JPH11353496A - Intersection search device for light ray tracing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an intersection search device which can be actualized at low cost by custom VLSI technology and is needed to generate images by a light ray tracing method fast enough for view as a moving picture. SOLUTION: The device consists of a voxel tracing part 2 which finds out a voxel on an optical path, an intersection calculation part 5 which performs intersection calculation, and an output buffer 4 and the respective parts process one light-ray packet in each clock cycle through pipeline processing. The voxel tracing part 2 performs parallel processes by voxel plane set (slide) for a three- dimensional grating which is made hierarchical. Further, three-dimensional DDA which perform three two-dimensional DDAs in parallel finds out a voxel having an intersection with a light ray without using floating-point operation. To remove repetition of the intersection calculation, information determined by the inclusion relation between a set of bodies included in each voxel and a set of three-dimensional bodies included in a voxel positioned nearby it is used.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for generating a three-dimensional shaded image at high speed using a ray tracing method (ray tracing method) to the extent that it can be viewed as a moving image, making full use of a custom VLSI technology. It is about.

[0002]

2. Description of the Related Art The ray tracing method is a method of generating a highly realistic shaded image based on a database storing information on three-dimensional objects (for example, triangles, spheres, free-form surfaces, etc.). is there. The processing that requires the most calculation time in the ray tracing method is to check for each ray whether there is a three-dimensional object (hereinafter, referred to as an object) that intersects the ray, and if there is an intersection, determine the position of the intersection. This is the process of asking. Hereinafter, this processing is called an intersection search processing. In the intersection search processing, when one ray intersects with a plurality of objects, or when there are a plurality of intersections with the same object, the position of the intersection closest to the start point of the ray is usually obtained. How to speed up this intersection search process is an important key for speeding up the entire image generation.

The simplest method of searching for an intersection is to calculate the position of the intersection directly by solving a simultaneous equation of an equation representing the ray and an equation representing the object for each ray (hereinafter, this calculation is referred to as a direct calculation). (Referred to as intersection calculation) for all objects, and the intersection closest to the ray start point is selected from among them. However, when an image is generated by this method, the amount of calculation for intersection search processing becomes enormous.

Therefore, as a method of performing the intersection search processing with a relatively small amount of calculation, Fujimoto et al., "ARTS: Accesse
rated Ray-Tracing System
m ", IEEE Computer Graphics
and Application Magazine, Volume 6, Volume 4
No., pages 16 to 26, published in 1986, is known. In this method, as shown in FIG. 5, a space where an object exists is divided into small rectangular parallelepipeds called voxels, and a list of objects included in each voxel is obtained in advance. When performing the intersection search process, scan the voxels on the optical path in order from the ray starting point, and when encountering a voxel including the object, perform a direct intersection calculation on the object included in the voxel. Do. For example, in the intersection search processing for the ray 601, first, the contents of the voxel 602 are checked. Since voxel 602 does not contain an object, voxel 60
Since 3 is checked and is again empty, processing proceeds to voxel 604. Since the voxel 604 includes the object 609 and the object 610, a direct intersection calculation is performed on both objects, and as a result, an intersection 611 with the object 609 is found. In this example, the intersection 611 is the intersection closest to the ray start point 612. Object 613 and object 61
Note that a direct intersection calculation with 4 is not required. For the sake of simplicity, FIG. 5 shows a two-dimensional diagram, but it is actually three-dimensional.

[0005] In the above-mentioned document, as a method of quickly finding voxels on an optical path, there is disclosed a method called 3DDDA method, in which multiplication / division is not used except for initial calculation by using incremental calculation. In this 3DDDA method,
The coordinate axis where the absolute value of the component of the direction vector of the light ray is the maximum is defined as the drive axis, the other two coordinate axes are defined as the passive axes, and two two-dimensional DDAs (digital differential analyzers) between the drive axis and each passive axis are determined. Execute in parallel with synchronization. The biggest problem in making a two-dimensional DDA three-dimensional is that, as shown in FIG. 6, when moving in the direction of a drive shaft (shown as a main drive shaft), both passive shafts (see FIG. In this case, the light beams are voxels 621 and 62 when traveling in the directions at the same time.
The path of 2,624 and voxels 621,623,62
This is how to determine which of the four paths is taken (points 629 and 630 in the figure indicate the intersection of the ray with the plane of the voxel perpendicular to the main drive axis). In the 3DDDA method described in the above document, the DDA error term is set in the drive axis direction, thereby making it possible to determine a path only by comparing the magnitude of the error term. However, when the DDA error term is set in the direction of the drive axis, the range of the value of the error term is greatly widened, so that fixed-point representation is difficult, and as a result, a floating-point operation is required.

Although the number of direct intersection calculations is greatly reduced by the voxel space division method compared with the case where direct intersection calculations are performed for all objects, in order to keep this number low, as the number of objects increases, the number of voxel Must be finely divided. Therefore, when the number of objects is large,
A large amount of memory is required to store the list of objects for each voxel, and the amount of calculation for tracking voxels on the optical path is increased. In order to solve these problems, a method of performing division hierarchically is called S
nyder et al., “Ray Tracing Compl.
exModels Containing Surfa
ce Tessellations ”, ACM Com
puter Graphics, Vol. 21, No. 4,
Pp. 119-128, published in 1987, the specific effects of which are described in Klimaszewski.
Others, "Faster Ray Tracing Usi
ng Adaptive Grids ", IEEE C
Omputer Graphics and Appl
Ications Magazine, Vol. 17, No. 1, 42-51
Page, published in 1997. In this method, a local three-dimensional grid is set for a portion where objects are aggregated, and finer voxel division is performed in the grid. This grid is for the voxel of the grid above it,
It is registered in the object list in the same way as a normal object.
The well-known octree is a special form of this method. When performing the intersection search processing, first, the outermost grid is scanned with voxels on the optical path in order from the closest to the ray start point. When a local grid is included in the voxel, the intersection search processing is recursively applied to the local grid. In this specification, the recursive intersection search processing applied to the local grid is called an indirect intersection calculation, and the direct intersection calculation and the indirect intersection calculation are collectively called an intersection calculation.

[0007] By the above-described hierarchical grid method, the number of direct intersection calculations can be kept down considerably even when the number of objects increases. Nevertheless, in order to generate an image as fast as it can be seen as a moving image, 1
Since it is necessary to finish the intersection search processing for the light beam within about several cycles of the processor clock, the performance of a single general-purpose processor at present is insufficient.
Therefore, a dedicated device for intersection search processing is required.

As a conventional dedicated device for executing the intersection search processing, a structure in which a plurality of general-purpose processors are connected by a network is disclosed in Kobayashi et al., "Load Balanci".
ng Strategies for a Paraal
ll Ray-Tracing System Ba
sed on Constant Subdivisi
on ", The Visual Computer,
Vol. 4, No. 4, pp. 197-209, published in 1988. In this device, processing is assigned to processors in units of voxels, and then intersection search processing is performed in parallel using a voxel space division method. The processor exchanges a data packet, which is called a light packet, in which various pieces of information about one light beam are recorded.
In this apparatus, a huge number of processors are required to generate an image at such a high speed that it can be viewed as a moving image.
In addition, since the communication between the processors is irregular, there is a problem that many hardware resources are required for the communication buffer and the communication control circuit. Further, another disadvantage is that it does not support a hierarchical grid.

The intersection search process mainly consists of a process of finding a voxel on the optical path and a process of performing a direct intersection calculation. Of these processes, the process of finding a voxel on the optical path is a process of generating an image by a volume rendering method. In the device for the dedicated hardware associated with the present invention can be found. Unlike the ray tracing method, the volume rendering method does not use a database in which objects are stored, directly defines colors and transparency for voxels, and generates an image based on the color and transparency. The following two paragraphs describe the prior art of devices intended for volume rendering.

First, a dedicated processor for finding a voxel on an optical path at high speed by utilizing a skewed voxel information memory described in Kaufman et al., US Pat. No. 5,594,842 is known. However, this device is designed on the assumption that the starting points of all rays are the same, and if it is used for purposes such as ray tracing where the starting points are not constant, processing for switching the starting points is performed. Is very time consuming. Therefore, it is difficult to use this technique for ray tracing purposes.

On the other hand, Akashi et al., "Architecture of Parallel Computer for Volume Rendering", Information Processing Society of Japan Computer Architecture Study Group Report No. 97-15, No. 11
A dedicated processing device for finding a voxel on an optical path for a light beam in an arbitrary direction by duplicating a voxel memory described in pages 3 to 120, published in 1992 is known. In this apparatus, a coordinate axis in which the absolute value of the component of the direction vector of a light ray is the maximum is determined as a main axis, and processing is assigned to a processing apparatus based on the main axis component of voxel coordinates. This device differs from that of Kaufman et al.
Even if the starting point is random, the processing speed does not decrease, and it is possible to find a voxel on the optical path for a ray in an arbitrary direction during one memory access cycle. In addition, since communication is regular, much hardware resources are not consumed for a communication buffer and a communication control circuit. However, since this apparatus is intended for the volume rendering method, the following problems occur when used for the ray tracing method. In the ray tracing method, all voxels having an intersection with a ray (for example, voxel 6 in FIG. 5)
02, 603, 604, 605, 606, 607, 60
8), but in this apparatus, one voxel (for example, in the case of FIG. 5, voxels 602, 60
4, 605, 607, 608). Therefore, it cannot be used for the purpose of ray tracing unless this point is improved.

One of the unique problems of the voxel space partitioning method is the elimination of the intersection calculation. This is to avoid performing an intersection calculation with the same object more than once for one ray. For example, in the situation shown in FIG.
When the voxel 641 is tracked, the intersection between the ray 642 and the object 643 is calculated, and the intersection 644 is found. But,
Since the intersection point 644 is located outside the voxel 641, there is a possibility that the intersection point 644 may intersect with another object in a subsequent voxel. Therefore, it is not possible to determine the intersection point 644 as a final intersection point at this time. Therefore, next voxel 645
Is transferred to the object 64 in the process.
There is no need to perform an intersection calculation with 3. By avoiding overlapping intersection calculations in this way, a reduction in calculation time is achieved. Also, if the same duplication elimination is applied to the indirect intersection calculation, the calculation time can be significantly reduced.

As a method for performing the above-described intersection calculation duplication elimination, a method for a single processor is described in J. Org. Am
anatides et al., "A Fast Voxel T
raveral Algorithm for Ray
Tracing ", Proceedings of
Eurographics '87, pp. 3-10, 19
The method described in 1987 is known. In this method, the identification number of the ray whose intersection calculation was performed last is stored for each object. However, when this method is used in an apparatus that processes a large number of rays in parallel, it is not sufficient to store one identification number for each object. The set of identification numbers for the calculated rays must be stored. This is difficult to realize by dedicated hardware, and even if it can be realized, there is an upper limit to the number of elements in a set.

[0014]

As described above, if a device for generating a three-dimensional shaded image at a high speed such that it can be seen as a moving image using the ray tracing method is constituted by a plurality of general-purpose processors, the number of processors becomes enormous. In addition, since many hardware resources are required for the communication buffer and the communication control circuit, there is a problem that the cost increases.

In addition, there has been an attempt to construct voxels on the optical path, which is a part of the intersection search processing, by using dedicated hardware. However, the voxels discovered by the conventional hardware are missing. There was a problem that there is.

In addition, since the conventional 3DDDA method for finding voxels on the optical path requires a floating-point operation, considering the realization with dedicated hardware,
There is a problem that the circuit scale becomes large.

Further, there is a problem that the conventional method for eliminating duplication of intersection calculation is difficult to realize with dedicated hardware.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to realize a moving image by a ray tracing method which can be realized at low cost by a custom VLSI technology. An object of the present invention is to provide an intersection search device necessary for generating an image.

[0019]

According to the present invention, the above objects are achieved by the means as set forth in the appended claims.

That is, based on various information of a three-dimensional object group defined in the three-dimensional virtual space, various parameters of light rays are received as input, and one or more elements having hexahedral voxels as elements in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light ray and the three-dimensional object group, information indicating the necessity of calculating an intersection of each of the voxels with a three-dimensional object included in the voxel One or more voxel information storage means for storing the information indicating the necessity of the intersection calculation from the voxel information storage means, and voxels that intersect with the ray and are considered to need the intersection calculation And one or more voxel tracking circuits that output an index of A plurality of voxel plane sets for each coordinate axis in each of the three-dimensional grids are arranged in at least one of the banks, and the voxel tracking circuit includes a plurality of voxel tracking units. And each of the voxel tracking units is responsible for a predetermined set of voxel planes among a group of voxel plane sets related to the coordinate axis where the absolute value of the direction vector component of the light ray is the largest. Another object of the present invention is to provide an intersection search apparatus characterized in that it has a function of specifying voxels included in some voxel plane sets and intersecting with the light ray without omission.

Further, when the voxel tracking circuit sets k to be a constant of 2 or more, it outputs the index of at most k voxels at a time. k is one of the parameters for determining the hardware configuration.

Further, a local three-dimensional grid is set for a part of the space occupied by the three-dimensional grid, and the local three-dimensional grid is used as a kind of the three-dimensional object. Are organized hierarchically.

The voxel tracking units are interconnected linearly by communication means.

When the main drive axis, the sub drive axis, and the passive axis are set in the descending order of the absolute values of the components of the direction vector of the light beam with respect to the three coordinate axes, the two-dimensional DDA between the main drive axis and the sub drive axis is used. (Digital differential analyzer)
And a two-dimensional DDA between the main driving axis and the passive axis and a two-dimensional DDA between the sub-driving axis and the passive axis are executed in synchronization and in parallel to find a voxel having an intersection with the ray. .

Further, the bank stores information indicating the necessity of the intersection calculation for the voxel and information indicating the necessity of the intersection calculation for the voxels around the voxel in the same word. I do.

The voxel further includes one or more object set storage means for storing a set of three-dimensional objects included in the voxel. The object set storage means includes a three-dimensional object included in each voxel. Regarding the object, duplicate removal information that records whether or not each voxel located near the voxel includes the same three-dimensional object as the three-dimensional object is further held.

The voxel includes one or a plurality of voxel information storage means for storing information indicating the necessity of intersection calculation with each of the three-dimensional objects included in the voxel. The information to be shown is assumed to have information defined by the inclusion relationship between the set of three-dimensional objects included in the voxel and the set of three-dimensional objects included in voxels located near the set.

The voxels located in the vicinity are six adjacent voxels.

