JP2003099419A - High-speed interpolation computing element - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform matrix computing and interpolation computing by using a single computing unit. SOLUTION: A data string consists of the combination of a header for designating a matrix to be used for matrix computing and data to be matrix- computed. The data is inputted and a header part and a data part are identified from the data string. The matrix is selected by the header. The selected matrix and the identified data are computed by a computing part, and interpolation computing is performed with the result of the matrix computing and an interpolation ratio. Since the matrix to be used for computing and data to be computed are designated by the header, a series of matrix computing can continuously be performed. Further, by rewriting the header, the matrix to be used for computing can be changed and interpolation computing can simultaneously be selected without changing an instruction.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a three-dimensional space simulation technique using a computer, and more particularly to interpolation calculation processing using a matrix calculator.

[0002]

2. Description of the Related Art In game systems and the like using a computer, a "3D game" in which characters and graphics appearing in the game are expressed in three dimensions has become mainstream. In the 3D game, in order to display an object composed of polygons, the game system performs a huge amount of calculation such as the position of the object. Among them, matrix operations are often used for rotating / moving objects, changing viewpoints, and the like. When objects have a connection relationship like a model of a node connecting humans with joints such as limbs, the relationship can also be defined by matrix calculation. For example, there are vertex groups A, B, and C that form nodes that are connected to each other by joints. The vertex group B is connected to the vertex group A, and the vertex group C is connected to the vertex group B. The local matrices that define the relationships between the nodes are Ma, Mb, and Mc. The world matrix for converting the coordinates of the vertices to the world coordinate system must be defined. As the matrix used for this calculation, it is necessary to convert the local matrices Ma, Mb, Mc into world matrices Ma ′, Mb ′, Mc ′ respectively by the following formula.

[Equation 1] Ma '= Ma Mb '= Mb.Ma' = Mb.Ma Mc '= Mc.Mb' = Mc.Mb.Ma

The position vector V (x, y, z) defining the coordinates of each vertex of the vertex group in three dimensions is converted into a matrix corresponding to any one of the above-mentioned matrices Ma ', Mb', Mc '. When the coordinate conversion is performed by substituting it in M, the coordinates of the new vertex V ′ (x ′,
y ', z') can be obtained. As the formula used for the matrix calculation for this coordinate conversion, the calculation between the vertex group and the matrix as shown in the following conversion formula is often used. In this calculation, the vector V of the vertex coordinates to which the redundant data (0 or 1) is added and the matrix M are used for convenience. By using the data obtained by ignoring the redundant data from V'obtained by this calculation result, the conversion of rotation / enlargement / reduction and parallel movement can be performed at one time.

[Equation 2] New vertices can be created using the coordinate vectors of some of these vertices. This newly created vertex is used and often used to interpolate vertices. Also, various effects can be obtained by using this interpolation, and it is also used in games. An example is given below.

FIG. 1 is a diagram showing an interpolation result vertex group 30 which is the result of an interpolation calculation using the interpolated vertex groups A10 and B20. As can be seen from this figure, assuming that the lower part of the paper is the lower part of each vertex group, the interpolation result vertex group 30 is the interpolated vertex group A1 at the bottom.
0, at the top is the group of vertices to be interpolated B20. In the interpolation result vertex group 30, the ratio of the proportions of the interpolated vertex group A10 to the interpolated vertex group B20 increases from the bottom to the top. When the coordinates of the vertices on the interpolated vertex group A10 are V _A and the coordinates of the vertices on the interpolated vertex group B20 are V _B , the interpolated vertex V can be expressed as follows.

## EQU3 ## V = αV _A + βV _B (α + β = 1,0 ≦ α ≦ 1,0 ≦ β ≦ 1) where α and β are interpolation rates when performing interpolation. When constructing the interpolation result vertex group V using this formula, by changing α from 1 to 0 and β from 0 to 1 from the lower part to the upper part of the interpolation result vertex group V, the interpolation result as shown in FIG. Vertex group 30
Can be configured. When the interpolated vertices or vertices are converted from, for example, the local coordinates described above to the world coordinates, the interpolation calculation and the coordinate conversion can be performed at the same time. 3 of the interpolation vertex before conversion _{V 1,} V 2, _{V 3}
Assuming that the points are points and the matrix used for coordinate conversion for each interpolated vertex is M ₁ , M ₂ , and M ₃ , the interpolated vertex is calculated as follows.

V ′ = αV ₁ · M ₁ + βV ₂ · M ₂ + γV ₃ · M ₃ Here, α, β, and γ are interpolation rates when performing interpolation. Conventionally, there has not existed one that performs such coordinate conversion and performs matrix calculation by using one calculation unit for interpolation calculation.

[0005]

SUMMARY OF THE INVENTION An object of the present invention is to perform a matrix operation and an interpolation operation by using one arithmetic unit in a simulation of a three-dimensional space used in a video game.

