JPH06187312A - Processing method and its device in multi-cpu system - Google Patents

Abstract

PURPOSE:To perform smooth and efficient parallel decentralized processes of many data. CONSTITUTION:Traverser threads 1, 2, and 3 which performs processes based upon programs are generated corresponding to CPUs 11-14. Process completion flags showing which traverser thread processed the data are added to the data that the traverser threads process and stored in a memory 15. Each traverser thread selects unprocessed data by referring to the process completion flags. The flag of the selected data is changed into a state indicating that the data are already processed. Each traverser thread processes the selected data according to the program.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a processing method and apparatus in a multi-CPU system.

[0002]

2. Description of the Related Art In a multi-CPU system, a task is divided into a plurality of threads for processing. Threads correspond to CPUs and share application programs, data, etc. in tasks.

As a typical example of the processing executed in the multi-CPU system, a graphics display processing is given. The following method is known for this processing.

One of them is to prepare a plurality of threads for performing geometric transformation processing of graphic data for displaying graphics on a display device, and perform the processing in parallel for each graphic data unit. However, according to this method, since multiple threads process graphic data in parallel, multiple threads can share the data and two or more threads can access the same graphic data simultaneously or before and after. is there. In this case, there will be a problem that the processed data will be processed again.
There is a problem in managing data or mutual exclusion between threads.

Another method is to divide the geometric transformation processing into processing units such as coordinate transformation processing, clipping processing, ... And perform the divided processing by different threads. In this case, the graphic data is passed to each thread and processed. However,
In order to transfer graphic data between threads, it is necessary to communicate data between threads, and for that purpose it is necessary to synchronize. Therefore, there is a problem that a waiting time occurs in one of the threads when synchronizing and an overhead occurs.

[0006]

DISCLOSURE OF THE INVENTION An object of the present invention is to enable parallel and distributed processing in a multi-CPU system to be carried out efficiently and smoothly.

A processing method in a multi-CPU system according to the present invention, in a system having a plurality of CPUs, uses a plurality of threads for processing according to a preset program generated corresponding to the plurality of CPUs. , Is a method for parallel and distributed processing of a large number of data, and sets a processed flag indicating in advance whether or not the data is processed by any of the plurality of threads corresponding to each of the plurality of data. A state in which the unprocessed data that has not been processed among the large number of data is selected by referring to the processed flag of the data, and the thread indicates that the processed flag of the selected data has been processed. And the thread will process the selected data. That.

A processing device in a multi-CPU system according to the present invention includes a plurality of CPUs, and a plurality of threads are formed by a plurality of threads for performing processing according to a preset program generated corresponding to the plurality of CPUs. An apparatus for performing parallel and distributed processing of data, comprising a storage means for storing in advance a large number of data and a processed flag corresponding to each of the plurality of threads for processing. The thread changes the processed flag of the selected data to a state indicating that the processed flag of the selected data is processed, by referring to the processed flag of the data and selecting unprocessed data that has not been processed. And means for executing processing for the selected data.

According to the present invention, a plurality of threads can select the unprocessed data to be processed by each thread by checking the processed flag attached to the data. Therefore, multiple threads do not process the same data redundantly. Data management becomes easier, and mutual exclusion between threads becomes unnecessary. Therefore, the processing speed can be increased. Furthermore, since it is not necessary to communicate data between threads, the problem of overhead is solved.

In a preferred embodiment of the invention,
Data indicating a priority is set in advance for each of the large number of data, and the thread selects the unprocessed data with the highest priority among the unprocessed data when selecting the unprocessed data.

According to this embodiment, even if the data has priority, the thread can process the data in the order of the priority.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the structure of a multi-CPU system.

The multi-CPU system is composed of a plurality of CPUs.
And a memory 15 and a display device 16. In the embodiment, four CPUs 11, 12, 13 and 14 are provided. The CPUs 11 to 14 may be CPUs having the same function or CPUs having different functions. The memory 15 stores an OS (operating system) for operating the system, application programs, data to be processed and processed, and other information. The display processing 16 is, for example, a CRT display.

In a multi-CPU system, a task is divided into a plurality of threads for processing. The task is
It is an execution environment in which one or more threads operate, and is a basic unit of resource (CPU, memory, etc.) allocation. A thread is a unit of work that is executed while using the resources of a task, and is the minimum unit assigned to each CPU. All resources of a task are shared by multiple threads in the task. Specifically, areas such as application programs and data are shared by all threads, and program counters, stack pointers, registers, etc. are provided unique to each thread. Different threads can call the same program in a task.

The graphics display processing will be described below as an example.

FIG. 2 is a flow chart showing an outline of the graphics display processing. FIG. 3 is a table showing the contents of the memory 15. Figure 4 shows traverser threads 1-3
It is a table which shows an example of the data processed by.

When the main thread for displaying graphics is started, the main thread generates traverser threads 1, 2, and 3 (thread dispatch processing; step 21). Traversa
The threads 1 to 3 perform traverse processing, geometric conversion processing, and display request management processing, which will be described later.
Preferably, the traverser threads 1 to 3 and the main thread correspond to the CPUs 11 to 14 one to one.

