KR101662769B1 - Apparatus for quick sorting and the method thereof - Google Patents

Abstract

A fast alignment apparatus and method are disclosed. The sorting target data can be divided into a plurality of groups and the time required for sorting the data belonging to each of the divided groups can be calculated and the sorting target data can be re-divided. And, the sorted time can be shortened by rearranging and recombining the re-divided data, respectively.

Description

[0001] APPARATUS FOR QUICK SORTING AND THE METHOD THEREOF [0002]

A fast alignment apparatus and method are disclosed. In particular, a high-speed alignment apparatus that performs sharing and performing a sorting operation performed in a control apparatus is disclosed.

Generally, processing techniques such as sorting are required to use digital data. At this time, as the amount of data increases, the calculation time for data sorting is consumed.

Normally, data sorting time can be reduced by research that increases the computational speed of a computer.

However, in the case of geographical division of terrain data, the amount of data is very large, so there is a difficulty in sufficiently shortening the alignment time only by increasing the computer operation speed.

Accordingly, there is a need for a technique capable of shortening the computation time by distributing computations that are performed in a centralized control apparatus.

The fast sorting method includes: calculating first and second estimated operation times required for sorting a plurality of pieces of data to be sorted by different techniques, respectively; calculating a plurality of sorting times using the first operation estimation time and the second operation estimation time; Dividing the target data into a first group and a second group, calculating a transmission expected time required to transmit data belonging to the second group, recalculating a second calculation expected time based on the estimated transmission time, Dividing the plurality of sorting target data into first and second groups based on the first calculation expected time and the recalculated second calculation expected time, and sorting the sorting target data belonging to the subdivided group can do.

Here, data belonging to the first group is sorted by the CPU, data belonging to the second group is sorted by the GPU.

Extracting an index and an address corresponding to the plurality of pieces of sorting target data, transmitting sorting target data belonging to the sorted second group, and transmitting the sorting target data belonging to the first group arranged based on the extracted index and address And rearranging the data to be sorted belonging to the second group that is aligned with the data to be sorted belonging to the data.

In addition, the recombining step may rearrange the alignment target data belonging to the second group aligned with the alignment target data belonging to the aligned first group using a merge sort.

The fast alignment apparatus may further include a calculation time calculation unit for calculating first and second estimated operation times required to align the plurality of pieces of alignment target data in different techniques, A transmission time calculation unit for calculating a transmission expected time required to transmit sorting target data belonging to the second group, A calculation time recalculation unit for recalculating a second calculation expected time required for sorting the data belonging to the second group, a plurality of sorting target data based on the first calculation expected time and the recalculated second calculation expected time, And a re-division unit for re-division into a second group, and an arrangement unit for arranging the data to be sorted belonging to the re-divided group.

In this case, the calculation time calculation unit may calculate a first calculation expected time required for quick sorting of a plurality of sorting target data in the CPU and a second calculation expected time required for a second sorting target data to be used for bi- It is possible to calculate the estimated operation time.

The computation time recalculation unit may recalculate a second computation expected time required for sorting the data belonging to the second group based on the estimated transmission time.

The high-speed alignment apparatus can process the data alignment operation in parallel in the central control device and the graphic control device, thereby shortening the calculation time.

1 is a diagram showing a configuration of a high-speed alignment apparatus.
2 is a diagram for explaining an operation of extracting indexes and addresses of data to be sorted.
FIG. 3 is a diagram for explaining a definition of sorting and recombining data to be sorted by dividing the data into a plurality of groups. FIG.
Figures 5 and 7 are diagrams provided to illustrate bichronic alignment.
6 is a flowchart provided to explain a method of fast sorting data to be sorted.
Figures 8 and 9 are diagrams provided to illustrate pipeline merge alignment.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a diagram showing a configuration of a high-speed alignment apparatus.

1, the fast alignment apparatus 100 includes an extraction unit 110, a calculation time calculation unit 120, a division unit 125, a transmission time calculation unit 130, a calculation time recalculation unit 140, A display unit 150, an arranging unit 160, a graphics processing unit 170, and a combining unit 180. [

The extracting unit 110 may extract an index and an address corresponding to the data to be sorted.

