KR101560283B1 - 3d image building method, 3d image building machine performing the same and storage media storing the same - Google Patents

Abstract

A 3D image building method may include the steps of: (a) dividing a given space into multiple sub-spaces (each sub-space forms a node of a KD-tree) after determining a division method for the given space; and (b) repeating the step (a) after setting each of the multiple sub-spaces as a given space. When the pertinent sub-space corresponds to an inside node of the KD-tree, the step (a) includes a step of checking whether the number of raw data in the inside node exceeds a threshold value.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a three-dimensional image generation method, a three-dimensional image generation apparatus and a recording medium storing the same,

The present invention relates to a three-dimensional image generation technique, and more particularly, to a three-dimensional image generation method, a three-dimensional image generation device, and a recording medium storing the three-dimensional image generation method, will be.

In general, a k-dimension tree is a k-dimensional tree which is used in generating a three-dimensional image and has a space-division data structure for arranging points (i.e., data) in a k- Tree. ≪ / RTI > The KD-tree can be useful for searching using multi-dimensional search keys (eg, range search and proximity search), and the KD-tree is a kind of binary space partitioning (BSP) tree.

Korean Patent Laid-Open No. 10-2011-0124834 relates to a method of generating a KD-tree using hardware, and a location where the cost of a node is minimized by SAH (surface area heuristic) Is generated. The KD-tree generation method refers to the index of the list data, classifies the left and right of the list data based on the triangle data read from the memory, finds the position where the cost of the current node becomes the minimum in the SAH method, If the minimum location is not found, a leaf node can be created.

Korean Patent Laid-Open No. 10-2012-0090543 discloses a method for constructing a KD-tree based on the visibility of a voxel. The method of constructing the KD-tree is to construct a KD-tree by dividing the voxel into two sub-voxels using a cost function based on the voxel visibility to construct the KD-tree and using a plane minimizing the cost function .

Korean Patent Publication No. 10-2011-0124834 Korean Patent Publication No. 10-2012-0090543

One embodiment of the present invention is a method for generating a three-dimensional image capable of generating a tree at a high speed while maintaining the quality of a tree by generating an upper node in a binning manner and a lower node in a sorting manner in parallel Method.

In an embodiment of the present invention, an upper node and a lower node are distinguished based on the number of primitive data, and when the number of primitive data is the number in which an alignment method can be performed, an alignment method is used. Otherwise, To generate a tree at a high speed while maintaining the quality of the tree.

The present invention provides a method of generating a three-dimensional image by storing all the data necessary for generating a tree in the internal memory and reducing accessing to an external memory when a tree is created.

One embodiment of the present invention is to provide a three-dimensional image generation method capable of burst accessing by storing nodes in a memory in order of searching for nodes, and thereby achieving maximum memory access effect.

Among the embodiments, a three-dimensional image generation method includes the steps of: (a) determining a partitioning method of a given space and dividing the given space into a plurality of sub-spaces (each forming a node of a KD-tree) ) Repeating the step (a) by setting each of the plurality of sub-spaces as the given space, and in the step (a), if the sub-space corresponds to an internal node of the KD-tree, And checking whether the number of primitive data in the node exceeds a threshold value.

In one embodiment, the threshold may be determined based on the size of the internal memory required when the corresponding sub-space is divided in an alignment manner.

In one embodiment, the internal node may be composed of a plurality of sub-spaces divided by a division plane and a division plane of the given space.

In one embodiment, the step (a) may further include generating the plurality of sub spaces by dividing the sub space in a binning manner if the number of the raw data exceeds the threshold value .

In one embodiment, the step (a) may further include, if the number of the raw data does not exceed the threshold, dividing the corresponding sub-space into a plurality of sub-spaces by sorting .

In one embodiment, the step (a) may include generating the plurality of sub-spaces by performing the division of the sorting scheme for the sub-space in parallel.

In one embodiment, the step (a) may include converting the division order of the plurality of sub-spaces into the visit order of the KD-tree, and storing the plurality of sub-spaces in a memory.

In one embodiment, the step of storing the plurality of subspaces in a memory comprises the steps of receiving an address of the node and data of the node generated through the division of the given space, and a step of receiving a burst node entry node entry, wherein the step of searching for the location of the burst node entry may include comparing the node address to the burst address.

In one embodiment, storing the plurality of sub-spaces in a memory may include searching for a burst node entry in which there is no room for node data to be stored.

