula using the difference (dXi) and a perspective Z-coordinate (SZ) on the XY perspective plane surface in relation to the Z-coordinate of the reference point, are carried out. The approximate formula uses a formula that a progression obtained by expanding the Maclaurin's series related to a differential function f(dZ)=1/(Z+dZ) is cut at an absolute term and a primary term.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to perspective transformation (projection transformation) relating to three-dimensional graphics processing, and more particularly to a technique effective when applied to processing of fine images.

[0002]

2. Description of the Related Art In the field of three-dimensional graphics, three-dimensional objects are represented by a group of polygons called polygons, and image processing is applied to them. For reference on image processing for polygons, see Addison
A document published in 1982 by Wesley, Inc. (by JD Foley & A. Van Dam)
Fundamentals of Interacti
ve Computer Graphics ".

As a method of expressing an object represented by polygons with high definition and smoothness, it is conceivable to make the polygons constituting the object finer and increase the number of necessary polygons. However, by increasing the number of polygons,
The amount of calculation increases.

[0004] In particular, the present inventors have studied the amount of operation of perspective transformation in three-dimensional graphics processing requiring division. In perspective transformation of three-dimensional graphics processing,
As described on page 113 of the Computer Graphics Lecture (Yonao Aoki, Corona), vertices in a three-dimensional space that is a world coordinate system are projected onto a two-dimensional space that is a perspective coordinate (screen coordinate) system. Perform the following processing.
In this perspective transformation, a geometric operation of SXi = Xi * SZ / Zi SYi == Yi * SZ / ZI may be performed. Where (Xi,
(Yi, Zi) is the vertex coordinates to be converted, and (SXi, SY)
i) is the coordinate after conversion, and SZ is the Z coordinate of a perspective plane (screen plane) which is a perspective coordinate system. In perspective transformation, 3
The above operation is performed on all vertices of the three-dimensional object, and the three-dimensional object is projected on a two-dimensional screen plane.

[0005]

In the perspective transformation, division must be performed for each vertex every time, as is apparent from the above equation. In general, division requires a larger number of operation cycles than addition and multiplication. In short, the calculation cost is high for a data processing device such as a processor. That is, in the above perspective transformation method, one division and two multiplications are required for each vertex. Even when the object is far from the perspective plane and the precision is not so much required for the perspective transformation, accurate processing is performed by the perspective transformation by one division and two multiplications as described above.

An object of the present invention is to speed up perspective transformation by approximating division by an operation having a lower operation cost.

Another object of the present invention is to speed up the entire graphics processing by speeding up the processing in the perspective transformation.

Another object of the present invention is to provide a high-speed graphics processing apparatus by reducing the operation cost of graphics processing.

The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0010]

The following is a brief description of an outline of a typical invention among the inventions disclosed in the present application.

[1] The data processing device according to the present invention
An arithmetic control unit capable of performing perspective transformation processing of three-dimensional graphics for projecting a set of a plurality of vertices (Xi, Yi, Zi) constituting a dimensional shape onto an XY perspective plane. The arithmetic control means includes a first process of setting a reference point (Xc, Yc, Zc) in the set of vertices as a perspective transformation process, a Z coordinate (Zc) of the reference point, and a Z coordinate (Zi) of the vertex. ), And a Z coordinate (Zc) of the reference point, a perspective Z coordinate (XY) on the XY perspective plane with respect to the difference (dZi) and a Z coordinate (Zc) of the reference point. SZ), a perspective X coordinate (SXi) and a perspective Y coordinate (S
Yi) and a third process for calculating Yi).

The approximate expression is, for example, a differentiable function f (d
This is obtained by applying a calculation formula obtained by terminating a series obtained by expanding a macro-Lin series regarding Z) = 1 / (Z + dZ) with a constant term and a first-order term.

As described above, the approximation calculation is used in the perspective transformation which requires the computation cost in the graphics processing, so that the computation cost is reduced. In addition, since the calculation cost in the perspective transformation is reduced, the calculation speed of the entire graphics processing can be improved.

