KR101568204B1 - Manicore system based cpu/gpu and method for determining the number of cores of a multicore cpu for cpu/gpu concurrent processing - Google Patents

Abstract

A many-core system based on a CPU/GPU according to the present invention includes a CPU including multiple cores, a GPU, and a control unit, which controls the CPU and the GPU. The control unit selects cores to perform tasks from the cores, included in the CPU, based on time for copying data from the CPU to GPU, allocates tasks to corresponding cores according to the number of selected cores, allocates the tasks to the whole cores, included in the CPU, if the time for copying the data is shorter than a predetermined threshold value, and allocates the tasks to the cores except the cores for copying the data to the GPU which are included in the CPU if the time for copying the data is longer than the predetermined threshold value.

Description

Technical Field [0001] The present invention relates to a method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU-based Mannier system and a CPU / GPU,

The present invention relates to a method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU-based Mannier system and a CPU / GPU.

Due to the recent development of computer architecture, general PC or mobile computing environment has multicore CPU / GPU structure. As a result, research on parallel processing techniques to improve the speed by using a multicore CPU or a GPU has become an important issue. In addition, there are various kinds of distributed processing systems for cloud computing and big data processing As the heterogeneous computing environment develops, researches utilizing CPU and GPU are being actively carried out.

On the other hand, when CPU and GPU are used at the same time, GPU accelerator has a large overhead of copying data, and when CPU and GPU are processed at the same time, data must be copied from CPU to GPU. It features a very large parallel overhead when parallelizing tasks.

Therefore, when a multi-core CPU and a GPU are used at the same time, it is common to use all of the CPU's cores in order to appropriately divide tasks and process tasks therefor. However, when the size of the task is small, the data copying time is relatively longer than the time for processing the task. Particularly, in the case of video streaming, the data copying rate varies depending on the size of a video image. Accordingly, there is a problem that a task execution time becomes long when all cores are used without considering the size of a task.

Such object tasking requires a small amount of work but has a large amount of work when applied to a streaming service, so that it is necessary to shorten the execution time. In addition, object tasking has data dependency in each frame because it refers to the characteristics of the whole image, so parallelization must be performed at a task level having a small amount of work. When a CPU and a GPU are simultaneously used for such a small task, a method of efficiently performing parallel processing is required.

In this connection, Korean Patent Laid-Open Publication No. 2012-0091550 (a high-speed image processing apparatus and method using a parallel processor of a general-purpose graphics processing apparatus) is applied to parallel processing using a general-purpose video card equipped with a GPU, Discloses a technique for improving the operation speed required for data conversion.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a CPU / GPU-based ManiCore system and a CPU / GPU for efficiently performing parallel processing of small tasks simultaneously using a CPU and a GPU And a method for determining the number of cores of a multicore CPU for simultaneous processing.

According to a first aspect of the present invention, there is provided a Mannich system based on a CPU / GPU, comprising: a CPU including a plurality of cores; And a control unit for controlling the GPU and the CPU and the GPU, wherein the control unit selects a core to perform an operation among cores included in the CPU based on a data copy time from the CPU to the GPU, Assigns a task to a corresponding core according to a predetermined threshold value, and allocates a task to all of the cores included in the CPU when the data copy time is less than a predetermined threshold, All of the cores are allocated to the cores other than the core for copying data to the GPU.

According to a second aspect of the present invention, there is provided a method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU, comprising: selecting a core to perform an operation based on a data copying time from a CPU to a GPU; Wherein the step of selecting a core to perform the task comprises: if the data copy time is less than a predetermined threshold value, allocating an entire core included in the CPU to the corresponding core, And selects, as a core to be executed, the remaining cores except the cores for copying data to the GPU among the cores included in the CPU when the threshold is greater than the preset threshold value.

According to the above-described embodiments of the present invention, it is possible to improve the processing speed of the CPU by determining the number of cores of the CPU to perform a job based on the time required for copying data with the GPU.

1 is a block diagram of a manifold system according to one embodiment of the present invention.
2 is a diagram showing an example of a method of determining the number of cores according to data copying time.
3 is a flowchart of a method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU according to an embodiment of the present invention.
4 is a graph showing the execution time according to the image frame size.
FIG. 5 is a graph comparing speed enhancement levels when a CPU and a GPU are simultaneously processed.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . Also, when an element is referred to as "comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

FIG. 1 is a block diagram of a Mann-Core system 100 based on a CPU / GPU according to an embodiment of the present invention.

