CN110362785A - Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream - Google Patents
Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream Download PDFInfo
- Publication number
- CN110362785A CN110362785A CN201910511081.4A CN201910511081A CN110362785A CN 110362785 A CN110362785 A CN 110362785A CN 201910511081 A CN201910511081 A CN 201910511081A CN 110362785 A CN110362785 A CN 110362785A
- Authority
- CN
- China
- Prior art keywords
- equation
- subregion
- initial
- multifile
- stream
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
- G06F17/12—Simultaneous equations, e.g. systems of linear equations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- General Engineering & Computer Science (AREA)
- Operations Research (AREA)
- Computing Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of any Diagonal Linear equation group Parallel implementation method and system based on multifile stream, including four steps: (1) band-like equation being divided into several initial subregions, and each initial subregion data information is individually stored in corresponding file;(2) each main core individually reads each initial subregion system equation data simultaneously, and respectively from karyomorphism at each assembling subregion system equation, and polycondensation eliminates its built-in variable;(3) each main core Parallel implementation interface equation;(4) each initial subregion solves built-in variable according to each main core interface equation solution result back substitution.It is terminated if iteration terminates, otherwise again since second step.The present invention realizes communication localization to greatest extent on the basis of realizing Distributed Storage, through layered communication, and the design feature that can make full use of high-performance calculation system promotes the parallel efficiency calculation of extensive any Diagonal Linear equation group.
Description
Technical field
The present invention relates to a kind of parallel Diagonal Equation group method for solving, specifically based on any Diagonal Linear of multifile stream
Equation group Parallel implementation method and system.
Background technique
A kind of powerful of the finite element method as engineering numerical analysis, is widely used in aerospace, energy exploration
With the fields such as exploitation, communications and transportation, more and more extensive attention is obtained in engineering design and analysis, it is multiple to have become solution
The effective way of miscellaneous Engineering computing problems.However for large-scale complicated system, since calculation scale is huge, complicated, use is solved
Conventional serial finite element method is difficult to overcome required memory headroom big, solves the problem of time length.Utilize parallel computer
Research and develop corresponding parallel algorithm then for such issues that solution provide practicable method.
In terms of parallel algorithm, be broadly divided into two classes: one kind is formd from the concurrency of Finite Element Equations itself
Domain decomposition parallel calculation method;Another kind of is the solution of time-consuming most system of linear equations from Finite Element Equations, is sought
Seek the parallel calculating method of Efficient Solution system of linear equations.Domain Decomposition Method individually completes the base of net region segmentation at the same time
The equation to be solved that all subregion is formed on plinth eliminates built-in variable by static condensation to reduce interface order of equation number, asks
Solve after the equation of interface all subregion can back substitution simultaneously solve built-in variable, the scale-dependent of interface equation is in the region of grid point
It cuts.However for extensive finite element numerical analysis, with the increase of boundary node number, the scale of interface equation is therewith sharply
Increase, to bring difficulty to solution.And in the parallel calculating method of system of linear equations, traditional parallel calculating method is often
The excessive problem of global communication is faced, thus the increase of the communication overhead that each process generates in solution procedure and synchronization overhead
Also parallel efficiency calculation can be greatly reduced.And in the sparse vectors of practical problem, Diagonal Equation group is then that its is heavy
Want component part.
For from hardware, massively parallel computer uses isomery many-core distributed storage architecture, allusion quotation at present
Type is characterized in: being connected between the different processor in same node by local bus, traffic rate is very high;And it is located at not
It is connected with the processor between node by high performance network, traffic rate is relatively low.Therefore to improve large scale scale heterogeneous many-core
The parallel efficiency calculation of parallel computer reduces different nodes it is necessary to which communication is limited in each intra-node as far as possible as far as possible
Between communication.How to construct suitable parallel algorithm and is allowed to architecture with current large scale scale heterogeneous many-core parallel computer
It is adapted to improve parallel efficiency calculation as problem in the urgent need to address.