[2] The error due to the approximate calculation of the perspective transformation is smaller as the vertex is closer to the reference point. With this in mind,
In order to prevent the image quality from deteriorating, in the perspective transformation process, the arithmetic control unit performs a fourth process of determining whether a difference between the coordinates of the vertex and the coordinates of the reference point is smaller than a predetermined value. When the determination result is small, the third processing by the approximation calculation is performed, and when the determination result by the fourth processing is large, the perspective transformation by division using the similarity of the geometry is performed instead of the third processing. Go to perspective X coordinate (SX
i) and a fifth process of calculating the perspective Y coordinate (SYi) may be performed.

[3] Higher precision is required for perspective transformation as the object approaches the perspective plane. Therefore, the accuracy of the perspective transformation has a correlation with the distance from the perspective plane of the perspective transformation to the reference point. From this viewpoint, in order to prevent the deterioration of the image quality, in the perspective transformation processing, the arithmetic control unit determines whether or not the distance from the perspective plane of the perspective transformation to the reference point is larger than a predetermined value. 6 processing, the sixth
When the determination result by the processing is large, the third processing by the approximation calculation is performed, and when the determination result by the sixth processing is small, the perspective transformation by division using a similarity of geometrics instead of the third processing To perform the perspective X coordinate (SX
i) and a seventh process of calculating the perspective Y coordinate (SYi) may be performed.

[4] Further, the error caused by the approximate calculation of the perspective transformation may be directly compared with a normal calculation result to make the determination. From this point of view, in order to prevent image quality deterioration,
In the perspective transformation processing, the arithmetic control means
For the representative vertex of the plurality of vertices constituting the dimensional object, the third processing by the approximation calculation is performed to calculate the perspective coordinates, and the perspective transformation is performed by the division using the similarity of the geometry. An eighth process for calculating the perspective coordinates by using the second process, a ninth process for determining the magnitude of the error between the two calculation results obtained by the eighth process, and the remaining The third processing by the approximation calculation is performed on the vertices, and when the determination result is larger than a predetermined value, the remaining vertices are divided by a division using a similarity of geometry instead of the third processing. It is preferable to perform a tenth process of calculating perspective X coordinates (SXi) and perspective Y coordinates (SYi) by performing perspective transformation.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS <Principle of Perspective Transformation Using Approximation Formula> FIG. 3 shows an example of a set of vertices forming a three-dimensional image. Here, an object surface in a three-dimensional space is simulated by a set of triangles. Such a triangle is called a polygon. In FIG. 3, if a polygon having three vertices P1, P2, and P3 is described as a polygon P1P2P3, FIG. 3 shows an object surface having a polygon P1P2P3, a polygon P2P3P4, and a polygon P3P4P5. Such geometric information is obtained by arranging the vertices P1, P2, P3, P4,
It can be expressed by giving P5 in this order. That is, with the exception of the first two vertices P1 and P2, the polygon P1P2P3 corresponds to the next vertex P3, and the next vertex P4
And two new vertices of the immediately preceding polygon, ie, vertices P2 and P3, form a polygon P2P.
3P4 can be expressed. Similarly, the polygon P3P4P5 is represented by the next vertex P5. Such a data expression that constitutes a new polygon each time one vertex is added is called a strip expression. The position of each vertex is determined by its three-dimensional coordinates, that is, the X coordinate value, the Y coordinate value, and the Z coordinate value. Each coordinate value is represented by a floating point number. For example, the X coordinate value, Y coordinate value, and Z coordinate value of vertex P1 are X
1, Y1, Z1 and each coordinate value is a floating point number. Each vertex has attribute data such as color attributes and normal vector information in addition to the three-dimensional coordinate values.

In FIG. 3, five vertices P1, P
2, P3, P4, and P5 are shown. The set of vertices determined in advance in this way is called a cluster. A cluster is used to represent a vertex set when a vertex set constituting a complex three-dimensional object is divided into some simpler subsets. Examples of these clusters include a collection of vertices forming an object having a rigid characteristic, that is, a characteristic that does not change its shape due to movement in space.