[0006]

In order to achieve the above object, the present invention inputs a data string formed by a combination of a header designating a matrix and data to be matrix-operated, and performs an interpolation operation. A high-speed interpolation calculator that performs matrix calculation including: a matrix storage section that stores a plurality of matrices used for matrix calculation; an interpolation rate storage section that stores an interpolation rate used for interpolation calculation;
Data analysis in which the header and data are identified from the input data string, the matrix stored in the matrix storage unit is selected by the header, and in the case of interpolation calculation, the interpolation rate and data are selected and output. / Matrix selection unit, a matrix operation unit for performing matrix operation, and a matrix operation result storage unit for temporarily storing the operation result, and when performing only matrix operation, the matrix operation unit selects When performing a matrix operation of the selected matrix and data and performing an interpolation operation, the matrix operator stores the result of the matrix operation of the selected matrix and data in the matrix operation result storage unit. , The matrix calculation result storage unit temporarily stores the calculation result and the interpolation rate storage unit. Enter the between rates, performs interpolation calculation. Thereby, the matrix calculation and the interpolation calculation can be performed at high speed by using one arithmetic unit.

The execution of the interpolation calculation is designated by the data in the header, and is detected by the data analysis / matrix selection unit to perform the interpolation calculation. Further, a buffer may be provided between the matrix storage unit and the matrix operation unit, and the buffer may be used as a matrix cache and / or the matrix operation result storage unit. The interpolation calculation can be performed using two data or three data, and a different matrix can be designated for each data used for interpolation. Further, a plurality of the stored interpolation rates can be stored, and the number of times of use of each interpolation rate can be designated.

[0008]

Embodiments of the present invention will be described in detail with reference to the drawings. The present invention relates to a three-dimensional space simulation technique for games and the like, and has a configuration in which a matrix used for calculation can be selected in header information that specifies vertex data. As a result, it is possible to continuously perform a series of matrix operations, which are specified by a data string composed of a header specifying a matrix and data to be matrix-operated,
For example, it is suitable for processing a group of vertices of a joint structure that continuously uses a plurality of matrices. In addition, it is possible to specify whether or not to perform interpolation calculation by the header information. The specified vertex data is used as the vertex to be interpolated, and a matrix operation for performing an interpolation operation can be performed. For these, see Figure 2.
And it demonstrates with reference to FIG.

The data structure of the present invention will be described with reference to FIG.
FIG. 2A is a diagram explaining the data structure used in the present invention. In FIG. 2A, the vertex group data of the joint structure is shown as an example in which the coordinate data of one vertex is one word, and the numeral attached to the side indicates the number of words. Figure 2
As shown in (a), a plurality of vertex data items are combined into a continuous vertex group data by adding a 1-word header to each. In this header, the minimum necessary information for performing calculation is embedded. Such a combination of a plurality of headers and vertex group data will be referred to as "vector data" for the sake of explanation. FIG. 2B shows the header in the vector data. As the first information embedded in the header area, as shown in FIG.
As in (i), the number of data words in which the information in this header affects the operation is embedded. Further, the information to be stored in the header is not limited to the number of words, but may be any information that can identify the header and the data from the vector data. As another example, as shown in FIGS. 2 (b) and (ii), a flag indicating a header may be used. It may also be an offset to the next header. As the second information embedded in the header area, information designating a matrix used for calculation of the data after the header is embedded. In this example, a matrix number indicating a storage location in the matrix storage area, which will be described later with reference to FIG. 3, is designated. Instead of the matrix number used in the calculation, a reference number may be determined in advance and an offset value from the reference number may be embedded. It is also possible to select the matrix used for the calculation based on the number obtained by adding the offset value and the reference number included in this header. For example, if the reference number M _base is 0x11 and the offset value Mn included in the header is 0x0E, the number of the matrix used for the calculation is 0x1F. As third information embedded in the header area, an ON / OFF flag for interpolation calculation is embedded. When this flag is ON, the matrix operation is performed at the same time as the interpolation operation is performed by using the following data which is shown as the first information embedded in the header area and which is influenced by the header. On the contrary, when it is OFF, only the usual matrix calculation is performed. In the case of interpolation calculation, the matrix specified by the header is the first matrix. For example, in the case of interpolating at 3 points, if different matrices are used at each point, the matrix specified by the header will be 3 matrices. Is used. In addition, since the interpolation calculation calculates the data for one point based on the data for two points or three points, the number of data after the interpolation is targeted as the number of data affected by the header embedded in the header. You can also set In this case, when the data for one point is calculated based on the data for three points, the number of data is set to 3 in the header, but there are nine vertex group data as input vertex group data. is doing.