Find by prior art documents: Luo etc. is in Jourpal of Computer Research&
Developmept, 2000,37 (7): publish an article " Ap effective parallel algorithm on 802-807
Solvipg tridiagopal " (" a kind of effective Distributed Parallel Algorithm of periodic tridiagonal linear system ", computer research with
Development, 2000,37 (7): 802-807), this article independently forms the system equation of all subregion simultaneously by region segmentation, and passes through
It crosses polycondensation and eliminates each initial subregion built-in variable, obtain the interface equation of only each initial subregion relevant to Boundary Variables
The time is solved to reduce global communication, and then improves the parallel efficiency calculation of system entirety.But this method is in practical applications
Very big limitation is received, is on the one hand since the band-like equation bandwidth of practical problem is not fixed, this method can not solve;Another party
Frequency between different subregions is caused since this method does not fully consider the architecture feature of current domestic high-performance computer in face
Numerous communication, finally obtained parallel efficiency calculation are also not too ideal.
Summary of the invention
In view of the deficiencies of the prior art, the present invention proposes a kind of, and any Diagonal Linear equation group based on multifile stream is parallel
Method for solving and system.
A kind of any Diagonal Linear equation group Parallel implementation method based on multifile stream provided according to the present invention, including
Following steps:
Partitioning step: by band-like equation subdivision being p initial subregions by subregion software, and will each initial subregion
Coefficient matrix, equation right-hand vector individually save in one file, wherein p be individual node machine inner treater nucleus number integer
Times;
Solution procedure: enabling the main core in each Node station read file parallel, from core according to coefficient matrix, the equation in file
Right-hand vector forms system equation, and after eliminating to the built-in variable value of system equation, then back substitution solves the conduct of built-in variable value
Solving result output.
Preferably, the solution procedure includes:
Condensation polymerization step: the corresponding initial subregion of main core parallel reading simultaneously in each Node station for carrying out parallel computation is enabled
The file of file, the initial subregion that the slave core in each Node station is stored by each Node station independently forms each initial sub-district
The system equation in domain, and the initially built-in variable value of subregion is eliminated each by polycondensation, it obtains only containing each of Boundary Variables
The interface equation of initial subregion;
Interface equation solution step: the main core in each Node station is by being in communication with each other and being asked using parallel SuperLU algorithm
Interface equation is solved, the Boundary Variables of each initial subregion are obtained;
Back substitution step: after obtaining the Boundary Variables of each initial subregion, the main core in each Node station will include coefficient square
Data information including the inner equations coefficient and oneself boundary variate-value of battle array gives the slave core in same Node station, from core
It receives data information and extracts from bound variable value at one's side, then back substitution solves built-in variable value;It is exported if iteration terminates as a result, no
Then, then solution is re-executed.
A kind of any Diagonal Linear equation group Parallel implementation system based on multifile stream provided according to the present invention, including
Following module:
Division module: by band-like equation subdivision being p initial subregions by subregion software, and will each initial subregion
Coefficient matrix, equation right-hand vector individually save in one file, wherein p be individual node machine inner treater nucleus number integer
Times;
It solves module: enabling the main core in each Node station read file parallel, from core according to coefficient matrix, the equation in file
Right-hand vector forms system equation, and after eliminating to the built-in variable value of system equation, then back substitution solves the conduct of built-in variable value
Solving result output.
Preferably, the solution module includes:
Polycondensation module: the corresponding initial subregion of main core parallel reading simultaneously in each Node station for carrying out parallel computation is enabled
The file of file, the initial subregion that the slave core in each Node station is stored by each Node station independently forms each initial sub-district
The system equation in domain, and the initially built-in variable value of subregion is eliminated each by polycondensation, it obtains only containing each of Boundary Variables
The interface equation of initial subregion;
Interface equation solution module: the main core in each Node station is by being in communication with each other and being asked using parallel SuperLU algorithm
Interface equation is solved, the Boundary Variables of each initial subregion are obtained;
Back-substitution module: after obtaining the Boundary Variables of each initial subregion, the main core in each Node station will include coefficient square
Data information including the inner equations coefficient and oneself boundary variate-value of battle array gives the slave core in same Node station, from core
It receives data information and extracts from bound variable value at one's side, then back substitution solves built-in variable value;It is exported if iteration terminates as a result, no
Then, then solution is re-executed.