For example, the data to be sorted can be stored in a memory of a CPU (Computer Processing Unit) s. Referring to FIG. 2, the extracting unit 110 may extract an address 230 of a memory and an index 220 of each of the data stored in the data to be aligned 210.

The calculation time calculation unit 120 may calculate first and second estimated operation times required for sorting a plurality of sorting target data by different techniques.

For example, the calculation time calculator 120 may calculate a first calculation expected time (CPU _t ) required for quick sorting a plurality of sorting target data in the CPU using the following equation (1). Also, the calculation time calculation unit 120 may calculate a second calculation expected time (GPU _t ) required for bi-tonic sorting the alignment target data in the GPU using the following equation (2).

In Equations (1) and (2), n is the number of data to be sorted.

In this case, when calculating the first calculation expected time using the Quick Sort, the calculation time calculation unit 120 calculates the time complexity (nlogn (QuickSort)) of the quick alignment by flops (Floating-point Operations Per Second) The first calculation expected time can be calculated.

In addition, when calculating the second calculation expected time using the biphonic sort, the calculation time calculation unit 120 divides the time complexity (nlogn (n ² ) (BitonicSort)) of the bionic arrangement by flops The second calculation expected time can be calculated.

The division unit 125 may divide the sorting target data into first and second groups using the first and second estimated operation time.

At this time, the partitioning unit 125 may divide the data to be sorted into a first group for performing sorting in the CPU and a second group for performing sorting in the graphic processing unit 170. [

For example, the dividing unit 125 may calculate the size of the divided first group among the data to be sorted by using Equation (3) below.

Here, Array is the total number of data to be sorted, and CPU _array is the size of the first group.

Also, the partitioning unit 125 can calculate the size of the second group (GPU _array ) as a difference between the total number of arrays to be sorted (Array) and the size of the first group (CPU _array ) have. Then, the division unit 125 may divide the data to be sorted into the first and second groups based on the calculated size of the first and second groups.

The transmission time calculation unit 130 may calculate a transmission expected time required for transmitting data belonging to the divided second group to the graphic processing unit 170 through the bus 10. [

At this time, the transmission time calculation unit 130 may calculate the transmission expected time using the size of the data array belonging to the second group (GPU _array ) and the bus speed of the graphics card (GPUBUS (SPEED)).

At this time, data belonging to the second group can be transmitted to the graphic processing unit 170 through the bus 10. [ Then, the data belonging to the second group arranged in the graphic processing unit 170 can be received by the combining unit 180 through the bus 10. Accordingly, the transmission time calculation unit 130 can calculate the transmission expected time based on the time that the data belonging to the second group is transmitted twice through the bus 10 before and after the alignment.

For example, the transmission time calculation unit 130 calculates the transmission expected time TRANSMIT _t by dividing the size (GPU _array ) of the data array belonging to the second group by the bus speed of the graphics card, as shown in Equation (5) .

The computation time recalculation unit 140 can recalculate the second computation expected time based on the estimated transmission time.

For example, the calculation time recalculation unit 140 may recalculate the second calculation expected time as the sum of the second calculation expected time and the transmission expected time calculated using Equation (2). At this time, the recalculated second calculation expected time (final GPU _t ) may be expressed by Equation (6) below.

The re-division unit 150 divides the sorting target data based on the first calculation estimated time calculated by the calculation time calculating unit 120 and the second calculation estimated time (final GPU _t ) recalculated in the calculation time recalculating unit 140 And can be subdivided into first and second groups.

For example, the re-divider 150 calculates the number of arrays to be sorted (Array), the first calculation estimated time (CPU _t ), and the recalculated second calculation estimated time (final GPU _t ) to recalculate the size of the first group.

Here, Array is the total number of data to be sorted, and CPU _array is the size of the first group.

Also, the re-divider 150 can recalculate the size of the second group by a difference between the total number of arrays to be sorted (Array) and the recalculated size of the first group (CPU _array ) have.

Then, the re-division unit 150 may re-divide the data to be sorted into the first and second groups based on the size of the re-computed first and second groups.