FIG. 1 is a view of a three-dimensional space viewed from a direction perpendicular to the Z axis. The intersection of a straight line connecting the point Pc and the origin and a two-dimensional perspective plane (XY perspective plane) 10 is represented by Wc. Obtaining the intersection Wc between the line connecting the point Pc and the coordinate origin and the perspective plane 10 in this way is called perspective transformation, and Wc is called the perspective point (perspective coordinate).

The X, Y, and Z coordinate values of the transparent viewpoint Wc are represented as SXc, SYc, and SZ, respectively. 2D perspective plane 10
Since the coordinate value of the upper point is a plane satisfying the equation “Z = SZ”, the Z coordinate value of the point on this plane is always SZ. Wc
The X and Y coordinate values SXc and SYc can be calculated by equations 3 and 4 in FIG. Here, the value Zf1 included in Expressions 3 and 4 is calculated by Expression 1.
That is, in order to calculate the coordinate values of the perspective viewpoint, one division and two multiplications are sufficient.

At this time, if the point Pi is sufficiently close to the point Pc, in other words, if the vector from the point Pi to the point Pc is small, the X coordinate value SXi and the Y coordinate value SYi of the perspective point Wi of the point Pi are calculated. It can be calculated with an approximate expression that does not use division. This approximate expression is a series obtained by “macrolin series expansion” of a differentiable function “1 / z” in the field of mathematics near z, in other words, a macrolein series of a function F (dZ) = 1 / (Z + dZ) ( This is obtained by applying an expression obtained by truncating Expression 10) in FIG. 2 with a constant term and a first-order term. In short, as shown in Equation 11 in FIG. 2, SZ / {Zc + (Zi−Zc)} is subjected to a macrolein series expansion truncated by a constant term and a first-order term. Thus, if the point Pi in FIG. 1 is sufficiently close to the point Pc, the X and Y coordinate values of the perspective point Wi of the point Pi can be approximated by SXi and SYi in Expressions 5 and 6 in FIG. (Zi-Zc) is the Z component dZi of the relative vector from the point Pc to the point Pi.

Here, the values Zf1 and Zf2 included in Equations 5 and 6 in FIG. 2 are calculated in advance by Equations 1 and 2 in FIG.
If 1-Zf2 * (Zi-Zc) is also a value once calculated, the division is not required each time in the approximate calculation of the coordinate values of the perspective point, and the number of times of multiplication is reduced each time. (Where the symbol “*” represents multiplication).
When the number of operation cycles required for division by the processor is much larger than multiplication or addition, the above-described approximation calculation is applied to the perspective transformation target point Pi sufficiently close to the point Pc to obtain the entirety. The calculation processing time is reduced. For example, if the division of floating-point data takes 15 cycles and the multiplication and addition of floating-point data take 2 cycles each, if approximation calculation is not performed, the division becomes 1
Times, multiplication twice and addition twice, a total of 23 cycles. On the other hand, if the approximate calculation is performed, for example, three multiplications and three additions are required, for a total of 12 cycles.

The premise of such an approximate calculation is that the point Pi is sufficiently close to the point Pc. More directly, the Z component dZi of the relative vector from the point Pc to the point Pi is obtained.
(= Zi−Zc) is small. Therefore, for a set of vertices of the cluster as shown in FIG. 3, a reference point such as the point Pc described with reference to FIG. 1 is set at a position relatively close to each vertex, and this is used for perspective transformation. FIG.
The approximation calculation described in is adopted. As the reference point, for example, a center of gravity of a shape defined by a set of vertices of a cluster or one vertex closest to the center of gravity can be adopted.

<< Graphic Processor >> A graphic processor to which the principle of perspective transformation using the above approximate expression is applied and a specific example of a processing procedure thereof will be described.