Preferably, the initial subregion is to obtain by subregion software to band-like equation subdivision, and each initial son
The bandwidth of interface equation in region is not influenced by subdivision.
Preferably, the built-in variable value, which refers to, is subordinated to the exclusive variate-value of initial subregion.
Preferably, the Boundary Variables refer to while being subordinated to the variable of two or more initial subregions.
Preferably, the interface equation refer to initial subregion by polycondensation eliminate built-in variable after obtain only with boundary
The related equation of variable, the equivalent coefficient matrix for the initial subregion that interface equation left end obtains for cohesion, right-hand vector
For the equivalent equation right-hand vector for agglomerating obtained initial subregion.
Preferably, when the parallel SuperLU algorithm Parallel implementation interface equation each interface equation equivalent coefficient matrix
With equivalent equation right-hand vector still distributed storage on each Node station, results of intermediate calculations therein is with matrix and vector product
The storage of formal distribution formula.
Preferably, described any its form of Diagonal Linear equation group is as follows:
Each initial subregion, if M, N are respectively the initial row of initial subregion, termination row, according to the diagonal side of the meaning
Journey group dotted line, which divides, constitutes Block matrix { Ai,j, i=1 ... 5;J=1 ... 5;
The polycondensation cancellation is obtained by following formula:
Wherein, FIAnd FIIFor equivalent equation right-hand vector,
Compared with prior art, the invention has the following advantages that
1, the data information of initial subregion can be stored on corresponding each Node station by way of multifile stream,
To utilize realization storage localization the characteristics of massively parallel computer distributed storage;
2, on the basis of periodic tridiagonal linear system solves, the solution of any Diagonal Linear equation group is extended to;Interface
The scale of equation is not influenced by grid division, only related with the bandwidth of equation;
3, since assembling and back substitution process solution procedure carry out in same Node station, a large amount of local communications can be concentrated
Inside each Node station, the higher advantage of traffic rate in massively parallel processor (MPP) Node station is given full play to;
4, all in same Node station will form top layer after the assembling of initial subregion from core and assemble subregion,
Node each in this way participates in global communication with regard to only one main core and solves the interface equation ultimately generated, to reduce as far as possible
Global communication between different Node stations.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 is that the present invention is based on any Diagonal Linear equation group Parallel implementation flow charts of multifile stream;
Fig. 2 is the level-one subregion organigram in the embodiment of the present invention.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, various modifications and improvements can be made.These belong to the present invention
Protection scope.
The present invention realizes communication part to greatest extent on the basis of realizing Distributed Storage, through layered communication
It is parallel that the characteristics of changing, can make full use of current large scale scale heterogeneous many-core concurrent computer architecture promotes extensive finite element
Computational efficiency.
Concrete operations, which include: (1), is divided into several initial subregions for band-like equation, and will each initial sub-district numeric field data
Information is individually stored in corresponding file;(2) each main core simultaneously individually read each initial subregion system equation data, respectively from
Karyomorphism is at each assembling subregion system equation, and polycondensation eliminates its built-in variable;(3) each main core Parallel implementation interface equation;(4)
Each initial subregion solves built-in variable according to each main core interface equation solution result back substitution.It is terminated if iteration terminates, otherwise
Again since (2).
In specific implementation, any Diagonal Linear equation group Parallel implementation method based on multifile stream includes the following steps:
Band-like equation subdivision is p initial subregions by subregion software by the first step, and by each initial subregion
Coefficient matrix, equation right-hand vector information individually save in one file, and wherein p is the whole of individual node machine inner treater nucleus number
Several times;
The initial subregion is all subregions obtained by subregion software to the direct subdivision of band-like equation, this is a little
Equation scale in area is roughly equal, and the bandwidth of interface equation is not influenced by subdivision.
