CN102004644A - Optimal evaluation method for embedded software source program structure-level energy consumption - Google Patents

Optimal evaluation method for embedded software source program structure-level energy consumption Download PDF

Info

Publication number
CN102004644A
CN102004644A CN2010105905613A CN201010590561A CN102004644A CN 102004644 A CN102004644 A CN 102004644A CN 2010105905613 A CN2010105905613 A CN 2010105905613A CN 201010590561 A CN201010590561 A CN 201010590561A CN 102004644 A CN102004644 A CN 102004644A
Authority
CN
China
Prior art keywords
evaluation
grey
class
matrix
estimate
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
Application number
CN2010105905613A
Other languages
Chinese (zh)
Inventor
郭兵
沈艳
胡俊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan University
Original Assignee
Sichuan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Sichuan University filed Critical Sichuan University
Priority to CN2010105905613A priority Critical patent/CN102004644A/en
Publication of CN102004644A publication Critical patent/CN102004644A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention provides a structure-level hierarchical grey evaluation model of an embedded software source program based on researching the present evaluation method, comprising the following steps of: firstly constructing a judgment matrix; secondly constructing an evaluation sample matrix, determining evaluation grey type, and calculating grey evaluation weight vector and matrix; finally calculating the comprehensive evaluation value. Aiming at five topic C language source programs, which are bubble sort, matrix multiplication, Josephus loop algorithm, Fibonacci number sequence and eight queens algorithm, the source programs are optimized by loop unrolling and loop invariable unswitching method. The experiment results show that the evaluation grey type level of the loop unrolling optimization method is the second, belonging to the better effect; and the evaluation grey type level of the loop invariable unswitching method is the third, belonging to the medium effect. The effect of the loop unrolling optimization method is better than that of the loop invariable unswitching method, and the loop unrolling optimization method can make the system have great energy-saving effect and higher running speed.