FIG. 3 is a diagram showing the configuration and operation of a computing unit 300 that performs a matrix computation using the data structure described with reference to FIG. The arithmetic unit 300 will be described as an example of a configuration that is independent of the CPU and operates as an arithmetic unit for executing a matrix arithmetic instruction. When the interpolation calculation is performed by the ON / OFF flag of the interpolation calculation of the header,
A case where a normal matrix calculation is performed without performing an interpolation calculation can be selected. As shown in FIG. 3A, the configuration of the arithmetic unit 300 for performing this matrix operation is such that the matrix storage area 31 for storing the matrix used for the operation
0, a vector / data analysis / matrix selector 350 for analyzing a header, a matrix calculator 370, an interpolation value storage area 380, and a matrix calculation result storage area 390. The matrix calculator 370 can perform the matrix calculation shown in [Equation 2].

First, a case will be described in which the interpolation calculation is performed by using different matrices at the three points shown in [Equation 4]. First, the operation of the arithmetic unit 300 is set. This setting can change the operation of the arithmetic unit 300, and by preparing data corresponding to the operation, it is possible to select the operation of the interpolation calculation described later. And FIG.
As shown in (b), in the matrix storage area 310,
A matrix M1 used in the interpolation calculation of the vertex group A in FIG.
M2 and M3 are stored in the locations of matrix numbers 1, 2, and 3, respectively. The interpolation value storage area 380 stores interpolation rates α, β, and γ used for interpolation calculation. After preparing for the operation in this way, the matrix operation instruction is issued.

By the matrix operation instruction, vector data is input to the operator 300, as shown in FIG. The arithmetic unit 300 inputs the input vector data to the vector / data analysis / matrix selector 350, and the vector / data analysis / matrix selector 350.
Separates each word of the input vector data from the header or the vertex data. The first input word is identified as a header, and the vector / data analysis / matrix selector 350 refers to the matrix number of the header as the matrix to be used for calculation, and the matrix storage area 3
Select from 10 and input the selected matrix to one of the matrix calculators 370. Further, since the interpolation calculation is performed in this example, the ON / OFF flag of the interpolation calculation of the header is ON, and the calculator 300 recognizes that the interpolation calculation is performed from the header.

Since the word to be input next is the first vertex data, the vertex data is sent to the matrix calculator 370, and the matrix calculation is performed on the selected matrix (M1) and the matrix calculator 370. The calculation result is output to and stored in the matrix calculation result storage area 390. Similarly, the second and third vertex data are also subjected to matrix calculation with the selected matrix (M2, M3), and the calculation result is stored in the matrix calculation result storage area 390.
To store. As shown in FIG. 3D, the interpolated value storage area 38 is formed by using the operation result of the vector data of three points stored in the matrix operation result storage area 390 as a matrix.
A matrix calculation is performed by the calculator 370 as a vector (α, β, γ) having the interpolation values α, β, γ stored in 0 as components. Then, the calculation result is output. Thereafter, similar processing is performed on the input vertex data until the next header is identified. When the next header is input to the vector / data analysis / matrix selector 350, the operation of the calculator 300 is determined by determining whether the interpolation calculation is to be performed by the header. In this way, by specifying the matrix in the header and specifying whether to perform interpolation calculation,
The matrix operation on the vector data shown in (a) can be completed.

In the above description, the case where the interpolating operation is performed using the three different matrixes has been described. However, the interpolating operation at the two points can be performed by setting the operation of the arithmetic unit 300 before issuing the operation instruction. You can also do it. At that time, the calculator 300 performs an interpolation calculation using two interpolated vertices as the input vertex data. Therefore, as the corresponding data, it is necessary to prepare in advance a set of two points for interpolation, which is input to derive one interpolation vertex. Similarly, by setting the operation of the arithmetic unit 300, it is possible to use the same matrix for all three points and two points of the interpolated vertices. In this case, depending on the setting, the calculator 300 selects only M1 from the matrix storage area 310 described above without selecting M2 and M3. As another embodiment of calculation such as coordinate conversion by interpolation at two points, the calculator 300 is fixed to always perform interpolation calculation at three points, and setting such as interpolation calculation at two points is not performed. You can also At this time, as the vector data to be input, a set of 3 points of the interpolated vertex 2 points and the dummy 1 point is prepared. If the interpolation value γ = 0 is set in the interpolation register and the interpolation value γ is multiplied by the dummy data, the interpolation calculation by substantially two points is performed.

Next, a case where no interpolation calculation is performed will be described. When the interpolation calculation is not performed, the “high-speed matrix calculator (Japanese Patent Application No. 2001-26
6423) ”, which can be used to continuously perform matrix calculation processing used for coordinate conversion, interpolation calculation, and the like. First, as shown in FIG.
The matrixes M1, M2 and M3 used in the calculation of the vertex groups A, B and C of FIG. 2 are stored in 0. Then, the arithmetic instruction is issued.