Second step carries out the corresponding initial subregion the file information of each main core parallel reading simultaneously of parallel computation, Mei Gejie
In point machine each is by what the initial subregion the file information that each Node station stores independently formed each initial subregion from core
It unites equation, and eliminates each initial subregion built-in variable by polycondensation, obtain only relevant to equation bandwidth only containing boundary
Each initial sub-region interface equation of variable;
The built-in variable, which refers to, is subordinated to the built-in variable that the exclusive variable of a sub-regions is known as subregion;Boundary becomes
Amount, which refers to while being subordinated to the variables of two or more subregions, is known as the Boundary Variables of subregion.
The interface equation refers to that subregion is only related with Boundary Variables by obtaining after polycondensation cancellation built-in variable
Equation, the subregion equivalent coefficient matrix that interface equation left end obtains for cohesion, right-hand vector are the sub-district that cohesion obtains
Domain equivalent equation right-hand vector.
Main core in each Node station of third step is by being in communication with each other and solving assembling sub-district using SuperLU_DIST algorithm
Domain interface equation;
Any Diagonal Linear equation group Parallel implementation method based on multifile stream, the SuperLU_DIST are calculated
The equivalent coefficient matrix of each interface equation and equivalent equation right-hand vector still distributed storage are each when method Parallel implementation interface equation
On a Node station, results of intermediate calculations is also with the storage of the formal distribution formula of matrix and vector product.
4th step, after obtaining all subregion Boundary Variables, main core in each Node station will comprising inner equations coefficient and
Required result including Boundary Variables value is sent to the slave core in same Node station, extracts itself in Cong Hecong reception result
Boundary Variables value, then back substitution solve built-in variable value;It is exported if iteration terminates as a result, otherwise being held since second step again
Row.
As shown in Figure 1, any Diagonal Linear equation group Parallel implementation method based on multifile stream integrally specifically includes that head
First structural finite element model subdivision is p initial subregions by sectorized software, and by the coefficient matrix of each initial subregion
It is individually saved in one file with equation right-hand vector information;Then each process reads initial subregion the file information parallel simultaneously
The system equation of each initial subregion is formed, and eliminates its built-in variable by polycondensation;Then it is calculated using SuperLU_DIST
Method solves assembling sub-region interface equation;Finally, required result to be sent to the initial sub-district of bottom in same Node station
Domain, each initial subregion of bottom extract from bound variable value at one's side from reception result, then back substitution solves built-in variable value.
In this embodiment, consider that bandwidth is the n rank Band system of 2m+1, form is as follows
The hardware condition of parallel computation are as follows: starting p/64 platform Node station, each Node station is interior to have 65 processor cores,
It is criticized comprising 1 main core and 64 from by high performance network interconnection, main core between core, Node station and from dma mode is passed through between core
Amount access main memory.
It is successively described in detail below as the sequencing of system solution:
The first step, using subregion software by the n rank square matrix in formula (1) according to p region mean allocation, and will be each initial
The coefficient matrix of subregion, equation right-hand vector information individually save in one file.The band-like equation of completion as shown in formula (2)
Subregion:
The initial subregion of p that the subdivision as shown in formula (2) generates is respectively as follows: initial subregion 0, initial subregion
1 ... ..., initial subregion p.
Second step numbers 1 main core in Node station for mi, i is initial subarea number;By 64 in Node station
It is numbered respectively from core are as follows: p0+i╳64、p1+i╳64、……、p63+i╳64, it is a initial then the p for being 0 to p-1 will to be numbered shown in formula (2)
Subregion is respectively matched to number with it and individually handle on identical p Node station.Each node owner core is parallel simultaneously to read this
Node station corresponding 64 initial subregion the file informations form the system equation of each initial subregion, then use Gauss
The elimination to sub- coefficient of region matrix carry out decompose realize polycondensation process to eliminate its built-in variable, obtain only with Boundary Variables phase
The interface equation of each initial subregion closed.Each subregion, system equation can divide according to dotted line shown in Fig. 2 and constitute block
Shape matrix, as shown in formula (3), M, N are respectively the initial row in the region, termination row:
{Ai,j, i=1 ... 5;J=1 ... 5 (3)
For each subregion, corresponding reduction equation are as follows:
In formula, FIAnd FIIFor equivalent equation right-hand vector, can be calculated according to the following formula:
Third step, core m0, m1... ..., mp, it is in communication with each other and solves assembling sub-region interface using SuperLU_DIST algorithm
Equation obtains Boundary Variables.