The manifold system 100 according to an exemplary embodiment of the present invention includes a CPU 110, a GPU 120, and a controller 130 for controlling the CPU 110 and the GPU 120. At this time, the CPU 110 includes a plurality of cores for performing tasks.

1 refers to a hardware component such as software or an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit), and performs predetermined roles .

However, 'components' are not meant to be limited to software or hardware, and each component may be configured to reside on an addressable storage medium and configured to play one or more processors.

Thus, by way of example, an element may comprise components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, Routines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables.

The components and functions provided within those components may be combined into a smaller number of components or further separated into additional components.

The control unit 130 selects a core included in the CPU 110 to perform a job based on the data copying time from the CPU 110 to the GPU 120. [ At this time, the data copying time may be determined based on the size of the image provided to the video streaming service.

The control unit 130 assigns a job to a corresponding core according to the number of selected cores. At this time, if the data copying time is less than the preset threshold value, the task is allocated to all of the cores included in the CPU 110.

On the contrary, if the data copying time is greater than the predetermined threshold value, the CPU 110 allocates the jobs to the cores other than the core for copying data to the GPU 120 among the entire cores included in the CPU 110.

At this time, the predetermined threshold value can be set based on the execution time of the core. For example, it is also possible to compare the data execution time with a data execution time by setting a general task execution time to be processed by the core for copying data to the GPU 120 as a threshold value. As a result of the comparison, if the data copying time is shorter than the normal job execution time, the job is assigned to the entire core since it hardly affects the execution speed of the CPU.

The controller 130 determines whether to use all N cores or N-1 cores of the CPU 110 according to the ratio of the data copying time, and can determine this using the following equation (1).

[Equation 1]

That is, the control unit 130 determines whether the sum of the total work amount to be processed by the CPU 110 by the above equation, that is, the discrimination formula, by the number N of cores included in the CPU 110, If the total workload is greater than the value obtained by dividing the total workload by the number of remaining cores except the core for copying data to the GPU 120, N-1 cores can be selected as the core to be operated. Hereinafter, a method for selecting the number of cores will be described with reference to FIG.

2 is a diagram showing an example of a method of determining the number of cores according to data copying time.

Referring to FIG. 2, it can be seen that the execution time is sensitively affected by the data copy time (CPU-GPU data copy). For example, if the data copying time is long, a faster execution time can be obtained in FIG. 2 (b) than in FIG. 2 (a). That is, when the data copying time is long, if the same workload is allocated to each of the four cores of the CPU 110 as shown in FIG. 2A, the execution time of the CPU 110 is longer than the time It gets caught. Therefore, when a work to be performed by 'Core 4' is distributed to 'Core 1' to 'Core 3' without allocating a job to 'Core 4' as in the case of FIG. 2 (b) The entire execution time can be shortened.

On the other hand, if the data copying time is short, a faster execution time than that of FIG. 2 (d) can be obtained in FIG. 2 (c). That is, if the data copying time is short, even if a job is allocated to a core for copying data, the total execution time does not have a great influence. Therefore, Can be shortened.

Meanwhile, when selecting the core to perform the task, the optimal execution amount is allocated to the CPU 110 and the GPU 120, thereby shortening the overall execution time.

First, when the amount of work to be allocated to the CPU 110 and the GPU 120 for one job is W _CPU and W _GPU , the total work amount (W _TOTAL ) can be expressed by Equation (2) below.

&Quot; (2) "

At this time, the W _CPU and the W _GPU can be expressed as the ratio of the total workload. For example, if the total workload (W _TOTAL ) is 100%, the W _CPU and W _GPU can be expressed as 80% and 20%, respectively.

Next, the simultaneous processing execution time (T _TOTAL ) between the CPU 110 and the GPU 120 can be expressed by Equation (3).

&Quot; (3) "

At this time, the T _CPU (W) represents the execution time of the CPU 110, and the T _GPU (W) represents the execution time of the GPU 120. Carried out in the CPU (110) and perform simultaneous processing of the GPU (120) time (T _TOTAL) is performed times of the calculated _{CPU (110) (T CPU (} W CPU)) and the calculated _GPU (120) (T GPU ( W _GPU )).