By issuing the operation instruction, vector data is input to the operation unit 300 as shown in FIG. The arithmetic unit 300 inputs the input vector data to the vector / data analysis / matrix selector 350, and the vector / data analysis / matrix selector 350.
Separates each word of the input vector data from the header or the vertex data. When the input word is identified as a header, the vector / data analysis / matrix selector 350 selects the matrix to be used for the operation from the memory 310 according to the matrix number of the header, and the selected matrix is the matrix operator. Input to one side of 370. Further, since the interpolation calculation is not performed in this example, the ON / OFF flag of the interpolation calculation of the header is OFF, and the calculator 300 recognizes that the interpolation calculation is not performed.

Since the next word to be input is the vertex data, this vertex data is sent to the matrix calculator 370, and the matrix selected by the header is subjected to matrix calculation in the matrix calculator 370. Figure 3
In (c), the result of this matrix calculation is output to the matrix calculation result storage area 390, but since the interpolation calculation is not performed here, the calculation result is output to the outside as it is. Thereafter, similar processing is performed on the input vertex data until the next header is identified. When the next header is input to the vector / data analysis / matrix selector 350, the vector / data analysis / matrix selector 350 newly selects the matrix specified by the header and inputs it to the matrix calculator 370. To do. The subsequent vertex data is input to the matrix calculator 370 by the vector / data analysis / matrix selector 350, and matrix calculation is performed with the selected matrix. In this way, the matrix operation on the vector data shown in FIG. 2A can be completed.

As shown in FIG. 2, in the present invention, a header specifies a matrix used for calculation, ON / OFF of interpolation calculation, and data to be calculated by the matrix. Thus, by using the arithmetic unit shown in FIG. 3, it is possible to perform interpolation calculation and coordinate conversion for a plurality of vertex groups, which conventionally must be divided for each vertex group forming a node, by performing a single calculation process (calculation instruction). Issue). With this configuration, data such as joint structure can be calculated at high speed, and the data can be created easily and intuitively. Also, since vertex data can be centrally managed,
It is easy to understand the relationship between the vertices that compose a node. Therefore, it is easy to create the vertex data, and the vector data also has the matrix designation information, so that the vertex data can be easily managed. Further, by rewriting the header, it is possible to change the matrix used for the calculation and switch ON / OFF of the interpolation calculation without changing the command. Therefore, it is possible to represent a joint structure that is interpolated on the way.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Specific data and circuit configuration examples will be described below with reference to FIGS. FIG. 4 is an example of data for explaining the operation in this circuit configuration, and is a diagram showing vector data of an interpolation calculation using one type of matrix. FIG. 4A shows the first vertex group 1-1 and 1-2.
-1, 3-1 (shown in black in the figure), second vertex group 1
-2, 2-2, 3-2 (shown in white in the figure), 3rd
Vertex m ₁ (vertex 1-performed by 3-point interpolation calculation by the vertex group 1-3, 2-3, 3-3 (shown in gray in the figure)
1, 1-2, 1-3 interpolation), vertex m ₂ (vertex 2
The data of the vertices has a structure such that -1, 2, 2 and 2-3 interpolate) and a vertex m ₃ (interpolates vertices 3-1, 3-2 and 3-3). The vertex group used for all of these interpolations needs to be subjected to coordinate conversion such as rotation with respect to the origin of the world coordinate system by the matrix M1. This matrix M1 is the world that defines the relationship between each vertex group and the origin of the world coordinate system.
It is a matrix. FIG. 4B shows the vertex group of FIG. 4A created as vector data. The first word is the header, and the other words are the coordinate data of each vertex used for the interpolation calculation. Since it is a three-point interpolation calculation, the data of three points used for the interpolation becomes the data of one vertex after the interpolation. The header has a matrix number (1) used for the calculation, an ON / OFF flag for the interpolation calculation (1 = O
N) and the data number (3) of the vertices after the interpolation which the header influences. Since the number of data affected by this header is three-point interpolation, it is 3 × 3 = 9 words. The operation will be described later using the circuits shown in FIGS. 6 and 7 using the above world matrix and vector data as an example.

FIG. 5 is another diagram of data for explaining the operation in this circuit configuration, and is a diagram showing vector data of an interpolation operation using an individual matrix. FIG. 5A shows the first vertex group 1-1, 2-1, 3-1.
(Indicated in black in the figure), second vertex group 1-2, 2-
2, 3-2 (shown in white in the figure), third vertex group 1-
Vertexes m ₁ (vertices 1-1, 1-2, 3-2) which are interpolated by three points 3, 2, 3, 3-3 (shown in gray in the figure)
1 to 3), vertex m ₂ (vertices 2-1 and 2-
2, 2-3 interpolated), vertex m ₃ (vertices 3-1 and ₃
The data of the vertices having a structure such that (interpolated by -2, 3-3) is shown. It is necessary to perform coordinate conversion such as rotation with respect to the origin of the world coordinate system by the matrix M1, the second vertex group by the matrix M2, and the third vertex group by the matrix M3. That is, these matrices M1, M2
M3 is a world matrix that defines the relationship between each vertex group and the origin of the world coordinate system.