4th step, core miIt include its built-in variable and Boundary Variables by result required by subregion 0+64 × i~63+64 × i
It is sent to initial subregion 0+64 × i in Node station i~initial subregion 63+64 × i, initial subregion 0+64 × i~just
Beginning, subregion 63+64 × i extracted from bound variable at one's side from reception result respectively, then can be with solver built-in variable.
On specific embodiments of the present invention are described.It is to be appreciated that the invention is not limited to above-mentioned spies
Determine embodiment, those skilled in the art can make various deformations or amendments within the scope of the claims, this has no effect on
Substantive content of the invention.
Claims (10)
1. a kind of any Diagonal Linear equation group Parallel implementation method based on multifile stream, which is characterized in that including walking as follows
It is rapid:
Partitioning step: passing through subregion software for band-like equation subdivision is p initial subregions, and is by each initial subregion
Matrix number, equation right-hand vector individually save in one file, and wherein p is the integral multiple of individual node machine inner treater nucleus number;
Solution procedure: enabling the main core in each Node station read file parallel, from core according to coefficient matrix, the equation right end in file
Item forms system equation, and after eliminating to the built-in variable value of system equation, then back substitution solves built-in variable value as solution
As a result it exports.
2. any Diagonal Linear equation group Parallel implementation method according to claim 1 based on multifile stream, feature
It is, the solution procedure includes:
Condensation polymerization step: the text for reading corresponding initial subregion parallel simultaneously of the main core in each Node station for carrying out parallel computation is enabled
The file of part, the initial subregion that the slave core in each Node station is stored by each Node station independently forms each initial subregion
System equation, and eliminate by polycondensation the built-in variable value of each initial subregion, obtain only containing Boundary Variables it is each just
The interface equation of beginning subregion;
Interface equation solution step: the main core in each Node station is by being in communication with each other and solving boundary using parallel SuperLU algorithm
Face equation obtains the Boundary Variables of each initial subregion;
Back substitution step: after obtaining the Boundary Variables of each initial subregion, the main core in each Node station will include coefficient matrix
Data information including inner equations coefficient and oneself boundary variate-value gives the slave core in same Node station, receives from core
Data information simultaneously extracts from bound variable value at one's side, then back substitution solves built-in variable value;It is exported if iteration terminates as a result, otherwise,
Then re-execute solution.
3. a kind of any Diagonal Linear equation group Parallel implementation system based on multifile stream, which is characterized in that including such as lower die
Block:
Division module: passing through subregion software for band-like equation subdivision is p initial subregions, and is by each initial subregion
Matrix number, equation right-hand vector individually save in one file, and wherein p is the integral multiple of individual node machine inner treater nucleus number;
It solves module: enabling the main core in each Node station read file parallel, from core according to coefficient matrix, the equation right end in file
Item forms system equation, and after eliminating to the built-in variable value of system equation, then back substitution solves built-in variable value as solution
As a result it exports.
4. any Diagonal Linear equation group Parallel implementation system according to claim 3 based on multifile stream, feature
It is, the solution module includes:
Polycondensation module: the text for reading corresponding initial subregion parallel simultaneously of the main core in each Node station for carrying out parallel computation is enabled
The file of part, the initial subregion that the slave core in each Node station is stored by each Node station independently forms each initial subregion
System equation, and eliminate by polycondensation the built-in variable value of each initial subregion, obtain only containing Boundary Variables it is each just
The interface equation of beginning subregion;
Interface equation solution module: the main core in each Node station is by being in communication with each other and solving boundary using parallel SuperLU algorithm
Face equation obtains the Boundary Variables of each initial subregion;
Back-substitution module: after obtaining the Boundary Variables of each initial subregion, the main core in each Node station will include coefficient matrix
Data information including inner equations coefficient and oneself boundary variate-value gives the slave core in same Node station, receives from core
Data information simultaneously extracts from bound variable value at one's side, then back substitution solves built-in variable value;It is exported if iteration terminates as a result, otherwise,
Then re-execute solution.