On the other hand, the execution time (T _CPU (W _CPU )) of the CPU 110 is calculated based on the sequential processing execution time (T _SEQ (W _CPU )) of the _CPU 110. The execution time T _GPU (W _GPU ) of GPU 120 corresponds to the sequential processing execution time T _SEQ (W _GPU ) of the core that copies data to GPU 120 and the data copy time CP (W _GPU )).

At this time, the execution time (T _CPU (W _CPU )) of the _CPU 110 and the execution time (T _GPU (W _GPU )) of the _GPU 120 can be represented by Equation 4 and Equation 5 have. The control unit 130 controls the execution time ( _CPU (W _CPU )) of the CPU 110 and the GPU 120 in consideration of the degree of speed increase (ES _CPU , ES _GPU ) relative to sequential processing of the CPU 110. [ (T _GPU (W _GPU )) can be calculated. In Equations (4) and (5), the ES _CPU and the ES _GPU mean the degree of speed increase with respect to sequential processing of the CPU 110 of the CPU 110 and the GPU 120, respectively.

&Quot; (4) "

&Quot; (5) "

The sequential processing execution time T _SEQ (W) of the CPU 110 can be expressed as Equation 6 using the linear interpolation method and the data copying time CP (W) of the GPU 120 can be expressed by Equation As shown in Fig.

&Quot; (6) "

&Quot; (7) "

At this time, the data copying time CP (W) of the GPU 120 corresponds to the data transfer time HtoD (W _GPU ) from the CPU 110 to the GPU 120 and the data transfer time from the GPU 120 to the CPU 110 Time (DtoH (W ' _GPU )). The data transfer time to the GPU 120 and the data transfer time to the CPU 110 can be calculated based on the work amount (W _GPU ) of the input data of the GPU 120 and the work amount (W ' _GPU ) of the output data, respectively have.

On the other hand, each parameter α, β, γ, and δ can be obtained experimentally, and each parameter corresponds to a constant. In this way, the optimum amount of work for simultaneous utilization of the CPU 110 and the GPU 120 can be calculated through the parameters obtained through experiments. As an example of each parameter, numerical values as shown in the following table can be obtained.

[table]

Next, the control unit 130 assigns the CPU 110 and the GPU 120 a work amount when the simultaneous processing execution time (T _TOTAL ) becomes minimum. In this case, the execution time of the CPU 110 and the GPU 120 are equal when the execution time of the concurrent processing becomes minimum, which can be expressed by Equation (8) below.

&Quot; (8) "

The optimal amount of work the _CPU OW OW best effort _GPU and CPU (110) of the GPU (120) through the equation (8) can be obtained.

It is possible to expect an effect that the execution time can be shortened by about 50% as compared with the single parallelization method of the CPU 110 or the GPU 120 when the optimal workload is calculated and allocated to the CPU 110 and the GPU 120 have.

3 is a flowchart of a method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU according to an embodiment of the present invention.

The method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU according to an embodiment of the present invention first selects a core to perform an operation based on a data copying time from the CPU 110 to the GPU 120 (S110).

Next, a core to perform an operation is selected based on the data copying time (S120). That is, when the data copying time is smaller than the predetermined threshold value, the entire core included in the CPU 110 is selected as the core to perform the task. On the other hand, when the data copying time is greater than a preset threshold value, the remaining core excluding the core for copying data to the GPU 120 among the cores included in the CPU 110 is selected as the core to perform the task. On the other hand, the predetermined threshold value may be set based on the execution time of the core, and the data copying time may be determined based on the size of the image provided to the video streaming service.

Specifically, when selecting a core to perform an operation, it is possible to select the entire core or select one core in the entire core to perform the operation. That is, the sum of the total work amount for processing by the multi-core CPU 110 divided by the number N of cores included in the CPU 110 and the data copy time is calculated by multiplying the total work amount by the GPU 120 If the number of cores is greater than the number of cores other than the core divided by N-1, then one less N-1 cores can be selected as the core to perform the operation.

In accordance with the number of cores selected by this method, tasks are allocated to corresponding cores.

4 and 5, the execution time and speed enhancement degree in the core number determination method of the multicore CPU for simultaneous processing of the CPU / GPU-based Mannier system 100 and the CPU / GPU according to the present invention .

