CN109522126A - Data optimization methods and device towards Thread-Level Parallelism in shared drive coenocytism - Google Patents
Data optimization methods and device towards Thread-Level Parallelism in shared drive coenocytism Download PDFInfo
- Publication number
- CN109522126A CN109522126A CN201811376636.0A CN201811376636A CN109522126A CN 109522126 A CN109522126 A CN 109522126A CN 201811376636 A CN201811376636 A CN 201811376636A CN 109522126 A CN109522126 A CN 109522126A
- Authority
- CN
- China
- Prior art keywords
- privatization
- array
- data
- scalar
- coenocytism
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/5018—Thread allocation
Abstract
The invention belongs to computer coenocytism parallel optimization technique fields, in particular to a kind of data optimization methods and device towards Thread-Level Parallelism in shared drive coenocytism, this method includes: being directed to data to be optimized, detecting and determining that its all circulation carries to rely on whether there is the advantageous implementations of privatization；And determined according to detection as a result, implementing data privatization processing.The present invention carries out the profitable of privatization processing to data first and detects for parallel thread is recycled in multicore hardware configuration, and circulation is then eliminated by data privatization and carries the multi-threaded parallel for relying on, and then realizing program；And scalar privatization and array privatization further are divided into data privatization, make full use of the multicore hardware configuration advantage of shared storage, improve thread execution efficiency, and improve the utilization rate of processor or even computer system performance, operation expense is reduced, there is important directive significance to shared drive coenocytism Thread-Level Parallelism processing technique.
Description
Technical field
The invention belongs to computer coenocytism parallel optimization technique fields, in particular to one kind is towards shared drive multicore
The data optimization methods and device of Thread-Level Parallelism in structure.
Background technique
Currently, parallel architecture has not been the privilege of supercomputer, and modern processors generally use coenocytism
To obtain higher performance.In order to make full use of the hardware performance of this coenocytism, need to develop effective Thread-Level Parallelism
Program.OpenMP programming model is the effective means of Thread-Level Parallelism.The execution model of OpenMP is as shown in Fig. 2, starting to hold
When row, only main thread is being run, and main thread in the process of running, when encountering the region for needing to carry out parallel computation, is sent
(Fork creates new thread or wakes up existing thread) multiple threads are born to execute parallel task, when parallel execute,
Main thread and derivation thread work together, and after parallel end of extent execution, derive from thread and exit or hang up, no longer work, control
Process processed returns in individual main thread (Join, i.e. multithreading converge).Circulation is realized with can be convenient by OpenMP
Multi-threaded parallel, that is, different loop iterations is assigned to different threads and is executed up.Sequence is executed due to per thread
Be it is random, be equal to upset original iteration execute sequence, this requires that cannot have any dependence between each iteration,
The subsequent dependence by between iteration, which is referred to as to recycle to carry, to be relied on.Any circulation, which carries the presence relied on, will prevent the parallel of program
Change, and then influences thread execution efficiency, coenocytism hardware runnability and expense.
Summary of the invention
For this purpose, the present invention provides a kind of data optimization methods and dress towards Thread-Level Parallelism in shared drive coenocytism
It sets, makes full use of the hardware performance of coenocytism, meet the multi-threaded parallel demand of circulation, realize the thread-level of loop-around data simultaneously
The validity that row executes improves multi-core processor data execution efficiency.
According to design scheme provided by the present invention, a kind of data towards Thread-Level Parallelism in shared drive coenocytism
Optimization method includes: being directed to circulation to be optimized, detecting and determining that its all circulation carries to rely on whether there is privatization profitability
Situation；And determined according to detection as a result, implementing data privatization processing.
Above-mentioned, can the advantageous implementations of privatization refer to that all circulations carry and rely on and be carried out by data privatization
It eliminates.
Above-mentioned, it detects and determines that all circulations carry the advantageous implementations of privatization relied on, include following content:
For the target circulation layer of data to be optimized, to wherein each scalar, if each in the target circulation layer
Reference carries out assignment by current iteration, then determines to rely on privatization processing caused by the scalar, otherwise, it is determined that the scalar is not necessarily to
Data privatization processing；To wherein each array, if the array rely on be carried by the circulation layer inside target circulation layer, and
Each reference of each array element in target circulation stacking generation is detected by current iteration assignment, then determines that the array is drawn
The dependence privatization processing risen, otherwise, it is determined that the array is not necessarily to data privatization.