As shown in FIG. 3, the main thread and the traverser threads 1 to 3 respectively include a program counter, a stack pointer, a register, etc. (main thread area, traverser threads 1,
(In a few areas)
The program (geometric conversion processing program etc.) area and shared data area are shared. Multiple shared can access the shared data area at the same time.

The generation of a thread is to secure in the memory 15 information unique to a thread such as a program counter, a stack pointer, and a register.

As shown in FIG. 4, a plurality of graphic data A to
N is processed by any one of the traverser threads 1 to 3, and these graphic data are stored in the shared data area of the memory 15 together with the processed flag. The processed flag is set to 1 if the graphic data is processed by any of the traverser threads, and is 0 if not processed. The graphic data is also accompanied by a priority (priority) of the order of display on the display device 16.

The traverser threads 1 to 3 check the processed flag of the graphic data stored in the shared data area and select unprocessed graphic data in descending order of priority (traverse process; step 22).

The traverser threads 1 to 3 access an application program (geometric conversion program) and perform geometric conversion of the figure data selected according to the program (geometric conversion process; step 2).
3).

The traverser threads 1 to 3 transmit the geometrically transformed data to the display device 16 in the order of the priority of the graphic data (display request management process; step 24).

The display device 16 includes traverser threads 1 through 1.
The geometric transformation data transmitted from any one of 3 is drawn and displayed on the screen (rendering process; step 25).

FIG. 5 and FIG. 6 are flow charts showing the detailed processing procedure of the thread dispatch processing (FIG. 2, step 21).

When the main thread for displaying graphics is started, the main thread creates the traverser threads 1, 2, and 3 as described above. The main thread temporarily stops the generated traverser thread (FIG. 5, step 31). In response to this, each of the traverser threads 1 to 3 generated by the main thread stops the processing and returns a stop signal indicating the stop to the main thread (FIG. 6, step 41). The traverser thread waits for the restart signal from the main thread (Fig. 6, step 42).

The main thread receives the stop signal returned from each of the traverser threads 1 to 3 and confirms that the traverser thread has been generated (FIG. 5, step 32), and the traverser thread 1 The restart signal is transmitted to 3 to 3 (step 33 in FIG. 5,).

The traverser threads 1 to 3 are the main
When a resume signal is received from the thread (Fig. 6, step
42), restart processing. The traverser thread which has resumed processing transmits a start signal indicating that the processing has been resumed to the main thread (FIG. 6, step 43). After that, the traverser thread moves on to traverse processing.

The main thread confirms that the traverser thread has resumed processing by receiving the start signal returned from each of the traverser threads 1 to 3 (step 34 in FIG. 5).

FIG. 7 shows the details of the processing procedure in the traverser thread. This process is common to all traverser threads 1-3.

When the traverser thread restarts processing, it first checks the processed flags associated with the graphic data in the shared data area in descending order of priority to determine whether there is data to be processed (step 51). If all graphic data flags are 1 (NO in step 51),
The traverser thread 1 finishes the process.

If there is unprocessed (processed flag is 0) graphic data, the graphic data with the highest priority is the graphic data to be processed by the traverser thread. The traverser thread idle-reads the graphic data that has already been processed (the processed flag is 1) by one of the traverser threads (step 52).

The traverser thread reads the unprocessed graphic data with the highest priority (step 53), and sets the processed flag of the read graphic data to 1 (step 54). The above processing is the traverse processing (step 22 in FIG. 2,).

Next, the traverser thread accesses the geometry conversion program and performs the geometry conversion process (FIG. 2, step 23) of the graphic data read in step 53 according to this program (step 55). ).

The graphic data geometrically transformed by the traverser thread is graphically displayed on the display device 16 in accordance with the priority. The display request management process (step 24 in FIG. 2) will be described below.

A queue used in the display request management process is realized in the shared data area of the memory 15. Figure 8 (A)
, (B) and (C) show an example of a queue.
One queue has a pointer indicating that there is a next queue, a CFP indicating priority, and a data area for storing converted graphic data. A queue with a CFP of 1 has no data area. In the queue at the end, dummy data is stored in the data area, and CFP has the lowest priority. The traverser thread recognizes by the CFP in the queue whether or not there is graphic data to be displayed on the display device 16. Only the initial queue with CFP = 1 exists when the graphics display process is activated.

FIG. 8C shows that in the queue shown in FIG. 8B, the CFP is (n
It shows a state of inserting the queue of -1). The last queue has the lowest priority, so
Inserting a new queue before this tail queue requires no special handling. As shown in Fig. 8 (B), when data with priority n is stored in the queue,
The queue of priority (n-1) is inserted before the priority n as shown in FIG. 8 (C).

It is assumed that the traverser threads 1, 2, 3 are processing the data A, B, C having the priorities 1, 2, 3, respectively.