3, if the sizes of the first and second groups are recalculated to 4, the re-divider 150 divides the data 4, 6, 2, and 8 into the first group 231, the data 3, 1, 7, 5 can be subdivided into a second group 232

At this time, in the case of the bi-tonic sort, optimal efficiency can be obtained for 2 ⁿ matrices. Thus, when performing bi-tonic sorting in the GPU, the subdivider 150 can check whether the calculated second group size (GPU _array ) is squared or not. If the size of the second group is not a power of two, the redistribution unit 150 may adjust the size of the first group (CPU _array ) so that the size of the second group is a power of two.

The re-division unit 150 may transmit the data belonging to the re-divided second group to the graphic processing unit 170 via the bus 10.

The sorting unit 160 may perform quick sorting on the data belonging to the first divided group. The data belonging to the first group can be sorted in the order of 2, 4, 6, and 8 through quick alignment.

The graphic processing unit 170 may perform vitalonic alignment on the data belonging to the second group. Data belonging to the second group can be sorted in order of 1, 3, 5, and 7. For example, in the case of a bi-tonic sorting, the graphic processing unit 170 can vithorise-align the sorting data as shown in FIG.

7, the graphics processing unit 170 may include a plurality of thread blocks 171 and a memory (global memory) 172. In this case, Here, the thread block 171 may include a shared memory 1712 and a plurality of threads 1711.

Then, the graphics processor 170 may control the thread blocks to divide the data belonging to the second group into a plurality of arrays of a two-dimensional mesh shape to perform vitalonic alignment. At this time, the graphic processing unit 170 can simultaneously control the thread blocks 171 using the following Equation (9).

Here, Thread _max is the maximum number of threads that can be accommodated in one thread block of, M _col is the number of columns of the two-dimensional mesh, M _row is the number of rows of the two-dimensional mesh, SharedMemory _size is the size of the shared memory, Element _size Is the size of an Element contained in a floating-point 2D array.

For example, if the thread _max is 256, the SharedMemory _size is 512 Kbytes, and the Element _size is 8 bytes, a 16 × 16 mesh can be a unit. In this case, when the number of elements in the entire array, that is, the number of data belonging to the second group is 1024, the graphic processing unit 170 divides the data belonging to the second group into four unit meshes, .

For example, the graphic processing unit 170 may divide data belonging to the second group into a plurality of meshes, and perform bi-tonic sorting, as shown in Table 1 below.

At this time, when the mesh is composed of 8x8, the graphic processing unit 170 performs the vitalonic alignment only for the eight data in the horizontal direction and does not perform the vitalonic alignment for the eight data in the vertical direction have.

malloc total _ num _ element  * 2 matrix on the GPU
memcopyHostToDevice ( deviceMatrix , hostMatrix );
alloc threadBolck ( M _col ? M _row ) on the GPU ;

Sort ( float origin _ matrix [ total _ num _ element ][2],
float sorted _matrix [ total _ num _ element ][2]){
Mesh [ M _col ] [ M _row ] = copied matrix from global Memory to shared memory
for  (j = 0; j <M; j + +)
ParallelBitonicSort (Mesh [j] [0])
}

In Table 1, prevMeshCount is a variable used for confirming the division status when the data belonging to the second group is divided into a plurality of meshes.

When the bi-tonic sorting is completed, the graphic processing unit 170 integrates the sorted data using a pipeline merge sort on the data constituting the column as shown in FIG. 8 . In this case, when there are 4 meshes constituted by 8 × 8 and one row of the mesh is assumed to be one subarray, the subarray corresponding to one mesh row may have 8 arrays × 4 = 32.

For example, if the mesh is composed of 8.times.8, the eight data in the horizontal direction are aligned, and the eight data in the vertical direction are not aligned, the data in the vertical direction can be processed as an independent array. That is, as shown in FIG. 8, one row of meshes can be processed into one independent array.