In addition to the voxel information storage means and the voxel tracking circuit, one or more objects for determining a three-dimensional object to be cross-calculated based on voxel indices output from the voxel tracking circuit. A decision circuit, and one or more intersection calculation circuits for performing a direct intersection calculation between the ray and the three-dimensional object to be calculated by the object determination circuit, each of the voxel tracking circuits; And each of the intersection calculation circuits is provided with a separate object determination circuit.

The meanings of some terms used in the above description are as follows. The voxel index means a code for uniquely identifying a voxel in one grid. One example is a triple integer corresponding to a subscript when voxel data is stored in a three-dimensional array. The voxel plane set means a set of voxels having the same coordinate value with respect to one coordinate axis in a coordinate system in which a grid is defined. The set of voxel planes on a certain coordinate axis means a set of voxels having the same coordinate value on the coordinate axis.

[0031]

In the voxel information storage means, arranging any set of voxel planes for each coordinate axis in each of the three-dimensional grids in at least one of the banks,
The number of voxel references per bank required to find a voxel that intersects a ray in any direction without omission can be reduced to at most 3 by preparing a sufficient number of banks.
There is an effect that it becomes possible to hold down. As a result, the processing time for one light beam in each bank and each voxel tracking unit becomes substantially equal, which has the effect of reducing communication buffers and communication control circuits between voxel tracking units.

In the voxel tracking circuit, k is set to a constant of 2 or more and an index of at most k voxels is output at a time, that is, a plurality of voxels which may be subjected to intersection calculation are found at a time. The thing is
The number of times of processing by the voxel tracking circuit is reduced, and as a result, higher speed processing can be performed with the same amount of hardware. Conversely, limiting the number of voxels that can be found at one time to k or less has the effect of making the size of the light packet fixed and thus simplifying the hardware for communication. When the voxel tracking circuit needs to track more than k voxels, the processing is performed twice or more. That is, at the time of the first tracing, the voxel including the ray start point is set as the tracing start voxel, and voxels beyond the tracing start voxel are traced. This is called new tracking. If the next voxel needs to be tracked, the voxel located farther than that voxel is tracked, with the last processed voxel as the tracking start voxel. This is called continuous tracking.
The continuous tracking is performed repeatedly as necessary.

The introduction of the hierarchical grid has the effect of reducing the storage capacity required for the voxel information storage means and greatly reducing the number of required voxel tracking units.

The linear interconnection of the voxel tracking units has the effects of simplifying the hardware and facilitating the VLSI configuration, as compared to a connection such as a tree.

The error terms in three two-dimensional DDAs executed in parallel and synchronously in the voxel tracking circuit are:
In the two-dimensional DDA between the main drive axis and the sub-drive axis, in the sub-drive axis direction, in the two-dimensional DDA between the main drive axis and the passive axis, in the passive axis direction, and in the two-dimensional DDA between the sub-drive axis and the passive axis, in the passive axis direction. Each is set. The two-dimensional DDA between the auxiliary drive shaft and the passive shaft, which did not exist in the prior art, is used for the purpose of performing the route determination described with reference to FIG. In the known technique, the error term is set in the direction of the main drive axis, and as a result, a floating-point operation is required. However, such a disadvantage does not occur in the present invention. That is, it is possible to use a fixed-point representation for the error term, and as a result, it is possible to greatly reduce the circuit scale when realization by dedicated hardware is considered. Further, since voxels on the optical path can be found without any omission, it is effective for solving the problem that voxels to be found in dedicated hardware for a known volume rendering method are missing.

In the bank of the voxel information storage means,
Storing the information indicating the necessity of the intersection calculation for the voxel and the information indicating the necessity of the intersection calculation for the voxel around the voxel in the same word requires finding the voxel on the optical path without omission. It is effective for performing high speed. Reading out one voxel from each bank was sufficient for dedicated hardware for the well-known volume rendering method, but in order to find a voxel on the optical path without missing, up to three voxels per bank were required. It needs to be read. However, if they are stored in the same word as described above, the reading of these three voxels can be performed by one access.

The duplicate elimination information in the object set storage means is effective in eliminating duplicate intersection calculations. When a ray enters a voxel and examines the intersection with an object contained in the voxel, if the ray has already visited some of the voxels in the vicinity, it also contains the same object In this case, the intersection calculation is already performed, so it can be omitted. In other words, before performing the intersection calculation,
The duplication elimination can be realized by examining the information corresponding to the voxel visited in the past among the duplication elimination information. This duplication elimination method is suitable for hardware implementation.

The information indicating the necessity of the intersection calculation in the voxel information storage means is determined by the inclusion relationship between a set of three-dimensional objects included in the voxel and a set of three-dimensional objects included in voxels located near the voxel. Using the information provided has implications for the elimination of duplicate intersection calculations described above. As a result of performing the duplicate elimination based on the duplicate elimination information, all the intersection calculations for one voxel may be omitted. The information defined by the inclusion relations described above serves to avoid such situations in advance. That is, the above situation occurs when the set of three-dimensional objects included in the voxel is a subset of the set of three-dimensional objects included in voxels visited in the past. Therefore, as information indicating the necessity of intersection calculation, it is recorded whether or not the set of three-dimensional objects included in the voxel is a subset of the set of three-dimensional objects included in voxels located near the voxel. In the voxel tracking circuit, a voxel in the above situation can be identified with a voxel containing no object. As a result, the number of voxel indices output from the voxel tracking circuit is greatly reduced, and high speed is achieved.

The most basic embodiment of the above-described duplication elimination information and the information indicating the necessity of intersection calculation is that adjacent voxels are used as adjacent voxels in six adjacent directions (+ x, -x, + y, -y, +
(v, z, -z direction). In this case, which of the neighboring voxels the light ray has visited in the past can be determined based on from which surface the light ray has entered the voxel, and the circuit scale can be small. However,
Not all duplicate intersection calculations can be eliminated. As in the example shown in FIG. 8, there are cases where the object cannot be removed unless the shape of the object is obtained by scaling the shape of the voxel. However, since local grids that have the greatest effect of reducing the amount of calculation due to the duplication removal can always be removed, they are considered to be practically sufficient. Note that it is possible to consider voxels that are farther away as nearby voxels in order to increase the removal efficiency.

Providing a separate object determination circuit for each of the voxel tracking circuits and each of the intersection calculation circuits simplifies the communication network in the apparatus and reduces the communication buffer for the object determination circuit. Has the effect of making it unnecessary.

[0041]

Embodiments of the present invention will be described below with reference to the drawings.

In the following description, a normal object and a grid which are the targets of direct intersection calculation are collectively referred to as an object. Objects other than the grid are called primitives.

In the following embodiment, an orthogonal coordinate system is used as a coordinate system, and it is assumed that the shape of a voxel is a cube. It is also assumed that the size of voxels is uniform within one grid. Further, the number of voxel divisions in each lattice is limited to n or less for each axis. The hierarchical level of the hierarchical grid is limited to two levels, ie
Only one global grid and some local grids set therein are used, and a local grid B is not set inside a certain local grid A and registered as an object in the local grid A (however, It is possible to register the local grid B in the global grid). Further, it is assumed that a local grid rotated with respect to the global grid is not considered. In addition, the shape of the primitive is limited to only a triangle. An embodiment that does not impose these restrictions will be described later.

FIG. 1 is a diagram showing a configuration of an intersection searching apparatus of the present invention and a configuration of an apparatus for generating an image by a ray tracing method using the same. The intersection search device 1 is an embodiment of the intersection search device of the present invention, receives a light ray packet including a start point coordinate and a traveling direction of a light ray as an input, and is closest to a start point among intersections of the light ray and the primitive. A ray packet including the position and the like of one is output. The initial ray generator 51 generates a ray passing through each pixel on the screen passing through the viewpoint, and sends the ray packet to the intersection searcher 1. Shading and next generation ray generator 52
Calculates the normal vector at the intersection, calculates the influence of the light on the apparent color of the primitive based on the vector, and accumulates the result in the frame buffer 53. In addition, the shading and next generation ray generator 52 may intersect with primitives having specular or transmissive properties,
A new light ray is generated starting from the intersection point in the reflection direction or the transmission direction, and the light ray packet is sent to the intersection point search device 1. Also, when performing the shadowing process, a new light ray is generated in the light source direction and sent to the intersection search device 1. CRT
The display device 54 displays the composite image stored in the frame buffer 53.

The outline of the structure of the intersection searching device 1 will be described below. The intersection search device 1 includes a voxel tracking unit 2, an intersection calculation unit 3, and an output buffer 4.
The voxel tracking unit 2 includes a voxel information storage unit, a voxel tracking circuit, an object set storage unit, and an object determination circuit in the claims, and is mainly responsible for processing for finding a voxel that intersects a ray. The intersection calculation unit 3 includes two circuits having the same configuration. The intersection calculation unit 3 includes the intersection calculation circuit, the object set storage means, and the object determination circuit in the claims, and mainly handles direct intersection calculation. The output buffer 4 collects the ray packet including the result of the final intersection search processing sent from the voxel tracking section 2 and the intersection calculation section 3, and sends the packet to the shading and next-generation ray generation device 52.

The voxel tracking section 2 and the intersection calculation section 3 are driven by a common clock signal. The speed of this clock signal is set such that access to any memory in the intersection search device 1 can be completed within one clock cycle. Then, both the voxel tracking unit 2 and the intersection calculation unit 3 process one ray packet per clock cycle by pipeline processing unless one of the following stagnation conditions is satisfied. There are two stagnation conditions, one is when there is no ray packet to be processed at the input of each part,
The other case is when packets cannot be output because the output destination buffer of each unit is full. Of the clock cycles, a normal one that does not satisfy the stagnation condition is referred to as an effective clock cycle in the following description.

FIG. 2 is a diagram showing the internal structure of the voxel tracking unit 2. FIFO (First-In First)
-Out) Buffers 201, 202, 203, and 204 temporarily store light beam packets sent from each unit.
These FIFO buffers improve the performance of the system by receiving all of the light packets even if two or more light packets arrive at the same time as long as the buffers do not overflow. Multiplexer 205 outputs F
One of the IFO buffers 201, 202, 203, 204 or the output destination determination circuit 217
And passes it to the DDA initialization circuit 23. The DDA initialization circuit 23 calculates a voxel for starting tracking, an error term in the DDA, and the like.

The voxel information storage means 24 is a group of memories holding information indicating the necessity of intersection calculation, and a bank 241 of memories holding different storage contents with the same circuit configuration.
Are provided. Each bank can read data from a different address.

The voxel tracking unit 25 sequentially examines voxels having intersections with rays by synchronizing and executing three two-dimensional DDAs in parallel, and reads information indicating the necessity of the intersection calculation from the corresponding bank 241. While finding voxels deemed to require up to k intersection calculations as a whole. There are provided n voxel tracking units 25 having the same circuit configuration, each of which is in charge of processing relating to one voxel plane set.
Each voxel tracking unit examines up to three voxels on the optical path at a time during one valid clock period.

The grid information memory 26 stores the starting point coordinates, the number of voxel divisions, the size of voxels, and the like for each grid.

The tracking end processing circuit 211 performs data reduction of the tracking result and the like. The DDA initialization circuit 23, the voxel tracking unit 25, and the tracking end processing circuit 211 together form a voxel tracking circuit 22.

When the voxel obtained by the voxel tracking circuit 22 has advanced ahead of the voxel including the intersection which has been found so far, the overrun check circuit 212 terminates the processing relating to the grid currently being processed. There is.

The L1 range inspection circuit 214 performs processing such as determining whether or not the intersection search can be terminated. For example, in FIG.
The intersection calculation between 2 and the object 643 is performed. As a result, the intersection 644 is found, but the intersection search processing cannot be terminated at this stage. This is because the intersection point 644 is located outside the voxel 641, and may intersect another object in a subsequent voxel. Conversely, if the intersection 644 is located inside the voxel 641, the intersection search process can be terminated at that point and the intersection 644 can be made the final intersection. In such a situation, it is the role of the range inspection circuit to determine whether the intersection search process can be terminated. L1 range inspection circuit 21
4 makes this determination for the global grid.

The object set storage means 27 stores a set of objects included in each voxel of each grid. FIG. 9 (a)
Shows the data structure of the stored contents. As shown in the figure, in this embodiment, a set is represented by a linear list. The list header array 273 holds the address of the head element of the list for each voxel. List element 274
Exists for each object included in each voxel, and the type 27
5, a primitive or lattice identification number 276, duplicate removal information 277, and a next address 278. The type 275 is 1-bit information for distinguishing whether an object is a primitive or a local lattice. The primitive or lattice identification number 276 is either the primitive serial number or the lattice serial number. Duplicate removal information 27
Reference numeral 7 denotes information indicating whether the same object is included in six adjacent voxels, the details of which are as shown in FIG. 9B. Deduplication bits 277 for the objects contained in the voxel whose index is (ix, iy, iz)
1, 2772, 2773, 2774, 2775, 277
6 are (ix-1, iy, iz) and (ix +
1, iy, iz), (ix, iy-1, iz), (i
x, iy + 1, iz), (ix, iy, iz-1),
It becomes 1 when the same object is included in the voxel having the index of (ix, iy, iz + 1). FIG.
The next address 278 in (a) holds the address of the next list element 279. When the next list element does not exist, as in the case of the list element 279, a special address that cannot exist (hereinafter, this address is referred to as “φ”) is added to the next address part. Store.

The list header memory 271 in FIG. 3 holds the list header array 273 and
72 holds a list element 274. The list element memories 272 are arranged with r (r ≧ 1) duplications, and all hold the same contents.

The list head element determination circuit 215 reads the address of the first list element from the list header memory 271 based on the voxel index.

The object determination circuit 216 determines an object to be cross-calculated by reading the list element 274 from the list element memory 272. Circuits having the same configuration are connected in r stages, and it is possible to check at most r continuous list elements for one light packet. If there is at least one of the r objects indicated by the r list elements, which is determined from the duplication elimination information 277 and has not yet been subjected to the intersection calculation, the object to be subjected to the intersection calculation is determined. success. However, if all the r objects have already been subjected to the intersection calculation, the determination of the object to be subjected to the intersection calculation fails, and the determination is carried over when the same ray packet passes through the object determination circuit again.

The output destination determination circuit 217 determines where the light packet should be delivered next based on the results of voxel tracking and object determination.