(A) dividing a given space into a plurality of sub-spaces (each of which forms a node of a KD-tree) by determining a partitioning mode of a given space; and (b) And a tree creation unit for setting the each of the plurality of sub-spaces as the given space and repeating the step (a) It is checked whether or not the number of the original data in the node exceeds the threshold value.

In one embodiment, the threshold may be determined based on the size of the internal memory required when the corresponding sub-space is divided in an alignment manner.

In one embodiment, the internal node may be composed of a plurality of sub-spaces divided by a division plane and a division plane of the given space.

In one embodiment, the tree generating unit may include a first tree generating unit for generating the plurality of sub-spaces by dividing the corresponding sub-space in a binning manner when the number of the raw data exceeds the threshold value have.

In one embodiment, the tree generating unit may include at least one second tree generating unit that generates a plurality of sub-spaces by dividing the corresponding sub-space in a sorting manner if the number of the raw data does not exceed the threshold value .

In an embodiment, the second tree generating unit may generate the plurality of sub-spaces by performing the partitioning of the sorting scheme for the corresponding sub-space in parallel.

In one embodiment, the tree creation unit may include a node scheduler that converts the division order of the plurality of sub-spaces into a visit order of the KD-tree and stores the plurality of sub-spaces in a memory.

In one embodiment, the node scheduler receives the address of the node and data of the node generated through the division of the given space, and receives the push node entry for searching the location of the burst node entry in which the data of the node is to be stored. And the push-entry selector may compare the node address and the burst address.

In one embodiment, the node scheduler may include a pop entry selector for searching for a burst node entry where there is no room for node data to be stored.

Among the embodiments, a recording medium on which a computer program relating to a three-dimensional image generating method is recorded includes (a) a method of determining a division method of a given space and dividing the given space into a plurality of subspaces (each of which forms a node of a KD- (B) repeating the step (a) by setting each of the plurality of sub-spaces as the given space, and the step (a) And a function of checking whether the number of primitive data in the internal node exceeds a threshold value if it is an internal node of the internal node.

In the method of generating a three-dimensional image according to an embodiment of the present invention, an upper node is generated in a binning manner and a lower node is generated in a sorting manner in parallel to generate a tree at a high speed while maintaining the quality of a tree .

According to an embodiment of the present invention, an upper node and a lower node are classified based on the number of primitive data, and when the number of primitive data is the number that can be performed, Otherwise, the tree can be generated at a high speed while maintaining the quality of the tree by generating a tree using the binning method.

The 3D image generation method according to an embodiment of the present invention can save all the data needed to generate a tree in the internal memory and reduce the case of accessing the external memory when generating the tree.

The three-dimensional image generation method according to an embodiment of the present invention enables burst access by storing nodes in a memory in order of searching for a node, thereby achieving the maximum memory access effect.

1 is a block diagram illustrating a tree generating unit according to an embodiment of the present invention.
Fig. 2 (a) is an embodiment of a dividing plane generated in a binning manner for one axis.
Fig. 2 (b) is an embodiment of the dividing plane created by the alignment method for one axis.
3 is a block diagram illustrating a ray tracing apparatus including the tree generating unit of FIG.
4 is a block diagram illustrating a working memory in accordance with one embodiment of the present invention.
5A is a block diagram illustrating a node scheduler and a node memory according to an embodiment of the present invention.
5 (b) is a block diagram showing an embodiment of a node scheduler.
6 is a block diagram illustrating a process of ray tracing.
Figure 7 is a block illustrating geometry and acceleration structures used in the disclosed technique.
8 is a flowchart illustrating a tree generating unit according to an embodiment of the present invention.

The description of the present invention is merely an example for structural or functional explanation, and the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, the embodiments are to be construed as being variously embodied and having various forms, so that the scope of the present invention should be understood to include equivalents capable of realizing technical ideas. Also, the purpose or effect of the present invention should not be construed as limiting the scope of the present invention, since it does not mean that a specific embodiment should include all or only such effect.

Meanwhile, the meaning of the terms described in the present application should be understood as follows.