FIG. 4 shows a graphic processor as an example of the data processing apparatus according to the present invention. Although the graphic processor 1 is not particularly limited, the graphic processor 1 may be made of single-crystal silicon by a CMOS integrated circuit manufacturing technique or the like.
It is formed on individual semiconductor chips (semiconductor substrates). The graphic processor 1 has, for example, a central processing unit (CPU) 3 and an accelerator 4 as the arithmetic and control unit 2. Although not specifically shown, the CPU 3 has instruction control means for decoding a fetched instruction to generate a control signal, and instruction execution means for performing arithmetic processing in response to the control signal. I do. The accelerator 4 is an arithmetic circuit for reducing the arithmetic processing load on the CPU 3 or realizing arithmetic processing that is difficult for the CPU 3. In the graphics processor 1 that supports three-dimensional graphics processing, the accelerator 4
For example, a floating-point arithmetic unit capable of floating-point arithmetic,
Alternatively, it is configured as dedicated hardware for performing processing such as affine transformation and perspective transformation. The image information processed by the accelerator 4 can be transferred to a display control system having a frame buffer (not shown) via the graphics interface buffer 9.

The cache unit 5 is a high-speed memory having an associative memory structure for temporarily storing information on addresses (instructions, data) used once by the CPU 3 near addresses of information. A bus state controller 6 for controlling an external bus cycle in response to a cache miss or the like by the cache memory unit 5 is provided, and access to an external bus can be controlled via a bus interface buffer 8. Peripheral circuits 7 such as a timer and a DMAC (Direct Memory Access Controller) are connected to the bus state controller 6.

3 using the graphic processor 1
The image data to be subjected to the three-dimensional image processing is composed of a collection of vertices in space, and each vertex has three-dimensional coordinate values and vertex normal data (normal vector) for indicating brightness.
And other attribute data. The graphic processor 1 performs the perspective transformation on these coordinate values and attribute data,
Further, processing such as affine transformation is performed.

<< Perspective Transformation Processing >> Whether to perform perspective transformation by applying the above-described approximation calculation or to perform perspective transformation by division can be arbitrarily determined according to a three-dimensional object. Sets the reference point. Thus, when an error due to approximation is large, such as an object close to the screen, normal perspective transformation by division may be performed.

If the range in which the approximation calculation is performed for one object is a range in the vicinity of the reference point, the roughness of the image can be set by arbitrarily setting the range in the vicinity. Hereinafter, a specific example of the perspective transformation processing will be described while considering them.

FIG. 5 shows a procedure of a perspective transformation process executed by the arithmetic and control unit 2, particularly a perspective transformation process by approximation.

First, vertex data of one cluster as one three-dimensional object is read, and a reference point Pc (Xc, Yc, Zc) is set for the vertices of those clusters (S1). The reference point is, for example, one vertex coordinate closest to the center of gravity of the figure formed by all vertices of the cluster. For example, assuming the cluster shown in FIG. 3, the vertex P3 is used as a reference point, and the other vertices P1 and P
2, P4, and P5 are set as points near the reference point. In short, the other vertices P1, P2, P with respect to the reference point P3
4, P5 are so close that errors in the approximation calculation do not matter.

Next, using the distance SZ from the viewpoint (the coordinate origin in FIG. 1) to the screen (the see-through plane 10 in FIG. 1) and the Z coordinate Zc of the reference point Pc, equations 1 and 2 in FIG. (Step S2).

It is assumed that one cluster to be processed includes i neighboring points. First, i is initialized to "0" (S3).

Then, using the Z coordinate Zc of the reference point Pc (Xc, Yc, Zc) set in step S1 and the Z coordinate Zi of the vertex included in the object, the reference point Pc
Is obtained (S4). That is,
Calculate dZi = Zi-Zc.

Next, the obtained approximation coefficient and the X of the vertex
The X coordinate SXi on the screen is calculated from the coordinates. That is, instead of SXi = Xi * SZ / Zi, Equation 5 in FIG.
SXi ≒ ｛Zf1-Zf2 * dZ
i｝ * Xi is calculated (S5). Similarly, the Y coordinate SYi on the screen is obtained from the approximation coefficient and the Y coordinate of the vertex.
Is replaced by SYi = Yi * SZ / Zi, and SYi {Zf1-Zf2 * dZi} by an approximate expression of Expression 6 in FIG.
* Yi is calculated and obtained (S6).

As a result, the vertex Pi (Xi, Xi,
Yi, Zi), the perspective coordinates (SXi,
SYi) is obtained. The above processing is performed for all neighboring points (S7).