FIG. 3 is a diagram showing the internal structure of the intersection calculation unit 3. The FIFO buffer 301 temporarily stores the light packet transmitted from the voxel tracking unit 2. The multiplexer 305 receives one light packet from either the FIFO buffer 301 or the output destination determining circuit 317 every valid clock cycle, and passes it to the intersection calculating circuit 32.

The intersection calculation circuit 32 reads out the shape data from the object shape information storage means 33 which holds data relating to the shape of the primitive, and performs a direct intersection calculation by solving a simultaneous equation of the primitive and the light ray.

The L2 range inspection circuit 313 performs processing such as determining whether or not the intersection search similar to the L1 range inspection circuit 214 can be terminated for the local grid.

The other parts of the intersection calculation unit 3 are the same as those of the voxel tracking unit 2. That is, the grid information memory 3
6, the grid information memory 26, the L1 range inspection circuit 314 and the L
One range inspection circuit 214, list head element determination circuit 315
And the list head element determination circuit 215 and the object determination circuit 316
And the object determination circuit 216, the list header memory 371 and the list header memory 271, the list element memory 372 and the list element memory 272, the output destination determination circuit 317 and the output destination determination circuit 217 are completely equivalent memories and circuits.

FIG. 4 is a diagram showing the internal structure of the output buffer 4. The FIFO buffers 401, 402, and 403 temporarily store the light packet transmitted from each unit.
These FIFO buffers can receive all ray packets as long as the buffer does not overflow, even when two or more ray packets arrive at the same time, or when the shading and next generation ray generator 52 is not ready to receive data. There is a role to do. The multiplexer 404 includes FIFO buffers 401, 402, and 403.
, And sends it to the shading and next-generation light generating device 52.

FIG. 10 shows an example of the whole image of the information indicating the necessity of the intersection calculation stored in the voxel information storage means 24 for one grid, and the case where the number of voxel divisions is n for each axis. FIG. The voxel information storage unit 24 uses x for each voxel in the grid when the x-axis component among the components of the direction vector of the light ray has the maximum absolute value (that is, the main drive axis is the x-axis).
The main axis voxel information 242, the y main axis voxel information 243 used when the main driving axis is the y axis, and the main driving axis z
It holds three types of different information called z-axis voxel information 244 used in the case of an axis. The x main axis voxel information group 245 includes n sets of voxel plane sets 2451, 2452,. . . , 2453, and the i-th voxel plane set is stored in two places, i.e., the i-th bank 241 and the (n−i + 1) -th bank 241 (where 1 ≦ i).
≤n). For example, the voxel plane set 2451 is stored in the first and nth banks 241, and the voxel plane set 2452 is stored in the second and (n-1) th banks 241. The data stored in the i-th bank 241 is used when the x component of the direction vector is positive,
The data stored in the (ni + 1) -th bank 241 is used when the x component of the direction vector is negative.
Similarly, the y-major axis voxel information group 246 includes the voxel y
Voxel plane sets 2461, 246 by coordinates
2,. . . , 2463, one of which is
The voxel plane set is the i-th bank 241 and (n
−i + 1) The data is redundantly stored in two places of the 241st bank 241. The z-major axis voxel information group 247 includes voxel plane sets 2471 and 247 based on z coordinates of voxels.
2,. . . , 2473, of which i
The voxel plane set is composed of the first bank 241 and (n
−i + 1) The data is redundantly stored in two places of the 241st bank 241.

FIG. 10 shows a case where the number of voxel divisions in the grid is n for each axis. However, even when the number of divisions is less than n, the voxel information is empty for a portion exceeding the number of divisions. Assuming that there is, apply the same rules. For example, when n is 8 and the number of divisions on the x axis is 2, the first, second, seventh, and eighth banks 241 are the first, second, second, and first voxel plane sets 2451, 24, respectively.
52, 2452, and 2451, and the third, fourth, fifth, and sixth banks 241 do not hold the x-spindle voxel information.

FIG. 11A shows z-axis voxel information 244.
FIG. 3 is a diagram showing the detailed contents of FIG. z-spindle voxel information 24
4 is 15 intersection calculation request bits 24401, 244
02,. . . , 24415. Some of them are determined based on the inclusion relationship between the set of objects included in the own voxel and the set of objects included in surrounding voxels. FIG. 11B shows a voxel 651 targeted by the z-axis voxel information 244 and voxels 652, 653,. . . , 661 are shown. Although the z-axis voxel information 244 is shown in FIG. 11, the details of the x-axis voxel information 242 are obtained by replacing z with x, x with y, and y with z in FIG. Are the same as those in FIG. 3 except that z is replaced with y, y is replaced with x, and x is replaced with z.

The c bit 24401 in FIG.
Is 1 when the set of objects included in the voxel 651 is non-empty. The + z bit 24402 becomes 1 when the set of objects included in the voxel 651 is not a subset of the set of objects included in the voxel 652, that is, when the voxel 651 includes an object different from the object included in the voxel 652. . The −z bit 24403 becomes 1 when the set of objects included in the voxel 651 is not a subset of the set of objects included in the voxel 653. + X
Bit 24404 is 1 when the set of objects included in voxel 654 is not a subset of the set of objects included in voxel 651. Similarly, intersection calculation request bit 24
405, 24406, and 24407 become 1 when the set of objects included in voxels 655, 656, and 657 is not a subset of the set of objects included in voxel 651, respectively.

The + x + y bit 24408 is for voxel 6
When the set of objects included in 58 is not a subset of the set of objects included in voxel 654, the value is 1. Similarly, intersection calculation request bits 24409, 24410, 2441
1, 24412, 24413, 24414, 24415
Are voxels 659, 660, 661, 65, respectively.
The set of objects included in 8, 660, 659, 661 is the voxel 654, 655, 655, 656, 656, 65.
7 and 1 if it is not a subset of the set of objects included in 657.

In determining the intersection calculation request bit,
In the case near the grid boundary, if the surrounding voxels do not exist because they are outside the range of the grid, the set of objects for the surrounding voxels is regarded as an empty set.

Of the above intersection calculation request bits, c bit 2
Fourteen, except 4401, are used to determine if a crossing calculation is needed when a ray travels from one voxel to the next. c bit 24401 is
It is used to determine the need for intersection calculations for voxels containing the starting point of the ray.

In FIG. 10, z-axis principal-axis voxel information group 2
A set of voxel planes 2471 and 247 constituting the 47
2,. . . , 2473 actually have voxel information one voxel extra around n × n voxel information, and store a total of (n + 2) × (n + 2) voxel information in the bank 241. Therefore, as a whole z principal axis voxel information group 247, n × (n + 2) × (n + 2)
The pieces of voxel information 244 are stored in the bank 241. According to the voxel information added to the surroundings, a set of objects included in the center voxel 651 is regarded as an empty set, and intersection calculation request bits 24401 and 2440 are set.
2,. . . , 24415 are determined. This is because the x-axis voxel information group 245 and the y-axis voxel information group 246
The same applies to.

The outline of the operation of the intersection searching device 1 will be described below by taking as an example a case where a light beam and an object have a positional relationship as shown in FIG. For the sake of simplicity, the intersection search process in a two-dimensional space will be described here. However, a three-dimensional intersection search process is actually performed. Further, in the description of the operation outline, it is assumed that the number n of voxel tracking units is 4.

When the ray packet of the ray 671 is input to the intersection search device 1, the ray packet is sent to the voxel tracking section 2, passes through the FIFO buffer 203 and the multiplexer 205, and is delivered to the voxel tracking circuit 22.
The voxel tracking circuit 22 searches the global grid 672 for a voxel with a possibility of intersection calculation.

FIG. 12B shows the information indicating the necessity of intersection calculation stored in the voxel information storage means 24 for each grid of FIG. It is. In this example, since the main drive axis is the z axis, the z main axis voxel information group 247 is referred to.
In the z-spindle voxel information group 247, 15
Intersection calculation request bit 24401, 2440
2,. . . , 24415, but in FIG. 12 (b), only the corresponding intersection calculation request bits corresponding to the directions of incidence of light rays on the respective voxels are selected and displayed.

The voxel tracking circuit 22 works to find at most k voxels whose intersection calculation request bit shown in FIG. Here, the case where k is 2 is considered. Voxel tracking for the global grid 672 is performed as follows. The first voxel tracking unit 25 examines the voxel 673, and stores the index of the voxel 673 in the ray packet and passes it to the second voxel tracking unit 25 since the intersection calculation request bit is 1. Second voxel tracking unit 2
5 examines voxel 674 and voxel 675 and additionally stores the index of voxel 674 whose intersection calculation request bit is 1 in the ray packet. At this point, since two voxels have already been found, the third and fourth voxel tracking units 25 pass the ray packets. Thus, the voxel tracking circuit 22 sends the voxel 67
A ray packet containing the index of 3 and voxel 674 is output.

The above-mentioned light ray packet passes through the overrun inspection circuit 212 and the L1 range inspection circuit 214 as they are, and the list head element determination circuit 215 determines the address of the list element 274 for the voxel 673, and the object determination circuit 216 In, the list element 274 is read. Since voxel 673 includes only local grid 678, this list element is that of local grid 678. The object determination circuit 216 further includes the duplication removal information 27
7, the local grid 678 is determined as the next object to be cross-calculated, and the ray packet is sent to the output destination determining circuit 217. The output destination determining circuit 217 returns the ray packet to the voxel tracking circuit 22 via the multiplexer 205 to perform voxel tracking on the local grid 678 next.

In the voxel tracking for the local grid 678, as shown in FIG. 12 (b), there is no voxel through which the ray passes, and the intersection calculation request bit is 1, so that the voxel tracking circuit 22 Is passed to the overrun inspection circuit 212 without adding the index of Overrun check circuit 212 and L
The one-range inspection circuit 214 determines that the processing on the local grid 678 has been completed, returns the processing target to the global grid 672, and
The next voxel in the global grid, voxel 674
The light packet is updated so that the processing target is moved to (the index still remains in the light packet) and sent to the list head element determination circuit 215. The list head element determination circuit 215 and the object determination circuit 216 form a voxel 674
The list element 274 is checked for the local grid 678 and the local grid 679 included in. The local grid 678 is found to be an overlapping intersection calculation based on the overlap elimination information 277. Therefore, here, the local grid 679 is determined to be an object to be subjected to the intersection calculation, and the light packet is output to the output destination determination circuit 2
Send to 17. The output destination determining circuit 217 determines whether the local grid 679
In order to perform the voxel tracking at, the ray packet is returned to the voxel tracking circuit 22 via the multiplexer 205 again.

In the voxel tracking for the local grid 679, only the voxel 683 is found. Overrun inspection circuit 212 and L1 range inspection circuit 21
4 allows the light packet to pass through. The list head element determination circuit 215 calculates the list element 2 for the voxel 683.
74 and an object determination circuit 216
Determines that the object 685 is an object to be cross-calculated, and sends the ray packet to the output destination determination circuit 217. The output destination determination circuit 217 delivers the ray packet to the intersection calculation unit 3 this time.

The light packet delivered to the intersection calculation unit 3 passes through the FIFO buffer 301 and the multiplexer 305 and is sent to the intersection calculation circuit 32. Intersection calculation circuit 32
Reads the shape information of the triangle related to the object 685 from the object shape information storage means 33, calculates the coordinate value of the intersection 686 by performing a direct intersection calculation, and sends the result to the L2 range inspection circuit 313. The L2 range inspection circuit 313 confirms that the intersection 686 is outside the voxel 683,
Therefore, an attempt is made to move the processing target to the next voxel in the local grid 679. In this case, since it is known that there is no longer a voxel whose intersection calculation request bit is 1 in the local grid 679, the processing target is set to the global grid. Returning to 672, the light packet is passed to the L1 range inspection circuit 314.
The L1 range inspection circuit 314 determines that the intersection 686 is the voxel 67
4 and therefore the global grid 672
Is attempted, the voxel before voxel 674 needs to be tracked, so the ray packet is updated to perform continuous voxel tracking in global grid 672. The light packet is transmitted to the list head element determination circuit 31.
5. The output destination determination circuit 3 passes through the object determination circuit 316
The data is delivered to the voxel tracking unit 2 via the server 17.

The light packet delivered to the voxel tracking unit 2 is transmitted to the FIFO buffer 201 and the multiplexer 205
To the voxel tracking circuit 22. In continuous voxel tracking ahead of voxel 674 for global grid 672, only voxel 677 is found,
The voxel tracking circuit 22 converts the light packet including the index of the voxel 677 into an overrun inspection circuit 212.
Send to The overrun inspection circuit 212 determines that the processing for the global grid 672 has been completed, since the voxel 677 is ahead of the voxel 676 including the intersection 686, and uses the intersection 686 as the final intersection. Update the ray packet to end. This light packet passes through the L1 range inspection circuit 214, the list head element determination circuit 215, and the object determination circuit 216, and is delivered to the output buffer 4 via the output destination determination circuit 217. Then, the signal passes through the FIFO buffer 403 and the multiplexer 404 and is finally output from the intersection search device 1 as an intersection search processing result.

While the above description has shown the processing of one ray packet, it is actually performed by pipeline processing to process many ray packets in parallel. It should be noted that no new light packet is generated inside the intersection searching device 1 and no light packet disappears. Therefore, the number of ray packets input to the intersection search device 1 and the number of ray packets output therefrom match.

Hereinafter, the structure and operation of each part in the intersection search device 1 will be described in more detail. First, as a preparation for that, a configuration of a light packet which is a unit of data exchange will be described.

FIG. 13 shows a basic part of a light ray packet used in the intersection point searching device 1. The ray packet to be input to the intersection search device 1 is obtained by extracting only the element from the element 701 storing the ray identification number rayid to the element 704 storing the direction vector component code S in the basic part. The ray packet output from the intersection searching device 1 and the ray packet used inside the output buffer 4 are ray identification numbers rayi of the basic part.
The elements from the element 701 storing d to the element 708 storing the intersection primitive identification number pid are extracted. The ray packets used inside the voxel tracing unit 2 and the intersection calculation unit 3 and the ray packets exchanged between them basically include all the core parts. However, as described later, some packet elements may be omitted depending on the location.

FIG. 14 shows a more detailed structure of a part of the main part of the light ray packet shown in FIG.

FIG. 15 shows the structure of an additional portion for a light packet used inside the voxel tracking circuit 22. Inside the voxel tracking circuit 22, a light packet having a structure in which the main part shown in FIG. 13 and this additional part are connected is used.