Table 2 below illustrates the pipeline merge alignment of four subarrays for one mesh.

element  A, B = elements from subArray or childNode
if  ( firstStage ) {
A = GetFirstElementFromSubArray (n)
B = GetFirstElementFromSubArray (n + 1)
}
if  (A! = NULL  && B! = NULL ) {
each thread Compare (A, B) / * assume  A is bigger  * /
if  ( lastStage )
PutElementOnGlobalMemory (A)
PutElementOnParentNode (A)
A = NULL
}

According to Table 2, according to Table 2, the graphics processor 170 uses the data constituting the sub-array of the mesh when the thread is the initial stage, and when the thread is the final stage, &Lt; / RTI &gt; in the final alignment array of FIG.

For example, the graphics processor 170 may perform pipeline merge sorting by comparing the data constituting the sub-array on a stage-by-stage basis.

The graphics processing unit 170 may transmit the data belonging to the second group, which is pipeline-merge-aligned, to the combining unit 180 through the bus 10. [ For example, a GPU (Graphic Processing Unit) can be used as the graphic processor 170.

The combining unit 180 combines the data belonging to the first and second groups arranged in the arranging unit 170 and the graphic processing unit 180 using the indexes and addresses of the data to be sorted extracted by the extracting unit 110, can do.

For example, the combining unit 180 may use merge sort to store data sorted by using the extracted index in the corresponding addresses again, as shown in FIG.

More specifically, referring to FIG. 5, the combining unit 180 may compare and sort the data 510 belonging to the sorted first group and the data 520 belonging to the second group, respectively. That is, the combining unit 180 may perform sorting by comparing data 2 belonging to the first group and data 1 belonging to the first group. Then, the combining unit 180 may merge the data 1 into the memory based on the address (& 0009) of the data 1. At this time, the combining unit 180 can improve the performance of the high-speed alignment apparatus by using pipelining in merge alignment.

6 is a flowchart provided to explain a method of fast sorting data to be sorted.

Referring to FIG. 6, the calculation time calculation unit 120 may calculate a first calculation expected time and a second calculation estimated time (S610). Here, the first calculation expected time is a time required for quick sorting of the sorting target data, and the second calculation expected time is a time required for biotonic sorting the sorting target data.

For example, quick alignment may be performed on the CPU, and vitalic alignment may be performed on the GPU. Then, the calculation time calculation unit 120 may calculate the time required for sorting the data by the CPU as the first calculation estimated time. Similarly, the calculation time calculation unit 120 may calculate a time required for sorting the sort target data in the GPU as a second calculation expected time.

The dividing unit 125 may divide the data to be sorted into first and second groups using the first and second estimated operation time (S620). In one example, the first group includes data for which sorting is performed in the CPU, and the second group may include data for which sorting is performed in the GPU.

Then, the transmission time calculation unit 130 may calculate a transmission expected time required to transmit data belonging to the second group (S630). Here, the data belonging to the second group can be transmitted to the device performing the graphic operation. For example, a GPU (Graphic Processing Unit) can be used as an apparatus for performing a graphic operation.

Then, the calculation time recalculation unit 140 can recalculate the second calculation expected time based on the estimated transmission time (S640).

For example, the calculation time calculation unit 140 can recalculate the second calculation expected time (final GPU _t ) as the sum of the transmission estimated time TRANSMIT _t and the second calculated estimated time GPU _t in step S610 have.

Subsequently, the re-division unit 150 may re-divide the sorting target data into first and second groups based on the first calculation expected time and the re-calculated second calculation expected time (S650). At this time, data belonging to the second group can be transmitted to the graphic processing unit 170 through the bus 10. [

Then, the graphic processing unit 170 can sort the data belonging to the second group using the bionic arrangement. At this time, the sorting unit 160 may sort the data belonging to the re-divided first group using quick alignment (S660).

Then, the combining unit 180 may recombine data belonging to the sorted first and second groups using merge sorting (S670).

More specifically, the combining unit 180 may receive data belonging to the second group arranged in the step S660 through the bus 10. [ 5, the combining unit 180 can reassemble data belonging to the first and second groups by referring to the indexes and addresses of the data to be sorted.

In the above description, the quick alignment is used in the CPU and the biphonic alignment is used in the GPU. However, this corresponds to the embodiment, and not only the bionic array in the CPU, quick alignment in the GPU can be used, Other alignments may be used.