5. any Diagonal Linear equation group Parallel implementation method or right according to claim 1 based on multifile stream
It is required that any Diagonal Linear equation group Parallel implementation system described in 3 based on multifile stream, which is characterized in that the initial son
Region be band-like equation subdivision is obtained by subregion software, and the bandwidth of the interface equation in each initial subregion not by
Subdivision influences.
6. any Diagonal Linear equation group Parallel implementation method or right according to claim 1 based on multifile stream
It is required that any Diagonal Linear equation group Parallel implementation system described in 3 based on multifile stream, which is characterized in that the internal change
Magnitude, which refers to, is subordinated to the exclusive variate-value of initial subregion.
7. any Diagonal Linear equation group Parallel implementation method or right according to claim 2 based on multifile stream
It is required that any Diagonal Linear equation group Parallel implementation system described in 4 based on multifile stream, which is characterized in that the boundary becomes
Amount refers to while being subordinated to the variable of two or more initial subregions.
8. any Diagonal Linear equation group Parallel implementation method or right according to claim 2 based on multifile stream
It is required that any Diagonal Linear equation group Parallel implementation system described in 4 based on multifile stream, which is characterized in that the interface side
Journey refers to that initial subregion eliminates the equation only related with Boundary Variables obtained after built-in variable, the interface side by polycondensation
Cheng Zuoduan are the equivalent coefficient matrix of initial subregion that cohesion obtains, right-hand vector be agglomerate obtained initial subregion etc.
Efficacious prescriptions journey right-hand vector.
9. any Diagonal Linear equation group Parallel implementation method or right according to claim 2 based on multifile stream
It is required that any Diagonal Linear equation group Parallel implementation system described in 4 based on multifile stream, which is characterized in that described parallel
The equivalent coefficient matrix of each interface equation and equivalent equation right-hand vector are still distributed when SuperLU algorithm Parallel implementation interface equation
It is stored on each Node station, results of intermediate calculations therein is with the storage of the formal distribution formula of matrix and vector product.
10. any Diagonal Linear equation group Parallel implementation method or power according to claim 2 based on multifile stream
Benefit require 4 described in any Diagonal Linear equation group Parallel implementation system based on multifile stream, which is characterized in that it is described any
Its form of Diagonal Linear equation group is as follows:
Each initial subregion, if M, N are respectively the initial row of initial subregion, termination row, according to the meaning Diagonal Equation group
Dotted line, which divides, constitutes Block matrix { Ai,j, i=1 ... 5;J=1 ... 5;
The polycondensation cancellation is obtained by following formula:
Wherein, FIAnd FIIFor equivalent equation right-hand vector,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910511081.4A CN110362785A (en) | 2019-06-13 | 2019-06-13 | Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910511081.4A CN110362785A (en) | 2019-06-13 | 2019-06-13 | Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110362785A true CN110362785A (en) | 2019-10-22 |
Family
ID=68217234
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910511081.4A Pending CN110362785A (en) | 2019-06-13 | 2019-06-13 | Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110362785A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111651208A (en) * | 2020-05-08 | 2020-09-11 | 上海交通大学 | Modal parallel computing method and system for heterogeneous many-core parallel computer |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324850A (en) * | 2013-06-21 | 2013-09-25 | 上海交通大学 | Finite element two-stage partition and twice polycondensation parallel method based on multiple document flows |
CN103914428A (en) * | 2014-03-25 | 2014-07-09 | 上海交通大学 | Efficient communication method of structural analysis under multi-core distributed computing environment |