Hereinafter, the meaning of each element of the light ray packet will be described in order. First, the core part will be described. Element 7
The ray identification number rayid stored in 01 is a number for uniquely identifying the ray being processed in the intersection search device 1. This number is given outside the intersection search device 1, and the intersection search device 1 outputs the number when input without any reference or change thereto. This number is output from the intersection search device 1 and the attribute data (for example, pixel coordinates and light color) of the light beam not included in the light packet in the initial light beam generating device 51 and the shading and next-generation light beam generating device 52. It is used to take correspondence with the ray packet.

The ray start point coordinates E = (Ex, Ey, Ez) stored in the elements 702 to 704, the direction vector component absolute value D = (Dx, Dy, Dz), and the direction vector component code S = (Sx , Sy, Sz) represent the starting point coordinates and direction of the light beam. The ray equation is expressed as follows: x = Ex · t + Vx y = Ey · t + Vyz = Ez · t + Vz where t is a parameter variable. V = (Vx, Vy, Vz) is the direction vector of the light ray. The above D and S have the following relationship with V. Dx = | Vx | Dy = | Vy | Dz = | Vz | When Sx = “positive”, Vx ≧ 0 == “negative”, when Vx <0 Sy = “positive”, when Vy ≧ 0 = When "negative", Vy <0 Sz = "positive", when Vz≥0, = When "negative", Vz <0

The intersection point existence flag isf stored in the element 705 is 1-bit information which becomes 1 when one or more intersection points have already been found, and becomes 0 when not yet found. This element 705 is initialized to 0 when the ray packet input to the intersection point searching device 1 is first sent to the voxel tracking unit 2.

The intersection coordinates P =
(Px, Py, Pz) are coordinates of the intersections found so far in which the parameter variable t of the ray equation is positive and closest to the ray start point E (that is, t is the smallest).

The intersection parameter value tp stored in the element 707 is the value of the parameter variable of the ray equation that gives the intersection coordinates P. That is, Px = Ex.pt + Vx Py = Ey.pt + Vy Pz = Ez.pt + Vz

The intersection primitive identification number pid stored in the element 708 is the identification number of the primitive intersecting at the intersection coordinates P.

Tracking level 1 stored in element 709
“evel” is an integer that becomes 1 when tracking is performed on the global grid and 2 when tracking is performed on the local grid. This element 709 is initialized to 1 when the ray packet input to the intersection search device 1 is first sent to the voxel tracking unit 2.

The grid identification number gid stored in the element 710 is the identification number of the grid currently being processed (global grid or local grid), that is, a number for uniquely identifying which grid. This element 710 is initialized to the identification number of the global grid when the ray packet input to the intersection search device 1 is first sent to the voxel tracking unit 2.

The intersection calculation request primitive identification number rpid stored in the element 711 is the identification number of the primitive for which the intersection calculation circuit 32 will perform the next direct intersection calculation. This element 711 includes the object determination circuit 216
Or 316 and starting therefrom, through an output destination determining circuit 217 or 317,
It is used only during the route to the intersection calculation circuit 32 via 01, and is omitted in other places.

The second half processing state lstate stored in the element 712 takes an integer value from 0 to 6, and is output from the overrun inspection circuit 212 to the output destination determination circuit 217 and from the L2 range inspection circuit 313 to the output destination. Decision circuit 3
It is used to transmit the processing status of the former stage to the latter stage up to 17. It is omitted in the light packet in other places.

The voxel tracking mode mode stored in the element 713 takes either the value of “new tracking” or “continuous tracking”, and the grid (hereinafter referred to as grid gid) indicated by the grid identification number gid (element 710). ) To inform the voxel tracking circuit 22 whether the voxel tracking is new tracking or continuous tracking described in the section of the operation. This element 713 is set to “new tracing” when the ray packet input to the intersection search device 1 is first sent to the voxel tracking unit 2. This element 7
Reference numeral 13 is used only during a path from when a light packet is input to the voxel tracking unit 2 through the FIFO buffers 201 to 204, passes through the multiplexer 205, and reaches the DDA initialization circuit 23, and is omitted in other places.

Each of the element 714 and the element 715 has the same structure, and stores information on the progress and result of voxel tracking (hereinafter referred to as tracking voxel information). Element 7
Reference numeral 14 stores tracking voxel information on the grid gid, and an element 715 is used to save the tracking voxel information on the global grid while performing processing on the local grid. Figure 1 shows the internal structure of each element.
It is shown in FIG.

Elements 7141, 7142,. . . , 714
Voxel indexes U (1), U stored in
(2),. . . , U (k) are the indices of the k voxels obtained by voxel tracking. This index is represented by a set of integers (ix, iy, iz). Regarding the index of a real voxel, when the number of voxel divisions in each axial direction is Gx, Gy, Gz, 0 ≦ ix ≦ Gx−1 , 0 ≦ iy ≦ Gy−1, 0
≤ iz ≤ Gz-1. For the sake of calculation, indices beyond these ranges are also used. U (1) holds the index of the tracking start voxel when the ray packet is input to the voxel tracking circuit 22. Also, U
(1) In the voxel tracking circuit 22, a three-dimensional D
The index of the position currently being tracked by the DA is held.

Elements 7144, 7145,. . . , 714
Voxel incident direction dir (1), d stored in
ir (2),. . . , Dir (k) are the voxel indices U (1), U (2),. . . , U (k). dir (i) (1 ≦ i ≦ k) indicates which of the six surfaces constituting the voxel U (i) passes through which surface the light beam enters the voxel. Its value is "D
A "," SD "," PA ", or" φ ", where" DA "means that a light beam has entered through a voxel plane perpendicular to the main drive shaft. Either of the voxel planes perpendicular to a certain main drive axis is determined by the direction vector component code S. "SD", "PA"
Means that the light beam has entered through the plane of the voxel perpendicular to the auxiliary drive axis and the passive axis, respectively. "Φ" means that voxel tracking starts from voxel U (i).

The number of effective voxels nv stored in the element 7147 is the number of voxels obtained by voxel tracking, where 0 ≦ nv ≦ k. Voxel indexes U (1), U (2),. . . , U (k) and voxel incident directions dir (1), dir (2),. . . , D
Of the ir (k), valid information is held by U
(1), U (2),. . . , U (nv) and dir
(1), dir (2),. . . , Dir (nv) only.

The voxel tracking end flag eos stored in the element 7148 is 1-bit information, and becomes 1 when no voxel whose intersection calculation request bit is 1 already exists on the optical path.

The next list element address next stored in element 7149 is the next list element 27 to be referenced.
4 address. If there is no next list element to refer to, “φ” is stored.

Next, an additional portion for the inside of the voxel tracking circuit 22 shown in FIG. 15 will be described.

The voxel tracking state state stored in the element 721 indicates the processing status in the voxel tracking circuit 22. Take one of these values.

Axis replacement information a stored in element 722
xi is information indicating the correspondence between the x, y, and z axes and the main drive axis, the auxiliary drive axis, and the passive axis.

The DDA error terms .epsilon.sd, .epsilon.pa, and .delta.pa stored in the elements 723, 724, and 725 are error terms of three two-dimensional DDAs executed in parallel. εsd is a two-dimensional DDA error term between the main drive axis and the sub drive axis, εpa is a two-dimensional DDA error term between the main drive axis and the passive axis, and δpa is a two-dimensional DDA error between the sub drive axis and the passive axis. Term. FIG. 16 is a diagram illustrating the geometric meaning of the three error terms. The point 691 in the figure is the intersection of the ray with the voxel plane perpendicular to the main drive axis. Point 692 is the intersection of the ray with the voxel plane perpendicular to the nearest sub-drive axis that is on the optical path before point 691. εsd ′ represents the distance from the point 691 to the voxel plane perpendicular to the next sub-drive axis on the optical path. εpa ′ is the distance from point 691 to the next voxel plane perpendicular to the passive axis. δpa ′ is the distance from point 692 to the next voxel plane perpendicular to the passive axis. There is a relationship between these values and the DDA error term described above: εsd = εsd ′ · Dda εpa = εpa ′ · Dda δpa = δpa ′ · Dsd. Here, Dda and Dsd mean the main drive axis component and the sub drive axis component of the direction vector component absolute value D, respectively.

The third DDA integer part jpa stored in the element 726 is the passive axis component of the index of the voxel including the point 692 in FIG. This corresponds to a third DDA, ie, an integer part in a two-dimensional DDA between the auxiliary drive shaft and the passive shaft.

In the present embodiment, all the elements of the light packet are in integer or fixed-point representation, and no floating-point representation is used. As a result of repeated simulation experiments, it was confirmed that the required precision (the number of bits) in the fixed-point representation was practically acceptable in performing image generation as follows. The components of the coordinate values, that is, the components of the ray start point coordinates E and the intersection coordinates P, use an 18-bit fixed-point representation. Direction vector component absolute value D
Use a 12-bit fixed-point representation for each component of. DD
A 12-bit fixed-point representation is used for the A error terms εsd, εpa, and δpa. A 24-bit fixed-point representation is used for the intersection parameter value tp.

The details of the operation of each unit in the intersection search device 1 will be described below with reference to flowcharts. In these flowcharts, the content of processing executed by each circuit is expressed by a sequence of processes that are executed sequentially, and in an actual circuit, some processes in the flowchart are executed in parallel. You. In the following description, the direction vector component absolute value D and the error term ε
sd and the like indicate the corresponding element in the ray packet. The voxel index U (1), the voxel incident direction dir (1), and the like indicate elements in the tracking voxel information 714 unless otherwise specified.

FIG. 17 is an equivalent flowchart showing the processing executed by the DDA initialization circuit 23. Process 23
01 sets three coordinate axes as a main drive axis, a sub drive axis, and a passive axis in descending order of the component based on the value of the direction vector component absolute value D, and sets a direction vector component absolute value D, a direction vector component code S, And voxel index U (1), U
(2),. . . , U (k), the order of the vector components is rearranged in the order of the main drive axis, the auxiliary drive axis, and the passive axis. Also, the rearrangement information is recorded in the axis replacement information axis so that the original order can be restored later.

In the process 2302, a normalization start point coordinate F temporarily necessary for obtaining the DDA error term is obtained. The coordinates F of the normalization start point are obtained by transforming the coordinates E of the light ray start point into a normal lattice coordinate system. What is a normal grid coordinate system?
The vertex having the minimum coordinate value among the vertices of the rectangular parallelepiped corresponding to the existence range of the grid gid is set as the origin, and the voxel has a size of 1
Is a rectangular coordinate system that is a cube of.

The process 2303 clears the external start point flag, which becomes 1 when a ray having a start point outside the grid is newly traced, to 0. The process 2304 is a process of examining the voxel tracking mode mode, and in the case of “continuous tracking”, obtaining the index of the tracking start voxel.
Steps 5 to 2309 are skipped. This is because in the case of continuous tracking, the index of the tracking start voxel is given from the beginning as an input. Process 230
5 calculates the tracking start voxel index U (1) based on the value of the normalized start point coordinate F by the following equation. Uda (1) = floor (Fda) Usd (1) = floor (Fsd) Upa (1) = floor (Fpa) Here, floor (x) means the largest integer not exceeding x. The process 2306 is based on the Uda determined in the process 2305.
Check the range of (1). When Gda is the number of voxel divisions in the main drive axis direction of the grid gid, if Uda (1) is equal to or larger than 0 and smaller than Gda, the control is transferred to step 2310. When Uda (1) is less than 0, if the direction vector main drive axis component code Sda is "positive", the process jumps to step 2308; Control is transferred to 2307. Also,
If Uda (1) is equal to or greater than Gda, if the direction vector main drive axis component code Sda is "positive", it is clear that no light beam intersects with the grid gid as in the above, so the process jumps to step 2307 and returns to "negative". If "", control is transferred to step 2308. The process 2307 sets the voxel tracking state to “
"Tracking skip" is set, and control is transferred to step 2320.
The process 2308 performs range correction on Uda (1) as follows. When new Uda (1) = 0, old Uda (1) <0, = Gda−1, old Uda (1) ≧ Gda. Processing 2309 sets the external start point flag to 1.

In step 2310, the error terms εsd, εp
a, a provisional tracking start voxel auxiliary drive axis index Wsd, a provisional tracking start voxel passive axis index W
Find pa. First, εsd and Wsd are obtained as follows. Tsd = (Uda (1) −Fda) · Vsd + Fsd · Dda, when Sda = “positive” == (Fda−Uda (1) −1) · Vsd + Fsd · Dda, when Sda = “negative” Wsd = floor (Tsd / Dda) εsd = Dda− (Tsd mod Dda), when Ssd = “positive”, = Tsd mod Dda) When Ssd = “negative” In the above equation, mod is an operator for performing a remainder operation. .
Vsd is equal to Dsd when Ssd is "positive",
When d is "negative", the value is equal to -Dsd. Tsd is a temporary variable, and its value is determined when the main drive axis coordinate in the normal grid coordinate system is Uda (1) (when Sda = “positive”) or Uda (1) +1 (when Sda = “negative”). )
Is equal to the value obtained by multiplying the sub-drive axis coordinate of the light beam by Dda. Tsd may be negative. εpa and W
pa is also obtained in the same manner as above. Tpa = (Uda (1) −Fda) · Vpa + Fpa · Dda, when Sda = “positive” == (Fda−Uda (1) −1) · Vpa + Fpa · Dda, when Sda = “negative” Wpa = floor (Tpa / Dda) εpa = Dda− (Tpa mod Dda), when Spa is “positive”, when = Tpa mod Dda, Spa = “negative”

The process 2311 classifies cases according to whether the external start point flag is "1". If this is 1, that is, if a ray having a start point outside the grid is to be newly traced, step 2312 is executed next, and if it is 0, step 2313 is executed next. In step 2312, the provisional tracking start voxel sub-driving axis index Wsd and the provisional tracking start voxel passive axis index Wpa obtained in step 2310 are finalized, respectively, and the tracking start voxel sub-driving axis index Usd (1) is used. Substitute into the tracking start voxel passive axis index Upa (1). In step 2313, the error terms εsd and εpa obtained in step 2310 are corrected as necessary. Step 2313
Is executed, the index U (1) of the final tracking start voxel has already been determined and cannot be changed. However, if these and the provisional index obtained in step 2310 are different, If no measures are taken, incorrect voxel tracking will be performed. Therefore, the error term is corrected as in the following equation. When new εsd = old εsd, Usd (1) = Wsd, old εsd + Dda, Usd (1) ≠ Wsd When new εpa = old εpa, Upa (1) = Wpa, old εpa + Dda, Upa (1) ≠ Wpa In this correction, the difference between Wsd and Usd (1), that is, the value of (Wsd−Usd (1)) is 0 or −1 when the direction vector sub-drive axis component code Ssd is “positive”. Either of these uses the property of being either 0 or 1 when Ssd is "negative" (the same applies to the passive axis).