Preferably, foundation detection determines as a result, dependence privatization processing caused by progress scalar, includes following content:
For dependence caused by the scalar handled to privatization, first every sentence in traversal loop body, searches and collect
It is wherein directed to the write operation of the scalar, and is added it in set W, if set W is sky, without carrying out scalar privatization,
Otherwise, traversal set W detects whether it is specification variable, if so, by specification for each scalar being related in set
Change of variable is induction variable；Set W is traversed, privatization processing is carried out to each scalar being related in set.
Further, array or privatization clause modification are extended using scalar to carry out privatization processing to scalar.
Preferably, foundation detection determines as a result, dependence privatization processing caused by progress array, includes following content:
For dependence caused by the array handled to privatization, array of traversal loop weight every first accesses sentence,
The write operation for being wherein directed to array is searched and collected, is added it in set AW；If set AW is sky, without carrying out array
Otherwise privatization traverses set AW, for each array reference being related in set, detect it with the presence or absence of specification, if depositing
It is being then to conclude by reduction transformation；Set AW is traversed, privatization processing is carried out to each array being related in set.
Further, privatization processing is carried out to array using array dimension extension or privatization clause modification.
A kind of data-optimized device towards Thread-Level Parallelism in shared drive coenocytism includes test module and optimization
Module, wherein
Test module, for being directed to data to be optimized, detecting and determining that its all circulation carries dependence with the presence or absence of privately owned
Change advantageous implementations, determines that the situation for needing privatization to handle is fed back to optimization module by result according to detection；
Optimization module, for implementing data privatization processing according to test module feedback result.
In above-mentioned device, test module includes detection sub-module one and detection sub-module two, wherein
Detection sub-module one, for being directed to the target circulation layer of data to be optimized, to wherein each scalar, if in the mesh
The each reference marked in circulation layer carries out assignment by current iteration, then determines to rely on privatization processing caused by the scalar, no
Then, determine that the scalar is handled without data privatization；
Detection sub-module two, for being directed to the target circulation layer of data to be optimized, to wherein each array, if the array
Dependence is to be carried by the circulation layer inside target circulation layer, and it is every in target circulation stacking generation to detect each array element
Secondary reference is then determined to rely on privatization processing caused by the array, otherwise, it is determined that the array is without number by current iteration assignment
According to privatization.
In above-mentioned device, optimization module includes optimization submodule one and optimization submodule two, wherein
Optimize submodule one, for according to feedback result, for dependence caused by the scalar handled to privatization, first time
Every sentence in loop body is gone through, search and collects the write operation for being wherein directed to the scalar, and is added it in set W, if collection
Closing W is sky, then without carrying out scalar privatization, otherwise, traverses set W for each scalar being related in set and detect it
It whether is specification variable, if so, being induction variable by specification change of variable；Set W is traversed, to each of being related in set
Scalar carries out privatization processing；
Optimize submodule two, for according to feedback result, for dependence caused by the array handled to privatization, first time
The circulation array of weight every access sentence is gone through, searches and collects the write operation for being wherein directed to array, add it to set AW
In；If set AW is sky, without carrying out array privatization, otherwise, set AW is traversed, for the every number being related in set
Group reference detects it with the presence or absence of specification, and if it exists, is then to conclude by reduction transformation；Set AW is traversed, to being related in set
Each array carry out privatization processing.
Beneficial effects of the present invention:
The present invention carries out the profitability of privatization processing for parallel thread is recycled in multicore hardware configuration to data first
Then detection is eliminated circulation by data privatization and carries the multi-threaded parallel for relying on, and then realizing program；And further will
Scalar privatization and array privatization are divided into data privatization, are made full use of the multicore hardware configuration advantage of shared storage, are mentioned
High thread execution efficiency, and the utilization rate of processor or even computer system performance is improved, operation expense is reduced, to shared drive
Coenocytism Thread-Level Parallelism processing technique has important directive significance.