FIG. 5B shows the vertex group of FIG. 5A created as vector data. The first word is the header, and the other words are the coordinate data of each vertex used for the interpolation calculation. Similar to the vector data of FIG. 4B, since it is a three-point interpolation calculation, the 3
The point data becomes the data of one vertex after interpolation. In the header, the top matrix number of the three used for the calculation, the ON / OFF flag of the interpolation calculation, and the number of data points of the vertices after the interpolation which the header affects are described.

FIG. 6 is a circuit diagram of the high speed interpolation calculator 600. This is a "high-speed matrix calculator (Japanese Patent Application No. 2000-26642) previously applied by the inventor of the present invention.
3) ”, but the difference is that there is a feedback path through which the operation result from the multiplier / adder is written back to the buffer and the configuration of the interpolation register and interpolation value selector is added. is there. High-speed interpolation calculator 60
0 is a header / data classifier 610 for classifying into a header and data, a header analysis / arithmetic unit controller 620, an interpolation value register 700 for storing an interpolation value, data from the header / data classifier 610 and an interpolation value register 700. An interpolated value selector 690 that selects data and sends it to a computing unit,
Memory 630 for storing matrix data, buffer 1 642, buffer 2 644, buffer 3 6 for use as a temporary cache of matrix data
46, buffer 4 648, MtoB for selecting which of the memory and the multiplier / adder takes data into the buffer
The multiplexers 641, 643, 645, 647 have multiplexers 650 for selecting four buffers. An arithmetic unit that actually performs a matrix operation is composed of three power adders 660, 670, and 680. Each of the multipliers / adders 660, 670, 680 is composed of three multipliers and one adder. For example, the multiplier / adder 68
0 is composed of multipliers 682, 684, 686 and an adder 688, and one multiplication / adder can realize an operation for one column. Since there are three multiply-adders, the operations for the three columns necessary for the coordinate conversion shown in [Equation 2] can be performed once.

The header / data classifier 610 receives the vector data input from the path va to the high-speed interpolation calculator 600 and separates it into a header and data. Is output to the interpolation value selector 690. The interpolation value selector 690 is
It is a selector that selects the data from the header / data classifier 610 and the data from the interpolation value register 700 and sends it to each of the power adders 660 to 680 via the route vb. Memory 6
The matrix 30 can store, for example, 64 matrices from the route ma. The matrix to be stored is designated by the matrix numbers 1 to 64 corresponding to the storage locations of the memory 630. Buffer 1 to buffer 4
Is used as a cache that can store four pieces of matrix data used in calculations. Further, at the time of interpolation calculation, the buffer 4 is used as a temporary, so that the cache operation is limited. At the time of this interpolation calculation, the calculation results of the multipliers / adders 660 to 680 are fed back to the buffer 4 used as a temporary. This path is the buffer entrance at the MtoB multiplexer 641, 64
3, 645 and 647 select a memory path. The management of this buffer is performed by the header analysis / operation unit controller 620 using a management table which stores the correspondence between the buffer and the matrix number and the temporary use during the interpolation operation of the buffer. The high-speed interpolation calculator 600 operates as a calculator independent of the CPU in order to execute the matrix calculation instruction.

Next, the circuit of the interpolation value register 700 in the high speed interpolation calculator 600 will be described with reference to FIG.
FIG. 7 is a block diagram showing the interpolation value register 700 of the high speed interpolation calculator. The interpolated value register 700 includes an interpolated value register controller 710, an interpolated value storage interface 720 that is an interface for writing an interpolated value from the outside, and registers 731 to 739 for storing interpolated values and the like through the interpolated value storage interface 720.
The stored interpolation value is selected and the interpolation value selector 69 is selected.
It has an nto1 multiplexer 740 which outputs to 0. The interpolation value register 700 is a register that can be written from the outside through the path pa, and is a group of registers that holds the interpolation rates α, β, γ. In this example, the interpolation value α,
It is possible to store 16 sets of interpolation values in the registers 731 to 739, with 4 sets of β, γ, and a value num representing how many times the interpolation values α, β, γ are used for calculation. ing. The interpolation value register controller 710 is
By operating the nto1 multiplexer 740, one set of interpolation values α, β, γ from 16 sets of registers is input to the interpolation value selector 690, and num is an interpolation value register controller 71.
Send to 0.