The terms "first "," second ", and the like are intended to distinguish one element from another, and the scope of the right should not be limited by these terms. For example, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

It is to be understood that when an element is referred to as being "connected" to another element, it may be directly connected to the other element, but there may be other elements in between. On the other hand, when an element is referred to as being "directly connected" to another element, it should be understood that there are no other elements in between. On the other hand, other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

It is to be understood that the singular " include " or "have" are to be construed as including the stated feature, number, step, operation, It is to be understood that the combination is intended to specify that it does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

In each step, the identification code (e.g., a, b, c, etc.) is used for convenience of explanation, the identification code does not describe the order of each step, Unless otherwise stated, it may occur differently from the stated order. That is, each step may occur in the same order as described, may be performed substantially concurrently, or may be performed in reverse order.

The present invention can be embodied as computer-readable code on a computer-readable recording medium, and the computer-readable recording medium includes all kinds of recording devices for storing data that can be read by a computer system . Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and also implemented in the form of a carrier wave (for example, transmission over the Internet) . In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined. Commonly used predefined terms should be interpreted to be consistent with the meanings in the context of the related art and can not be interpreted as having ideal or overly formal meaning unless explicitly defined in the present application.

1 is a block diagram illustrating a tree generating unit according to an embodiment of the present invention.

Referring to FIG. 1, a tree generating unit 100 may include a first tree generating unit 110 and at least one second tree generating unit 120.

The first tree generating unit 110 generates a tree using a binning method. Here, the binning method divides the bounding box space by a predetermined number of times for the three axes (x, y, z) to calculate the surface area heuristic (SAH) value only for the divided surface. Here, the SAH is a method for generating an acceleration structure for ray-tracing, which includes values for searching for primitives encountering arbitrary light rays (for example, the number of node visits, The number of times to calculate whether or not it is possible to calculate the number of times of calculation), and divides the space based on the division plane having the best value among the calculation results.

At least one second tree generating unit 120 generates a tree using a sorting method. Here, the alignment method extracts the start value and the end value of the entire raw data for three axes (x, y, z) and divides the three axes into a start value and an end value to generate six pieces of list data. Here, the list data is an array composed of the indices of the raw data. After generating the list data, it is sorted for each list data, and the division plane is generated from the start value and the end value of the sorted original data. Next, the SAH value is calculated from the generated division plane, and the space is divided based on the division plane having the best value.

More specifically, the at least one second tree generating unit 120 may obtain raw data information through the first tree generating unit 110, the plurality of second tree generating units 120 may operate in parallel, And stores the node and list data (AS data) received from at least one second tree generating unit 120 in an external memory 340 in connection with an arbiter 140.

A working memory 130 is used when a tree is generated in the first tree generating unit 110 and the at least one second tree generating unit 120. The work memory 130 will be described with reference to FIG.

Fig. 2 (a) is an embodiment of a dividing plane generated in a binning manner for one axis.

Referring to FIG. 2 (a), the bounding box 210a may include a dividing surface 220a and raw data 230a. In one embodiment, the dividing surface 220a is predetermined with five dividing surfaces, and the SAH value at each dividing surface 220a can be calculated for the six pieces of raw data 230a. As a result of calculation, it is possible to divide the bounding box 210a by selecting the dividing surface having the best value.

Fig. 2 (b) is an embodiment of the dividing plane created by the alignment method for one axis.

Referring to FIG. 2 (b), the bounding box 210b may include a dividing surface 220b and raw data 230b. In one embodiment, the dividing surface 220b is generated at the aligned start and end values of the raw data 230b and may calculate the SAH value at each dividing surface 220b for the raw data 230b. As a result of calculation, it is possible to divide the bounding box 210b by selecting the dividing surface having the best value.

3 is a block diagram illustrating a ray tracing apparatus including the tree generating unit of FIG.

3, the ray tracing apparatus 300 includes a central processing unit (CPU) 310, a system memory 320, a dynamic ray tracing accelerator (DRTX) 330, an external memory, (340).

The CPU 310 can process a 3D application and includes at least one of an application 311 such as a 3D game engine, an API (Application Programming Interface) 312, and a scene manager (Scene Manager) 313 .

The system memory 320 can store graphics data information required for a 3D application and includes a PSS area 321 for storing a primitive static scene, a PDS area for storing a primitive dynamic scene 322 for storing texture maps, and a texture map area 323 for storing mip-maps (MIP-MAP) for texture mapping. Since the primitive fixed scene is the primitive data information that does not change with respect to the entire frame, it is needed only once when the tree is generated. Therefore, the fixed acceleration structure, which is the tree creation result when driving the 3D application, is stored in the external memory 340 of the DRTX 330, Lt; / RTI > Since the primitive data information changes in each frame of the operation scene, a tree is generated through the tree creation unit 100 every time, and a motion acceleration structure, which is a result of generation, is sent to the external memory 340 of the DRTX 330.