Detailed description of the invention:
Fig. 1 is data optimization methods flow diagram in embodiment；
Fig. 2 is OpenMP Parallel Execution Model schematic diagram in embodiment；
Fig. 3 is data-optimized schematic device in embodiment；
Fig. 4 is test module schematic diagram in embodiment；
Fig. 5 is optimization module schematic diagram in embodiment.
Specific embodiment:
To make the object, technical solutions and advantages of the present invention clearer, understand, with reference to the accompanying drawing with technical solution pair
The present invention is described in further detail.
The execution model of OpenMP is as shown in Fig. 2, when starting to execute, and only main thread is being run, and main thread exists
In operational process, when encountering the region for needing to carry out parallel computation, derive (Fork, create new thread or wake up it is wired
Journey) multiple threads execute parallel task, and it when parallel execute, main thread and derives from thread and works together, in parallel area
Terminate execute after, derive from thread exit or hang up, no longer work, control flow return in individual main thread (Join, i.e., it is more
Thread converges).The multi-threaded parallel of circulation is realized by OpenMP with can be convenient.In circulation multi-threaded parallel, due to
Per thread execution sequence be it is random, carry this requires cannot have circulation between each iteration and rely on, otherwise can prevent journey
The parallelization of sequence, and then the runnability for influencing execution efficiency and multi-core processor etc..In consideration of it, the embodiment of the present invention, referring to
Shown in Fig. 1, a kind of data optimization methods towards Thread-Level Parallelism in shared drive coenocytism are provided, include following content:
S101, data to be optimized are directed to, detecting and determining that its all circulation carries to rely on whether there is privatization profitability
Situation；
S102, determined according to detection as a result, implementing data privatization processing.
In the embodiment of the present invention, whether the test of progress privatization profitability is first can eliminate institute when judging privatization
Have and hinder parallel dependence, then implements to rely on by data privatization for test result and eliminate work, abolish between iteration
Dependence, realize the Thread-Level Parallelism of circulation, make full use of the hardware runnability of coenocytism, improve program and execute effect
Rate.
When detecting and determining that all circulations carry the advantageous implementations of privatizations relied on, in further embodiment of the present invention,
The detection decision process is designed as comprising following content:
For the target circulation layer of data to be optimized, to wherein each scalar, if each in the target circulation layer
Reference carries out assignment by current iteration, then determines to rely on privatization processing caused by the scalar, otherwise, it is determined that the scalar is not necessarily to
Data privatization processing；To wherein each array, if the array rely on be carried by the circulation layer inside target circulation layer, and
Each reference of each array element in target circulation stacking generation is detected by current iteration assignment, then determines that the array is drawn
The dependence privatization processing risen, otherwise, it is determined that the array is not necessarily to data privatization.
Because data privatization can consume memory headroom, the effect that might not have been brought.Therefore data privatization it
Before, it detects whether that all circulations carry and relies on and can be eliminated by data privatization；If it can, so continuing to implement
Otherwise real data privatization stops data privatization processing, avoids unnecessary program executive overhead.Test scalar draws
The carrying risen relies on test, the circulation layer to parallelization is known as target circulation layer, for each scalar, if followed in target
Quoting in circular layer iteration each time is all the value calculated by current iteration, then relying on caused by declared amount can lead to
Privatization is crossed to eliminate, and continues profitable test in next step.For each array rely on, need to check carrying according to
Circulation layer to parallelization is known as target circulation layer by bad circulation layer, and there are three kinds of situations: (1) relying on is by target circulation
What the circulation layer except layer carried；(2) rely on is carried by target circulation layer；(3) relying on is by inside target circulation layer
What circulation layer carried；For the first case, it is carried due to relying on by outer loop, target circulation layer is relied on without carrying, array
Privatization is not needed；For second situation, dependence is carried by target circulation layer, if by the dependence followed by target
What circular layer carried, then the dependence can not be eliminated by array privatization, profitable test crash；For the third situation,
If it is what is carried by the circulation layer inside target circulation layer, each array element is continued to test, if it is in target circulation
The value quoted each time in stacking generation is calculated by current iteration, then the dependence can be disappeared by privatization
It removes, otherwise, can not eliminate, profitable test crash.