FIG. 4 is a graph illustrating execution time according to an image frame size, and FIG. 5 is a graph comparing speed enhancement levels when the CPU 110 and the GPU 120 are simultaneously processed.

We used Intel Core i5-3570 CPU and GeForce GTX 660 GPU with four cores to determine execution time and speed improvement according to the simultaneous processing of CPU 110 and GPU 120. And 120 image frames having image sizes of 3000 x 2000, 1440 x 1080, 1280 x 720, 800 x 600, 720 x 480 and 640 x 480 were used. In addition, OpenMP and OpenCL are used to extract the parallel characteristics of the CPU 110 and the GPU 120.

In order to predict the characteristics according to the present invention, FIG. 4 shows the comparison of Estimated execution time and Measured value. Referring to FIG. 4, it can be confirmed that the estimated value of the execution time according to the optimum workload ratio of the CPU 110 and the GPU 120 is close to the measured value. It can be seen that the optimal workload ratio of the CPU 110 and the GPU 120 is 80%: 20% in an image frame of 3000 × 2000 size and 75%: 25% in an image frame of 640 × 480 size.

Thus, the optimal amount of workload of the GPU 120 in a relatively small image frame is increased in a large image frame, which reduces the influence of overhead due to the data copying time of the GPU 120 This is to provide improved performance.

Referring to FIG. 5, since the GPU 120 has an overhead according to the data copying time, when only the GPU 120 is used (GPU-Olny) in various image frame sizes (CPU- Only, you can get better performance. Meanwhile, the method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU-based Mannier system 100 and a CPU / GPU according to the present invention can be realized by simultaneously using the CPU 110 and the GPU 120 (CPU / (In the case of a 3000 × 2000 image frame size) relative to the case where only the GPU 120 (Proposed)) and the GPU 120 are used, and 141% (640 × 480 Image frame size) for improved performance.

One embodiment of the present invention may also be embodied in the form of a recording medium including instructions executable by a computer, such as program modules, being executed by a computer. Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Communication media typically includes any information delivery media, including computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, or other transport mechanism.

While the methods and systems of the present invention have been described in connection with specific embodiments, some or all of those elements or operations may be implemented using a computer system having a general purpose hardware architecture.

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: Mania core system 110: CPU
120: GPU 130:

Claims (7)

In a Mannich system based on a CPU / GPU,
A CPU including a plurality of cores;
GPU and
And a controller for controlling the CPU and the GPU,
Wherein,
Selects a core to perform an operation among the cores included in the CPU based on a data copy time from the CPU to the GPU,
Allocating a job to a corresponding core according to the number of selected cores, allocating a job to all cores included in the CPU when the data copying time is less than a preset threshold value, And allocating a job to the cores other than the core for copying data to the GPU among the entire cores included in the CPU. The method according to claim 1,
Wherein the preset threshold value is set based on the execution time of the core. The method according to claim 1,
Wherein the control unit determines whether the sum of the total work amount for processing by the CPU by the number N of cores included in the CPU and the data copy time exceeds the total work amount excluding the core for copying data to the GPU The number of remaining cores divided by the number of N-1, N-1 cores to be selected as the core to perform the task. The method according to claim 1,
Wherein the data copying time is determined based on a size of an image provided to a video streaming service. A method for determining the number of cores of a multicore CPU for simultaneous processing of a CPU / GPU performed by a Mann-Core system based on a CPU / GPU,
Wherein the control unit of the manifold system selects a core to perform an operation based on a data copy time from the CPU to the GPU,
And allocating a job to a corresponding core according to the number of the selected cores,
Wherein selecting a core to perform the operation comprises:
If the data copying time is less than a preset threshold value, selecting all of the cores included in the CPU as a core to perform an operation,
And selects a core other than a core for copying data to the GPU among the cores included in the CPU as a core to perform a task if the core is larger than the preset threshold value. 6. The method of claim 5,
And the predetermined threshold value is set based on the execution time of the core. 6. The method of claim 5,
Wherein selecting a core to perform the operation comprises:
Calculating a sum of a value obtained by dividing the total work amount for processing by the multicore CPU by N, the number of cores included in the multicore CPU, and the data copy time, excluding the core for copying the entire work amount to the GPU Wherein if the number of cores is greater than the number of cores divided by the number of N-1, then N-1 cores are selected as cores to perform the operation.

Images