The DRTX 330 includes a tree build unit 100 of FIG. 1 and includes a bus interface unit 331, a primitive 332, acceleration structure data AS Data, A cache memory 333, a working memory 130, an AS cache 335, a texture cache 336, a color result buffer 337, a stack memory A stack memory 338, and a ray tracing unit 339. [0031] In one embodiment, a spatial segmentation structure is constructed based on the graphic data information, ray tracing is performed based on the generated spatial segmentation structure, the result of the performed ray tracing is transmitted to the CPU 310, So as to reconstruct the space division structure for the graphic data information.

The external memory 340 may temporarily store information processed in the DRTX 330 and may include a geometric information storage area 341, a static thin acceleration storage structure storage area 342, a dynamic thin acceleration storage structure storage area 343, A map storage area 344, and a color information storage area 345. [

4 is a block diagram illustrating a working memory according to one embodiment of the present invention.

Referring to FIG. 4, a working memory 130 includes primitive data 410, a primitive index 420, a temporary space 430, Acceleration structure data 440 may be stored.

The x, y, and z coordinate information of the raw data received from the at least one second tree generator 120 is stored in the raw data 410 memory, and the x, y, The index numbers of the raw data are stored in the order of the indexes. Also, the index number of the raw data divided into left and right is temporarily stored in the memory of the temporary space 430 while the classification is performed on the basis of the division plane, and at least one second tree generating unit is performed in the memory of the acceleration structure data 440 And the generated node and list data are stored. The work memory 130 corresponds to an internal memory.

5A is a block diagram illustrating a node scheduler and a node memory according to an embodiment of the present invention.

5A, a node scheduler 510 includes a Node Data Manager 511, a Push Entry Selector 512, and a Pop Entry Selector 513 and the node memory 520 may include a plurality of data array fields 521. [ The burst node entry 530 may include an existing field, a full field and a burst field and a data array field 521 of the node data manager 511.

The node data manager 511 manages the information of the data array fields 521 existing in the node memory and manages information on a plurality of existing fields, a full field and a burst field . Herein, the presence field has information on whether or not node data exists in the data array field 521, and the full field has information on whether data can be further stored in the data array field 521, and the burst field The node data has an external memory address for burst access.

The push entry selector 512 selects a burst node entry 530 in which the node data generated in the first and second tree generating units 110 and 120 is to be stored and outputs the burst node entry 530 to the node scheduler 510 And receives information on node data and node address from the outside. Here, the node address means an external memory address in which the node data is to be stored in the order of searching for the node. The push-entry selector 512 receives the information of each burst node entry 530 from the node data manager 511 in the node scheduler 510. Then, the push-entry selector 512 determines the position of the burst node entry 530 in which the node data is to be stored based on the input data. When the position is selected, the push-entry selector 512 selects the push- After the data offset is confirmed, the node data is stored in the data array field 521 of the corresponding burst node entry 530. Here, the node address includes the burst address and the data address, and the data offset has the position information in the data array field 521 where the node data is to be stored.

The pop entry selector 513 outputs the data array field 521 of the selected burst node entry 530 to the outside. More specifically, the pop entry selection unit 513 receives information in the full field of all the burst node entries 530 from the node data manager 511, and confirms whether there is valid information. If there is valid information, the burst field and data array field 521 of the corresponding burst node entry 530 are output to the outside.

The data array field 521 has the same size as the burst access size of the external memory 340 as a memory space capable of storing a plurality of node data. The existence field, the full field, and the burst field of the node data manager 511 together with the data arrangement field 521 of the node memory 520 constitute one burst node entry 530.

More specifically, the node scheduler 510 receives the node address and the node data as input, and locates the burst node entry 530 in which the node data input from the push-entry selector 512 is to be stored. Here is how to find the location. After all the existence fields are checked, the burst node entries 530 in which the node data exists are searched. The burst field of the next burst node entries 530 is compared with the burst address of the input node address to find a burst field having the same address. If the burst field is found, node data is stored in the corresponding burst node entry 530 in the push data. If the same burst field is not found, the burst node entry 530 in which node data does not exist is selected and the burst field is changed. The selection method takes precedence over the highest burst node entry 530 where the presence field is not valid. The changing method is to store the burst address of the node address entered in the burst field of the selected burst node entry 530. [ And then stores the node data in the data array field 521 of the burst node entry 530 initialized in the push data.