According to detection judgement as a result, carrying out relying on privatization processing caused by scalar, in another embodiment of the present invention, it is somebody's turn to do
Scalar privatization processing includes following content:
For dependence caused by the scalar handled to privatization, first every sentence in traversal loop body, searches and collect
It is wherein directed to the write operation of the scalar, and is added it in set W, if set W is sky, without carrying out scalar privatization,
Otherwise, traversal set W detects whether it is specification variable, if so, by specification for each scalar being related in set
Change of variable is induction variable；Set W is traversed, privatization processing is carried out to each scalar being related in set.
Further, array is extended using scalar or privatization clause modifies, such as private, scalar carried out private
Having processing.
According to detection judgement as a result, carrying out relying on privatization processing caused by array, in another embodiment of the present invention, it is somebody's turn to do
Array privatization processing includes following content:
For dependence caused by the array handled to privatization, array of traversal loop weight every first accesses sentence,
The write operation for being wherein directed to array is searched and collected, is added it in set AW；If set AW is sky, without carrying out array
Otherwise privatization traverses set AW, for each array reference being related in set, detect it with the presence or absence of specification, if depositing
It is being then to conclude by reduction transformation；Set AW is traversed, privatization processing is carried out to each array being related in set.
Further, it is modified, such as private, array is carried out private using array dimension extension or privatization clause
Having processing.
When array detects whether as specification, can by determine its whether be " ARRAY [i]=ARRAY [i] op XX " shape
Reduction transformation is to conclude if there is specification caused by array by formula.For the array of each privatization, if array element
Read operation is only existed in loop body, then each array element being related to is initialized, and the array of each privatization is silent
Recognizing value is all 0, the situation for avoiding may not being inconsistent with initial value etc., and then influences executing efficiency.
Based on above-mentioned data optimization methods, the embodiment of the present invention also provides one kind towards shared drive coenocytism middle line
The parallel data-optimized device of journey grade, it is shown in Figure 3, it include test module 101 and optimization module 102, wherein
Test module 101, for being directed to data to be optimized, detecting and determining that its all circulation carries to rely on whether there is private
The advantageous implementations of having determine that the situation for needing privatization to handle is fed back to optimization module by result according to detection；
Optimization module 102, for implementing data privatization processing according to test module feedback result.
Shown in Figure 4 in above-mentioned device, test module 101 includes detection sub-module 1 and detection sub-module
2 1002, wherein
Detection sub-module 1, for being directed to the target circulation layer of data to be optimized, to wherein each scalar, if
Each reference in the target circulation layer carries out assignment by current iteration, then determines to rely at privatization caused by the scalar
Reason, otherwise, it is determined that the scalar is handled without data privatization；
Detection sub-module 2 1002, for being directed to the target circulation layer of data to be optimized, to wherein each array, if should
It is to be carried by the circulation layer inside target circulation layer, and detect each array element in target circulation stacking generation that array, which relies on,
Each reference by current iteration assignment, then determine to rely on privatization processing caused by the array, otherwise, it is determined that the array without
Need data privatization.
Shown in Figure 5 in above-mentioned device, optimization module 102 includes optimization submodule 1 and optimization submodule
2 2002, wherein
Optimize submodule 1, it is first for dependence caused by the scalar handled to privatization for foundation feedback result
Every sentence in first traversal loop body is searched and collects the write operation for being wherein directed to the scalar, and adds it in set W,
If set W is sky, without carrying out scalar privatization, otherwise, set W is traversed, for each scalar being related in set, inspection
Survey whether it is specification variable, if so, being induction variable by specification change of variable；Set W is traversed, to what is be related in set
Each scalar carries out privatization processing；
Optimize submodule 2 2002, it is first for dependence caused by the array handled to privatization for foundation feedback result
First array of traversal loop weight every accesses sentence, searches and collects the write operation for being wherein directed to array, add it to collection
It closes in AW；If set AW is sky, without carrying out array privatization, otherwise, set AW is traversed, it is every for what is be related in set
A array reference detects it with the presence or absence of specification, and if it exists, is then to conclude by reduction transformation；Set AW is traversed, to relating in set
And each array arrived carries out privatization processing.
