CN105933702A - Power consumption control method based on task sensitivity - Google Patents

Power consumption control method based on task sensitivity Download PDF

Info

Publication number
CN105933702A
CN105933702A CN201610245186.6A CN201610245186A CN105933702A CN 105933702 A CN105933702 A CN 105933702A CN 201610245186 A CN201610245186 A CN 201610245186A CN 105933702 A CN105933702 A CN 105933702A
Authority
CN
China
Prior art keywords
task
processor
power consumption
video coding
mode
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
Application number
CN201610245186.6A
Other languages
Chinese (zh)
Other versions
CN105933702B (en
Inventor
曹倩
刘新亮
孙践知
李海生
王雯
王世民
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.)
Beijing Technology and Business University
Original Assignee
Beijing Technology and Business 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 Beijing Technology and Business University filed Critical Beijing Technology and Business University
Priority to CN201610245186.6A priority Critical patent/CN105933702B/en
Publication of CN105933702A publication Critical patent/CN105933702A/en
Application granted granted Critical
Publication of CN105933702B publication Critical patent/CN105933702B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/103Selection of coding mode or of prediction mode
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/189Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the adaptation method, adaptation tool or adaptation type used for the adaptive coding
    • H04N19/196Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the adaptation method, adaptation tool or adaptation type used for the adaptive coding being specially adapted for the computation of encoding parameters, e.g. by averaging previously computed encoding parameters
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/42Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computing Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

The present invention relates to a power consumption control method based on task sensitivity. With a TMS320C6678 processor as a platform, the method comprises the steps of (1) establishing a task-sensitive processor resource distribution model through abstractly extracting a multi-mode video encoding parameter, wherein the model describes the relation between a task amount and a processor utilization rate, and a technical basis is provided for the system power consumption control method, (2) designing an adaptive power consumption controller according to the task-sensitive processor resource distribution model, dynamically adjusting a processor frequency, a voltage and a processor working core number through the task mode of the real-time monitoring system, and achieving the purpose of saving system power consumption.