Description

A kind of embedded software source program structure level energy optimization evaluation method
Technical field
The present invention relates to embedded software power consumption assessment technique field, especially relate to a kind of evaluation method at source program structure level energy optimization---level grey evaluation model.
Background technology
Advocate under the background of " energy-saving and emission-reduction " in country at present, the energy consumption of embedded system is a hot issue that causes that day by day people pay close attention to, and has been subjected to government agencies at all levels and software/hardware developer's great attention in the industry.
Energy optimization can launch at all levels, and the energy saving space that high more design level is provided is big more, and design efficiency is also high more.The energy optimization of embedded system mainly concentrates on hardware layer in advance, comprises levels such as material level, process level, circuit stages, gate leve, RTL level, algorithm level and microstructure level.Along with the continuous development of microelectric technique, the appearance and the application of the advanced hardware energy optimization of various bottoms technology make the optimised power consumption technology of high layer software aspect progressively become the important means of control system power consumption.At present, the energy optimization technology of software layer can be divided into source program structure level, algorithm level and three levels of software architecture level.Because source program structure level optimization method is the important foundation that software energy consumption is optimized, this paper mainly carries out evaluation study to source program structure level optimization method.
The present invention at first sets up level grey evaluation model, then, adopts the emulation experiment method, and the result before and after the program optimization is compared, and by energy consumption, working time, three indexs of number of instructions, estimates the resultant effect and the affiliated grade of optimization method.Feedback result according to estimating can further instruct and improve corresponding embedded software power consumption optimization method, for the multi-level overall evaluation of embedded software power consumption lays the first stone.
Summary of the invention
The object of the present invention is to provide a kind of embedded software source program structure level energy optimization evaluation method.
It is as follows that the present invention solves the step that level gray model that its technical barrier adopts sets up:
1) structure judgment matrix
At first, according to the structure of the 1-9 scaling law in AHP method judgment matrix, obtain the weight of index objectively as far as possible.Importance to Article characteristic relatively is divided into 5 grades, to conclusion relatively get usually 1,3,5,7,9 and inverse measure the intermediate value of the above-mentioned adjacent judgement of 2,4,6,8 expressions.Relatively set up judgment matrix mutually according to importance between evaluation index:
Figure 41088DEST_PATH_IMAGE001
Wherein, The ratio of the important property of expression element i and element j.If matrix A is a positive matrices, this moment, necessarily there was unique non-vanishing eigenvalue of maximum in matrix A , at this moment claim the A matrix to satisfy consistance.Can calculate the proper vector of n rank A matrix by the root method of general linear algebra, at first calculate (1)
Figure 883908DEST_PATH_IMAGE004
, (2)
Figure 511329DEST_PATH_IMAGE005
,
Figure 206228DEST_PATH_IMAGE006
It is proper vector
Figure 535578DEST_PATH_IMAGE007
I component, calculate maximum characteristic root then
Figure 547528DEST_PATH_IMAGE008
, constitute a weight vectors
Figure 278723DEST_PATH_IMAGE009
But in practical problems, because the limitation that affairs complicacy and people judge, when affairs relatively, can not accomplish the crash consistency of judging and have error, this will cause the deviation of eigenwert and proper vector, and the weight that try to achieve this moment is the vector that has deviation.Excessive for fear of error, we have introduced the consistency check of matrix, and when matrix during more near the consistance matrix, eigenwert also approaches the eigenwert of positive matrices more.Have only by check, could illustrate that judgment matrix logically is reasonably, could continue the result is analyzed.The conforming step of test matrix is as follows:
Table 1 mean random consistance table
The matrix exponent number 1 2 3 4 5 6 7 8
RI 0 0 0.52 0.89 1.12 1.26 1.36 1.41
The first step is calculated coincident indicator CI(Consistency Index):
Figure 85137DEST_PATH_IMAGE010
，
Second step, the mean random coincident indicator RI(Random Index of the 1 definite different rank of tabling look-up).
In the 3rd step, calculate consistance ratio CR(Consistency Ratio) and judge.Wherein
Figure 714832DEST_PATH_IMAGE011