Processing 2314 includes another error term δpa
Is calculated. It is calculated as follows: Ssd
= “Positive” and Spa = “positive”, δpa = (Upa (1) + 1−Fpa) · Dsd + (F
sd−Usd (1)) · Dpa, and when Ssd = “negative” and Spa = “positive”, δpa = (Upa (1) + 1−Fpa) · Dsd + (U
sd (1) + 1−Fsd) · Dpa, and Ssd = ″
When “positive” and Spa = “negative”, δpa = (Fpa−Upa (1)) · Dsd + (Fsd
−Usd (1)) · Dpa, and when Ssd = “negative” and Spa = “negative”, δpa = (Fpa−Upa (1)) · Dsd + (Usd
(1) + 1−Fsd) · Dpa.

The process 2315 includes the third DDA integer part jpa
Is set to the value of the tracking start voxel passive axis index Upa (1). The process 2316 selects the subsequent process according to the value of the voxel tracking mode mode. In the case of the new tracking mode, the process 2317 is executed, in which the voxel incident direction dir (1) is set to “φ”, and in the process 2318, the voxel tracking state state
Is set to "new tracking not started". In the case of the continuous tracking mode, in step 2319, the voxel tracking state “sta” is set.
te is set to "continuous tracking not started". Process 2320
Initializes the number of valid voxels nv to 0.

Next, the operation of the voxel tracking unit 25 will be described. FIG. 18 is an equivalent flowchart showing the processing contents executed by the voxel tracking unit 25. Processing 2
501 determines whether the voxel tracking state state is "track skip", and if so, skips all further processing to allow light packets to pass through. The process 2502 is a state where the voxel tracking has not yet started, that is, the voxel tracking state state
Is "new tracking not started" or "continuation tracking not started", and if so, the control is transferred to step 2503. The process 2503 determines whether it is time to start voxel tracking. Here, numbers from 1 to n are given to the n voxel tracking units 25 from the left side of FIG. 2 and they are called unit numbers. When the direction vector main drive axis component code Sda is “positive”, the process 2503 is executed.
Is the tracking start voxel main drive axis index Uda
When (1) matches (unit number -1), it is determined that the timing of tracking start is determined, and the control is transferred to processing 2504. When the direction vector main drive axis component code Sda is “negative”, the control is transferred to the process 2504 when Uda (1) matches (n-unit number). If it is not the timing to start tracking, all the subsequent processes are skipped. The process 2504 determines that the voxel tracking state state is "".
It is determined that "tracking" and the number of effective voxels nv is equal to k, that is, that the voxel tracking processing has been completed.
If so, all subsequent processing is skipped.

The processing 2505 is executed by the voxel information storage unit 2
4 bank 241, x-axis voxel information group 245,
y-spindle voxel information group 246, z-spindle voxel information group 2
For one of the 47 corresponding to the main drive axis, the voxel information 242 or 243 or 244 at the position corresponding to the voxel index U (1) is read. At this time, if at least one of the auxiliary drive axis index Usd (1) and the passive axis index Upa (1) is the grid gid.
And the corresponding voxel information does not exist, the intersection calculation request bits 24401 and 2440
2,. . . , Voxel information in which all 24415 are set to 0 is read. As described above, the voxels corresponding to the range from -1 as the sub-drive axis voxel division number Gsd to the sub-drive axis voxel division number Gsd and from -1 to the passive axis direction voxel division number Gpa as the sub-drive axis index are stored in the bank 241. Since the information is stored, the absence of the corresponding voxel information as described above is because Usd (1) <−
1, Usd (1)> Gsd, Upa (1) <-1, Up
a (1)> when any of Gpa is satisfied.

The error term update processing 2506 advances the value of the error term by one step. A step 2507 determines whether the voxel tracking state state is “continuation tracking not started”, and if so, transfers control to a step 2510. At this point, the state is “continuous tracking not started” when the continuous tracking has just started. As described in the above operation section, in the case of new tracking, voxels beyond the tracking start voxel are targeted, but in the case of continuous tracking, voxels located far from the tracking start voxel must be targeted. Therefore, the process 2507 skips the processes 2508 and 2509 so that the tracking start voxel itself is not set as a voxel tracking target when the continuous tracking has just started.

The process 2508 is a voxel visit process 254.
And if the corresponding intersection calculation request bit is 1, it is determined that a voxel has been found, and U (1) is recorded at an appropriate position in the ray packet, and the number of effective voxels nv is increased by one. Processing 2509 determines whether the number of valid voxels nv has reached k. If k voxels have already been obtained, the value of U (1) (the index of the last voxel found) is determined by subsequent processing. ) Is transferred to the process 2522 so as not to be changed.

The process 2510 determines the sign of the new error terms εsd and εpa updated in the process 2506, and classifies them into four cases. When both εsd and εpa are positive or 0, the voxel 621 in FIG.
In this case, a light beam penetrates from the voxel 625 to the voxel 625.
Transfer control to In the case where only εsd is negative, the light beam passes from the voxel 621 to the voxel 627 in FIG. 6 and passes through the voxel 623 on the way.
After the step is advanced in the sub-drive axis direction by the SD-direction step processing 2511, the control is transferred to the processing 2520. When only εpa is negative, voxel 621 to voxel 6 in FIG.
In this case, since the light beam passes through the voxel 622 on the way, the control is shifted to the passive axis direction by the PA direction step processing 2512 and then the control is transferred to the step 2520. When both εsd and εpa are negative, it means that the light beam passes from voxel 621 to voxel 628 in FIG. 6, and in this case, voxel 6
21, 622, 624, 628 and voxels 621, 62
Two different paths are considered, 3,624,628. Therefore, the control is shifted to the process 2513, and a determination is made as to which route the route is. This route determination is
Third DDA integer part jpa and passive axis index Upa
This is performed depending on whether (1) matches. If they match, it is known that the routes are voxels 621, 623, 624, and 628.
Then, the control is shifted in the sub-drive axis direction, and then in the PA direction step processing 2516, the step is advanced in the passive axis direction. The process 2515 located between the processes 2515
Similarly to 09, when the number of valid voxels nv reaches k, the control is transferred to the process 2522 to terminate the tracking. If jpa and Upa (1) do not match, it is known that the path is a voxel route of voxels 621, 622, 624, and 628.
Therefore, the step is first stepped in the passive axis direction, and then in the SD step step 2519, the step is stepped in the sub-drive axis direction.
Process 2518 and process 2520 are both process 2515
Works the same as. The DA stepping process 2521 is a process for stepping in the main drive axis direction. The process 2522 sets the voxel tracking state state to “tracking”.

FIG. 19 is an equivalent flowchart showing details of the error term update processing 2506. The process 2531 subtracts the direction vector sub-drive axis component absolute value Dsd from the error term εsd. The process 2532 subtracts the direction vector passive axis component absolute value Dpa from the error term εpa. The process 2533 checks whether the error term εsd is negative. If the error is negative, that is, if it advances in the sub-drive axis direction, the control is transferred to the process 2534, and if it is non-negative, the error term updating process 2506 is ended. The processing 2534 to the processing 2539 includes the error term δpa related to the two-dimensional DDA between the sub-drive axis and the passive
This is a process of updating the DA integer part jpa. Process 2534 includes
From the error term δpa, the direction vector passive axis component absolute value Dpa
Subtract Step 2535 is based on the result of step 2534,
Check whether pa has become negative. If negative, process 2
The control is shifted to 536, and if non-negative, the error term update processing 250
6 is ended. The process 2536 adds the direction vector sub-drive axis component absolute value Dsd to the error term δpa. Processing 2
537 selects the subsequent processing according to the direction vector passive axis component code Spa, and if it is "positive", processing 2538 is selected.
, The third DDA integer part jpa is increased by one, and “negative”
In the case of, jpa is reduced by 1 in step 2539.

FIG. 20 is an equivalent flowchart showing details of the voxel visit processing 254. Voxel visit processing 2
Reference numeral 54 denotes not only the process 2508 in FIG. 18 but also the SD-direction stepping processes 2511, 2514, and 251.
9. It is also called from among the PA direction stepping processes 2512, 2516, and 2517. Step 2541 is equivalent to step 2505
The voxel information read out in (1) is the intersection calculation request bits 24401 and 2440 shown in FIG.
2,. . . , 24415), one corresponding intersection calculation request bit is extracted based on the value of the voxel incident direction dir (1) and the direction vector component code S. Hereinafter, the main drive axis is the z axis, and the sub drive axis is the x axis.
Taking the case of an axis as an example, a rule for selecting intersection calculation request bits will be described. The selection rule depends on where it was called from, and voxel visit operation 254 is performed by operation 250 in FIG.
8 or SD stepping process 2
511, 2514, PA direction step processing 2512, 251
7, one of c bits 24401 to -y bit 24407 is selected as follows. When dir (1) = “φ”, c bit 24401, when dir (1) = “DA”, Sda = “positive”, + z
Bit 24402, when dir (1) = “DA” and Sda = “negative”, −z
Bit 24403, when dir (1) = “SD” and Ssd = “positive”, + x
Bit 24404, when dir (1) = “SD” and Ssd = “negative”, −x
Bit 24405, when dir (1) = “PA” and Spa = “positive”, + y
Bit 24406, when dir (1) = “PA”, Spa = “negative”, −y
Bit 24407 Voxel visit processing 254 is performed in step SD direction processing 251
9. When called from among the PA direction step processing 2516, + x + y bit 24408 to -yx bit 24
Any one of 415 is selected as follows. dir (1) = “PA”, Ssd = “positive”, Spa = ”
When “positive”, + x + y bits 24408, dir (1) = “PA”, Ssd = “positive”, Spa = ”
When negative, + xy bits 24409, dir (1) = "PA", Ssd = "negative", Spa = "
When "positive", -x + y bit 24410, dir (1) = "PA", Ssd = "negative", Spa = "
When negative, -xy bit 24411, dir (1) = "SD", Ssd = "positive", Spa = "
When “positive”, + y + x bits 24412, dir (1) = “SD”, Ssd = “negative”, Spa = ”
When “positive”, + y−x bits 24413, dir (1) = “SD”, Ssd = “positive”, Spa = ″
When "negative", -y + x bit 24414, dir (1) = "SD", Ssd = "negative", Spa = "
-Yx bit 24415 when negative "

The process 2542 checks the intersection calculation request bit extracted above, and transfers control to the process 2543 if it is 1, and terminates the voxel visit process 254 if it is 0. Process 2543 determines whether the number of valid voxels nv is equal to (k-1), that is, whether it is the last found voxel, and if so, skips processes 2544 and 2545. Process 2544 is performed for voxel index U (1)
To save the value of U (k-1) to U (k) in order to store the value of
Is transferred to U (k-1), and similarly in the following, U (1)
Is repeated until the value of U is transferred to U (2). This operation does not change the value of U (1). The process 2545 applies the same operation as the process 2544 to the voxel incident direction dir. That is, the value of dir (k-1) is set to dir
(K), the value of dir (k-2) is transferred to dir (k-1), and so on.
Repeat until transfer to (2). The process 2546 increases the number of valid voxels nv by one.

FIG. 21 shows the stepping processes 2511 and 25 in the SD direction.
It is an equivalent flowchart which showed the details of 14 and 2519. The process 2551 adds the direction vector main drive shaft component absolute value Dda to the error term εsd. The process 2552 selects the subsequent process according to the direction vector sub-drive axis component code Ssd. If it is “positive”, the sub-drive axis index Usd (1) is incremented by 1 in process 2553, and “negative”.
In the case of, Usd (1) is reduced by 1 in step 2554. The process 2555 includes a voxel incident direction dir (1)
Is set to “SD” representing the sub drive shaft. Process 2556
Invokes the voxel visit process 254.

FIG. 22 shows the PA-direction step processing 2512, 25.
It is an equivalent flowchart which showed the details of 16 and 2517. This process is the same as the stepwise process in the SD direction in FIG.

FIG. 23 is an equivalent flowchart showing details of the DA-direction stepping processing 2521. In the process 2563, the subsequent process is selected by the direction vector main drive axis component code Sda.
If "negative", Uda (1) is decremented by 1 in step 2565. Step 2566 is executed in the voxel incident direction dir.
(1) is set to “DA” representing the main drive shaft.

Next, the operations from the tracking end processing circuit 211 to the output destination determining circuit 217 will be described in order. FIG. 24 is an equivalent flowchart showing the processing content executed by the tracking end processing circuit 211. The process 21101 includes the number nv of voxels found by the voxel tracking unit group 25.
To select the subsequent processing. If nv is less than k, it is known that there are no more voxels for which the intersection calculation request bit is 1, so the processing 21102 sets the voxel tracking end flag eos to 1. nv
Is k, eos is set to 0 by the processing 21103 because such a voxel may still exist earlier. The process 21104 includes a voxel index U
(1),. . . , U (nv) in reverse order. The processing 21105 is performed in the voxel incident direction d.
ir (1),. . . , Dir (nv), in the same way as the voxel index. Process 211
06 is the direction vector component absolute value D, the direction vector component sign S, and the voxel indices U (1), U
(2),. . . , U (k), by referring to the axis replacement information axis, the order of the vector components rearranged in the order of the main drive axis, the auxiliary drive axis, and the passive axis by the process 2301 is represented by the x axis, the y axis, and the z axis. Revert to axis order. Processing 21
107 initializes the latter half processing state lstate to 0.

FIG. 25 is an equivalent flowchart showing the processing executed by the overrun inspection circuit 212. Process 21201 determines whether no voxel was found by voxel tracking circuit 22, ie, whether the ray passed through grid gid, and if so, process 2 to terminate the process for grid gid.
Control is transferred to 1204. Process 21202 determines whether the voxel obtained by voxel tracking circuit 22 has advanced beyond the voxel containing the intersection found so far, and if so, process 21204 to abort the process for grid gid. Transfer control to Specifically, when the intersection existence flag isf is 1 and the voxel U (1) is located farther on the optical path than the voxel including the intersection (the coordinates of which are the intersection coordinates P), the process 2
Control is transferred to 1204. A process 21203 changes the latter half processing state lstate to 1 in order to skip the next processing by the L1 range inspection circuit 214. Process 212
04 checks whether the tracking level is 1, that is, tracking for the global grid. If leave
If 1 is 1, the second half processing state lstate is changed to 6 by the processing 21205 to end the entire intersection search processing.
Change to If level is not 1, pop
Processing 21206 switches from tracking for the local grid to tracking for the global grid.