When requesting the interpolation values α, β, γ in the interpolation calculation, the header analysis / calculator controller 620 is used.
Is the interpolation value register
Send to controller 710. The interpolation value register controller 710 counts the number of times of this request, and if the value matches the stored value of num _k (1 ≦ k ≦ 16), k is incremented by 1 to output the next interpolation value. To do. As a result, the interpolated values are α _{k + 1} , β _{k + 1} , γ
_{k + 1} is output, and this value is used for the next calculation. For example, when num ₁ is 2, when the interpolated value request comes to the interpolated value register controller 710, the interpolated values α ₁ , β ₁ , and γ ₁ are sent to the interpolated value selector 690. Even if the next interpolation value request comes, the interpolation values α ₁ , β ₁ ,
Send γ ₁ . num _1, the value of the two-time interpolation value α ₁ ,
After sending β ₁ , γ ₁ , the interpolated value register controller 710 prepares to output the next set of interpolated values and sends α ₂ , β ₂ , γ ₂ on the next request. The number of times the second set of interpolation values is sent is only the number of times the value of num ₂ . Continuing in this way, the system of the present embodiment has a maximum of α ₁₆ , β.
_{16 up} to γ ₁₆ . Note that the interpolation calculation proceeds and k = 1
When it becomes 6, the interpolation value register controller 7
Depending on the flag in 10, then either k = 1 or k =
You can choose 16 or not. This flag is selected and set in advance before issuing the operation instruction.

Next, the operation of the high-speed interpolation calculator 600 shown in FIG. 6 when only one type of matrix used for calculations such as coordinate conversion shown in FIG. 4 is used will be described in detail with reference to FIGS. 8 and 9. . FIG. 8 and FIG. 9 are diagrams showing the operations of the multiplier / adder and the temporary buffer for storing the operation result when performing the interpolation operation. This temporary buffer uses buffer 4648 in this embodiment. This buffer can store a matrix of 4 rows and 3 columns. The buffer 4648 stores a storage area 649a in the first row of the matrix, a storage area 649b in the second row, a storage area 649c in the third row, and a storage area 649c in the third row. It has a storage area 649d for the row. Three buffers other than the buffer 4, which is a temporary buffer, are used as a matrix cache for temporarily storing the matrix used for the interpolation calculation. Matrix cache is the matrix number stored in the table when the matrix data was previously transferred from the memory to the buffer, and as long as the matrix data is stored in the buffer, it will be used from the next time. When the same number matrix is requested,
This is a method of directly specifying the buffer in which the matrix is stored without performing the matrix data transfer from the memory to the buffer. By doing this, when the same matrix is used, the memory transfer time is saved, so that the operation can be performed at high speed.

In FIG. 6, the matrix M1 used in the interpolating operation and the interpolated values α, β, γ and num are stored in the memory 630 of the high speed interpolating operation unit 600 as the matrix number 1
And the interpolation value register 700. It also sets the operation of the high-speed interpolation calculator 600 and sets the flag in the interpolation value register controller 710 described above. In order to execute the example shown in FIG. 4, the high-speed interpolation calculator 600 is set so as to perform a three-point interpolation calculation using a matrix in which all the interpolated vertices are the same. After that, when the high-speed interpolation calculator 600 is activated by the matrix calculation command,
The high-speed interpolation calculator 600 operates to perform an interpolation calculation using the same matrix of three points. After that, the route va
From the vector data is input to the high speed interpolation calculator 600. In this example, the 10-word vector data shown in FIG. 4B is input. Since the first word is the header, it is sent to the header analysis / operation unit controller 620 through the header classifier 610. The matrix M1 used in the operation at the location of the matrix number 1 in the memory 630, which is shown in the vector data header, is stored in the memory 63.
Transfer from 0 to buffer 1642. At the same time, the header analyzer records the number written in the header. Since the vector data is the case of FIG. 4B, "3" is written in the header of the first word, and this 3 is recorded.

Referring to FIG. 8 and FIG. 8A, the vertex data V for the first interpolation operation which is vector data.
1 is input to the multipliers / adders 660 to 680 through the route vb. Similarly, the matrix M1 is also input from the buffer 1642 to the power adders 660 to 680 via the path mc.
Then, the following matrix operations are performed by the multipliers / adders 660 to 680.

## EQU00005 ## V1 '= V1.M1 and x1', y which are the components of the calculated result V1 '
1 ′ and z1 ′ are MtoB multiplexers 641 and 64
3, 645 and 647 are operated to store them in the storage area 649a in the first row of the matrix of the buffer 4648 used as a temporary for the interpolation calculation. The contents of the temporary buffer 648 after the calculation are as shown in FIG. Also, the storage area 649d in the fourth row of the matrix
Since it is not used in the calculation, (0,0,0) is entered.

Next, similarly, FIG. 8 (b) and FIG. 9 (c)
As shown in, the following matrix operation is performed on the vertex data V2 and V3 for the second and third interpolation operations, and the result is stored in the second row 649 of the buffer 4648.
b, stored in the third row 649c.