In addition, the fast alignment apparatus and method according to embodiments of the present invention include a computer-readable medium including program instructions for performing various computer-implemented operations. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The media may be program instructions that are specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

110:
120: calculation time calculation unit
125: quarter installment
130: transmission time calculation unit
140: calculation time recalculation unit
150: Subdivision
160:
170: Graphics processor
180: Combination unit

Claims (10)

A high-speed alignment method performed by a high-speed alignment apparatus including a calculation time calculation unit, a division unit, a transmission time calculation unit, a calculation time recalculation unit, a subdivision unit, and an alignment unit,
Calculating the first and second estimated operation time required for arranging the plurality of sorting target data in different techniques;
Dividing the plurality of sorting target data into a first group and a second group using the first calculation expected time and the second calculation expected time;
Calculating a transmission expected time required for the transmission time calculation unit to transmit data belonging to the second group;
The operation time recalculation unit recalculating the second operation expected time based on the expected transmission time;
Dividing the plurality of sorting target data into the first and second groups based on the first calculation expected time and the recalculated second calculation expected time; And
Wherein the sorting unit arranges sorting data belonging to the re-divided group
/ RTI &gt; The method according to claim 1,
The high-speed alignment apparatus includes:
Further comprising a graphics processing unit,
The data belonging to the first group are sorted by the CPU by the sorting unit,
Wherein the data belonging to the second group are sorted by the GPU by the graphics processing unit. 3. The method of claim 2,
The high-speed alignment apparatus includes:
An extracting unit, and a combining unit,
Extracting an index and an address corresponding to the plurality of sorting target data;
Transmitting the sorting data belonging to the sorted second group to the graphics processing unit; And
The combination unit recombining the data to be sorted belonging to the sorted first group and the data to be sorted belonging to the sorted second group based on the extracted index and address
Further comprising: The method of claim 3,
Wherein the recombining step comprises:
Wherein the combining unit recombines the data to be sorted belonging to the sorted first group and the data to be sorted belonging to the sorted second group using a MERGE SORT. The method according to claim 1,
Wherein the step of calculating the expected transmission time comprises:
Wherein the transmission time calculation unit calculates the transmission expected time using the size of the divided second group and the bus speed of the GPU. 3. The method of claim 2,
Wherein the calculating step comprises:
Wherein the calculation time calculation unit calculates the first calculation expected time required for quick sorting the plurality of sorting target data and the second calculation used for bitwise sorting the plurality of sorting target data Estimated time is calculated,
Wherein the aligning comprises:
The sorting unit quick-aligns sorting data belonging to the first group,
Wherein the graphics processing unit arranges the sorting data belonging to the second group vigorously. A calculation time calculating unit for calculating first and second estimated operation times required for sorting a plurality of pieces of alignment target data by different techniques;
A dividing unit dividing the plurality of sorting target data into first and second groups using the first calculation expected time and the second calculation expected time;
A transmission time calculation unit for calculating a transmission expected time required to transmit data to be sorted belonging to the second group;
A calculation time recalculation unit for recalculating a second calculation expected time required for sorting the data belonging to the second group based on the transmission expected time;
A subdivision unit for subdividing the plurality of sorting target data into the first and second groups based on the first calculation expected time and the recalculated second calculation expected time; And
And a sorting unit for sorting the sorting target data belonging to the re-
/ RTI &gt; 8. The method of claim 7,
And a graphics processing unit for sorting the sorting target data belonging to the re-divided second group
Further comprising:
The alignment unit may include:
And sorting the data to be sorted belonging to the re-divided first group. 9. The method of claim 8,
An extracting unit for extracting an index and an address corresponding to a plurality of sorting target data, respectively; And
A combination unit for recombining the data to be sorted belonging to the sorted first group and the data to be sorted belonging to the sorted second group based on the extracted index and address,
Further comprising: 8. The method of claim 7,
The calculation time calculator calculates,
Wherein the first calculation expected time required for quick sorting the plurality of sorting target data by the CPU and the second calculation estimated time used for bitwise sorting the plurality of sorting target data in the GPU Lt; / RTI &gt;
The calculation time recalculation unit,
And recalculates the second estimated operation time required for sorting the data belonging to the second group based on the estimated transmission time.

Images