FIG. 26 shows the L1 range inspection circuits 214 and 314.
6 is an equivalent flowchart showing the contents of the processing executed in step S1. In the process 21401, when the latter half process status lstate is other than 0, the process after FIG. 26 is skipped.
The process 21402 determines that the next list element address next is "
Check if it is φ ”, and if not“ φ ”,
Since the intersection calculation needs to be performed for the next object in the same voxel before performing the range inspection, the control is shifted to step 21403, where the second half processing state lstate is set to 2. Process 21404 determines whether the intersection found so far is inside or before voxel U (1), and if so, knows that the process for the global grid can be aborted. Transfer control to Specifically, the intersection existence flag isf is 1, and the index of the voxel including the intersection (the coordinates of which are the intersection coordinates P) is equal to U (1) or U (1).
(1) If it is determined on the optical path to be nearer, processing 21
Control is transferred to 405. The process 21405 changes the latter half processing state lstate to 6 in order to end the entire intersection search processing. If it is determined in the process 21404 that the process related to the global grid cannot be terminated yet, the process proceeds to a process 21406 to check the next voxel. The process 21406 decrements the effective voxel number nv by one. Process 21407 checks whether the resulting nv is zero, and if it is zero, ie, if the next voxel index is no longer present in the ray packet,
Control is transferred to processing 21408. A process 21408 checks the value of the voxel tracking end flag eos. If the value is 1, the process 21408 terminates the process related to the global grid.
Control is transferred to 405, and if it is 0, control is transferred to step 21409. The process 21409 is for the purpose of further tracking the voxel ahead on the optical path, so that the latter half process state lstat
Change e to 5. In processing 21407, nv is 0
If not, the process moves to processing 21410. Process 214
10 moves U (2) to shift the processing target to the next voxel
Is moved to U (1), the value of U (3) is moved to U (2), and so on until the value of U (nv + 1) is moved to U (nv). Processing 21411 is performed in the voxel incident direction d.
The same operation as the process 21410 is applied to ir.
That is, the value of dir (2) is changed to dir (1),
The value of (3) is moved to dir (2), and d
This is repeated until the value of ir (nv + 1) is transferred to dir (nv). Process 21412 applies the same overrun check as process 21202 to the new value of U (1). That is, if the intersection existence flag isf is 1 and the voxel U (1) is located on the optical path farther from the voxel including the intersection (the coordinates of which are the intersection coordinates P), the control is transferred to the process 21405, and The entire search process ends. The processing 21413 is the second half processing state lstate
To 1.

FIG. 27 shows a list head element determination circuit 215,
315 is an equivalent flowchart showing the processing content executed in 315. The processing 21501 is performed in the latter half processing state lstat.
When e is other than 1, the subsequent two processes are skipped. The processing 21502 is performed in the list header array 27 relating to the grid gid stored in the list header memory 271.
The array element is read from U.3 using U (1) as an index, and the value is stored in the next list element address next. The processing 21503 is performed in the latter half processing state lstat.
Change e to 2. The process 21504 initializes the intersection calculation request primitive identification number rpid with “φ” indicating that there is no primitive requesting the intersection calculation.

FIG. 28 is an equivalent flowchart showing the contents of processing executed by the object determination circuits 216 and 316. In the process 21601, when the latter half processing state lstate is other than 2 to 4, the subsequent processing is skipped. The process 21602 determines that the next list element address next is “φ”.
, That is, when the list element no longer exists,
Subsequent processing is skipped. The process 21603 reads the list element 274 shown in FIG. 9 from the corresponding list element memory 272 using the address value of the next list element address next. The process 21604 is executed for the list element 2
Based on the overlap elimination information 277 in 74, the overlap of the intersection calculation is determined. Specifically, the voxel incident direction dir (1)
And the direction vector component code S, the overlap removal information 2
77, the six duplicate elimination bits 2771 to 277
The corresponding one of the six is taken out, and if it is 1, control is transferred to a process 21610 as an overlapped intersection calculation, and if it is 0, the process proceeds to the next process 21605.
A process 21605 checks whether or not the second half processing state lstate is 2, and if not, that is, lstate
Is 3 or 4, the subsequent processing is skipped. This is because when lstate is 3 or 4, the object to be cross-calculated has already been determined, so that only the check for duplicate elimination is performed and the object to be cross-calculated is not updated. The process 21606 stores the primitive or lattice identification number 276 in the list element 274 read out in the process 21603 in the intersection calculation request primitive identification number rpid. At this time, when the type of the object is a local lattice, the identification number of the lattice is temporarily stored instead of the primitive in the rpid. However, in the next output destination determination circuits 217 and 317, such mismatches are stored. Is quickly resolved. A process 21607 examines the type 275 in the list element 274 read in the process 21603, and selects a subsequent process. If the type is not the local lattice, that is, if the type is a primitive, the second half processing state lstate is set to 3 by processing 21608. If the type is the local lattice, the second half processing state lstate is set to 4 by processing 21609. Process 2161
0 sets the value of the next address 278 in the list element 274 read in the process 21603 to the next list element address ne.
xt.

FIG. 29 is an equivalent flowchart showing the contents of processing executed by the output destination determining circuits 217 and 317.
Processing 21701 selects the subsequent processing according to the value of the second half processing state lstate. lstate is 2 or 3
If so, control is transferred to step 21702. If lstate is 2 at this time, the object determination circuit 2
16 and 316, the object determination has failed, and it is necessary to continue the processing in the object determination circuit again. When lstate is 3, the next object to be cross-calculated is a primitive. In either case, the ray packet is forwarded to one of the two intersection calculators 3. A process 21702 determines which intersection calculation unit 3 to forward the ray packet to. This determination method differs between the output destination determination circuit 217 and the output destination determination circuit 317. In the case of the output destination determination circuit 217,
The determination is made in consideration of the load distribution so as to increase the operation rate of the intersection calculation unit 3. There are various decision methods, FIFO
Processing based on the number of ray packets stored in the buffer 301, based on the grid identification number gid, based on the voxel index U (1), based on the intersection calculation request primitive identification number rpid, and so on One based on the number of ray packets, one based on a random number, a combination thereof, and the like can be considered. In the case of the output destination determination circuit 317, its own intersection calculation unit 3 is always selected due to the structure of the device. A process 21703 transfers the ray packet to the intersection calculation unit 3 determined above.

In process 21701, lstate is 4
Is the case where the next object to be intersected is a local grid, and in this case, push processing 21704 and thereafter are executed. The push process 21704 switches from tracking for the global grid to tracking for the local grid. The process 21705 sets the grid identification number gid to the value of the intersection calculation request primitive identification number rpid. Processing 2
Step 1706 sets the voxel tracking mode mode to "new tracking". A process 21707 transfers the ray packet to the voxel tracking unit 2.

In process 21701, lstate is 5
If so, process 217 to perform continuous voxel tracking
08, where the voxel tracking mode mod
e is set to "continue tracking". Then, processing 2170
At 7, the ray packet is forwarded to the voxel tracking unit 2.

In process 21701, lstate is 6
Is when the intersection search process has been completed,
Control transfers to operation 21709, where the light packet is forwarded to the output buffer 4.

FIG. 30 is an equivalent flowchart showing details of the push process 21704. Process 21901
Is to save the tracking voxel information 714 to save the tracking voxel information for the global grid during tracking for the local grid.
Are copied to the evacuation tracking voxel information 715. The process 21902 changes the tracking level level to 2.

FIG. 31 shows pop processing 21206, 313
It is an equivalent flowchart which showed the detail of 06. Processing 2
1903 copies all elements of the evacuation tracking voxel information 715 to the tracking voxel information 714 in order to restore the tracking voxel information related to the global grid. Process 21904
Sets the identification number of the global grid to the grid identification number gid. The process 21905 changes the tracking level level to 1.

FIG. 32 is an equivalent flowchart showing the processing executed by the intersection calculation circuit 32. Process 320
1 indicates that the intersection calculation request primitive identification number rpid is "
If it is φ ″, that is, if the object determination circuits 216 and 316 cannot determine the object to be cross-calculated in the previous processing, the subsequent processing is skipped.

Processing 3202 is a step of actually performing a direct intersection calculation. If the object is a triangle, it is calculated as follows. Let the vertex coordinates of the triangle be P0, P1, and P2. The surface normal vector N and the values a, L and M corresponding to twice the area expressed by the following equation are obtained in advance by the pre-calculation and stored in the object shape information storage means 33. deep. A = P1−P0 B = P2−P0 a = | A × B | N = (A × B) / a L = B × N M = N × A In the process 3202, first, the intersection of the plane including the triangle and the light ray , The parameter value tq of the ray and the coordinates Q of the intersection point
Is determined by the following equation. tq = (N · (P0−E)) / (N · V) Q = E + V · tq where V is a signed direction vector of the light ray, and can be obtained from the direction vector component absolute value D and the direction vector component code S. it can. Next, values u and v for determining whether the intersection is inside the triangle are determined by the following equations. u = L * (Q-P0) v = M * (Q-P0) The process 3203 determines whether or not there is an intersection of a ray and a triangle. Specifically, the following five conditions are checked. tq> 0, u> 0, u <a, v> 0, u + v <a An intersection exists only when all of these five conditions are satisfied, and control is transferred to step 3204. If there is no intersection, the intersection calculation processing ends. Note that in an actual circuit, the multiplication and division included in the above calculation is realized using a pipeline multiplier and a pipeline divider so that one light packet can be processed for each effective clock period. .

The process 3204 checks the value of the intersection existence flag isf, and if 0, that is, if no intersection has yet been found, the process 3205 is skipped. Processing 3
205 compares the parameter value tq obtained in the process 3202 with the parameter value tp of the intersection found so far to leave only the intersection closest to the ray start point,
If tq is not smaller, the subsequent processing is skipped. Process 3206 includes an intersection primitive identification number pid
Is set to the value of the intersection calculation request primitive identification number rpid. The process 3207 replaces the intersection coordinates P with the coordinates Q of the newly found intersection. The process 3208 replaces the intersection parameter value tp with the newly found intersection parameter value tq. The process 3209 sets the intersection existence flag isf to 1.

FIG. 33 is an equivalent flowchart showing the processing executed by the L2 range inspection circuit 313. Processing 3
1301 initializes the latter half processing state lstate to 0. In the process 31302, when the tracking level is 1, the subsequent processes are skipped. Process 3130
3 checks whether the next list element address next is "φ", and if it is not "φ", the intersection calculation needs to be performed for the next object in the same voxel before performing the range check. Control is transferred to step 31304, where the second half processing state lstate is set to 2. Processing 3
1305 determines whether or not the intersection found so far is located inside or before voxel U (1), and if so, knows that the processing on the grid gid can be terminated. Transfer control to Specifically, the intersection existence flag isf is 1,
If the index of a voxel including the intersection (the coordinates of which are the intersection coordinates P) is equal to U (1) or is determined to be closer to the optical path than U (1), pop processing 313 is performed.
Transfer control to 06. The pop process 31306 switches from tracking for the local grid to tracking for the global grid.
If it is determined in the process 31305 that the process related to the lattice gid cannot be terminated yet, the process proceeds to a process 31307 to check the next voxel. Processing 3
Step 1307 decrements the number of effective voxels nv by one. Processing 31
Step 308 checks whether the subtraction nv has become 0. If it is 0, that is, if the next voxel index is no longer present in the ray packet, process 313 is performed.
Control is transferred to 09. The process 31309 checks the value of the voxel tracking end flag eos. If the value is 1, the pop process 31 is performed to end the process related to the grid gid.
The control is shifted to 306, and if it is 0, the control is shifted to step 31310. The process 31310 is for the purpose of further tracking the voxel located further on the optical path.
Change e to 5. In process 31308, nv is 0
If not, the flow shifts to step 31311. Process 313
11 moves U (2) to shift the processing target to the next voxel
Is moved to U (1), the value of U (3) is moved to U (2), and so on until the value of U (nv + 1) is moved to U (nv). The process 31312 includes a voxel incident direction d.
The same operation as the process 31311 is applied to ir.
That is, the value of dir (2) is changed to dir (1),
The value of (3) is moved to dir (2), and d
This is repeated until the value of ir (nv + 1) is transferred to dir (nv). The process 31313 applies the same overrun check as the process 21202 to the new value of U (1). That is, when the intersection existence flag isf is 1 and the voxel U (1) is located on the optical path farther from the voxel including the intersection (the coordinates of which are the intersection coordinates P), p
Control is transferred to op processing 31306. The process 31314 includes:
The second half processing state lstate is changed to 1.

The structure of the voxel tracking unit 25 will be described in detail below. FIG. 34 shows an example of the internal structure of the voxel tracking unit 25 when k is 2. This circuit is constituted by a two-stage pipeline and processes one light packet for each effective clock cycle. In other words, by this circuit, 1
The series of processes shown in FIGS. 18 to 23 for one light packet is completed during two valid clock periods.

The address calculation circuit 2571 performs the processing 250
In step 5, the address to be supplied to the bank 241 when the voxel information 242, 243, or 244 is read is calculated. The register 2572 stores the voxel information read from the bank 241.

The control circuit 2573 stores the value of the register 2572, the voxel incidence direction dir, the direction vector component code S, the number of effective voxels nv, and the voxel tracking state stat.
e, the axis replacement information axis, the outputs of the comparators 2584, 2585, and the sign of the updated error terms εpa, εsd, δpa as inputs, the voxel incident direction dir, the number of effective voxels nv, the new value of the voxel tracking state state, And a combinational logic circuit for determining the value of the selection signal of each multiplexer.

The 1-addition / subtraction circuit 2574 executes the processing 256
4 or a circuit for executing the process 2565, and the selection of addition or subtraction is determined by the direction vector main drive axis component code Sda. The multiplexer 2575 determines a new value of the main drive axis index Uda (1), and the upper side is selected under the condition that the DA stepping process 2521 is executed. The multiplexer 2576 determines a new value of the main drive axis index Uda (2), and the upper side is selected under the condition that the process 2544 is executed.