When the perspective transformation process is performed by the accelerator 4 shown in FIG. 4, as shown in FIG. 6, the accelerator 4 performs the three-dimensional object reference point Pc (Xc, Xc, Yc, Zc) coordinate data, distance SZ from viewpoint (coordinate origin) to screen
, And polygon data forming a point near the reference point. The accelerator 4 performs a perspective transformation based on the approximation calculation based on the given data, and outputs data after the perspective transformation.

<< Consideration of Neighboring Range with Reference Point >> The error caused by the approximate calculation of the perspective transformation is smaller as the vertex is closer to the reference point. Therefore, a near range with respect to the reference point can be set, and the approximate calculation can be applied in that range.

FIG. 7 exemplifies a perspective transformation process in consideration of the setting of a neighboring range which is a range of neighboring points with respect to a reference point. In step S10, (RX, RY, RZ) is set as a value in a range near the reference point. The coordinates of the reference point Pc are (Xc, Yc, Zc), the coordinates of the processing target vertex are Pi (Xi, Yi, Zi), and the reference point Pc
Relative coordinates of the vertex Pi with respect to (dX, dY, d
Z), dX = Xi-Xc, dY = Yi-Y
c, dZ = Zi−Zc are calculated (S11). For the relative coordinates, dXi <RX, dYi <RY, dZi <
It is determined whether or not RZ is satisfied (S12). If so, the perspective transformation using the approximate calculation is performed (S1).
3) If not satisfied, perform perspective transformation by division (S14).

By repeating steps S11 and S12 for each vertex, it becomes possible to selectively perform perspective transformation by approximation calculation only on the set neighborhood range.
Note that among the vertices in the cluster, for example, the vertex Pi (Xi, Yi, Zi) farthest from the reference point is used as a representative point, the relative coordinates are obtained in step S11, and the determination in step S12 is performed. It is also possible to determine the processing for the vertices.

FIG. 8 is a schematic block diagram of an accelerator having a neighborhood range setting function and performing perspective transformation.
The value (RX, RY,
RZ) is set in the register 21 by the CPU 3, for example. The calculation of the relative position in step S11 is performed by the relative position calculation circuit 20. The comparison in step S12 is performed by the neighborhood value comparison circuit 22. The perspective transformation using the approximation calculation in step S13 is performed by the recent perspective transformation circuit 23, and step S1 is performed.
The perspective transformation by division of 4 is performed by the perspective transformation circuit 24. The calculation instruction of the perspective transformation is given from the CPU 3 by the control signal C1. The selector 25 selectively gives the operation instruction C1 to the perspective transformation circuit 24 or the approximate perspective transformation circuit 23 according to the comparison result signal S1 from the neighborhood value comparison circuit 22, and uses the perspective transformation by division or the approximate calculation. Select perspective transformation.

The relative position calculation circuit 20, neighborhood value comparison circuit 22, approximate perspective conversion circuit 23, and perspective conversion circuit 24
Can be realized by using a floating-point arithmetic unit without using dedicated hardware.

<< Consideration of Reference Point Position on Perspective Surface >> The closer the object is to the perspective surface, the higher the accuracy of perspective transformation is required. Therefore, the accuracy of the perspective transformation has a correlation with the distance from the perspective plane of the perspective transformation to the reference point. From this viewpoint, a Z coordinate threshold for the reference point is assumed. That is, Z
The coordinate threshold can be considered as a distance from the see-through surface. If the reference point is within the range of the Z coordinate threshold, in other words, if the reference point is closer to the perspective plane than the Z coordinate threshold, perspective transformation using division is performed on the cluster having the reference point. Conversely, if the reference point is farther from the perspective than the Z-coordinate threshold, perspective transformation by approximate calculation is performed on the cluster having the reference point.

FIG. 9 shows a perspective transformation processing procedure when the Z coordinate threshold value is considered. First, Z coordinate threshold data is set for the conversion target object (S20). The Z coordinate threshold value is determined in consideration of the enlargement / reduction ratio of the object. In short, when the image is enlarged, the Z coordinate threshold value is set relatively large so that approximation calculation can be performed at a position further away from the see-through surface, so that an error due to the approximation calculation is unlikely to appear.