Description

The power consumption control method that a kind of task based access control is sensitive
Technical field
The present invention relates to a kind of power consumption control method, particularly relate to the Low-power Technology of multi-mode video coding system, belong to Low-power-consumption embedded Video coding process field.
Background technology
Multi-mode video coding generally be directed to dissimilar and different resolution video image (visible ray, infrared and Digital photo etc.) carry out the real-time coding of various code rate, meet the application demand of different illumination conditions and transmission bandwidth, just Emergency processing in accident (such as vehicle accident, natural disaster etc.).Improving constantly however as system complexity, The requirement of low-power consumption is determined pursuing high performance while by embedded system, and power consumption control becomes one needs solution One of key issue.Power problems research worker for embedded system has been done substantial amounts of work and has obtained certain Achievement in research, the most representational is DVFS (Dynamic voltage and frequency scaling) technology, it Different to the demand calculating resource according to distinct program, dynamically adjust running frequency and the running voltage of processor, thus real The now control to energy consumption.Document [1] proposes to coordinate power consumption and performance by change input voltage and clock frequency.For standard The true power consumption efficiency describing asymmetric multinuclear flush bonding processor and extension trend, document [2] proposes a kind of novelty Power consumption-behavior extension method, they summarize the rule of power consumption-behavior extension, be contacted by with multinuclear autgmentability by DVFS Get up.Document [3] combines the multitask particular configuration at multinuclear processing platform, dynamically regulates the frequency of processor, reaches fall The purpose of low-power consumption.Document [4] proposes the DVFS framework of a kind of software management based on embedded gpu, by analyzing GPU processes the change of load thus confirms DVFS availability on GPU.
In recent years, a lot of researchs were absorbed in and were realized energy efficient, the task scheduling of this kind of power consumption sensitive by task scheduling Adjust voltage and frequency dynamically according to different tasks, provide feasible scheme for effective energy efficient.Document [5] In conjunction with dynamic memory management and DVFS technology, by real-time monitoring mechanism control DVFS Developing Tactics frequency and Voltage, reduces system power dissipation.Document [6] is for ARM embedded system, it is proposed that a kind of based on the DVFS calculating density Algorithm, the method adjusts optimum frequency according to application program to the situation of memory access.The results show the method is with existing Embedded system in consumption control method compare there is certain advantage.When considering the execution of task in actual applications Between suffer from the impact of the factor such as operation platform, input data set and be rendered as a kind of probability distribution, document [7] is for many Business processor system proposes a kind of schedule by probability mode method of power consumption sensitive, and the method research is meeting certain time probability In the case of, how to weigh power consumption and execution time.
According to current achievement in research, DVFS technology can effectively control the power consumption of embedded system, but for For some flush bonding processor, the work clock of its part peripheral interface has processor frequencies frequency dividing to obtain, therefore The excessive reduction of processor frequencies will affect the transmission of system normal data.The present invention is (following with TMS320C6678 It is called for short C6678) it is processing platform, when processing multi-mode video, owing to task load is different, use DVFS strategy Can by control processor frequencies reduce power consumption, but the frequency of too much reduction C6678 will to other resources (as EMIF interface, shared storage management etc.) produce impact, even cannot meet the demand of data transmission, ask for this Topic, herein towards multi-mode video coding knot, in conjunction with C6678 processor characteristic and DVFS technology propose a kind of based on The power consumption control method of task-sensitive (Task-aware Power Control Method based on DVFS, TAPCM-DVFS)。
List of references
[1].Nakai,M.,et al.,Dynamic voltage and frequency management for a low-power embedded microprocessor.Solid-State Circuits,IEEE Journal of,2005.40(1):p.28-35.
[2].Yoo,S.,An empirical validation of power-performance scaling:DVFS vs.multi-core scaling in big.LITTLE processor.IEICE ELECTRONICS EXPRESS,2015.12(201502368).
[3].Li,S.and F.Broekaert,Low-power scheduling with DVFS for common RTOS on multicore platforms.ACM SIGBED Review,2014.11(1):p.32-37.
[4].You,D.and K.Chung,Dynamic voltage and frequency scaling framework for low-power embedded GPUs.Electronics letters,2012.48(21):p.1333-1334.
[5].Anagnostopoulos,I.,et al.,Power-Aware Dynamic Memory Management on Many-Core Platforms Utilizing DVFS.ACM TRANSACTIONS ON EMBEDDED COMPUTING SYSTEMS,2013.131(40).
[6].Cho,S.J.,S.H.Yun and J.W.Jeon,A powersaving DVFS algorithm based on Operational Intensity for embedded systems.IEICE ELECTRONICS EXPRESS,2015.12(201411283).
[7]Deng Z,Zeng G and Wang W.Energy consumption analysis satisfying time-energy-probability constraints for modern DVFS microprocessor[J].IETE Technical Review,2015,32(4):260-272.
Summary of the invention
The technical problem to be solved: the present invention is with TMS320C6678 (hereinafter referred to as C6678) for processing Other, towards multi-mode video coding system, will be provided by platform for the reduction operating frequency using DVFS technology too much Source (such as EMIF interface, sharing storage management etc.) produces impact, the problem that even cannot meet the demand of data transmission, Propose the power consumption control method that a kind of task based access control is sensitive, when the little time of task, system merit can be reduced further Consumption.