The 1 addition / subtraction circuit 2577 performs processing 255
9 or a circuit for executing the process 2560. The selection between addition and subtraction is determined by the direction vector passive axis component code Spa. Multiplexer 2578 determines the new value of passive axis index Upa (1),
The upper side is selected under the condition that the A-direction step processing 2512, 2516, 2517 is executed. Multiplexer 2
A step 579 determines a new value of the passive axis index Upa (2).
The top input is selected when condition 2544 is executed in the voxel visit process 254 called by either 516, 2517 or the SD step process 2519, and is called from elsewhere. The center input is selected when the condition that the process 2544 is executed in the voxel visit process 254 is selected, otherwise the bottom input is selected.

The 1-addition / subtraction circuit 2580, the multiplexer 2581, and the multiplexer 2852 operate in the same manner as the 1-addition / subtraction circuit 2577, the multiplexer 2578, and the multiplexer 2579 with respect to the sub-drive axis index.

The one-addition / subtraction circuit 2583 performs processing 253.
8 or a circuit for executing processing 2539, and the selection of addition or subtraction is determined by the direction vector passive axis component code Spa. Both the comparator 2584 and the comparator 2585 implement the processing 2513. The reason why two comparators are necessary is that the comparison process is desired to be completed in the first stage of the pipeline for speeding up, but the value of the third DDA integer part jpa has not yet been determined at the time of the first stage. This is because all possible values of jpa are compared in the future. The multiplexer 2586 determines a new value of jpa, and the upper side is selected under the condition that the process 2538 or the process 2539 is executed.

Subtractor 2587, multiplexer 258
8. The adder 2589 and the multiplexer 2590 are in charge of processing relating to the error term εpa of the passive axis. Subtractor 258
7 executes a process 2532. Multiplexer 258
8, the lower side is selected when the determination condition of the processing 2501 to the processing 2504 is such that the error term updating processing 2506 is executed. Adder 2589 performs processing 2557. The multiplexer 2590 determines a new value of εpa, and performs a PA direction step process 2512,
The lower side is selected when the conditions 2516 and 2517 are executed.

A subtractor 2591 and a multiplexer 259
2. The adder 2593 and the multiplexer 2594 output the error term εsd of the sub-drive axis to the subtractor 2587, the multiplexer 2588, the adder 2589, and the multiplexer 2
Works the same as 590.

Subtractor 2595, multiplexer 259
6. The adder 2597 and the multiplexer 2598 are in charge of processing relating to the error term δpa. The subtractor 2595 is
The process 2534 is executed. Multiplexer 2596 is
The lower side is selected under the condition that the process 2534 is executed. The adder 2597 performs a process 2536. Multiplexer 2598 determines the new value of error term δpa, and the upper side is selected under the condition that operation 2536 is performed.

In the voxel tracking unit 25, in addition to the circuit shown in FIG. 34, elements (ray identification number rayid, ray start point coordinates E, intersection existence flag isf) of the ray packet which do not appear in FIG. , Intersection coordinate P, intersection parameter value tp, intersection primitive identification number pid, tracking level, evacuation tracking voxel information 715)
There is a register group which outputs the data after two clock delays.

As described above, a preferred embodiment of the present invention has been described, but the present invention is not limited to this embodiment. Hereinafter, some typical examples of other possible embodiments will be described.

In the previous embodiment, the intersection search device 1 shown in FIG. 1 has one voxel tracking unit and two intersection calculation units, but both units have one or more arbitrary numbers. It is possible to change the system scale according to the required performance.

In the previous embodiment, the intersection search device 1 finds one of the intersections closest to the ray start point.
An embodiment in which other intersection information is obtained is also conceivable.
For example, the present invention is also applied to an intersection search device that obtains the closest intersection to the start point by giving the second closest intersection to the start point or the distance from the start point, and finding intersections located farther away than that. it can. In addition, information related to the intersection (for example,
An embodiment in which the normal vector at the intersection or the barycentric coordinates based on the vertex when the object is a polygon is output together with the position of the intersection is also conceivable.

In the previous embodiment, the hierarchical level of the hierarchical grid is set to 2
Although the embodiment is limited to the levels, there can be considered an embodiment capable of handling three or more levels. However, the number of evacuation tracking voxel information 715 in the ray packet is increased to (maximum hierarchy level -1), and the range inspection circuit group 2
14, 313 and 314 also need to be increased in number as the maximum hierarchy level increases.

In the previous embodiment, the number of voxel divisions in each lattice is limited to n or less for each axis. However, an embodiment in which the number of voxel tracking units remains n and this restriction is not imposed is also conceivable. In this case, when the number of divisions of a certain grating in the main drive axis direction is p (p> n), m = floor (p
/ N) +1, and voxel tracking is performed m times. That is, the first voxel tracking tracks the first n voxel plane sets, the second voxel tracking tracks the next n voxel plane sets, and so on. If the number of voxels found exceeds k on the way, the number of voxel tracking may increase more than m times, and conversely, it is determined that tracking can be terminated by range inspection or overrun inspection on the way. In this case, the voxel tracking may be completed in less than m times. In order to remove the restriction on the number of voxel divisions, the voxel tracking unit 2 in the previous embodiment is used.
And the latter half of the intersection calculation unit 3 (from the tracking end processing circuit 211 to the output destination determination circuit 217 and the L2 range inspection circuit 313
To the output destination determination circuit 317) need to be changed so that a ray packet that has been processed only halfway through the grid is re-input to the voxel tracking unit 2. Also,
It is necessary to add to the ray packet information indicating to which group the set of n voxel planes has been tracked.

In the previous embodiment, the voxel information storage means 24 stores the first set of voxel planes in two places, that is, the first bank and the (ni + 1) th bank, and stores the direction vector main drive axis component. Either one was read by the sign of. However, if the direction vector main drive axis component is negative, if the voxel tracking circuit 22 is changed so that the n sets of voxel planes are checked in the opposite direction from the farthest from the ray start point, it is not necessary to overlap the two places. Disappears. An embodiment in which the amount of memory is reduced in this way is also conceivable.

In the previous embodiment, a cubic voxel was assumed. If the DDA initialization circuit 23 converts not only the ray starting point but also the direction vector of the ray into the normal lattice coordinate system, the rectangular or parallelepiped voxel can be obtained. This is an embodiment that can handle. In this case, in the voxel tracking circuit 22, it is necessary to store the direction vector after the coordinate conversion in the ray packet, and to add or subtract the component value of the direction vector to the DDA error term, it is necessary to use the one after the coordinate conversion. .
In addition, by using a coordinate transformation including rotation by the same method, it is possible to handle a local grid that is rotated with respect to the global grid.

In the previous embodiment, the shape of the primitive is limited to only a triangle. However, an embodiment in which other primitive shapes such as a polygon, a quadratic surface, and a free-form surface are used can be considered.
For this purpose, the intersection calculation circuit 32 may be replaced with a circuit that performs direct intersection calculation with primitives of other shapes. An embodiment in which two or more different primitive shapes can be handled by providing each of the plurality of intersection calculation units 3 with an intersection calculation circuit for another shape can be considered. In this case, FIG. As shown, the new communication paths 81 and 82 and the FIFOs are used so that the ray packet can be transferred from one intersection calculation unit 3a to another intersection calculation unit 3b.
Buffer 302 must be added.

In the previous embodiment, the list head element determination circuit 21
5 and 315 and the object determination circuits 216 and 316 are placed after the L1 range inspection circuits 214 and 314. These are placed before the DDA initialization circuit 23 in the voxel tracking section 2 and in the intersection calculation section 32 in the intersection calculation section 3. Is also conceivable. The advantage of such a structure is that the intersection calculation request primitive identification number rpid of the ray packet and the next list element address next in the evacuation tracking voxel information 715 are not required except inside the intersection calculation unit 3. Is slightly reduced, thereby slightly reducing the amount of hardware. vice versa,
A disadvantage of such a structure is that when primitives are registered in the global lattice, useless clock periods enter the voxel tracking unit 2, which slightly degrades performance.

In the previous embodiment, the voxel information storage means 24
, 15 intersection calculation request bits 24401, 24402,. . . , 24
Of the 415, from + x + y bits 24408 to -y-x
Up to bit 24415, it is determined by the inclusion relation of the surrounding voxel object set. However, embodiments are also conceivable in which more precise deduplication of intersection calculations is realized by further taking into account the set of objects contained in the central voxel. Specifically, for example, in the case of + x + y bit 24408 in FIG.
When the set of objects included in voxel 658 of (b) is not a subset of the set of objects included in voxel 654, 1
And had On the other hand, when the union of the set of objects included in voxel 651 and the set of objects included in voxel 654 is X, + x + y bits 2440 when the set of objects included in voxel 658 is not a subset of X
An embodiment in which 8 is 1 can be considered. This makes it possible to slightly reduce the number of cases where it is not possible to eliminate duplicate intersection calculations as shown in the example of FIG. However, the procedure for obtaining voxel information is complicated.

Voxel information storage means 2 in the previous embodiment
4, the object set storage means 27, 3
7, a small-capacity and high-speed storage means such as a cache connected to a larger-capacity and lower-speed storage means for all or a part of the lattice information memories 26 and 36 and the object shape information storage means 33. Is also conceivable. Since any storage means has access locality, the effect of shortening the average access time by using a cache can be expected. The bank constituting the voxel information storage means has a special effect of reducing the required total storage capacity. If the cache is not used, the i-th voxel plane set must be stored in the first bank and the (ni + 1) -th bank so as to overlap the voxel information as described above. However, if the banks are configured by caches and one low-speed memory is shared by the i-th bank and the (n−i + 1) -th bank, it is not necessary to store two redundant low-speed memories. . Further, the fifteen intersection calculation request bits 24401, 24402,. . . , 24415, the effect of the cache can be enhanced by putting only a part that is likely to be used in the future, instead of putting the whole in the cache. For example, if the direction vector component codes are all positive, c bits 24401,
+ Z bit 24402, + x bit 24404, + y bit 24406, + x + y bit 24408, + y + x
It is conceivable to place only 6 bits of bit 24412 in the cache. These 6 bits are compared to the other 9 bits
Since there is a very high possibility that the cache will be used depending on the light beam to be processed next, the cache can be used more effectively as compared with a case where all 15 bits are provided.

In the previous embodiment, all the parts of the intersection searching apparatus were constituted by dedicated hardware. However, for the parts other than the voxel tracking unit 25, an embodiment using a microprocessor is also conceivable. In this case, 1
It is possible that one microprocessor plays the role of multiple circuits. Further, in the previous embodiment, separate memories are provided for the bank 241 constituting the voxel information storage means 24, the respective memories of the object set storage means 27 and 37, the lattice information memories 26 and 36, and the object shape information storage means 33. Although the embodiment is used, an embodiment in which the same memory is shared for all or a part of them is also conceivable.

In the previous embodiment, the intersection searching apparatus according to the present invention is used for generating an image by a method of tracing from the viewpoint, which is the most typical ray tracing method, in the direction opposite to the actual traveling direction of light. , One that tracks forward from the light source side,
It can also be used for ray tracing in which backward tracing and forward tracing are used together. Further, the present invention can also be used for a ray casting method in which ray tracing is stopped from the viewpoint to the first object. Furthermore, rays do not necessarily correspond to physical light, and for example, by considering the progress of electromagnetic waves and sound waves as rays, it is also possible to use the intersection search device according to the present invention for simulation of reflection of electromagnetic waves and sound. It is.

[0167]

According to the present invention, the intersection search processing for one light ray can be performed in a time several times as long as the memory access time. This is so fast that conventional devices based on a plurality of general-purpose processors can never reach. As a result, it has become possible according to the present invention to generate a three-dimensional shaded image as fast as a moving image using the ray tracing method. Moreover, the intersection search device of the present invention does not require a floating point arithmetic unit, and can be realized at low cost by a custom VLSI.

According to a simulation experiment performed by the inventor, the specific value of the time required for the intersection search processing for one ray described above is as follows. The processing time shown here is n = 32, k = 2 in the embodiment described with reference to FIGS. 1 to 34, the number of stages of the object determination circuit is three for both the voxel tracking unit and the intersection calculation unit, and each FIF
The capacity of the O-buffer is obtained when the voxel tracing unit has 5 packets and the intersection calculation unit has 10 packets. In addition, a method of alternately transferring each clock cycle to another intersection calculation unit was used for assignment of the intersection calculation unit. The image to be generated has a size of 512 × 512 pixels and performs anti-aliasing. As a three-dimensional object group to be subjected to image generation, three different object groups from scene 1 to scene 3 were used. Scene 1 is composed of a sphere having two diffuse reflections, a plane having one specular reflection, and a cube having one transparency. The total number of triangles resulting from the triangulation is 1,934, and the number of local grids is one. Is 2, and the total number of rays traced (including rays traced in the specular reflection direction, the transmission direction, and the light source direction) is 859 or 773. Scene 2
Consists of a teapot with one diffuse reflection,
The total number of triangles resulting from the triangulation is 4,096,
8 local grids, 398, 56 total rays traced
There are eight. Scene 3 is a model of a doll composed of a total of 17 spheres and free-form surfaces.
One is a transparent body and the rest are diffuse reflectors. The total number of triangles resulting from the triangulation of scene 3 is 10,904, the number of local grids is 19, and the total number of rays traced is 1,129,3
There are 38. The time required for the intersection search processing under the above conditions is as follows. In scene 1, the entire intersection search process takes 2,002,928 clock cycles, and 1
The average is 2.33 clock cycles per ray. In scene 2, the entire intersection search process takes 2, 494, and 674 clock cycles, with an average of 6.26 clock cycles per ray. In scene 3, 4, 91 as a whole
It takes 1,408 clock cycles and averages 4 per ray.
This is 35 clock cycles. Therefore, assuming a clock speed of 50 MHz, the average time per ray for the intersection search processing is 46.6 nanoseconds and 1 average, respectively.
25.2 nanoseconds and 87.0 nanoseconds.

The operating rate of each unit when the intersection search processing is performed on the above three scenes, that is, the ratio of the effective clock cycle to the entire clock cycle is as follows. In the case of scene 1, the voxel tracking unit was 81.0%, and the intersection calculation unit was 83.7%. In the case of scene 2, the voxel tracking unit is 57.2%, and the intersection calculation unit is both 78.9%.
%Met. For scene 3, the voxel tracking unit is 87.3
% And the intersection calculation part were both 83.4%.

