JP2003346178A - Parallel processing method of radiosity and its device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a parallel processing method of a radiosity method and its device capable of realizing efficient load distribution by reducing communication data quantity under processing when simulating by using the radiosity method. <P>SOLUTION: When defining the surface of a shape model in a scene as patches and allocating the patches to each node, distribution is performed so that the total area of the patches becomes equal. Namely this method is characterized by adopting a total area identification standard as a distribution standard in parallel processing. <P>COPYRIGHT: (C)2004,JPO

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a radiosity parallel processing method and apparatus, and more particularly, to a radiosity parallel processing method and radiosity three-dimensional computer graphics processing capable of high-speed processing. Regarding the device.

[0002]

2. Description of the Related Art In BS digital broadcasting, HD (high definition) image processing communication technology (broadcasting technology) is a core technology, and high-definition real image processing communication (broadcasting) is realized. On the other hand, compared to "real images", computer graphics (CG) technology for realizing high image quality and real-time processing has not advanced much.

[0003] Network technology will transition from gigabit levels to tera levels within a few years. That is,
In the next generation of communication, ultra-high-definition images exceeding HD will be data-communicated using a wideband network as well as wireless as in current broadcasting. In such next-generation ultra-high-definition image data communication, not only real images,
For CG, ultra-high image quality and real-time processing transmission will be strongly required.

[0004] In recent years, the design of industrial products has been
I can't think without it. However, the current CG system is still insufficient in various points such as image quality (resolution) and processing speed. For example, in the case of car design, C
Comparing the G model with a still camera photograph, the former is still far inferior in terms of reality and texture. Also, in the field of architecture, there is a strong demand for a landscape simulation of a building or the like and an interior design function using a light simulation function having high reality.

Further, recently, three-dimensional CG has been used between remote locations.
Increasingly, data is transmitted to design industrial products. However, due to the large amount of data, it has not yet reached the level of performing this in real time.

To generate a realistic three-dimensional CG space,
It is essential to simulate the physical behavior of light on a computer and to pursue reality that approaches live-action video. The radiosity method, which is a technique for generating such a realistic three-dimensional CG space, calculates the physical behavior of light (reflection, diffusion, reflection, shadow, etc.) using a mathematical physical model on a computer. This is a method of generating a photorealistic space by simulating light due to indirect light.

[0007]

The radiosity method, which is one of the three-dimensional CG rendering methods, is an image generation method based on a global illumination model to which heat transfer engineering is applied, and generates a high-quality CG of indoor lighting. This is a method that is often used in some cases. In the radiosity method, the image is calculated taking into account not only the direct light from the light source but also the interdiffuse reflection between objects. The feature is that it is suitable for expression and can generate a very realistic image. However, in the radiosity method, a large part of the calculation time is required to determine the form factor. It is important to speed up form factor calculation by parallelization.

An object of the present invention is to provide a radiosity parallel processing that can reduce the amount of communication data being processed and realize efficient load distribution when performing a simulation using the radiosity method. It is an object of the present invention to provide a method and an apparatus therefor.

[0009]

In order to solve the above problems, a radiosity parallel processing method according to the present invention includes a step of defining a surface of a shape model in a scene as a patch, and a processor having a plurality of nodes. Determining the magnitude relationship of the total area of the patches already allocated to the respective nodes, and allocating the next patch to the node having the smallest total area of the patches already allocated to the respective nodes. Is the gist.

A radiosity parallel processing apparatus according to the present invention includes a processor having a plurality of nodes, means for defining a surface of a shape model in a scene as a patch,
Means for determining the magnitude relationship of the total area of the patches already allocated to each of the nodes, and means for allocating the next patch to the node having the minimum total area of the patches already allocated to the nodes. Make a summary.

The present invention relates to a method and an apparatus for grouping surfaces constituting a shape, distributing the surface groups to respective CPUs, and performing parallel processing. The feature of this parallel processing method is that
The data communication amount between U is small. That is, what is communicated between the CPUs is each face group (each C
PU), there is only surface data (one surface) at which the energy value of the light receiving surface is maximum.

In the present invention, a total area equalization criterion is employed as a criterion for equalizing the calculation amount. This criterion is a method of grouping faces such that the sum of the areas of the faces included in each face group is as identical as possible to other face groups. Therefore, the calculation amount of each CPU is equalized,
A high degree of parallel processing can be realized.

[0013]

Embodiments of the present invention will be described below in detail with reference to the drawings. In the present invention, a parallel computation model of the radiosity method taking into account load balancing is proposed and its characteristics are clarified. The basic algorithm uses a method of uniformly distributing the surfaces (hereinafter, referred to as patches) of the shape constituting the object when calculating the form factor. This method is considered to be effective for a distributed memory type parallel computer having a relatively low communication speed such as a PC cluster computer system because the communication amount between the processors does not need to be small.