When the storage of the node data is completed, the pop-entry selecting unit 513 checks all the full-field information and finds the burst node entry 530 where there is no more space for storing the node data. If there is no corresponding burst node entry 530, the node scheduler 510 is terminated. Otherwise, the data present in the burst field and data array field 521 of the corresponding burst node entry 530 are output and the node scheduler 510 is terminated.

5 (b) is a block diagram showing an embodiment of a node scheduler.

When the node scheduler 510 receives the information on the nodes in the order of 0, 1, 2, 3, ..., the nodes are stored in the node memory 520 through the process described with reference to FIG. Can be stored in the order of 7, 3, 1, 0, ... according to the search order. In addition, the burst access size is 4, so four node data can be stored in one data array field 521. [ That is, the order of creation of the nodes is 0, 1, 2, 3, ..., but the order stored in the node memory is stored in the search order as 7, 3, 1, 0, ....

6 is a block diagram illustrating a process of ray tracing.

Referring to FIG. 6, a primary ray P is generated from a position of a camera 610 for each pixel, and a calculation for searching an object 620 to meet the ray P is performed. When the object to be encountered with the ray P is an object 620 having refracting properties or an object 631 and 632 having a reflecting property, the refraction effect is obtained at the position where the ray P and the object meet A reflection ray F for reflection and a reflection ray R for reflection can be generated and a shadow ray S can be generated in the direction of the light 650. [ In one embodiment, when the shadow ray S meets another object 640, a shadow may be generated at the point where the shadow ray S is generated.

Figure 7 is a block illustrating geometry and acceleration structures used in the disclosed technique.

In FIG. 7, it is assumed that the Acceleration Structure (AS) uses a KD-tree. The KD-tree is a kind of Spatial Partitioning Structure and can be used for Ray-Triangle Intersection Test. For example, the KD-tree may include a box node 710, an inner node 720, and a leaf node 730. In one embodiment, the leaf node 730 may include a triangle list for pointing at least one triangle information included in the geometric data. For example, the triangle information may include vertex coordinates, normal vectors, and texture coordinates for the three points of the triangle. In one embodiment, when the triangle information included in the geometric data is implemented as an array, the triangle list included in the leaf node may correspond to the array index. In one embodiment, the inner node 720 has a spatial area based on a bounding box, and the spatial area can be divided into two areas and allocated to two lower nodes. In other words, the internal node 720 is composed of a sub-tree of a divided area and two divided areas thereof, and the leaf node 730 includes only a series of raw data. The division plane dividing the space must find a point at which the value for finding the raw data to be hit with an arbitrary light ray (for example, the number of node visits, the number of times of calculating whether intersecting with the raw data, etc.) (Surface area heuristic) can be a method used to find the point.

8 is a flowchart illustrating a tree generating unit according to an embodiment of the present invention.

First, it is checked whether or not the number of primitive data is larger than a threshold value (step S801). Here, the threshold value is the maximum number of raw data that one second tree generating unit 120 can process. As a result of checking, if the number is greater than the threshold value, the space is divided by the binning method (step S802), and the divided space is returned to step S801 to check again whether or not the number of raw data is larger than the threshold value.

If the number of raw data is smaller than the threshold value, the space is divided by the sorting method (step S803). (Step S804). If there is no space for the leaf node (step S804), the process ends. If not, the process returns to step S803 and the space is divided again by the sorting method.

In one embodiment, partitioning the space is to find a partition plane, which is the location for partitioning the space, and the partition plane is a value for finding raw data that collides with the ray (e.g., the number of node visits, The number of calculations, etc.) can be minimized. The surface area heuristic (SAH) calculation method is used to find the divided surface, and the SAH method has a binning and sorting method.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the present invention as defined by the following claims It can be understood that

100: tree generating unit 110: first tree generating unit
120: second tree generator 130: work memory
210a: Bounding box 220a:
230a: raw data 210b: bounding box
220b: splitting surface 230b: raw data
300: ray tracing device 310: CPU (Central Processing Unit)
320: System memory 330: Dynamic Ray Tracing Accelerator (DRTX)
340: External memory
410: raw data 420: raw data index
430: Temporary space 440: Accelerated structure data
510: Node Scheduler 520: Node Memory
530: burst node entry
610: Camera 620: Object
631: object 632: object
640: object 650: light
710: box node 720: internal node
730: leaf node