Based on the above content, in the embodiment of the present invention, further by can refer to pseudocode content and specific example do into
One step explanation, scalar, array privatization processing pseudocode design content and example are as follows:
Scalar privatization pseudocode:
Scalar privatization example:
Array privatization pseudocode:
Array privatization example is as follows:
Based on above-mentioned method, the embodiment of the present invention also provides a kind of server, comprising: one or more processors；It deposits
Storage device, for storing one or more programs, when one or more of programs are executed by one or more of processors,
So that one or more of processors realize above-mentioned method.
Based on above-mentioned method, the embodiment of the present invention also provides a kind of computer-readable medium, is stored thereon with computer
Program, wherein the program realizes above-mentioned method when being executed by processor.
Unless specifically stated otherwise, the opposite step of the component and step that otherwise illustrate in these embodiments, digital table
It is not limit the scope of the invention up to formula and numerical value.
The technical effect and preceding method embodiment phase of device provided by the embodiment of the present invention, realization principle and generation
Together, to briefly describe, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
It with the specific work process of device, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
In all examples being illustrated and described herein, any occurrence should be construed as merely illustratively, without
It is as limitation, therefore, other examples of exemplary embodiment can have different values.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
The flow chart and block diagram in the drawings show the system of multiple embodiments according to the present invention, method and computer journeys
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, section or code of table, a part of the module, section or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two continuous boxes can actually base
Originally it is performed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that
It is the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, can uses and execute rule
The dedicated hardware based system of fixed function or movement is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes by another way.The apparatus embodiments described above are merely exemplary, for example, the division of the unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, in another example, multiple units or components can
To combine or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or beg for
The mutual coupling, direct-coupling or communication connection of opinion can be through some communication interfaces, device or unit it is indirect
Coupling or communication connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in the executable non-volatile computer-readable storage medium of a processor.Based on this understanding, of the invention
Technical solution substantially the part of the part that contributes to existing technology or the technical solution can be with software in other words
The form of product embodies, which is stored in a storage medium, including some instructions use so that
One computer equipment (can be personal computer, server or the network equipment etc.) executes each embodiment institute of the present invention
State all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can be with
Store the medium of program code.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. a kind of data optimization methods towards Thread-Level Parallelism in shared drive coenocytism, characterized by comprising:
For circulation to be optimized, detecting and determining that its all circulation carries to rely on whether there is the advantageous implementations of privatization；And according to
Determined according to detection as a result, implementing data privatization processing.
2. the data optimization methods according to claim 1 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is that can the advantageous implementations of privatization refer to that all circulations carry and rely on and disappear by data privatization
It removes.
3. the data optimization methods according to claim 1 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is, detects and determines that all circulations carry the advantageous implementations of privatization relied on, includes following content:
For the target circulation layer of data to be optimized, to wherein each scalar, if each reference in the target circulation layer
Assignment is carried out by current iteration, then determines to rely on privatization processing caused by the scalar, otherwise, it is determined that the scalar is not necessarily to data
Privatization processing；To wherein each array, if it is to be carried by the circulation layer inside target circulation layer, and detect that the array, which relies on,
Each reference of each array element in target circulation stacking generation is by current iteration assignment, then caused by determining the array
Privatization processing is relied on, otherwise, it is determined that the array is not necessarily to data privatization.
4. the data optimization methods according to claim 3 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is, according to detection judgement as a result, carrying out relying on privatization processing caused by scalar, includes following content:
For dependence caused by the scalar handled to privatization, first every sentence in traversal loop body, searches and collect wherein
For the write operation of the scalar, and add it in set W, if set W is sky, without carrying out scalar privatization, otherwise,
It traverses set W and detects whether it is specification variable, if so, specification variable is become for each scalar being related in set
It is changed to induction variable；Set W is traversed, privatization processing is carried out to each scalar being related in set.
5. the data optimization methods according to claim 1 or 4 towards Thread-Level Parallelism in shared drive coenocytism,
It is characterized in that, array or privatization clause modification is extended using scalar to carry out privatization processing to scalar.