## EQU00006 ## V2 '= V2.M1 V3' = V3.M1 When the calculation of the vertex data for the third interpolation is completed and the storage area of the temporary buffer 648 is completely filled, the multiplexer 650 is operated to operate the route mc. Temporary
The contents of buffer 4648, which is a buffer, are output. At the same time, the interpolation value selector 690 is operated to output the interpolation values α, β, γ from the interpolation value register 700 to the route vb. Then, the multiplication / adders 660 to 680 perform matrix calculation, which is the following interpolation calculation.

## EQU00007 ## V '=. Alpha.V1' +. Beta.V2 '+. Gamma.V3' The interpolation result vertex V ', which is the result of the interpolation calculation including this coordinate transformation, is output to the route vc. After that, the operations of [Equation 5] to [Equation 7] are repeated in the same manner, and after the operation for the stored number is repeated, the next header is analyzed. In the case of the vector data of FIG. 5B used in this embodiment, after the operations of [Equation 5] to [Equation 7] are repeated three times,
The next 11th word header is analyzed.

Next, the operation of the high-speed interpolation calculator 600 shown in FIG. 6 when the matrix used for calculation such as coordinate conversion is switched for each interpolation data using vector data such as the vector data shown in FIG. To do. In the interpolation calculation using the three vertexes for interpolation, the matrix used for the calculation specified by the header is used as the matrix for the first vertex group, as described in the description of the data in FIG. That is, when the matrix number is n, n is the first
Used as a matrix for a group of vertices, and n + 1, n
+2 is used in the calculation as a matrix for the second and third vertex groups. Therefore, in the example of FIG. 5, "matrix used for calculation" is designated as 1 in the header, so M1 stored in the matrix number 1 for the first vertex group and matrix number 2 for the second vertex group. 2 and M3 stored in the matrix number 3 for the third vertex group are used as a matrix for coordinate conversion and the like. The basic operation is the same as the case where only one type of matrix used for the calculation such as the coordinate conversion is used as described above, but since a separate matrix is used for each vertex group, three matrices can be obtained by one interpolation calculation. Must be used. In this interpolation calculation, since the matrix to be used is stored in the buffer in advance, three of the four buffers are used. Further, since one of the buffers is used as a temporary buffer for interpolation calculation, all buffers are used for calculation.

The coordinate vectors of one vertex of the first to third vertex groups, which are the above-mentioned interpolated vertex groups, are V1 to V3, and the coordinate vector of each vertex after the coordinate conversion of these interpolated vertex groups is V.
1'-V3 'and the interpolation values α, β, γ are used to perform the following matrix calculation, and the interpolation result vertex V'which has been subjected to the interpolation calculation including the coordinate conversion can be obtained.

[Equation 8] V1 '= V1 · M1 V2 '= V2 · M2 V3 '= V3 · M3 V '= αV1' + βV2 '+ γV3'

FIG. 10 shows a vertex group of a joint structure and its vector data when an interpolation operation is performed only on two joints formed by the vertex group and a designated joint portion. Such a mixture with ordinary joint calculation is designated by the operation of the ON / OFF flag of the interpolation calculation embedded in the header in the vector data to specify an object having a joint structure composed of several sections. Interpolation calculation can be performed only on the indirect portion. FIG. 10A shows a vertex group having a joint structure in which vertices a, b, and c are nodes 1, vertexes d, e, and f are interpolation vertices, and vertices g, h, and i are nodes 2, and vertex d. , E, and f, the vertex m interpolated is shown. Then, the joint structure of the nodes 1 and 2 with the vertex m as a joint is provided, and the relationship between the origin of the world coordinate system and the node 1 is M1, the relationship between the node 1 and the vertex m is M2, and the node m is the node. The relationship of 2 is defined by the local matrix of M3. Figure 10
10B shows the vertex group of FIG. 10A created as vector data. The first word is header 1 of section 1
And the fifth word is the header 2 and 9 of the vertex group for interpolation calculation.
The word th is the header 3 of the section 2, and the other words are the coordinate data of the vertices and the interpolation vertices that form the section.
In each header, the node number and the matrix number used by the interpolation vertex group for calculation and the number of data affected by the header are described, and it is shown that only the header 2 is subjected to interpolation calculation. At this time, the local matrices M1, M2, and
M3 is converted into world matrices M1 ', M2', M3 ', and these are stored in respective locations of matrix numbers l, m, n on the memory 630 which is a matrix storage area.