When the traverser thread 2 finishes the geometric conversion processing of the figure data B having the priority 2 before the traverser thread 1, the traverser thread 2 becomes the priority 2 of the figure data B subjected to the geometric conversion processing. The CFP of the queue already formed (in this case, there is only the initial queue with CFP = 1) is compared (step 56). The priority of the data processed by the traverse thread 2 is 2, the CFP of the existing queue is 1, and the priority and CFP are different. Therefore, step 56 is NO. The traverser thread 2 creates a new queue, sets the CFP to 2, and stores the converted data of the converted graphic data B in the data area of this new matrix (see FIG. 8A) (step 60). ).

The traverser thread 2 returns to step 51, performs the traverse process, processes the next graphic data,
For example, the geometric conversion process of the graphic data D of priority 4 is performed.

Next, when the traverser thread 1 finishes the geometric conversion processing, the traverser thread 1 compares the priority of the converted graphic data A with the CFP of the initial queue. The CFP of the queue is 1, the priority of the graphic data A converted by the traverser thread 1 is 1, and the priority and CFP are equal. Therefore, step 56 is YES. The traverser thread 1 sends the converted graphic data to the display device 16, and the display device 16 displays graphics based on the converted data of the graphic data A sent from the traverser thread 1 (step 57). . When the traverser thread 1 sends the converted data to the display device 16, it increments the CFP in the queue (step 58). Here, CFP is 2.

The traverser thread 1 determines whether there is conversion data in another queue (step 5).
9). The conversion data of the graphic data B of the priority 2 which has been converted by the traverser thread 2 is already stored in the data area of the other queues where the CFP is 2. Therefore, the step
Since the answer in 59 is YES, the traverser thread 1 returns to step 57, similarly, the converted data of the graphic data B is sent to the display device 16, and the display device 16 performs the rendering process.
Traverser thread 1 increments the queue's CFP. The CFP is 3. The traverser thread 1 determines whether there is more conversion data to be displayed. The other queue no longer exists, so traverser thread 1 returns to step 51. The traverser thread 1 performs the traverse processing, and moves to the processing of the next graphic data, for example, the processing of the graphic data E of priority 5.

When the traverser thread 3 finishes the geometric conversion processing of the graphic data C having the priority 3, the traverser thread 3 compares the priority of the converted graphic data C with the CFP of the queue. The CFP is 3, and the priority is 3, which is equal. Therefore, traverser thread 3
Is a display device for displaying the converted data of the converted graphic data C.
Then, the display device 16 performs a rendering process. Traverser Thread 3 increments CFP.
The CFP is 3. Since there are no other queues, the traverser thread 3 returns to step 51 and proceeds to traverse processing.

As described above, the traverser threads 1 to 3 process all graphic data.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a system.

FIG. 2 is a flow chart showing a processing outline of graphics display processing.

FIG. 3 is a diagram showing the contents of a memory.

FIG. 4 is a table showing an example of graphic data.

FIG. 5 is a flow chart showing a detailed processing procedure of thread dispatch processing.

FIG. 6 is a flow chart showing a detailed processing procedure of thread dispatch processing.

FIG. 7 is a flow chart showing a processing procedure in a traverser thread.

8A, 8B, and 8C are diagrams each showing an example of a queue.

[Explanation of symbols]

 11, 12, 13, 14 CPU 15 Memory 16 Display device

Claims (4)

[Claims] 1. In a system having a plurality of CPUs, a large number of data are processed in parallel and distributed by a plurality of threads that perform processing according to a preset program generated corresponding to the plurality of CPUs. A processed flag indicating whether or not the data is processed by any one of the plurality of threads corresponding to each of the plurality of data is set in advance, and the thread sets the processed flag of the data. The unprocessed data that has not been processed is selected from the above-mentioned large number of data by reference, the thread changes the processed flag of the selected data to a state indicating that it has been processed, and the thread selects A processing method in a multi-CPU system for performing the above-mentioned data processing. 2. The data indicating the priority is set in advance for each of the plurality of data, and the thread selects the unprocessed data with the highest priority among the unprocessed data when selecting the unprocessed data. A processing method in the multi-CPU system according to claim 1. 3. A plurality of CPUs are provided, and the plurality of Cs are provided.
A device for performing parallel distributed processing of a large number of data by a plurality of threads generated corresponding to a PU and performing processing according to a preset program, and the plurality of data and the plurality of threads corresponding to the respective data. A processing unit that stores in advance a processed flag indicating whether or not it has been processed by any of the above, and each of the threads refers to the processed flag of the above-mentioned data, and has not been processed yet among the large number of data. In a multi-CPU system having means for selecting data, means for changing a processed flag of the selected data to a state indicating that the data has been processed, and means for executing processing for the selected data Processing equipment. 4. The data indicating the priority is set in advance for each of the plurality of data, and each thread selects the unprocessed data with the highest priority among the unprocessed data when selecting the unprocessed data. The multi-CPU according to claim 3, further comprising:
The processing unit in the system.