Claims (19)

(a) determining a partitioning mode of a given space and dividing the given space into a plurality of sub-spaces (each forming a node of a KD-tree); And
(b) repeating the step (a) by setting each of the plurality of subspaces to the given space,
In step (a), if the corresponding sub-space corresponds to an internal node of the KD-tree (K-Dimension tree), it checks whether the number of raw data in the internal node exceeds a threshold, tree node based on the number of primitive data of the sub-space.
2. The method of claim 1, wherein the threshold is
And determining the size of the KD-tree based on the size of the internal memory required when the sub-space is divided by the sorting method.
2. The method of claim 1, wherein the internal node
And a plurality of sub-spaces divided by a division plane and a division plane of the given space.
The method of claim 1, wherein step (a)
And generating the plurality of sub-spaces by dividing the sub-space in a binning manner if the number of the raw data exceeds the threshold value.
The method of claim 1, wherein step (a)
And generating a plurality of sub-spaces by dividing the sub-space in a sorting manner if the number of the raw data does not exceed the threshold value.
6. The method of claim 5, wherein step (a)
And generating the plurality of sub-spaces by performing the partitioning of the sorting scheme for the sub-space in parallel.
The method of claim 1, wherein step (a)
And converting the division order of the plurality of sub-spaces into a visited order of the KD-tree, and storing the plurality of sub-spaces in a memory.
8. The method of claim 7, wherein storing the plurality of sub-
Receiving address and node data of a node generated through division of the given space; And
Searching for a location of a burst node entry in which data of the node is to be stored,
Wherein the step of searching for the location of the burst node entry comprises comparing the node address to a burst address.
8. The method of claim 7, wherein storing the plurality of sub-
Searching for a burst node entry in which there is no space in which node data is to be stored.
(a) determining a partitioning method of a given space to divide the given space into a plurality of sub-spaces (each forming a node of a KD-tree)
(b) a tree generating unit for setting each of the plurality of sub-spaces to the given space and repeating the step (a)
If the corresponding sub-space corresponds to an internal node of the KD-tree, the tree generating unit checks whether the number of the original data in the internal node exceeds a threshold value, and checks whether the raw data of the sub-space (KD-tree node) And determining a division scheme of the sub-space based on the number of sub-spaces.
11. The method of claim 10, wherein the threshold is
And determining the size of the sub-space based on the size of the internal memory required when the sub-space is divided into the sub-spaces.
11. The method of claim 10, wherein the internal node
And a plurality of sub-spaces divided by a division plane and a division plane of the given space.
11. The apparatus of claim 10, wherein the tree generating unit
And a first tree generating unit for generating the plurality of sub-spaces by dividing the corresponding sub-space in a binning manner if the number of the raw data exceeds the threshold value.
11. The apparatus of claim 10, wherein the tree generating unit
And generating a plurality of sub-spaces by dividing the sub-space into a plurality of sub-spaces if the number of the raw data does not exceed the threshold. Device.
15. The apparatus of claim 14, wherein the second tree generator
Wherein the plurality of sub-spaces are generated by performing the division of the sorting scheme for the sub-space in parallel.
11. The apparatus of claim 10, wherein the tree generating unit
And a node scheduler for converting the division order of the plurality of sub-spaces into a visit order of the KD-tree and storing the plurality of sub-spaces in a memory.
17. The system of claim 16, wherein the node scheduler
And a push entry selector for receiving a node address and data of a node generated through the division of the given space and searching for a location of a burst node entry in which data of the node is to be stored,
Wherein the push-entry selector compares a node address with a burst address.
17. The system of claim 16, wherein the node scheduler
And a pop entry selector for searching for a burst node entry in which there is no space in which node data is to be stored.
(a) determining a partitioning mode of a given space to divide the given space into a plurality of sub-spaces (each forming a node of a KD-tree); And
(b) setting each of the plurality of sub-spaces as the given space, and repeating the step (a)
In step (a), if the corresponding sub-space corresponds to an internal node of the KD-tree, it is checked whether the number of raw data in the internal node exceeds a threshold, And determining a division method of the sub space based on the number of data.

Images