In the above description, whether or not the interpolation calculation is performed is specified by the header, but the calculation operation of the matrix calculation unit may be externally specified by an operation command or the like to execute the interpolation calculation. In this case, as shown in FIG. 10, it is not possible to handle only the matrix calculation and the matrix calculation and the interpolation calculation as one data. Also,
Although the operation specification of the interpolation calculation (the number of matrices used and the number of data of the interpolation calculation) was set to the arithmetic unit before issuing the operation instruction, even if the operation specification of this arithmetic unit is specified in the header. Good. In this case, the header analysis / operation unit controller 620 controls the operation of the operation unit according to the setting information in this header. In this way, when an object with a joint structure is expressed, it is possible to specify that a specific joint part should be subjected to interpolation calculation, and it is easy to express such that the joint part after interpolation calculation is bent smoothly like rubber. You can do it. For this,
The range of image expression for games and the like expands.

[0034]

According to the present invention, it is used in a video game.
In a simulation of a dimensional space or the like, the matrix calculation and the interpolation calculation can be performed using one calculator.

[Brief description of drawings]

FIG. 1 is a diagram showing a result of an interpolation calculation using a group of interpolated vertices.

FIG. 2 is vector data in which a header and vertex data are combined, which is one of the embodiments of the present invention.

FIG. 3 is a diagram showing a technique relating to coordinate conversion and interpolation calculation, which is one of the embodiments of the present invention.

FIG. 4 shows a group of vertices for interpolation calculation using one type of matrix and vector data thereof.

FIG. 5 shows a group of vertices for interpolation calculation using a separate matrix and vector data thereof.

FIG. 6 is a block diagram showing a high-speed interpolation calculator, which is an example of the present invention.

FIG. 7 is a block diagram showing an interpolation value register of a high speed interpolation calculator.

FIG. 8 is a diagram showing operations of a multiplier / adder and a temporary buffer when performing interpolation calculation.

FIG. 9 is a diagram showing operations of a multiplier / adder and a temporary buffer when performing interpolation calculation.

FIG. 10 shows a group of vertices of a joint structure and vector data thereof, which is a configuration in which interpolation calculation is performed only on two joints and designated joint portions.

[Explanation of symbols]

10 Interpolated vertex group A 20 Interpolated vertex group B 30 Interpolation result vertex group 300 Operator 310 Memory 350 Vector / data analysis / matrix selector 370 Matrix operator 380 Interpolated value storage area 390 Matrix operation result storage area 600 High-speed interpolation operation Device 610 Header / Data sorter 620 Header analysis / arithmetic unit controller 630 Memory 641, 643, 645, 647 MtoB
Multiplexer 642 Buffer 1 644 Buffer 2 646 Buffer 3 648 Buffer 4 649a to 649d Matrix operation result storage area 650 Multiplexer 660, 670, 680 Multiplier adder 682, 684, 686 Multiplier 688 Adder 690 Interpolation value selector 700 Interpolation value register 710 Interpolation value register / controller 720 Interpolation value storage interface 731 to 739 register 740 nto1 multiplexer

Continued front page    F-term (reference) 2C001 BA03 BC05 CB01 CB02 CB03                       CC01                 5B056 AA02 BB31 BB52 FF01 FF05                       HH03                 5B080 DA06 FA16

Claims (6)

[Claims] 1. A high-speed interpolation calculator for performing a matrix calculation including an interpolation calculation by inputting a data string composed of a header designating a matrix and data to be matrix-calculated. A matrix storage unit that stores a plurality of matrices used for the interpolation, an interpolation ratio storage unit that stores the interpolation ratio used for the interpolation calculation, and the header and the data are identified from the input data string, and the header Selects a matrix stored in the matrix storage unit, and in the case of interpolation calculation, selects and outputs an interpolation rate and data, a data analysis / matrix selection unit, a matrix calculation unit for performing matrix calculation, and the calculation described above. A matrix calculation result storage unit for temporarily storing results is provided. The calculator performs a matrix operation on the selected matrix and the data, and when performing the interpolation operation, the matrix operator outputs the result of the matrix operation on the selected matrix and the data to the matrix operation result. Stored in the memory,
A high-speed interpolation calculator, which performs interpolation calculation by inputting the calculation result and the interpolation rate stored in the interpolation rate storage section, which are temporarily stored in the matrix calculation result storage section. 2. The high-speed interpolation calculator according to claim 1, wherein the execution of the interpolation calculation is designated by the data in the header, and the interpolation calculation is performed by being detected by the data analysis / matrix selection unit. A high-speed interpolation calculator characterized in that 3. The high-speed interpolation arithmetic unit according to claim 1, further comprising a buffer provided between the matrix storage unit and the matrix arithmetic unit, the buffer being a matrix cache and / or the matrix. A high-speed interpolation calculator which is used as a calculation result storage unit. 4. The high-speed interpolation calculator according to claim 1, wherein the interpolation calculation can specify a different matrix for each data used for interpolation. 5. The high-speed interpolation calculator according to claim 1, wherein the interpolation calculation is performed using two data or three data. 6. The high-speed interpolation calculator according to claim 1, wherein a plurality of the stored interpolation rates can be stored, and the number of times of use of each interpolation rate can be designated. A high-speed interpolation calculator.