The technology of the present invention solution: the power consumption control method that a kind of task based access control is sensitive, with TMS320C6678 process Device is hardware platform, according to the task amount control system power consumption of multi-mode video coding, it is achieved step is as follows:
Step (1), abstract extraction multi-mode video coding parameter, video coding system needs to support L kind encoding tasks mould Formula, every kind of mission mode uses seven tuples to describe, and i.e. i-th kind mission mode is expressed as Mi=< hi,wi,bi,ci,fi,Tii>, wherein 1≤i≤L, hi, wi, bi, ciRepresent the resolution of video respectively Rate height, resolution width, code check, pixel bit wide, referred to as video coding parameter;fiRepresent video frame rate;TiRepresent single frames The average long process time of Video coding, its value is determined by the video frame rate in mission mode, i.e. Ti=1/fi;τiFor The average processor of single frame video coding processes the cycle, represents the task amount of i-th kind of pattern, in actual applications by real The processor process cycle of Shi Tongji continuous n frame Video coding averages again;
Step (2), different according to different mode Video coding task amount, set up the processor resource of a kind of task-sensitive Distribution model, the model describes the relation between task amount and processor working frequency and processor work check figure;
Step (3), distributes model according to the processor resource of task-sensitive, designs adaptive power eonsumption controller, logical Cross the mission mode of real-time monitoring system, dynamically adjust processor frequencies, voltage and processor work check figure, control system Power consumption.
In described step (2), task amount with the relation of processor working frequency is:
6 τ i f i η + λ ≤ F ≤ 6 τ i f i η - λ - - - ( 1 )
Wherein F is processor working frequency;
τiIt is the system real-time statistics value when task i, specific size and video coding parameter, encryption algorithm and video content Factor is relevant;fiBeing video frame rate, its value is determined by encoding tasks pattern;η and λ is respectively the processor that user sets Utilization rate value and range of error.
In described step (2), task amount with the relation of processor work check figure is:
Wherein NjIt is expressed as under Video coding mode based on parallel data processing, the work of DSP when forwarding task j to from task i Make check figure;WithThe general assignment amount of processor when task i of being respectively and task j, whereinRepresent and round downwards, 1≤Nj≤ 8,1≤Ni≤ 8 and Nj, NiBelong to set of integers.
The automatic state machine model of the described adaptive power eonsumption controller in step (3), usesTable Show, wherein:
(1) state setWherein S0State is detected in real time, in real time detection pattern switching life for event According to command context, order, judges that entering processor process performance adjusts state, S1Show that being carrying out resource adjusts, for appointing The mode of operation that business amount is bigger distributes more processor resource, improves processor process performance so that it is meet task needs, S2Show that being carrying out resource adjusts, and on the basis of meeting mission requirements, reduces the process performance of processor, saves energy Consumption;
(2) event setsx1And x2Represent that the system that detects is appointed to big from little task amount respectively The order that business cut-to-measure is changed with system switches to little task amount from big task amount is effective, x3And x4Represent that processor performance adjusts behaviour Complete;
(3) operational setr1And r2Show that performing processor resource adjusts operation, i.e. according to task need Seek the process performance improving or reducing processor;
(4) state transition functionShow that controller is determined according to the type of current state and event Fix the state at a moment;
(5) output functionShow that controller could be made that different operations under different conditions, specifically fixed Justice is φ (S1)=r1, φ (S2)=r2
Power eonsumption controller original state is S0, when event x1Or x2State S is entered during generation1Or S2, and make corresponding Control dynamic r1Or r2, work as r1Or r2After execution terminates, controller automatically returns to state S0, S1And S2For momentary status, They terminate automatically along with the end of control action, operational motion r1Represent the process performance improving processor;Operational motion r2Represent the process performance reducing processor.
Described operational motion r1Specific operation process be:
If work at present check figure NcurWhen the check figure that works less than 8 and after adjustment is less than or equal to 8, the work check figure of DSP N need to be adjusted toadj, the operating frequency of DSP is 800MHz;The work of DSP when check figure is more than 8 if worked after Tiao Zhenging Check figure need to be adjusted to 8, formula (1) operating frequency obtaining DSP is adjusted to Fadj;If work at present check figure is more than In 8 time, the work check figure of DSP is 8, formula (1) operating frequency obtaining DSP is adjusted to Fadj
Described operational motion r2Specific operation process be:
If present operating frequency more than 800MHz and adjust after operating frequency more than or equal to 800MHz time, DSP's Work check figure need to be adjusted to 8, formula (1) operating frequency obtaining DSP is adjusted to Fadj;If after DSP adjusts Operating frequency less than during 800MHz by formula (2) the work check figure of DSP is adjusted to Nadj, the operating frequency of DSP is 800MHz;If present operating frequency is equal to 800MHz, formula (2) the work check figure obtaining DSP is adjusted to Nadj, The operating frequency of DSP is 800MHz.
Present invention advantage compared with prior art is:
(1) present invention passes through analysis video encoding tasks, establishes the relational model between task amount and processor performance, Control to improve foundation for system power dissipation.
(2) according to the difference of Video coding task amount, the dynamically operating frequency of regulation DSP and work check figure, than tradition Use DVFS strategy only regulation processor frequencies compare can further control system power consumption.
(3) present invention realizes simply, the hardware designs that different increases are extra, and is conveniently transplanted to can dynamically regulate work On other processors of working frequency and work check figure.
Accompanying drawing explanation
Fig. 1 is the inventive method flowchart;
Fig. 2 is the state of automata machine model of middle controller of the present invention;
Fig. 3 is the operational approach improving DSP process performance in the present invention;
Fig. 4 is the operational approach of the reduction DSP process performance in the present invention;
Fig. 5 is that multi-mode video picture coding processes plate experiment porch.
Detailed description of the invention
The power consumption control method of task based access control of the present invention sensitivity is made up of two parts: (1) is by abstract extraction multimode Formula video coding parameter, sets up the processor resource distribution model of a kind of task-sensitive, the model describes task amount and place Relation between reason device utilization rate, provides technical foundation for system power dissipation control method.(2) according to the place of task-sensitive Reason device resource allocator model, designs a kind of adaptive power eonsumption controller, by the mission mode of real-time monitoring system, dynamic State adjusts processor frequencies, voltage and processor work check figure, reaches to save the purpose of system power dissipation.
The power consumption control method of task based access control of the present invention sensitivity is specific as follows:
From the power consumption characteristics of cmos circuit, DVFS technology is used to realize the master that optimised power consumption is processor Wanting power save mode, it reduces frequency and the voltage of little load processor within the specific limits so that meeting performance requirement On the basis of, reduce energy consumption.The embedded multi-core processor that TMS320C6678 saves as frequency-adjustable, can be by reducing Processor frequencies and minimizing work check figure reduce power consumption of processing unit.It is as shown in the table, at the average utilization processing core is 80%, operating temperature is 50 DEG C, under conditions of only opening DDR3 controller, SRIO and sharing storage resource, The performance of TMS320C6678 processor difference dominant frequency condition, power dissipation ratio are relatively.
The performance of table 1 C6678 processor difference dominant frequency condition and power consumption
Therefore, the present invention combines the task amount of different disposal pattern and dynamically adjusts the frequency of DSP and the nuclear volume that works, and is expiring On the basis of foot task processing requirements, it is to avoid DSP utilization rate is the least and the power wastage that causes, it is achieved reduce system power dissipation Target.
As it is shown in figure 1, the present invention is with TMS320C6678 as hardware platform, according to the task amount of multi-mode video coding Control system power consumption, is embodied as follows:
The first step, multi-mode task description
It is assumed for convenience of description that video coding system needs to support L kind encoding tasks pattern, every kind of mission mode can be used One seven tuple describes, and i.e. i-th (1≤i≤L) plants mission mode and be represented by Mi=< hi,wi,bi,ci,fi,Tii>, wherein hi, wi, bi, ciRepresent respectively video resolution (high, wide), Code check, pixel bit wide, this article is called video coding parameter;fiRepresent video frame rate;TiRepresent single frame video coding The average long process time, its value is determined by the video frame rate in mission mode, i.e. Ti=1/fi;τiCompile for single frame video The average DSP of code processes the cycle, can represent the task amount of i-th kind of pattern, in actual applications by real-time statistics even The DSP process cycle of continuous n frame Video coding averages again.Notice that the value of n can affect statistical accuracy and configuration The time delay adjusted, needs to carry out compromise process.For from little task amount tupe to the adjustment of big task amount tupe, In order to avoid the overweight problem of system load caused because of statistics task amount, in actual applications, can be by this case System configuration be first adjusted to peak performance, on the basis of ensureing systemic-function and performance, then carry out configuration and adjust.
On the basis of obtaining task amount, the average handling time t of single frame video coding in i-th kind of patterniIt is represented by:
t i = τ i F × 6 - - - ( 3 )
Wherein F is the operating frequency of DSP.
Second step, the resource distribution model of task-sensitive
Assume that the utilization rate of DSP is η within the execution cycle of task ii, convolution (3) can have:
η i = t i T i = 6 τ i FT i = 6 τ i f i F - - - ( 4 )
Wherein τiIt is system real-time statistics value, the factor such as specific size and video coding parameter, encryption algorithm and video content Relevant;fiBeing video frame rate, its value is determined by encoding tasks pattern.Utilization rate η of DSP is understood by formula (4)iWith DSP Dominant frequency between inversely.In actual applications, owing to video content is continually changing, can in conjunction with Video coding principle Knowing, the process cycle that the more violent video that moves is consumed is more, i.e. task amount is relatively big, and the video consumption moving mild The process cycle less, i.e. task amount is less, and therefore under same video task pattern, the utilization rate of DSP is not one Fixing value.The power wastage caused in order to avoid DSP utilization rate is the lowest, needs dynamically to adjust according to the change of task amount Its utilization rate is controlled in a metastable scope by joint DSP process performance, it may be assumed that | ηi-η |≤λ, wherein η and λ divides The DSP utilization rate value not set for user and range of error.In order to ensure to meet the demand of all tasks, can be according to maximum Task load set η and λ.Therefore can have:
| 6 τ i f i F - η | ≤ λ - - - ( 5 )
Can be obtained by formula (5):
6 τ i f i η + λ ≤ F ≤ 6 τ i f i η - λ - - - ( 6 )
Result of calculation according to formula (6) selects the dominant frequency of DSP to be adjusted from table, processes merit to meet task Can, if there being multiple qualified DSP dominant frequency, then select bigger being adjusted.
According to circuit dynamic power consumption characteristic, when task i of execution, the dominant frequency of DSP is the biggest, and its power consumption is the highest, instead As the same.But, when the dominant frequency of DSP is 800MHz, if the utilization rate of DSP is the most too small, it is contemplated that right The quantity of processor work core is adjusted, and i.e. reduces system power dissipation further by close portion divisional processing core.Due to DSP The change of work nuclear volume can affect the handling process of algorithm, therefore to reduce power consumption control method complexity as far as possible, under It is video coding algorithm based on data parallel that face work core adjusts the foundation of model.
Under Video coding mode based on parallel data processing, video data is by the quantity according to polycaryon processor work core Video is divided into corresponding data block be assigned to each process core encode.It is assumed that within the execution cycle of task i Utilization rate η of DSPiMeet | ηi-η | work check figure during≤λ condition is Ni(1≤Ni≤ 8), each work core appoint Business amount isThen general assignment amountWhen mission mode is adjusted to task j, the task of each work core Amount is(assume), then on the basis of meeting Video coding mission requirements, in order to reduce DSP further The power consumption penalty that free time causes, can adjust the work check figure of DSP so that the utilization of DSP according to the change of task amount Rate ηjAppoint and so meet | ηj-η|≤λ。
Because TMS320C6678 is isomorphism polycaryon processor, the performance of each work core is identical, and based on data Under parallel Video coding processing mode, the data volume distributing to each work core is essentially identical, therefore holding in task i The average task amount of each core in line periodWhen mission mode is adjusted to j by i, the total of DSP appoints Business amount reduced valueSo the work check figure that can reduce isWhereinRepresent to Under round.So in sum, when mission mode is j, DSP works check figure NjFor:
Wherein 1≤Nj≤ 8,1≤Ni≤ 8 and Nj, NiBelong to set of integers.From formula (7), by adjusting The work nuclear volume of DSP can change the utilization rate of DSP.ηjCan be obtained by formula (4), in actual applications, in order to After ensureing to adjust, the process performance of DSP can meet the demand of systemic-function, selects to meet in feasible solution |ηj-η | the maximum functional check figure under the conditions of≤λ.
3rd step, its design of task-sensitive power consumption control
In order to realize automatically controlling of system power dissipation, need to combine the Video coding of different mode and power consumption control modelling One self-adapted power consumption controller, can be with real-time monitoring system mission mode by this controller, and according to different mode Mission requirements dynamically adjust DSP frequency and work nuclear volume, reduce the purpose of power consumption.Controller automatic state machine mould Type is as in figure 2 it is shown, can useRepresent.Specific as follows:
1) state setWherein S0Detect state for event in real time, detect mode handover command in real time, Judge that entering DSP process performance adjusts state according to command context.S1Show that being carrying out resource adjusts, for task amount Bigger mode of operation distributes more DSP resource, improves DSP process performance so that it is meet task needs.S2Table The bright resource that is carrying out adjusts, and on the basis of meeting mission requirements, reduces the process performance of DSP, saves energy consumption;
2) event setsx1And x2Respectively represent detect system from little task amount to big task The order that cut-to-measure is changed with system switches to little task amount from big task amount is effective.x3And x4Represent the operation of DSP adjusting performance Complete;
3) operational setr1And r2Show that performing DSP resource adjusts operation, i.e. comes according to mission requirements Improve or reduce the process performance of DSP;
4) state transition functionShow controller according to the type decided of current state and event next The state carved;
5) output functionShow that controller could be made that different operations under different conditions, be specifically defined as φ(S1)=r1, φ (S2)=r2
The state of automata machine model of Fig. 1 controller, power eonsumption controller original state is S0, when event x1(or x2) State S is entered during generation1(or S2), and make the dynamic r of corresponding control1(or r2).Work as r1(or r2) perform knot Shu Hou, controller automatically returns to state S0。S1And S2For momentary status, they are tied automatically along with the end of control action Bundle.
Operational motion r1Represent the process performance improving DSP, as shown in Figure 3;Operational motion r2Represent the place reducing DSP Rationality energy, as shown in Figure 4.It is described in detail below.
Experimental comparison and interpretation of result
1) experimental program
In conjunction with the mission requirements of multi-mode real time video coding systems, its conventional mode of operation includes 7 kinds, i.e. L=7. Specific tasks attribute description is as shown in table 2.Under high definition visible light video real time processing conditions, owing to needing to be processed regarding Frequency amount is maximum, so on the basis of using identical encryption algorithm, it is desirable to the process performance of DSP is the highest.Video coding Task amount τiNeed to be obtained by system test, in the present invention by the DSP of the continuous 50 frame Video codings of real-time statistics The process cycle averages again.
Table 2 multi-mode Video Coding task attribute describes
As shown in Table 2, according to different mission modes, in conjunction with C6678 optimised power consumption characteristic, task treatability is being met On the basis of energy, power consumption control can be carried out by two ways: first, when performing task, detection DSP state in real time, When it is idle, reduce frequency and the resource distribution of DSP;Second, different according to mission mode, dynamically adjust DSP Frequency and resource distribution, improve DSP utilization rate, it is to avoid idle condition.Analyzing theoretically, both modes can Control DSP power consumption.But when concrete operations, decision processor duty transfer point is power consumption control method accurately The key effectively realized.For first kind of way, the DSP free time generally occurs in frame data and has processed and next frame Between starting to process, due to during this period of time at Millisecond, and frequently occur, if adjusting DSP frequency and money frequently Source can introduce the most extra expense, and therefore first kind of way is impracticable in engineering design;For the second way, only Have the adjustment carrying out DSP frequency when mission mode switches because the execution cycle of every kind of mission mode is long, short then A few minutes, long then several hours, and task switching realized by control command, the time point of switching easily determines, Convenient to operate, therefore select the second processing mode proper.
Experiment porch selects multi-mode Video Coding based on C6678 autonomous Design to process plate, builds special test Environment, sets up multi-mode video picture coding as shown in Figure 5 and processes plate experiment porch.This platform include special transmission board, Multi-mode Video Coding processes plate and receives plate, and the most special transmission board sends the video data of different mode in real time, Multi-mode Video Coding processes plate Real-time Collection video data and carries out coded treatment, is finally passed by the code stream after coding Enter to receive plate and be decoded display.When carrying out multi-mode Video Coding and processing plate design, the power supply of C6678 selects The UCD series of power managing chip of TI is controlled and monitors in real time, so the power consumption measuring DSP can be with the use of The Fusion Digital Power Designer software of PMBus interface facility and TI completes.
2) interpretation
The major parameter of every kind of mission mode includes video resolution, frame per second, pixel bit wide and code check etc..Coded treatment is adopted With up-to-date HEVC video coding algorithm, perform on C6678 after optimizing.In order to verify the power consumption that this article proposes The effect of control method, under the same conditions, respectively to not using power consumption control method, using DVFS method and use This article method is tested, and table 3 is the experimental result without and with DVFS power consumption control method, and table 4 is for using The experimental result of the power consumption control method of this article.According to experimental evaluation, set η and λ difference 0.7 and 0.1.
From experimental result, when not using power consumption control method, average power consumption produced by different task pattern also has Institute is different, and as when carrying out HD video coding, power consumption is the highest.When system just powers up, without task status power consumption Low, the average power consumption of remaining several mission mode is also not quite similar, but is totally more or less the same, and main reason is that CMOS The dynamic power consumption of circuit is relevant with the number of times of level upset in circuit, when lacking without task or need data volume to be processed, and DSP Load will be little, so the power consumption produced is relatively low.For using the experimental result of power consumption control method, it can be seen that its merit Consumption substantially reduces, and under without task status, uses TAPCM-DVFS method in this paper, by dsp processor core Closing, decrease system idle condition to the full extent, its average power consumption reduces 47.1% when controlling than idle.? Process minimum load mission mode M7Time power dissipation ratio process maximum load mission mode M2Power consumption save 37%.No When considering without mission mode, idle control method and the TAPCM-DVFS power consumption control method used herein, process The average energy consumption of single frame video saves 23.7%.The power consumption control method processor frequencies using DVFS strategy is 800MHz Above, and need work check figure be at full capacity time with this article method there is close effect.But it is as load further Reduce, be not easy to regulating frequency again due to C6678, and now use method minimizing work check figure that this invention proposes with Reduce system power dissipation further.
Table 3 is without and with DVFS power consumption control method experimental result
Table 4 uses the power consumption control method experimental result of TAPCM-DVFS
There is provided above example to be only used to describe the purpose of the present invention, and be not intended to limit the scope of the present invention.This Bright scope is defined by the following claims.Various equivalents made without departing from spirit and principles of the present invention and repairing Change, all should contain within the scope of the present invention.