However, if the initial plane is simply divided into the respective processors, there is a difference in the processing time between the processors due to the bias of the mesh (hereinafter, element) of the subdivision for the radiosity calculation. Therefore, it is necessary to allocate the processing time of each processor so as to be equal. This algorithm is implemented in a massively parallel computer PC-cluster, and the effectiveness of the algorithm and problems in executing it on massively parallel computing are examined.

1. In the massively parallel computer PC-cluster parallel computer, depending on the positional relationship between the CPU and the memory,
It is roughly classified into a shared memory type and a distributed memory type. In the shared memory type (FIG. 1), a plurality of processors are connected to this centering on one memory. The advantage of this type is
Since there is no need to take into account data division during programming, automatic parallelization can be easily performed. Furthermore, since there is no need for inter-memory communication, there is an advantage that performance is improved when the number of processors is small. However, when the number of processors is increased, there is also a disadvantage that communication is crowded due to memory access competition with other tasks and other jobs, and performance is reduced.

In the distributed memory type (FIG. 2), one memory and one processor are used as one node (node), and a plurality of nodes are connected by an interconnection network. The advantage of this type is that the communication is not crowded due to a memory access conflict with another task or another job, so that the overall performance is improved. However, there is also a problem that it is difficult to manage a plurality of memories.

In order to perform the parallel processing of the radiosity method, it was implemented in a massively parallel computer PC-cluster as shown in FIG. In addition, MPICH was used as a parallelized library. Table 1 shows the specifications of the implemented massively parallel computer PC-cluster system.
Shown in

[0018]

[Table 1]

2. Parallel Processing by Surface Distribution In the present invention, a method of dividing a patch, distributing the patch to each node, and parallelizing the patch is devised. In this method, processing is performed in parallel by a host that performs data management and a node that performs radiosity calculation.

First, the host reads input data of an object model constituting a scene and divides the input data into partial areas as patches, and creates data of each partial area thus created. Then, a node in charge of calculating the radiosity of each partial area is determined. Next, the data necessary for the radiosity calculation of each partial area is created, and the data for the radiosity calculation is transmitted to each of the determined nodes. After transmitting the data, wait for the radiosity calculation result to return. Then, after receiving the radiosity calculation results of all the partial areas, the radiosity value is updated to determine whether the radiosity values have converged. If the convergence has been achieved, the calculation is completed. If not, the process returns to the step of creating the radiosity calculation of each partial area and the same calculation is performed.

On the other hand, each node in charge of the radiosity calculation of the partial area receives the radiosity calculation data and performs the calculation, and after the calculation is completed, transmits the calculation result to the host and other nodes. This is repeated until convergence. The radiosity processing uses a computationally efficient progressive method.

2.1 Data Arrangement Processing to Each Processor In the arrangement of scene data, copies of polygon data are provided to all nodes, and patch data is distributed to each node so as to equalize the area. Element data is held by a processor in which patch data including the element data is placed. In this method, since all processors hold polygon data necessary for drawing in a scene, it is not necessary to transfer coordinate data between nodes when calculating a parallel form factor.

Further, the patch data and the element data are distributed and arranged at each node, and each node holds a search for a patch having a maximum radiosity value in the distributed patch and an update of the radiosity value of each element. The communication time and the memory can be saved by distributing the processing to the nodes that perform the communication.

2.2 Area Equalization of Distributed Patches The surface of a shape model in a scene is defined as quadrangular and triangular patches, and when this patch is assigned to each node,
Each patch is distributed so that the total area of the patches of each node is equal. For example, consider a method of assigning processors to a case where patch data of a scene as shown in FIG. 4 is parallelized with three nodes. However, (1) patches A to K are sequentially distributed. (2) All of this processing is performed only by the host processor.

Step 1. First, as shown in FIG. 5, patches are assigned to each node one by one in the order of the data arrangement. Step 2. Next, the next patch is assigned from the one with the smaller total area of the patches assigned to each node (see FIG. 6). When the above processing is repeated, it is assumed that the result is roughly as shown in FIG.

At this time, which patch is to be distributed and processed by which node is determined, and is reflected in the structure of the patch data holding the patch information as shown in Table 2.

[0027]

[Table 2]

When the above processing is completed, "data.rad"
And keep all other information in the file. This "data.rad" file is broadcast from the host processor to all node processors (FIG. 8).
reference). From the above, each node only needs to perform the radiosity calculation process for the patch assigned.

2.3 Parallel Form Factor Calculation First, the information of the patch having the maximum unradiated energy from the host, that is, the light source patch (hereinafter referred to as the “shot patch”) is broadcast to each node. Next, each node calculates the form factor of only the patch in charge based on the information of the received Shoot Patch.