Next, whether the distance (Zc-SZ) from the perspective plane of the perspective transformation to the reference point is greater than the Z coordinate threshold,
In other words, it is determined whether or not the reference point is closer to the see-through surface than the Z coordinate threshold (S21). When the determination result is large (Yes), the perspective transformation by the approximate calculation is performed (S2).
2) If the result of the determination is small (No), perspective transformation is performed by division using the similarity of the geometry (S2).
3).

FIG. 10 is a schematic block diagram of an accelerator having a Z coordinate threshold value setting function and performing perspective transformation. The data of the Z coordinate threshold is set in the Z coordinate threshold changing circuit 30. Here, the Z-coordinate threshold value changing circuit 30 uses the CP
The Z coordinate threshold setting data and the data of the enlargement / reduction ratio for the original object are input from U3, and the Z coordinate threshold adjusted according to the enlargement / reduction ratio is set. This setting is performed for the register 30A in the Z coordinate threshold value changing circuit 30. Distance from perspective plane of perspective transformation to reference point (Z
c-SZ) is greater than the Z coordinate threshold.
This is performed by a coordinate comparison circuit. Based on the selection signal S2 corresponding to the comparison result, the selector 34 selects whether the polygon data to be operated is supplied to the approximate perspective transformation circuit 32 or to the perspective transformation circuit 33 using division.

It is to be noted that the present invention is not limited to the control mode for controlling the supply destination of polygon data according to the determination result as shown in FIG.
Approximate perspective transformation circuit 32 or perspective transformation circuit 3 using division
Needless to say, it is possible to select any one of the three operations. The Z coordinate comparison circuit 31, the approximate perspective transformation circuit 32, and the perspective transformation circuit 33 using division include:
The present invention can be realized by using a floating point arithmetic unit without using dedicated hardware.

<< Perspective Transformation Considering Error of Representative Point >> FIG.
1 shows an example of a perspective transformation processing procedure for directly comparing an error caused by the approximate calculation of the perspective transformation with a normal operation result and determining an operation method of the perspective transformation. First, the representative points (FX, FY, F) are selected from the vertices included in the processing target object.
Z) is determined (S30). For example, the representative point is the vertex at the position farthest from the reference point among the vertices included in the cluster,
Alternatively, it may be a vertex located approximately at the center.

The representative point is subjected to perspective transformation processing by approximate calculation (S31), and then the representative point is subjected to normal perspective transformation processing by division (S32). Then, it is determined whether or not an error with respect to both perspective transformation processing results is within a predetermined value range (S33). If the error is within the predetermined value (Yes), approximate perspective transformation is performed on all vertices (S35). ), If it deviates from the default value (No), perspective transformation by division is performed on all vertices (S3).
4).

FIG. 12 is a schematic block diagram of an accelerator having a function of confirming an error of a representative point and performing perspective transformation. The result of the perspective transformation process by the approximate perspective transformation circuit 41 for the representative point and the result of the perspective transformation process by the perspective transformation circuit 42 using division for the representative point are compared by the data transformation circuit 43 after the perspective transformation, and according to the comparison result. The polygon data is selected by the selector 40 according to the selection signal S3, and supplied to the perspective transformation circuit 41 or the perspective transformation circuit 42 using division.

It should be noted that the present invention is not limited to the control mode for controlling the supply destination of the polygon data according to the determination result as shown in FIG.
Approximate perspective transformation circuit 41 or perspective transformation circuit 4 using division
2 may be selected.
Further, the perspective-transformed data comparison circuit 43, the approximate perspective transformation circuit 41, and the perspective transformation circuit 42 using division can be realized by using a floating-point arithmetic unit without using dedicated hardware.