, when CR＜0.1, think that the consistance of judgment matrix is an acceptable; When CR 〉=0.1, think that judgment matrix does not meet coherence request, need adjust till the consistance that reaches satisfied.
2) structure is estimated sample matrix and is determined to estimate grey class
For m estimator of structure to n evaluation of indexes collection sample matrix in, the individual different source program of m is represented m estimator, same optimization method changes the variation effect that brings behind the different source programs and can quantize with evaluation score.The present invention replaces expert's scoring to provide evaluation score by adopting experimental data, can avoid too much subjectivity to judge as far as possible, thereby construct the evaluation sample matrix:
Figure 514161DEST_PATH_IMAGE012
Wherein, evaluation of estimate
Figure 847666DEST_PATH_IMAGE013
Represent evaluation score.China issued " remarkable performance appraisal criterion " and " remarkable performance appraisal criterion implementation guide " two national standards in 2004 in succession, for remarkable indicator model provides evaluation criterion and implementation guide.Use for reference its method, in conjunction with the concrete condition of experimental situation, provide the standards of grading of this evaluation method in table 2, be divided into 4 and estimate grey classes, data are in its scope, can be in interval value.
The evaluation criterion of each grade of table 2
The assignment mark Whether energy-conservation Instruction number Performance Estimate grey class
3.5-4 Energy-conservation more than 35% Reduce more than 30% Execution time reduces more than 30% 1
2.5-3.4 Energy-conservation 20%-35% Reduce 20%-30% Execution time reduces 20-30% 2
1.5-2.4 Energy-conservation 10%-20% Reduce 10%-20% Execution time reduces 10-20% 3
0-1.4 Energy-conservation 5%-10% or energy consumption increase on the contrary Reduce 5-10% or increase on the contrary Execution time reduces 5-10% or the execution time increases on the contrary 4
For the reflected appraisal object belongs to the degree of certain class, need determine to estimate the number of degrees of grey class and the white function of grey class according to concrete evaluation problem.If estimating grey class sequence number is e=1,2 ..., g is g grey class, as g=3, can think that estimating grey class is divided into " on ", " in ", D score three classes; G=4 and for example, can with estimate grey class be divided into " excellent ", " very ", " in ", " poor " four classes, the albefaction weight function that grey class need determine to estimate grey class is described.Provide three kinds of form albefaction weight functions commonly used below.
The first grey class, grey number
Figure 456819DEST_PATH_IMAGE015
[
Figure 544991DEST_PATH_IMAGE016
,
Figure 250779DEST_PATH_IMAGE017
), the albefaction weight function
Figure 271432DEST_PATH_IMAGE018
As follows:
Figure 62670DEST_PATH_IMAGE019
The second grey class, grey number
Figure 954534DEST_PATH_IMAGE015
[0,
Figure 327877DEST_PATH_IMAGE020
, 2
Figure 161841DEST_PATH_IMAGE020
], its albefaction weight function is
Figure 925529DEST_PATH_IMAGE021
As follows:
Figure 870351DEST_PATH_IMAGE022
The 3rd grey class, grey number
Figure 95272DEST_PATH_IMAGE014
Figure 850869DEST_PATH_IMAGE015
[0,
Figure 351121DEST_PATH_IMAGE023
, 2
Figure 584787DEST_PATH_IMAGE023
], its albefaction weight function
Figure 119674DEST_PATH_IMAGE024
As follows:
Figure 108489DEST_PATH_IMAGE025
。
3) calculate grey evaluation weight vector and matrix
For the evaluation of estimate of every evaluation index of object to be appraised, estimate grey class e the grey evaluation coefficient of evaluation index be designated as:
Figure 843840DEST_PATH_IMAGE026
Every evaluation index belongs to each total grey evaluation coefficient of estimating grey class for index to be appraised and is designated as:
Figure 130465DEST_PATH_IMAGE027
Every evaluation index of object to be appraised advocates that the grey evaluation power of estimating grey class e is designated as:
Figure 67328DEST_PATH_IMAGE028
Consider that grey class has g, so estimate the grey evaluation weight vector of grey class e is:
Figure 164728DEST_PATH_IMAGE029
.
With each evaluation index for each estimate the grey evaluation weight vector of grey class comprehensive after, the affiliated index that can construct an object to be appraised for each grey evaluation weight matrix of estimating grey class is:
Figure 639571DEST_PATH_IMAGE030
?
4) calculate comprehensive evaluation value
In evaluation system, the weight vectors of evaluation index
Figure 480620DEST_PATH_IMAGE009
, calculate
Figure 724519DEST_PATH_IMAGE031
, the expression evaluation object is in the description of each integrated status ash class degree, can determine grey class hierarchy under it by maximum principle.But owing to this judgment principle sometimes drop-out lost efficacy too much, and can not be directly used in quality ordering between evaluation object, therefore, we also need further to handle to make its uniformization, calculate comprehensive evaluation value
Figure 622566DEST_PATH_IMAGE032
According to