After the form factor calculation processing, the energy held by the patch assigned to each node is obtained. Each node transmits “total energy of the assigned patch: En” and “maximum energy of the assigned patch: Emaxn” to the host. The host obtains “total energy: Etotal” by adding all the received Ens (Equation 1).

[0031]

Etotal = E1 + E2 +... + En

The host can know from Etotal how much energy has been attenuated. Also, by comparing Emaxn, the maximum unradiated energy value and the next Shoot Pa
Acquires the node ID of whether to assign tch. As an example (FIG. 9), when Emax3 <Emax1 <Emax2, the node 2 is in charge of the next Shooting Patch.

Next, the node in charge of the Shoot Patch is S
Send the information of the hoot patch to each node. Shoot Patch
The node not in charge of receiving the information of the Shoot Patch. This operation is repeated until the unradiated energy falls below the threshold. The resulting radiosity value is transmitted to the host processor, which renders the data transmitted from each node processor and displays the result.

[0034]

[Example] In this experiment, CornelBox, TestModel01, TestMo
The parallel processing of the radiosity method devised is evaluated using three kinds of data of del02. In the case of CornelBox, the scene data is very simple. From a certain area, the load of the partial form factor transmission time required because the number of nodes has increased from the radiosity processing time has increased, and the number of nodes has increased to 8 or more. Then, the entire processing time is delayed (see FIG. 10).

From FIGS. 11 and 12, it can be considered that the more complicated the scene data, the larger the number of times of emission is required. Therefore, since the ratio of the communication time is small, the speed is improved in accordance with the number of nodes. However, it can be seen that even if the number of nodes increases, the speed does not increase linearly.

FIG. 13 shows the shortest processing time T of all nodes.
The load balance when min is the longest processing time Tmax is shown. The value is 1 when the load is completely balanced, and decreases as the load becomes unbalanced. According to the method used in the present invention, a shape surface (patch) is distributed to each node, and each node performs a radiosity calculation process on a part which is in charge of itself. At this time, since the number of subdivided meshes (elements) is largely biased depending on the location of the patch, the processing time of each node becomes unbalanced as shown in FIG. In order to solve this problem, in the present invention,
After dividing the initial patch into the optimal size for load distribution by simple quadtree division, parallel processing of radiosity was performed. FIG. 14 shows the result.

As a result of performing parallel processing by this method based on the area of the shape surface (patch), the speed was improved about 8.5 times with 16 units. However, even if the number of nodes increases due to the problem of load distribution, linear speed improvement cannot be obtained.

From the above results, the parallel calculation of the radiosity method is performed by using the technique of uniformly distributing the area based on the area of the patch, which is about 8 times smaller than the sequential calculation in the normal radiosity calculation. It was found that a speed increase of 0.5 was obtained. However, a linear speedup was not obtained because the load distribution could not be balanced well.

Although the embodiments of the present invention have been described in detail above, the present invention is not limited to the above embodiments, and various modifications can be made without departing from the gist of the present invention. It is.

[0040]

According to the present invention, since the area is uniformly distributed on the basis of the patch area for the parallel calculation of the radiosity method, efficient load distribution can be realized. This makes it possible to perform high-speed processing as compared with the conventional method.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of a shared memory type parallel computer.

FIG. 2 is a schematic configuration diagram of a distributed memory type parallel computer.

FIG. 3 is a diagram illustrating a mounted massively parallel computer PC-cluster system.

FIG. 4 is a diagram illustrating an example of a shape model.

FIG. 5 is a schematic configuration diagram showing a method of assigning first patches to each node;

FIG. 6 is a schematic configuration diagram showing a method of assigning each patch to each node after the second time.

FIG. 7 is a schematic configuration diagram showing a result of assigning each patch to each node.

FIG. 8 is a block diagram illustrating a method of broadcasting held information.

FIG. 9 is a block diagram illustrating an example of parallel processing by three nodes.

FIG. 10 is a view showing experimental results of Cornel Box.

FIG. 11 is a diagram showing test results of Test Model 01.

FIG. 12 is a diagram showing test results of Test Model 02.

FIG. 13 is a diagram showing evaluation of load balance based on the number of nodes.

FIG. 14 is a diagram showing a result of load distribution by simple quadtree addition.

   ────────────────────────────────────────────────── ─── Continuation of front page    F term (reference) 5B045 AA01 BB17 BB28 BB42 GG02                       GG11                 5B080 AA13 CA03 GA08

Claims (2)

[Claims] A step of defining a surface of a shape model in a scene as a patch; a step of determining a magnitude relationship of a total area of the patches already assigned to each of the nodes of a processor having a plurality of nodes; Assigning the next patch to the node having the smallest total area of the assigned patches. 2. A processor having a plurality of nodes, means for defining a surface of a shape model in a scene as a patch, means for judging a magnitude relationship of a total area of the patches already assigned to each of the nodes, Means for allocating the next patch to the node having the smallest total area of the already allocated patches.