In the various perspective transformation processes described above, a reference point is provided for a plurality of vertices of a cluster, and data near the reference point is divided by using data as a relative position from the reference point. Is performed using the approximation coefficient without performing the perspective transformation. Therefore, the calculation cost of the perspective transformation processing in the three-dimensional graphics processing can be reduced. For example, the number of operation cycles required for addition of a floating-point operation is 3, the number of operation cycles required for multiplication is 3,
Assuming that the number of operation cycles required for division is 12, when one perspective and one multiplication are required when performing perspective transformation by division on one vertex, the number of operation cycles required for perspective transformation in this case is 15 cycles. Become. On the other hand, when the perspective transformation is performed by the approximate calculation using the approximation coefficient, the perspective transformation can be performed by two multiplications and one addition, and the number of operation cycles is nine. As described above, the more the data in the vicinity of the reference point is, the more noticeable the difference in calculation cost is. Therefore, as the processing is performed on a finer image, the effect of reducing the calculation cost is higher.

A register 3 for setting a Z coordinate threshold value
0A is provided, a comparison circuit 31 is provided for comparing the value of the register 30A with the reference point of the object and, when the reference point of the object is far, controlling to perform perspective transformation by approximation calculation. A high-quality graphics processing device can be configured while reducing costs.

The provision of the vicinity range setting register 21 which can set the range of the vertex which is the vicinity point with respect to the reference point allows the roughness of the image to be arbitrarily set.

Perspective transformation by approximation calculation and perspective transformation by division are performed on the representative point of the vertex data in the vicinity of the reference point, an error of the data is obtained, and the error in the vicinity of the reference point is determined in accordance with the error. Approximate calculation for vertices,
Alternatively, it is easy to maintain high image quality by making it possible to select perspective transformation by division.

Although the invention made by the present inventor has been specifically described based on the embodiment, it is needless to say that the present invention is not limited to the embodiment and can be variously modified without departing from the gist thereof. No.

For example, the arithmetic and control unit of the data processing device is not limited to a configuration including a CPU and an accelerator, and may be dedicated to a hard wired logic. Further, the vertex data of the polygon is not limited to the floating point data, and may be integer data as long as the calculation accuracy is not hindered.

[0058]

The effects obtained by typical ones of the inventions disclosed in the present application will be briefly described as follows.

A reference point is provided for a plurality of vertices of a cluster, and data as a relative position from the reference point is used for data in the vicinity of the reference point, and an approximate coefficient is used without performing perspective transformation by division. Since the perspective transformation is performed, the calculation cost in the three-dimensional graphics processing can be reduced. Since the perspective transformation can be speeded up by approximating the division for the perspective transformation by multiplication and addition with lower operation costs, the entire graphics processing can be speeded up. The more data in the vicinity of the reference point, the more noticeable the difference in computation cost. Therefore, a higher effect can be obtained for an image processing for a fine image.

A predetermined value such as a Z coordinate threshold value is set, and the value is compared with a reference point. If the reference point is far, control is performed to perform perspective transformation by approximation calculation, thereby reducing the calculation cost. In addition, a high-quality graphics processing device can be configured.

By setting a predetermined value such as a range of vertices that are neighboring points with respect to the reference point, the roughness of the image can be arbitrarily set.

The representative point of the data in the vicinity of the reference point is subjected to the perspective transformation by approximation calculation and the perspective transformation by division, and the approximation calculation or the perspective transformation by division is performed according to the error of the data. By selecting, it is easy to maintain high image quality.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram for explaining perspective transformation when a three-dimensional space is viewed from a direction perpendicular to a Z axis.

FIG. 2 is an explanatory diagram of an approximate expression for perspective transformation.

FIG. 3 is an explanatory diagram showing an example of a vertex set forming a three-dimensional image.

FIG. 4 is a block diagram of a graphic processor which is an example of a data processing device according to the present invention.

FIG. 5 is a flowchart illustrating a procedure of a perspective transformation process based on an approximate calculation executed by the arithmetic and control unit.

FIG. 6 is an explanatory diagram exemplifying input / output data when the perspective control processing of FIG. 5 is performed by the arithmetic and control unit.

FIG. 7 is a flowchart illustrating an example of a perspective transformation processing procedure in which a neighborhood range with respect to a reference point is considered;

FIG. 8 is a schematic block diagram of an accelerator having a near range setting function and performing perspective transformation.