Also, assuming that the initial ray generator 51 and the shading and next-generation ray generator 52 can perform processing at a sufficiently high speed so as not to make the intersection search processing apparatus according to the present invention wait, a clock speed of 50 MHz is assumed. , The images of scenes 1 to 3 are generated at a rate of 25.0 frames / sec, 20.0 frames / sec, and 10.2 frames / sec, respectively. These velocities are sufficient to be able to be viewed as a moving image.

[Brief description of the drawings]

FIG. 1 is a diagram showing an overall structure of an intersection search device and an example of application to an image generation device.

FIG. 2 is a diagram illustrating a structure of a voxel tracking unit 2.

FIG. 3 is a diagram showing a structure of an intersection calculation unit 3;

FIG. 4 is a diagram showing a structure of an output buffer 4;

FIG. 5 is a diagram illustrating a voxel space division method.

FIG. 6 is an explanatory diagram of a voxel crossing a light ray.

FIG. 7 is a diagram for explaining overlap removal in intersection calculation;

FIG. 8 is a diagram showing an example in which duplication elimination does not work;

FIG. 9 is a diagram showing the structure of data stored in object set storage means 27 and 37.

FIG. 10 is a diagram showing a structure of data stored in a voxel information storage unit 24;

FIG. 11 is a diagram showing a detailed structure of z-spindle voxel information 244.

FIG. 12 illustrates an example of an intersection search process.

FIG. 13 is a diagram showing a structure of a main part of a light ray packet;

FIG. 14 is a diagram showing a detailed structure of tracking voxel information 714 and evacuation tracking voxel information 715 which are a part of a ray packet.

FIG. 15 is a diagram showing a structure of an additional portion of a light packet.

FIG. 16 is a view for explaining the geometric meaning of a DDA error term;

FIG. 17 is an equivalent flowchart showing the operation of the DDA initialization circuit 23;

FIG. 18 is an equivalent flowchart showing the operation of the voxel tracking unit 25.

FIG. 19 is an equivalent flowchart showing the contents of an error term update process 2506.

FIG. 20 is an equivalent flowchart showing the contents of voxel visit processing 254.

FIG. 21 is an SD-direction incremental process 2511, 2514, 25;
Equivalent flowchart showing the contents of 19

[FIG. 22] PA direction step processing 2512, 2516, 25
Equivalent flowchart showing the contents of 17

FIG. 23 is an equivalent flowchart showing the contents of a DA stepping process 2521;

FIG. 24 is an equivalent flowchart showing the operation of the tracking end processing circuit 211.

FIG. 25 is an equivalent flowchart showing the operation of the overrun inspection circuit 212;

FIG. 26 is an equivalent flowchart showing the operation of the L1 range inspection circuits 214 and 314.

FIG. 27 is an equivalent flowchart showing the operation of the list head element determination circuits 215 and 315

FIG. 28 is an equivalent flowchart showing the operation of the object determination circuits 216 and 316.

FIG. 29 is an equivalent flowchart showing the operation of the output destination determination circuits 217 and 317

FIG. 30 is an equivalent flowchart showing the contents of push processing 21704.

FIG. 31 is an equivalent flowchart showing the contents of pop processing 21206 and 31306.

FIG. 32 is an equivalent flowchart showing the operation of the intersection calculation circuit 32;

FIG. 33 is an equivalent flowchart showing the operation of the L2 range inspection circuit 313

FIG. 34 is a diagram showing a detailed structure of a voxel tracking unit 25.

FIG. 35 is a diagram showing changes in the embodiment that handles two or more types of different shapes.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Intersection search device 2 Voxel tracking part 3 Intersection calculation part 4 Output buffer 22 Voxel tracking circuit 23 DDA initialization circuit 24 Voxel information storage means 25 Voxel tracking unit 26, 36 Grid information memory 27, 37 Object set storage means 32 Intersection calculation circuit 33 Object Shape Information Storage Means 51 Initial Ray Generator 52 Shading and Next Generation Ray Generator 53 Frame Buffer 54 CRT Display 81, 82 Communication Path 211 Tracking End Processing Circuit 212 Overrun Inspection Circuit 313 L2 Range Inspection Circuit 214, 314 L1 Range inspection circuit 215, 315 List head element determination circuit 216, 316 Object determination circuit 217, 317 Output destination determination circuit 241 Bank 271, 371 List header memory 272, 372 List element memory

Claims (23)

[Claims] 1. A method for receiving various parameters of light rays as input based on various information of a group of three-dimensional objects defined in a three-dimensional virtual space, and including one or more hexahedral voxels as elements in the three-dimensional virtual space. Set multiple 3D grids,
An apparatus for outputting intersection information between the ray and the group of three-dimensional objects, wherein one or a plurality of voxels storing information indicating the necessity of intersection calculation between each of the voxels and a three-dimensional object included in the voxel Information storage means;
One or more voxel tracking circuits for reading information indicating the necessity of the intersection calculation from the voxel information storage means and outputting an index of a voxel that intersects with the ray and is considered to need the intersection calculation; Wherein the voxel information storage means has a plurality of banks each readable by different addresses, and any set of voxel planes for any coordinate axis in each of the three-dimensional grids is stored in at least one of the banks. Arranged, the voxel tracking circuit has a plurality of voxel tracking units, each of the voxel tracking units being a predetermined one of a group of voxel plane sets related to coordinate axes where the absolute value of the direction vector component of the light ray is maximum. Is responsible for some voxel plane sets, and for the predetermined number of voxel planes Voxels having intersecting the in the set and the light beam, intersection search apparatus characterized by having a specific functions without omission. 2. The intersection searching apparatus according to claim 1, wherein said voxel tracking circuit sets k as a constant of 2 or more.
An intersection search apparatus, which outputs at most k voxel indices at a time. 3. The intersection searching apparatus according to claim 1, wherein a local three-dimensional grid is set for a part of a space occupied by the three-dimensional grid, and the local three-dimensional grid is set to the three-dimensional object. An intersection search apparatus characterized in that the three-dimensional lattice group is hierarchically organized by being one of the following. 4. An intersection search apparatus according to claim 1, wherein n
Is a constant of 2 or more, and has n number of the voxel tracking units and n number of banks, and i is an integer of 1 or more and n or less, the i-th voxel tracking unit is the i-th bank. , Where m is an arbitrary integer greater than or equal to 0, and the i-th bank is the (i + m.n) -th number of voxel plane sets for each axis counted from the smallest coordinate value. And an i + m · n-th set of voxel planes counted from the largest coordinate value. 5. The intersection search apparatus according to claim 4, wherein the number of voxel divisions on said three-dimensional grid is limited to n or less for each axis. 6. An intersection searching apparatus according to claim 1, wherein said voxel tracking units are interconnected in a straight line by communication means. 7. The intersection search device according to claim 1, wherein said voxel tracking circuit performs a pipeline process using a read time of one data in said bank as a pipeline pitch. apparatus. 8. The intersection searching device according to claim 1, wherein the voxel tracking circuit comprises: a main driving axis in an order of an absolute value of a component of a direction vector of the light beam with respect to three coordinate axes;
When the auxiliary drive axis and the passive axis are used, a two-dimensional DDA (digital differential analyzer) between the main drive axis and the auxiliary drive axis, a two-dimensional DDA between the main drive axis and the passive axis, and the auxiliary drive axis and the passive axis An intersection search device characterized in that a voxel having an intersection with the light ray is found by executing the two-dimensional DDA in parallel and in parallel with each other. 9. The intersection search apparatus according to claim 1, wherein the bank indicates the necessity of the intersection calculation for the voxel and the information indicating the necessity of the intersection calculation for the voxel around the voxel. Are stored in the same word. 10. The intersection search device according to claim 9,
The information indicating the necessity of the intersection calculation has information defined by an inclusion relationship between a set of three-dimensional objects included in the voxel and a set of three-dimensional objects included in voxels located in the vicinity of the set. Intersection search device. 11. Based on various information of a three-dimensional object group defined in a three-dimensional virtual space, various parameters of a light ray are received as input, and one or more elements having hexahedral voxels as elements in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, information indicating the necessity of calculating the intersection of each of the voxels with the three-dimensional object included in the voxel One or a plurality of voxel information storage means for storing the information indicating the necessity of the intersection calculation from the voxel information storage means, and voxels which intersect with the ray and are deemed to need the intersection calculation And one or more voxel tracking circuits that output an index of at least one k at a time when k is a constant of 2 or more. Intersection search unit and outputs the indices of voxels. 12. Based on various information of a group of three-dimensional objects defined in a three-dimensional virtual space, various parameters of a light ray are received as input, and one or more elements having hexahedral voxels as elements in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, information indicating the necessity of calculating the intersection of each of the voxels with the three-dimensional object included in the voxel One or a plurality of voxel information storage means for storing the information indicating the necessity of the intersection calculation from the voxel information storage means, and voxels which intersect with the ray and are deemed to need the intersection calculation And one or more voxel tracking circuits that output an index of the three-dimensional grid. And, intersection search apparatus characterized by hierarchically organize the 3-dimensional lattice by the said local three-dimensional lattice is a type of the three-dimensional object. 13. A three-dimensional virtual space, based on various information of a group of three-dimensional objects, receiving various parameters of light rays as input, and including one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, information indicating the necessity of calculating the intersection of each of the voxels with the three-dimensional object included in the voxel One or a plurality of voxel information storage means for storing the information indicating the necessity of the intersection calculation from the voxel information storage means, and voxels which intersect with the ray and are deemed to need the intersection calculation And one or more voxel tracking circuits that output an index of the light beam. The main drive shaft in descending order of the absolute value of the components of the Le, the auxiliary drive shaft, when the driven shaft, two-dimensional DDA between the main drive shaft and the secondary drive shaft (Digital Differential -
Analyzer) and two-dimensional DDA between the main drive axis and the passive axis
And a two-dimensional DDA between the auxiliary drive axis and the passive axis, which is executed in parallel while synchronizing, to find a voxel having an intersection with the light ray. 14. Based on various information of a three-dimensional object group defined in a three-dimensional virtual space, various parameters of light rays are received as input, and one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In an apparatus for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, one or a plurality of sets each storing a set of three-dimensional objects included in the voxel for each of the voxels Object set storage means, wherein for each three-dimensional object included in each voxel, each of the voxels located near the voxel includes the same three-dimensional object as the three-dimensional object. An intersection search apparatus characterized by further retaining duplication elimination information that records whether or not there is any intersection. 15. Based on various information of a group of three-dimensional objects defined in a three-dimensional virtual space, various parameters of light rays are received as input, and one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, information indicating the necessity of calculating the intersection of each of the voxels with the three-dimensional object included in the voxel One or a plurality of voxel information storage means for storing a set of three-dimensional objects included in the voxel and a set of three-dimensional objects included in the voxel located in the vicinity thereof. An intersection search device having information defined by an inclusion relationship with a set of intersections. 16. The intersection searching apparatus according to claim 15, further comprising one or more object set storage means for storing a set of three-dimensional objects included in each voxel for each of said voxels, wherein said object set storage is performed. The means further includes, for each three-dimensional object included in each voxel, duplicate removal information that records whether or not each of the voxels located near the voxel includes the same three-dimensional object as the three-dimensional object. An intersection search device characterized by holding. 17. The intersection search apparatus according to claim 14, wherein the voxels located in the vicinity are six adjacent voxels. 18. The intersection search apparatus according to claim 14, wherein a local three-dimensional grid is set for a part of the space occupied by the three-dimensional grid, and the local three-dimensional grid is set to the three-dimensional grid. An intersection search apparatus characterized in that a three-dimensional lattice group is hierarchically organized by being a kind of a three-dimensional object. 19. Based on various information of a three-dimensional object group defined in a three-dimensional virtual space, various parameters of light rays are received as inputs, and one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In a device for setting a plurality of three-dimensional grids and outputting intersection information between the light beam and the three-dimensional object group, information indicating the necessity of calculating the intersection of each of the voxels with the three-dimensional object included in the voxel One or a plurality of voxel information storage means for storing the information indicating the necessity of the intersection calculation from the voxel information storage means, and voxels which intersect with the ray and are deemed to need the intersection calculation And one or more voxel tracking circuits that output the index of the voxel, and based on the voxel index output from the voxel tracking circuit, 1 to determine the three-dimensional object to be calculated
One or more object determination circuits, and one or more intersection calculation circuits for performing a direct intersection calculation between the three-dimensional object to be calculated for intersection and the light ray determined by the object determination circuit, An intersection search device, comprising: a separate object determination circuit for each of the voxel tracking circuit and each of the intersection calculation circuits. 20. The intersection search apparatus according to claim 1, wherein said bank is a small-capacity and high-speed storage means connected to a larger-capacity and lower-speed storage means. apparatus. 21. A voxel tracking method for dividing a three-dimensional virtual space into a three-dimensional lattice having hexahedral voxels as elements, receiving various parameters of light rays as inputs, and finding the voxels intersecting with the light rays. When the main drive axis, the sub drive axis, and the passive axis are set in the descending order of the absolute value of the component of the direction vector of the light ray with respect to the coordinate axis, the two-dimensional DDA of the main drive axis and the sub drive axis, and the main drive axis and the passive axis Two-dimensional DDA between the auxiliary drive axis and the passive axis
And v. Finding a voxel having an intersection with the ray by executing the voxel in parallel. 22. Based on various information of a group of three-dimensional objects defined in a three-dimensional virtual space, various parameters of a light ray are received as input, and one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In the intersection search method for setting a plurality of three-dimensional grids and calculating intersection information between the light ray and the three-dimensional object group, for each three-dimensional object included in each voxel, each of the voxels located in the vicinity of the voxel An intersection search method characterized by using a table in which information indicating whether the same three-dimensional object as the three-dimensional object is included is used to reduce duplicate intersection calculations. 23. Based on various information of a group of three-dimensional objects defined in a three-dimensional virtual space, various parameters of a light ray are received as input, and one or more elements each having a hexahedral voxel as an element in the three-dimensional virtual space. In an intersection search method for setting a plurality of three-dimensional grids and calculating intersection information between the light beam and the three-dimensional object group, a set of three-dimensional objects included in the voxel for each voxel, and a set of three-dimensional objects located in the vicinity thereof An intersection search method characterized by using a table in which information defined by an inclusion relation with a set of three-dimensional objects included in voxels is used to reduce duplicate intersection calculations.