6. the data optimization methods according to claim 3 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is, according to detection judgement as a result, carrying out relying on privatization processing caused by array, includes following content:
For dependence caused by the array handled to privatization, array of traversal loop weight every first accesses sentence, searches
And the write operation for being wherein directed to array is collected, it adds it in set AW；If set AW is sky, privately owned without carrying out array
Change, otherwise, traverse set AW, for each array reference being related in set, detects it with the presence or absence of specification, and if it exists,
It is then to conclude by reduction transformation；Set AW is traversed, privatization processing is carried out to each array being related in set.
7. the data optimization methods according to claim 1 or 6 towards Thread-Level Parallelism in shared drive coenocytism,
It is characterized in that, privatization processing is carried out to array using array dimension extension or privatization clause modification.
8. a kind of data-optimized device towards Thread-Level Parallelism in shared drive coenocytism, which is characterized in that include test
Module and optimization module, wherein
Test module, for being directed to data to be optimized, detecting and determining that its all circulation carries dependence and has with the presence or absence of privatization
Sharp implementations determine that the situation for needing privatization to handle is fed back to optimization module by result according to detection；
Optimization module, for implementing data privatization processing according to test module feedback result.
9. the data-optimized device according to claim 8 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is that test module includes detection sub-module one and detection sub-module two, wherein
Detection sub-module one, for being directed to the target circulation layer of data to be optimized, to wherein each scalar, if being followed in the target
Each reference in circular layer carries out assignment by current iteration, then determines to rely on privatization processing caused by the scalar, otherwise, sentence
The fixed scalar is handled without data privatization；
Detection sub-module two, for being directed to the target circulation layer of data to be optimized, to wherein each array, if the array relies on
It is to be carried by the circulation layer inside target circulation layer, and detect each array element drawing in target circulation stacking generation every time
With by current iteration assignment, then determining that relying on privatization caused by the array is handled, otherwise, it is determined that the array is private without data
Having.
10. the data-optimized device according to claim 8 towards Thread-Level Parallelism in shared drive coenocytism, special
Sign is that optimization module includes optimization submodule one and optimization submodule two, wherein
Optimize submodule one, for according to feedback result, for dependence caused by the scalar handled to privatization, traversal first to be followed
Every sentence in ring body is searched and collects the write operation for being wherein directed to the scalar, and adds it in set W, if set W is
Otherwise sky, traverses set W for each scalar being related in set and whether detects it then without carrying out scalar privatization
For specification variable, if so, being induction variable by specification change of variable；Set W is traversed, to each scalar being related in set
Carry out privatization processing；
Optimize submodule two, for according to feedback result, for dependence caused by the array handled to privatization, traversal first to be followed
Ring body every array of weight accesses sentence, searches and collects the write operation for being wherein directed to array, add it in set AW；If
Set AW is sky, then without carrying out array privatization, otherwise, traverses set AW, draw for each array being related in set
With detecting it whether there is specification, and if it exists, be then to conclude by reduction transformation；Set AW is traversed, it is every to what is be related in set
A array carries out privatization processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811376636.0A CN109522126B (en) | 2018-11-19 | 2018-11-19 | Thread-level parallel data optimization method and device in shared memory multi-core structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811376636.0A CN109522126B (en) | 2018-11-19 | 2018-11-19 | Thread-level parallel data optimization method and device in shared memory multi-core structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109522126A true CN109522126A (en) | 2019-03-26 |
| CN109522126B CN109522126B (en) | 2020-04-24 |
Family
ID=65778244