CN108763151A (en) * | 2018-04-12 | 2018-11-06 | 山东科技大学 | A kind of inverse kinematics method in arbitrary three joint |
US20190026865A1 (en) * | 2017-11-08 | 2019-01-24 | Intel Corporation | Methods and apparatus to enable parallel processing when solving linear equations in a computer vision processing system |
-
2019
- 2019-06-13 CN CN201910511081.4A patent/CN110362785A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103324850A (en) * | 2013-06-21 | 2013-09-25 | 上海交通大学 | Finite element two-stage partition and twice polycondensation parallel method based on multiple document flows |
CN103914428A (en) * | 2014-03-25 | 2014-07-09 | 上海交通大学 | Efficient communication method of structural analysis under multi-core distributed computing environment |
US20190026865A1 (en) * | 2017-11-08 | 2019-01-24 | Intel Corporation | Methods and apparatus to enable parallel processing when solving linear equations in a computer vision processing system |
CN108763151A (en) * | 2018-04-12 | 2018-11-06 | 山东科技大学 | A kind of inverse kinematics method in arbitrary three joint |
Non-Patent Citations (2)
Title |
---|
ZHIJIAN DUAN: "Parallel Row Operation Algorithm for Banded Linear Systems", 《IEEE INTERNATIONAL CONFERENCE ON COMPUTER SCIENCE AND AUTOMATION ENGINEERING》 * |
骆志刚: "三对角线性方程组的一种有效分布式并行算法", 《计算机研究与发展》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111651208A (en) * | 2020-05-08 | 2020-09-11 | 上海交通大学 | Modal parallel computing method and system for heterogeneous many-core parallel computer |
CN111651208B (en) * | 2020-05-08 | 2023-06-20 | 上海交通大学 | Modal parallel computing method and system for heterogeneous many-core parallel computer |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Aloise et al. | Column generation algorithms for exact modularity maximization in networks | |
CN110059067A (en) | A kind of water conservancy space vector big data memory management method | |
Lustig et al. | Interior point methods for linear programming: Computational state of the art | |
Zhao et al. | Geographical information system parallelization for spatial big data processing: a review | |
Lawler | The quadratic assignment problem | |
Hosseini et al. | Combined heat and power economic dispatch by mesh adaptive direct search algorithm | |
CN103324850B (en) | Twice polycondensation parallel method of finite element two-stage subregion based on multifile stream | |
CN103488662A (en) | Clustering method and system of parallelized self-organizing mapping neural network based on graphic processing unit | |
Curien et al. | Percolation on random triangulations and stable looptrees | |
CN109902804A (en) | A kind of convolution algorithm method and device | |
CN107451097B (en) | High-performance implementation method of multi-dimensional FFT on domestic Shenwei 26010 multi-core processor | |
CN103326351A (en) | Distribution type parallel load flow calculation method based on hierarchical clustering automatic partition | |
CN108228970A (en) | The explicit asynchronous long parallel calculating method of structural dynamical model | |
Liu et al. | Graph computation and its applications in smart grid | |
CN110163333A (en) | The parallel optimization method of convolutional neural networks | |
CN104572109A (en) | Two-stage partitioned two-time polycondensation parallel computing system development method and parallel computing system | |
CN110473073A (en) | The method and device that linear weighted function is recommended | |
CN110362785A (en) | Any Diagonal Linear equation group Parallel implementation method and system based on multifile stream | |
Zang et al. | A parallel variable neighborhood search for solving covering salesman problem | |
Chen et al. | A parallel approximate ss-elm algorithm based on mapreduce for large-scale datasets | |
CN102722472A (en) | Complex matrix optimizing method | |
Fukazawa et al. | Performance measurement of magnetohydrodynamic code for space plasma on the various scalar-type supercomputer systems | |
Yang et al. | Nested Reduction Algorithms for Generating a Rank-Minimized ℋ 2-Matrix From FMM for Electrically Large Analysis | |
Hu et al. | Vis-top: Visual transformer overlay processor | |
CN106330559B (en) | Complex network topologies calculation of characteristic parameters method and system based on MapReduce |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191022 |