FIG. 9 is a flowchart illustrating a procedure of a perspective transformation process when a Z coordinate threshold value is considered.

FIG. 10 is a schematic block diagram of an accelerator having a Z coordinate threshold value setting function and performing perspective transformation.

FIG. 11 is a flowchart illustrating an example of a perspective transformation processing procedure in which an error due to an approximate calculation of the perspective transformation is directly compared with a normal operation result to determine an operation method of the perspective transformation.

FIG. 12 is a schematic block diagram of an accelerator having a function of confirming an error of a representative point and performing perspective transformation.

[Description of Signs] 1 Graphic Processor 2 Arithmetic Controller 3 CPU 4 Accelerator 5 Cache Memory Unit Pc Reference Point Pi Vertex of Polygon

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Koji Yamada, Inventor 5--20-1, Kamizuhoncho, Kodaira-shi, Tokyo Within the Semiconductor Group, Hitachi, Ltd. 5-20-1 F-term within Hitachi Semiconductor Co., Ltd. Semiconductor Group 5B046 DA10 FA17 JA07 5B056 BB11 BB51 HH03 5B080 BA04 DA06

Claims (5)

[Claims] A plurality of vertices (X) forming a three-dimensional shape
i, Yi, Zi) having an operation control unit capable of performing a perspective transformation process of three-dimensional graphics for projecting the set on an XY perspective plane, wherein the operation control unit performs a perspective transformation process based on the set of vertices. A first process for setting a point (Xc, Yc, Zc); a second process for calculating a difference (dZi) between a Z coordinate (Zc) of the reference point and a Z coordinate (Zi) of the vertex; The approximate coordinates using the perspective Z coordinate (SZ) on the XY perspective plane with respect to the Z coordinate (Zc) of the point, the difference (dZi), and the Z coordinate (Zc) of the reference point are converted to the X coordinate and the Y coordinate of the vertex. Corresponding perspective X coordinate (SXi) and perspective Y
A data processing device for performing a third processing for calculating coordinates (SYi). 2. The approximate expression is a differentiable function f (dZ) =
2. The data processing apparatus according to claim 1, wherein a calculation formula obtained by truncating a series of 1 / (Z + dZ) by a Macroleigh series expansion with a constant term and a first-order term is applied. 3. In the perspective transformation process, the arithmetic control unit performs a fourth process of determining whether a difference between the coordinates of the vertex and the coordinates of the reference point is smaller than a predetermined value, and the determination result is small. When the third processing by the approximation calculation is performed, and when the determination result by the fourth processing is large, a perspective transformation is performed by division using a similarity relationship of the geometry instead of the third processing to perform a perspective X coordinate ( 3. The data processing device according to claim 1, wherein a fifth process for calculating SXi) and a perspective Y coordinate (SYi) is performed. 4. In the perspective transformation process, the arithmetic control unit performs a sixth process of determining whether a distance from a perspective surface of the perspective transformation to a reference point is greater than a predetermined value. When the determination result is large, the third processing by the approximation calculation is performed, and when the determination result by the sixth processing is small, the perspective transformation by division using the similarity of the geometry is performed instead of the third processing. The data processing apparatus according to claim 1, wherein a seventh process of calculating a perspective X coordinate (SXi) and a perspective Y coordinate (SYi) is performed. 5. In the perspective transformation processing, the arithmetic control unit performs the third calculation based on the approximate calculation with respect to a representative vertex of a plurality of vertices constituting the three-dimensional object.
An eighth process of calculating perspective coordinates by performing processing and calculating perspective coordinates by performing a perspective transformation by division using a similarity relationship between geometrics, and an error of both calculation results obtained by the eighth process A ninth process for determining the size of the vertices, and when the error due to the ninth process is equal to or smaller than a predetermined value, the third process based on the approximate calculation is performed on the remaining vertices. In the tenth process, a perspective X coordinate (SXi) and a perspective Y coordinate (SYi) are calculated for the remaining vertices by performing perspective transformation by division using a similarity in geometry instead of the third process. 3. The data processing device according to claim 1, wherein the data processing is performed.