Figure 335438DEST_PATH_IMAGE033
Value determines to be commented the grade of object with reference to the quantized value of each grey class opinion rating, perhaps selects excellent by relatively sorting of a plurality of evaluation object comprehensive evaluation values.
Description of drawings
The process of setting up of Fig. 1 level grey evaluation model.
Embodiment
Below at 5 typical C linguistic source programs, be respectively that bubble sort (Bubble), matrix multiplication (Matrix), Joseph encircle algorithm (Josephus), Feibolaqi ordered series of numbers (Fibonacci) and eight queen's algorithms (Queen), use the outer extracting method of loop unrolling and loop invariant that source program is optimized respectively, by HMSim(a kind of instruction-level high precision embedded software power consumption simulator) draws the relevant data (as shown in Tables 3 and 4) after the optimization.
Table 3 loop unrolling optimization method is to the optimized proportion of source program
Numbering Energy consumption reduces ratio Instruction strip number reduces ratio Reduce ratio working time
Bubble 32.3% 15.2% 32.6%
Matrix 30.3% 15.3% 29.3%
Josephus 35.8% 22.6% 30.8%
Fibonacci 22.7% 12.1% 26.3%
Queen 33.0% 12.1% 30.2%
The outer extracting method of table 4 loop invariant is to the optimized proportion of source program
Numbering Energy consumption reduces ratio Instruction strip number reduces ratio Reduce ratio working time
Bubble 14.1% 9.4% 13.6%
Matrix 10.2% 5.3% 10.3%
Josephus 9.8% 5.6% 10.8%
Fibonacci 11.7% 6.9% 13.3%
Queen 15.0% 10.1% 17.2%
In experimental situation, energy consumption, number of instructions, working time are used respectively
Figure 963866DEST_PATH_IMAGE034
Expression is with they proportion in twos.Because the emphasis that we pay close attention to is energy consumption and working time, energy loss-rate number of instructions is important a little, so get
Figure 78583DEST_PATH_IMAGE035
Generally speaking, linear positive correlation of the energy consumption of embedded software and execution time, both are of equal importance, so get According to the 1-9 scaling law in the analytical hierarchy process, the consistent judgment matrix of the relation of energy consumption, number of instructions, execution time index is as follows:
Table 5 judgment matrix
? ?
Figure 763960DEST_PATH_IMAGE037
 ?
?
Figure 569214DEST_PATH_IMAGE037
 1 3 1
?
Figure 991099DEST_PATH_IMAGE038
 1/3 1 1/3
? 1 3 1
Calculate,
Figure 50639DEST_PATH_IMAGE040
=1.4422,
Figure 891032DEST_PATH_IMAGE041
=0.4807,
Figure 49481DEST_PATH_IMAGE042
=1.4422, draw through normalized
Figure 574134DEST_PATH_IMAGE034
Weight vector is respectively
Figure 314688DEST_PATH_IMAGE043
=(0.4286,0.1429,0.4286),
Figure 515862DEST_PATH_IMAGE003
=3.0001, CI=0.0001, table look-up RI=0.58.Because CR＜0.1, matrix satisfies coherence request, therefore,
Figure 912340DEST_PATH_IMAGE034
Weight vector can be defined as the weight vector of three indexs in our system.
According to table 3 and table 4 experimental result and standards of grading thereof, can draw the pairing evaluation sample matrix of above-mentioned two kinds of methods:
Figure 224372DEST_PATH_IMAGE044
Figure 82082DEST_PATH_IMAGE045
In system, we establish 4 grey class hierarchies, be respectively " excellent ", " very ", " in ", " poor ", grey class hierarchy value vector is=(4,3,2,1).It is as follows to expand the albefaction weight functions that obtain these four the grey classes of evaluation:
The first grey class (e=1), grey number
Figure 337931DEST_PATH_IMAGE015
[0,4,
Figure 266704DEST_PATH_IMAGE017
], its albefaction weight function is:
Figure 981851DEST_PATH_IMAGE047
The second grey class (e=2), grey number
Figure 334947DEST_PATH_IMAGE046
Figure 440437DEST_PATH_IMAGE015
[0,3,6], its albefaction weight function is:
The 3rd grey class (e=3), grey number [0,2,4], its albefaction weight function is:
The 4th grey class (e=4), grey number
Figure 581066DEST_PATH_IMAGE046
Figure 501748DEST_PATH_IMAGE015
[0,1,2], its albefaction weight function is:
To estimate sample matrix and white function and calculate, obtain the grey evaluation weight matrix respectively:
Figure 318187DEST_PATH_IMAGE051
Figure 219278DEST_PATH_IMAGE052
Calculating the Grey Comprehensive Evaluation weight vector is respectively:
Figure 361678DEST_PATH_IMAGE053
=(0.3158 0.4031 0.2544 0.02669)
=?(0.188 0.2507 0.3761 0.1852)
The calculating integrated value is respectively:
Figure 839244DEST_PATH_IMAGE055
=3.008,
Figure 160504DEST_PATH_IMAGE056
=2.442
According to calculating integrated value, can find out: the evaluation ash class hierarchy of loop unrolling optimization method is 2, belongs to good result; The evaluation ash class hierarchy of the outer extracting method of loop invariant is 3, belongs to medium effect.
Figure 787269DEST_PATH_IMAGE057
, the effect of loop unrolling optimization method is better than the outer extracting method of loop invariant, can make system that bigger energy-conservation and higher travelling speed is arranged.Further, available this kind method is estimated the effect that other optimization method or multiple combined optimization method reach.

Claims (1)

1. embedded software source program structure level energy optimization evaluation method is characterized in that the step of this method is as follows:
1) structure judgment matrix
At first, according to the structure of the 1-9 scaling law in AHP method judgment matrix, obtain the weight of index objectively as far as possible, importance to Article characteristic relatively is divided into 5 grades, to conclusion relatively get usually 1,3,5,7,9 and inverse measure, 2, the intermediate value of the above-mentioned adjacent judgement of 4,6,8 expressions, relatively set up judgment matrix mutually according to importance between evaluation index:
Figure 835659DEST_PATH_IMAGE001
Wherein,
Figure 853294DEST_PATH_IMAGE002
The ratio of the important property of expression element i and element j;
2) structure is estimated sample matrix and is determined to estimate grey class
Be that m estimator of structure is to n evaluation of indexes collection sample matrix, m different source program represented m estimator, same optimization method changes the variation effect that brings behind the different source programs and can quantize with evaluation score, the present invention replaces expert's scoring to provide evaluation score by adopting experimental data, can avoid too much subjectivity to judge as far as possible, thereby construct the evaluation sample matrix:
Figure 206915DEST_PATH_IMAGE003
Wherein, evaluation of estimate
Figure 864161DEST_PATH_IMAGE004
Represent evaluation score;
Use for reference " remarkable performance appraisal criterion " and " remarkable performance appraisal criterion implementation guide " two national standards, in conjunction with the concrete condition of experimental situation, in following table, provide the standards of grading of this evaluation method, be divided into 4 and estimate grey class, data are in its scope, can be in interval value:
The assignment mark Whether energy-conservation Instruction number Performance Estimate grey class 3.5-4 Energy-conservation more than 35% Reduce more than 30% Execution time reduces more than 30% 1 2.5-3.4 Energy-conservation 20%-35% Reduce 20%-30% Execution time reduces 20-30% 2 1.5-2.4 Energy-conservation 10%-20% Reduce 10%-20% Execution time reduces 10-20% 3 0-1.4 Energy-conservation 5%-10% or energy consumption increase on the contrary Reduce 5-10% or increase on the contrary Execution time reduces 5-10% or the execution time increases on the contrary 4
For the reflected appraisal object belongs to the degree of certain class, need determine to estimate the number of degrees of grey class and the white function of grey class according to concrete evaluation problem;
3) calculate grey evaluation weight vector and matrix
For the evaluation of estimate of every evaluation index of object to be appraised, estimate grey class e the grey evaluation coefficient of evaluation index be designated as:
Figure 594220DEST_PATH_IMAGE005
Every evaluation index belongs to each total grey evaluation coefficient of estimating grey class for index to be appraised and is designated as:
Figure 59836DEST_PATH_IMAGE006
Every evaluation index of object to be appraised advocates that the grey evaluation power of estimating grey class e is designated as:
Figure 318779DEST_PATH_IMAGE007
Consider that grey class has g, so estimate the grey evaluation weight vector of grey class e is:
Figure 728901DEST_PATH_IMAGE008
With each evaluation index for each estimate the grey evaluation weight vector of grey class comprehensive after, the affiliated index that can construct an object to be appraised for each grey evaluation weight matrix of estimating grey class is:
；
4) calculate comprehensive evaluation value
In evaluation system, the weight vectors of evaluation index , calculate
Figure 809355DEST_PATH_IMAGE011
The expression evaluation object is in the description of each integrated status ash class degree, can determine grey class hierarchy under it by maximum principle, but owing to this judgment principle sometimes drop-out lost efficacy too much, and the quality that can not be directly used between evaluation object sorts, therefore, we also need further to handle to make its uniformization, calculate comprehensive evaluation value
Figure 582139DEST_PATH_IMAGE012
, according to
Figure 998208DEST_PATH_IMAGE013
Value determines to be commented the grade of object with reference to the quantized value of each grey class opinion rating, perhaps selects excellent by relatively sorting of a plurality of evaluation object comprehensive evaluation values.
CN2010105905613A 2010-12-16 2010-12-16 Optimal evaluation method for embedded software source program structure-level energy consumption Pending CN102004644A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010105905613A CN102004644A (en) 2010-12-16 2010-12-16 Optimal evaluation method for embedded software source program structure-level energy consumption

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010105905613A CN102004644A (en) 2010-12-16 2010-12-16 Optimal evaluation method for embedded software source program structure-level energy consumption

Publications (1)

Publication Number Publication Date
CN102004644A true CN102004644A (en) 2011-04-06

Family

ID=43812026

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010105905613A Pending CN102004644A (en) 2010-12-16 2010-12-16 Optimal evaluation method for embedded software source program structure-level energy consumption

Country Status (1)

Country Link
CN (1) CN102004644A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102426662A (en) * 2011-11-08 2012-04-25 四川大学 Embedded software system structural level energy consumption modeling method
CN103606112A (en) * 2013-11-20 2014-02-26 广东电网公司电力科学研究院 Power plant operation performance determination method based on multilevel gray evaluation
CN113554311A (en) * 2021-07-23 2021-10-26 中煤新集能源股份有限公司 Method for evaluating engineering quality of Ordovician limestone water damage under ground directional hole grouting treatment push-coated body

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102426662A (en) * 2011-11-08 2012-04-25 四川大学 Embedded software system structural level energy consumption modeling method
CN102426662B (en) * 2011-11-08 2014-02-26 四川大学 Embedded software system structural level energy consumption modeling method
CN103606112A (en) * 2013-11-20 2014-02-26 广东电网公司电力科学研究院 Power plant operation performance determination method based on multilevel gray evaluation
CN113554311A (en) * 2021-07-23 2021-10-26 中煤新集能源股份有限公司 Method for evaluating engineering quality of Ordovician limestone water damage under ground directional hole grouting treatment push-coated body

Similar Documents

Publication Publication Date Title
CN108510006A (en) A kind of analysis of business electrical amount and prediction technique based on data mining
CN108763810B (en) Load arrangement and adjustment method for bridge static load test
CN110111024A (en) Scientific and technological achievement market valuation method based on AHP model of fuzzy synthetic evaluation
CN104318482A (en) Comprehensive assessment system and method of smart distribution network
CN106709625A (en) Electricity market demand response planning evaluation method
CN103971175B (en) Short-term load prediction method of multistage substations
Guiraud et al. A non-central version of the Birnbaum-Saunders distribution for reliability analysis
CN103530706A (en) Analysis method for comprehensive energy-saving potential of power distribution network
CN104299067A (en) Quantifiable enterprise development situation evaluation model
CN107092751B (en) Variable weight model combination forecasting method based on Bootstrap
Meng et al. Rank reversal issues in DEA models for China’s regional energy efficiency assessment
CN106548413A (en) A kind of power system energy storage fitness-for-service assessment method and system
CN108984830A (en) A kind of building efficiency evaluation method and device based on FUZZY NETWORK analysis
CN109325298B (en) Tire pattern design system based on three-dimensional design platform
CN102004644A (en) Optimal evaluation method for embedded software source program structure-level energy consumption
CN105224577A (en) Multi-label text classification method and system
CN108182511A (en) It is a kind of based on Demand Side Response reserve value assessment method of the sum of ranks than method
CN102509299A (en) Image salient area detection method based on visual attention mechanism
CN103117823A (en) Short wave channel model building method
CN104915485B (en) It is a kind of based on the product demand of effect to structure mapping method
CN103366090A (en) Index weight assessment method based on section rating of experts
CN101604340A (en) A kind of method of the timeliness n that obtains to inquire about
CN106528507A (en) Chinese text similarity detection method and device
CN105354737A (en) Computing method suitable for big data value evaluation
CN108268982B (en) Large-scale active power distribution network decomposition strategy evaluation method and device

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C12 Rejection of a patent application after its publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20110406