Claims (3)

1. the power consumption control method that a task based access control is sensitive, it is characterised in that: it is hard with TMS320C6678 processor Part platform, controls video coding system power consumption according to the task amount of multi-mode video coding, and step is as follows:
Step (1), abstract extraction multi-mode video coding parameter, it is assumed that video coding system needs to support that L kind coding is appointed Business pattern, every kind of mission mode uses seven tuples to describe, and i.e. i-th kind mission mode is expressed as Mi=< hi,wi,bi,ci,fi,Tii>, wherein 1≤i≤L, hi, wi, bi, ciRepresent the resolution of video respectively Rate height, resolution width, code check, pixel bit wide, referred to as video coding parameter;fiRepresent video frame rate;TiRepresent single frames The average long process time of Video coding, its value is determined by the video frame rate in mission mode, i.e. Ti=1/fi;τiFor The average processor of single frame video coding processes the cycle, represents the task amount of i-th kind of pattern, in actual applications by real The processor process cycle of Shi Tongji continuous n frame Video coding averages again;
Step (2), different according to different mode Video coding task amount, set up the processor resource of a kind of task-sensitive Distribution model, the model describes the relation between task amount and processor working frequency and processor work check figure, for Dynamically adjust processor frequencies, voltage and processor work check figure and foundation is provided.
2. according to the power consumption control method that the task based access control shown in claim 1 is sensitive, it is characterised in that: described step (2) in, task amount with the relation of processor working frequency is:
6 τ i f i η + λ ≤ F ≤ 6 τ i f i η - λ - - - ( 1 )
Wherein F is processor working frequency;τiIt is the system real-time statistics value when task i, specific size and video coding parameter, Encryption algorithm is relevant with video content factor;fiBeing video frame rate, its value is determined by encoding tasks pattern;η and λ is respectively The processor utilization value set for user and range of error.
3. according to the power consumption control method that the task based access control shown in claim 1 is sensitive, it is characterised in that: described step Suddenly in (2) task amount and processor work check figure relation be:
Wherein NjIt is expressed as under Video coding mode based on parallel data processing, the work of DSP when forwarding task j to from task i Make check figure;WithThe general assignment amount of processor when task i of being respectively and task j, whereinRepresent and round downwards, 1≤Nj≤ 8,1≤Ni≤ 8 and Nj, NiBelong to set of integers.
CN201610245186.6A 2016-04-19 2016-04-19 A kind of power consumption control method of task based access control sensitivity Active CN105933702B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610245186.6A CN105933702B (en) 2016-04-19 2016-04-19 A kind of power consumption control method of task based access control sensitivity

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610245186.6A CN105933702B (en) 2016-04-19 2016-04-19 A kind of power consumption control method of task based access control sensitivity

Publications (2)

Publication Number Publication Date
CN105933702A true CN105933702A (en) 2016-09-07
CN105933702B CN105933702B (en) 2019-06-04

Family

ID=56839354

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610245186.6A Active CN105933702B (en) 2016-04-19 2016-04-19 A kind of power consumption control method of task based access control sensitivity

Country Status (1)

Country Link
CN (1) CN105933702B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106527666A (en) * 2016-11-18 2017-03-22 宇龙计算机通信科技(深圳)有限公司 Control method of central processing unit and terminal equipment
CN109889896A (en) * 2019-03-27 2019-06-14 努比亚技术有限公司 Method, mobile terminal and the storage medium of dynamic regulation CPU running frequency
CN110275510A (en) * 2019-06-14 2019-09-24 百度在线网络技术(北京)有限公司 Processing frequency method of adjustment, device and the electronic equipment of vehicle control model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1776568A (en) * 2005-11-28 2006-05-24 北京中星微电子有限公司 Task-based dynamic CPU working frequency regulating method and system
CN1945498A (en) * 2006-10-12 2007-04-11 浙江大学 Low power consumption real time task parameter model dispatching method facing embedded system
CN101604198A (en) * 2009-07-10 2009-12-16 杭州电子科技大学 A kind of method that reduces power consumption of embedded system
CN102073543A (en) * 2011-01-14 2011-05-25 上海交通大学 General processor and graphics processor fusion system and method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1776568A (en) * 2005-11-28 2006-05-24 北京中星微电子有限公司 Task-based dynamic CPU working frequency regulating method and system
CN1945498A (en) * 2006-10-12 2007-04-11 浙江大学 Low power consumption real time task parameter model dispatching method facing embedded system
CN101604198A (en) * 2009-07-10 2009-12-16 杭州电子科技大学 A kind of method that reduces power consumption of embedded system
CN102073543A (en) * 2011-01-14 2011-05-25 上海交通大学 General processor and graphics processor fusion system and method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
I ANAGNOSTOPOULOS等: "Power-aware dynamic memory management on many-core platforms utilizing DVFS", 《ACM TRANSACTIONS ON EMBEDDED COMPUTING SYSTEMS, 2013》 *
SJ CHO等: "A powersaving DVFS algorithm based on Operational Intensity for embedded systems", 《IEICE ELECTRONICS EXPRESS, 2015》 *
Z DENG等: "Energy consumption analysis satisfying time-??energy-probability constraints for modern DVFS microprocessor", 《IETE TECHNICAL REVIEW, 2015》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106527666A (en) * 2016-11-18 2017-03-22 宇龙计算机通信科技(深圳)有限公司 Control method of central processing unit and terminal equipment
CN109889896A (en) * 2019-03-27 2019-06-14 努比亚技术有限公司 Method, mobile terminal and the storage medium of dynamic regulation CPU running frequency
CN109889896B (en) * 2019-03-27 2022-05-27 努比亚技术有限公司 Method for dynamically adjusting CPU operation frequency, mobile terminal and storage medium
CN110275510A (en) * 2019-06-14 2019-09-24 百度在线网络技术(北京)有限公司 Processing frequency method of adjustment, device and the electronic equipment of vehicle control model
CN110275510B (en) * 2019-06-14 2021-12-07 百度在线网络技术(北京)有限公司 Processing frequency adjusting method and device of vehicle control model and electronic equipment

Also Published As

Publication number Publication date
CN105933702B (en) 2019-06-04

Similar Documents

Publication Publication Date Title
US9618997B2 (en) Controlling a turbo mode frequency of a processor
CN101379453B (en) Method and apparatus for using dynamic workload characteristics to control CPU frequency and voltage scaling
CN104169832B (en) Providing energy efficient turbo operation of a processor
Gu et al. Energy efficient scheduling of servers with multi-sleep modes for cloud data center
CN103605418B (en) The regulating power consumption method and apparatus of cluster server
KR100661851B1 (en) Power management method for platform and that platform
CN101211215A (en) Performance of a processor in a managing data processing equipment
CN113535409B (en) Server-free computing resource distribution system oriented to energy consumption optimization
CN101382912B (en) Method for optimizing resource sensitive property orienting application server and system thereof
KR20020018309A (en) Cpu scheduling method
KR20110038036A (en) Sleep processor
CN104049716A (en) Computer energy-saving method and system combined with temperature sensing
Raghunathan et al. Adaptive power-fidelity in energy-aware wireless embedded systems
CN102622273A (en) Self-learning load prediction based cluster on-demand starting method
CN114710563A (en) Cluster energy saving method and device
CN105933702A (en) Power consumption control method based on task sensitivity
Xu et al. {Power-Aware} Throughput Control for Database Management Systems
Shafique et al. Agent-based distributed power management for Kilo-core processors: Special session:“Keeping Kilo-core chips cool: New directions and emerging solutions”
Kong et al. Optimal selection of power saving classes in IEEE 802.16 e
US20150261290A1 (en) Systems and Methods for Messaging-based Fine Granularity System-on-a-Chip Power Gating
Zhang et al. GreenDRL: managing green datacenters using deep reinforcement learning
CN101685335A (en) Application server based on SEDA as well as energy-saving device and method thereof
CN116643844B (en) Intelligent management system and method for automatic expansion of power super-computing cloud resources
US20240004706A1 (en) Methods and systems for scheduling energy-efficient execution of periodic, real-time, directed-acyclic-graph tasks
US20220342474A1 (en) Method and system for controlling peak power consumption

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information

Inventor after: Cao Qian

Inventor after: Liu Lihong

Inventor after: Liu Xinliang

Inventor after: Sun Jianzhi

Inventor after: Li Haisheng

Inventor after: Wang Wen

Inventor after: Wang Shimin

Inventor before: Cao Qian

Inventor before: Liu Xinliang

Inventor before: Sun Jianzhi

Inventor before: Li Haisheng

Inventor before: Wang Wen

Inventor before: Wang Shimin

CB03 Change of inventor or designer information
GR01 Patent grant
GR01 Patent grant