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811376636.0A Expired - Fee Related CN109522126B (en) | 2018-11-19 | 2018-11-19 | Thread-level parallel data optimization method and device in shared memory multi-core structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109522126B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795106A (en) * | 2019-10-30 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | Dynamic and static combined memory alias analysis processing method and device in program vectorization process |
| CN113778518A (en) * | 2021-08-31 | 2021-12-10 | 中科曙光国际信息产业有限公司 | Data processing method, data processing device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944014A (en) * | 2010-09-15 | 2011-01-12 | 复旦大学 | Method for realizing automatic pipeline parallelism |
| CN105242929A (en) * | 2015-10-13 | 2016-01-13 | 西安交通大学 | Design method aiming at binary program automatic parallelization of multi-core platform |
| CN107438829A (en) * | 2015-04-08 | 2017-12-05 | 华为技术有限公司 | Partitioned storage data set redoes log record |
| CN107885531A (en) * | 2017-11-28 | 2018-04-06 | 昆山青石计算机有限公司 | A kind of concurrent big data real-time processing method based on array privatization |
-
2018
- 2018-11-19 CN CN201811376636.0A patent/CN109522126B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944014A (en) * | 2010-09-15 | 2011-01-12 | 复旦大学 | Method for realizing automatic pipeline parallelism |
| CN107438829A (en) * | 2015-04-08 | 2017-12-05 | 华为技术有限公司 | Partitioned storage data set redoes log record |
| CN105242929A (en) * | 2015-10-13 | 2016-01-13 | 西安交通大学 | Design method aiming at binary program automatic parallelization of multi-core platform |
| CN107885531A (en) * | 2017-11-28 | 2018-04-06 | 昆山青石计算机有限公司 | A kind of concurrent big data real-time processing method based on array privatization |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795106A (en) * | 2019-10-30 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | Dynamic and static combined memory alias analysis processing method and device in program vectorization process |
| CN110795106B (en) * | 2019-10-30 | 2022-10-04 | 中国人民解放军战略支援部队信息工程大学 | Dynamic and static combined memory alias analysis processing method and device in program vectorization process |
| CN113778518A (en) * | 2021-08-31 | 2021-12-10 | 中科曙光国际信息产业有限公司 | Data processing method, data processing device, computer equipment and storage medium |
| CN113778518B (en) * | 2021-08-31 | 2024-03-26 | 中科曙光国际信息产业有限公司 | Data processing method, device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109522126B (en) | 2020-04-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3443458B1 (en) | A computer-implemented method, a computer-readable medium and a heterogeneous computing system | |
| CN102981807B (en) | Graphics processing unit (GPU) program optimization method based on compute unified device architecture (CUDA) parallel environment | |
| US11429855B2 (en) | Acceleration of neural networks using depth-first processing | |
| Cano et al. | High performance evaluation of evolutionary-mined association rules on GPUs | |
| Khorasani et al. | Efficient warp execution in presence of divergence with collaborative context collection | |
| CN101593132A (en) | Multi-core parallel simulated annealing method based on thread constructing module | |
| CN109522126A (en) | Data optimization methods and device towards Thread-Level Parallelism in shared drive coenocytism | |
| Van Luong et al. | GPU-based multi-start local search algorithms | |
| Ahmed et al. | RALB‐HC: A resource‐aware load balancer for heterogeneous cluster | |
| Breß et al. | A framework for cost based optimization of hybrid CPU/GPU query plans in database systems | |
| Sun et al. | Parallelizing recursive backtracking based subgraph matching on a single machine | |
| Serban et al. | Autonomic scheduling of tasks from data parallel patterns to CPU/GPU core mixes | |
| CN112527304A (en) | Self-adaptive node fusion compiling optimization method based on heterogeneous platform | |
| Falt et al. | Locality aware task scheduling in parallel data stream processing | |
| Al-Zoubi et al. | Towards dynamic multi-task schedulling of OpenCL programs on emerging CPU-GPU-FPGA heterogeneous platforms: A fuzzy logic approach | |
| Kouzinopoulos et al. | Performance study of parallel hybrid multiple pattern matching algorithms for biological sequences | |
| CN104679521B (en) | A kind of accurate calculating task cache WCET analysis method | |
| Sarkar et al. | A Hybrid Clone Detection Technique for Estimation of Resource Requirements of a Job | |
| Liebig et al. | Scalability via parallelization of OWL reasoning | |
| Baramkar et al. | Review for k-means on graphics processing units (gpu) | |
| Tomiyama et al. | Automatic parameter optimization for edit distance algorithm on GPU | |
| Shi et al. | Fast locality sensitive hashing for beam search on gpu | |
| Li et al. | CoCoPIE XGen: A Full-Stack AI-Oriented Optimizing Framework | |
| Nawata et al. | APTCC: Auto parallelizing translator from C to CUDA | |
| Qi et al. | A Parallel BMH String Matching Algorithm Based on OpenMP |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200424 Termination date: 20201119 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |