CN103761212B - The method for designing of mapping scheme and topological structure between task and node in network-on-chip - Google Patents
The method for designing of mapping scheme and topological structure between task and node in network-on-chip Download PDFInfo
- Publication number
- CN103761212B CN103761212B CN201410027528.8A CN201410027528A CN103761212B CN 103761212 B CN103761212 B CN 103761212B CN 201410027528 A CN201410027528 A CN 201410027528A CN 103761212 B CN103761212 B CN 103761212B
- Authority
- CN
- China
- Prior art keywords
- scheme
- group
- power consumption
- node
- response time
- 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.)
- Expired - Fee Related
Links
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses in a kind of network-on-chip the method for designing of mapping scheme and topological structure between task and node, technical points is: power consumption and response time are separated the preferential index as assessed form, but during finding optimal case, consider power consumption and response time simultaneously, accomplish that the scheme found out makes power consumption and system delay the most minimum, additionally selected to make power consumption and the topological structure of response time minimum.
Description
Technical field
The present invention relates to IC design field, be specifically related to for the method for designing of mapping scheme and topological structure between task and node in network-on-chip between multitask and multinode in wireless network-on-chip.
Background technology
Along with future communications terminal and the increase of the computation complexity of equipment, the integrated scale demand of complication system chip will increase rapidly in real time, on chip piece may integrated dozens of or hundreds of processing unit, signal processing platform based on multinuclear becomes the development trend of software radio.In such Integrated system, reliable design, at a high speed, communication system becomes challenge and the opportunity that SOC(system on a chip) (System on Chip, SoC) develops in the high performance chips of low-power consumption.The bus type communication structure of tradition multinuclear SoC cannot meet the demand of data exchange in extensive complicated digital signal processing procedure in real time, it is therefore proposed that interconnection and communication mechanism network-on-chip on the sheet of a new generation.
The system that network-on-chip (upper each node i.e. processor) realizes is mainly the complex communication system higher to requirement of real-time or digital information processing system.Communication and signal processing have real-time and a processing delay requirement, and therefore, the mapping scheme of NoC don't fail to make the time delay of system the least.It addition, network-on-chip is to realize on the chip of integrated circuit, in order to ensure the work that chip can be stable, the temperature of chip can not be high.The temperature of chip is to be determined by the power consumption of chip, and power consumption is the biggest, and the temperature of chip is the highest.Therefore, when Choose for user scheme, it is necessary to select to make try one's best low and time delay of power consumption to try one's best few scheme.
The topological structure of traditional wired network-on-chip determines that, when chip production out after, the topology in chip is decided by the metal connecting line of cmos circuit.And the node of wireless network-on-chip can construct specific virtual topology by wireless channel, if the access way of concept transfer, network topology changes therewith.So in wireless network-on-chip, the performance that can compare mapping scheme between task and node under different topology structure is good and bad, according to different mission requirements, the optimal topology finding performance to be better than wired network-on-chip selects and mapping scheme.
By Cha Xin with extensively collect documents and materials, it has been found that disclosed have following a few class for finding on network-on-chip the method for mapping scheme between task and node:
Document " Zhou Ganmin; Yin Yongsheng; Hu Yonghua etc. NoC based on ant colony optimization algorithm maps. computer engineering and application .2005; 41 (18): 7-10. " use one ant colony optimization algorithm effectively to achieve NoC mapping, while by duty mapping to network-on-chip node, decrease the communication power consumption of system as far as possible.The object function of the method is power consumption, the method approximately as: according to the number of Formica fusca cyclically for each unappropriated node according to probability assignments task (this probability is the probability that a node is assigned to certain task calculated by a kind of mode), every circulation primary will make all nodes all be assigned to corresponding task, then use two exchange processes to carry out Local Search, find out the optimal solution that power consumption is minimum, and the pheromone of Formica fusca corresponding to optimal solution is updated by the value of the power consumption number pheromone dough softening of optimal solution with current information element, after to said process successive ignition, finally select the allocative decision making power consumption minimum.The mapping scheme that this method obtains has less communication power consumption, but does not accounts for the delay performance of system.
Document " Yang Shengguang; Li Li; brilliant human relations etc. towards the NoC mapping method of energy consumption and time delay. electronic letters, vol .2008; 36 (5): 937-942 " based on lattice structure NoC platform, establish and based on ant group algorithm be intended to optimization system communication energy consumption and the unified target function of link load variance.The calculating formula of the object function cost of the method is:
Cost=λ × E+ (1-λ) × VAR
Wherein, l is proportionality coefficient, and E is energy consumption, and VAR is link load variance.The method approximately as: (1) according to the number of Formica fusca cyclically for each unappropriated node according to probability assignments task (probability is the probability that a node is assigned to certain task calculated by a kind of mode), every circulation primary will make all nodes all be assigned to corresponding task, then use two exchange processes to carry out Local Search, find out power consumption minimal solution, and the pheromone of Formica fusca corresponding to optimal solution is updated by the value of the power consumption number pheromone dough softening of optimal solution with current information element, after to said process successive ignition, finally select the allocative decision making object function cost minimum.This method does not accounts for communication energy consumption and the execution time is diverse physical quantity, it is thus achieved that mapping scheme not necessarily have excellence performance.
Document Tang lei, Shashi Kumar.A two-step genetic algorithm for mapping task graphs to a network on chip architecture.Proceedings of the Euromicro Symposium on Digital System Design 2003.Antalya, Turkey:IEEE, 2003:180-187. propose a kind of method using two step genetic algorithms to realize duty mapping problem.The method approximately as:
(1) assume initially that in Flow chart task, the time delay on all limits is equal, be a constant, i.e. average；For each task distribution function unit (such as processor, memorizer etc.) so that total time delay of system is minimum；
(2) use the limit time delay of real data flow diagram, optimize further, by functional unit allocation to the node of NoC.
The whole optimizing phase be divide into two steps by the method, thus reduces the complexity of calculating, shortens the calculating time, but the energy consumption of the mapping scheme obtained and time delay will not be the most minimum.
nullDocument Wenbiao Zhou,Yan Zhang,Zhigang Mao.Pareto based Multi-objective Mapping IP Cores onto NoC Architectures.IEEE Asia Pacific Conference on Circuits and Systems.2006:331-334. and Giuseppe Ascia,Vincenzo Catania,Maurizio Palesi.Multi-objective Mapping for Mesh-based NoC Architectures.International Conference on Hardware/Software Dodesign and System ynthesis,2004:182-187. proposes to use multi-objective Evolutionary Algorithm task on network-on-chip to be mapped on the node of on-chip network structure based on Mesh topology.The object function of the method has two, and an object function is average data bag jumping figure (hop), and calculating formula is as follows:
Another object function is hot(test)-spot temperature (T), calculating formula is: T=max (R ' YP), wherein, R is transferred-impedance matrix, utilize K.Skadron, M.R.Stan and W.Huang et al. to be published in the article " Temperature-Aware Micro-architecture " that ISCA ' 03 meeting paper is concentrated heat modeling tool HotSpot proposed to obtain.The method approximately as:
(1) some forecast scheme configuration initialization scheme groups are first randomly choosed, it is published in, according to N.Srinivas and K.Deb, the method that " the Multi-objective optimization function optimization using non-dominated sorting genetic algorithms " of Evolutionary Computation1995 volume Two propose to carry out the duplication of genetic algorithm of non-dominant, intersection and mutation operation and generate new scheme group, then repeat above-mentioned duplication, intersect with mutation operation for several times, select the scheme minimum so that object function.The non-dominated ranking computational complexity used in the method performance higher, that solve is not the most outstanding.
Document " Chen Yiou. towards on-chip network structure and the mapping techniques research of real-time complication system " propose to use multi-objective Evolutionary Algorithm task on wired network-on-chip to be mapped on the node of on-chip network structure based on Mesh topology.The object function of the method has two, and one is wired network-on-chip energy consumption, and calculating formula can be:
E=EPE+ENet
Wherein, E is energy consumption,PEE is the energy consumption sum of all process nodes on network-on-chip,NetE is the energy consumption connecting each communication network processing node on network-on-chip.The energy consumption of communication networkNetE can use Terry Tao Ye, Luca Benini, Giovanni De Micheli is published in the DAC ' proceeding of 02. year " Analysis of Power Consumption on Switch Fabrics in Network Routers " Calculation Method of Energy Consumption or J.Hu and R.Marculescu that propose in a literary composition be published in IEEE Transaction on
The method proposed in the articles of volume 24 in 2005 " Energy-and performance-aware mapping for regular NoC architectures " of Computer-Aided Design of Integrated Circuits and Systems calculates.
Another object function is corresponding time, i.e. data from being input to the longest time delay of output experience, and its calculating formula is:
D=P+TR+TW
Wherein, D is response time, and P is the process time sum being input to all nodes on outgoing route, and TR is the time delay sum being input to all links on outgoing route, and TW is to be input to the queuing of packet and waiting time sum on outgoing route.The method approximately as:
(1) colony is first randomly generatedtP, the method that the A " Fast Elitist Non-dominated Sorting Genetic Algorithm for Multi-objective Optimization:NSGA-II " being published on IEEE Transactions on Evolutionary Computation volume six according to Deb proposes carries out the selection of genetic algorithm of non-dominant, intersection and mutation operation and generates new scheme group, then above-mentioned selection, intersection and mutation operation are repeated, until reaching maximum iteration time, finally select the scheme minimum so that object function.The method, when processing the scheme with identical crowding distance, can only carry out random alignment, it is impossible to preferably reflect practical situation.
Power consumption and response time are separated the preferential index as assessed form by the present invention, but during finding optimal case, consider power consumption and response time simultaneously, accomplish that the scheme found out makes power consumption and system delay the most minimum, additionally select to make power consumption and the topological structure of response time minimum, can comprehensively solve the not enough problem of each scheme above-mentioned.
Summary of the invention
It is an object of the invention to overcome the above-mentioned deficiency in the presence of prior art, the method for designing of mapping scheme and topological structure between task and node is provided in a kind of network-on-chip, power consumption and response time are separated the preferential index as assessed form by the method, but during finding optimal case, consider power consumption and response time simultaneously, accomplish that the scheme found out makes power consumption and system delay the most minimum, additionally select to make power consumption and the topological structure of response time minimum, can comprehensively solve the not enough problem of each scheme of prior art.
In order to solve the problems referred to above, the technical solution used in the present invention is:
The method for designing of mapping scheme and topological structure between task and node in a kind of network-on-chip, with vector X=(x1, x2... xn..., xN) represent a kind of scheme that N number of task of network-on-chip maps, the x in the nth position of vector X to M noden(0≤xn≤ M-1) represent n-th duty mapping node serial number to certain node, i.e. xnIllustrate and the task of numbered n is allocated to xthnNode, comprise the steps:
1, topological structure T to be selected of mapping scheme between wireless network-on-chip task and node is generated1,T2,…,TG, different topological structure to be selected selects the mapping scheme of optimum；Concretely comprise the following steps step 2 and arrive step 20, initialize g=1；
2, to g topology Tg to be selected, stochastic generation one comprises scheme group J of K allocative decisiont, and be each allocative decision X=(x1, x2... xn..., xN) set a Probability p intersectediProbability q with variationi, and total iterations T, and make the variable t=0 of the number of times of record iteration；
3, scheme group J is obtainedtIn network-on-chip power consumption and response time under every kind of scheme:
Utilize calculating formula P=PN+PPEGo out power consumption P of wireless network-on-chip, wherein, PNIt is the power consumption fixed value that on only relevant with carrier number on electromagnetic waveguide transmission line wireless network-on-chip, RF-I link consumes, PPEIt it is the power consumption sum of all nodes on wireless network-on-chip；
Response time is the time interval being input to output from Flow chart task, and its calculating formula is D=TP+TD+TW, wherein, TPFor being input to the process time of all tasks on outgoing route, TDFor the data transfer time sum being input on outgoing route, TWFor transmitting procedure is queued up and the time waited due to the congested packet caused；
To scheme group JtIn every kind of scheme, it is judged that whether it meets following four condition: a task can only be mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time；If being unsatisfactory for, then its power consumption is set to scheme group JtMiddle power consumption maximum, response time is set to scheme group JtResponse time greatest length；
4, scheme group J is calculatedtIn the fitness value of each scheme, the priority of this fitness value size reflection scheme, it is the highest be worth the least priority, and what is called priority level is little for excellent by power consumption and response time；
5, by scheme group JtIn K scheme be randomly divided intoIndividual scheme group, wherein the scheme number in L Shi Mei scheme group, select one priority scheme group Q of prepreerence scheme composition in every scheme group according to the dominance relation determined in the 3rd stept；
6, by priority scheme group QtIn scheme press random pair, described pairing refers to be grouped according to two schemes one group, and with the Probability p of intersection set in step 2iThe numbering of the node in any one same position in two schemes in every pair of scheme is interchangeable, more all be combined by every pair of scheme after exchanging is obtained scheme group Qt′；
7, with the mutation probability q set in step 2iChange scheme group QtIn ', the numbering of the node on any one position in each scheme, obtains scheme group Gt；
8, the fitness value calculated by the 4th step order from small to large, from scheme group JtMiddle selectionIndividual scheme, with scheme group GtScheme combine, constitute and comprise new scheme group G of K schemet′；
9, by scheme group JtAnd Gt' it is merged into scheme group Rt；
10, numerical procedure group RtIn network-on-chip power consumption and response time under every kind of scheme；
11, by scheme group RtIn 2K scheme be divided into multiple group, each group represents a boundary set, and to little group #, the priority of the scheme numbering the group bigger than numbering of the scheme in less group is high；So-called priority level is little for excellent by power consumption and response time；
12, new scheme group F not having scheme is initializedt, then from the group that the 11st step has been divided, by the numbering of group order from big to small, successively the scheme in group is joined FtIn, until scheme group FtIn scheme number more than K, then by the scheme of group that is eventually adding from FtMiddle taking-up；
13, determining in the 11st step the order of priority of each scheme in each group, concrete grammar is as follows:
13-1 makes the distance value of each scheme in group be 0, and described distance value represents that the program is the tightst with contacting of other scheme；
The size of the power consumption of each scheme that 13-2 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that power consumption is less comes before the scheme that power consumption is bigger；
The response time size of each scheme that 13-3 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that response time is less comes before the scheme that response time is bigger；
13-4 is for scheme ordering in step 13-2, calculate the power consumption distance of each scheme, the power consumption distance of each scheme is come it below the 1st, the 2nd ..., the power consumption of i-th scheme respectively correspondingly deducts the 1st come before it, 2nd ..., the difference addition obtained by the power consumption of i-th scheme, then divided by the value of i gained, wherein the value of i meets relational expression: i≤log22K；
13-5 is for scheme ordering in step 13-3, calculate the response time distance of each scheme, the response time distance of each scheme is come it below the 1st, 2nd ..., the power consumption of i-th scheme respectively correspondingly deducts the 1st come before it, 2nd, ..., the difference addition obtained by the power consumption of i-th scheme, then divided by the value of i gained；
Response time distance and the power consumption of each scheme are obtained the distance value of scheme apart from addition by 13-6；
Scheme is ranked up by 13-7 according to distance size, and the scheme that distance value is bigger comes before the scheme that distance value is less, and the priority ratio so coming scheme above comes the priority height of scheme below；
14, in the group taken out the 12nd step after have passed through the 13rd step sequence, according to priority height selection scheme adds scheme group Ft, until scheme group FtIn scheme number be K；
15, by scheme group FtIn K scheme be randomly divided intoIndividual scheme group, wherein the scheme number in L Shi Mei scheme group, combine, according to the 10th step and the 12nd step, the dominance relation determined and select one priority scheme group F of prepreerence scheme composition in every scheme groupt′；
16, by priority scheme group FtScheme in ' random pair again, the Probability p of the intersection to set in step 2iThe numbering of the node in any one same position in two schemes in every pair of scheme is interchangeable, more all be combined by every pair of scheme after exchanging is obtained scheme group Ht；
17, the probability q of the variation to set in step 2iChange scheme group HtIn the numbering of the node on any one position in each scheme, obtain scheme group Ht′；
18, combine, according to the 11st step and the 13rd step, scheme group F determinedtThe order of priority of middle scheme, according to priority order from high to low, from scheme group FtMiddle selectionIndividual scheme, with scheme group Ht' scheme combine, constitute and comprise new scheme group J of K schemet+1；
If 19 t < T, then t=t+1, then return to step 3 repeat above-mentioned in steps, otherwise enter step 20；
20, scheme group J is checkedtIn each scheme whether meet following four condition: a task can only be mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time；If being unsatisfactory for, then the fitness value of the program is set to maximum；Otherwise, scheme group J is calculated according to the method described in step 4tIn the fitness value of each scheme, select the scheme that the scheme that fitness value is minimum, i.e. power consumption and response time are the most minimum, as the optimum task of g topological structure to be selected to internodal mapping scheme Jtg, enter step 21；
If 21 g < G, then g=g+1, enter step 2, repeat above-mentioned in steps；Otherwise, step 22 is entered；
22, topological structure T to be selected is obtained1,T2,…TgCorresponding G kind optimal case is respectivelyCalculateIn the fitness value of each scheme, select the scheme that fitness is minimum, as the task of wireless network-on-chip to internodal mapping scheme, using topological g corresponding for the program as the topological structure of wireless network-on-chip.
Compared with prior art, beneficial effects of the present invention:
The present invention is by separating the preferential index as assessed form using power consumption and response time, but during finding optimal case, consider power consumption and response time the most simultaneously, accomplish that the scheme found out makes power consumption and system delay the most minimum, so that using the performance of this network out of conceptual design the most out to reach best, and compare finding the task method with the priority scheme mapped between node under multiple object functions with other, there is complexity, realize simple feature, additionally select to make power consumption and the topological structure of response time minimum, solve the not enough problem of each scheme of prior art.
Accompanying drawing illustrates:
The task image that Fig. 1 is made up of 6 tasks；
The network-on-chip that Fig. 2 is made up of 16 nodes；
Fig. 3 is to be layered wireless interconnected structure；
Fig. 4 is 2D-Mesh topological structure；
Fig. 5 is 2D-DeBruijn topological structure；
Fig. 6 is 3D-Mesh topological structure；
Wherein 1 is task, and 2 is the portfolio between task, and 3 is the process node of network-on-chip, 4 is the routing node of network-on-chip, 5 is routing node and the line processed between node, and 6 is wireless routing node, and 7 is the metal wire that wireless routing node is connected with process node.
Detailed description of the invention
Below in conjunction with detailed description of the invention, the present invention is described in further detail.But this should not being interpreted as, the scope of the above-mentioned theme of the present invention is only limitted to below example, and all technology realized based on present invention belong to the scope of the present invention.
The present invention is by separating the preferential index as assessed form using power consumption and response time, but during finding optimal case, consider power consumption and response time the most simultaneously, accomplish that the scheme found out makes power consumption and system delay the most minimum, so that using the performance of this network out of conceptual design the most out to reach best, and compare finding the task method with the priority scheme mapped between node under multiple object functions with other, there is complexity, realize simple feature, additionally select to make power consumption and the topology of response time minimum
Structure, solves the not enough problem of each scheme of prior art.For the ease of describing, we use vector X=(x1, x2... xn..., xN) represent a kind of scheme that N number of task of network-on-chip maps, the x in the nth position of vector X to M noden(0≤xn≤ M-1) represent n-th duty mapping node serial number to certain node, i.e. xnIllustrate and the task of numbered n is allocated to xthnNode.
The step finding the energy consumption scheme the most minimum with response time is as follows:
1, topological structure T to be selected of mapping scheme between wireless network-on-chip task and node is generated1,T2,…,TG, different topological structure to be selected selects the mapping scheme of optimum；Concretely comprise the following steps step 2 and arrive step 20, initialize g=1；
2, to g topology Tg to be selected, stochastic generation one comprises scheme group J of K allocative decisiont, and be each allocative decision X=(x1, x2... xn..., xN) set a Probability p intersectediProbability q with variationi, and total iterations T, and make the variable t=0 of the number of times of record iteration；
3, scheme group J is obtainedtIn network-on-chip power consumption and response time under every kind of scheme:
3-1 according to document " Chen Yiou. towards the on-chip network structure of real-time complication system and mapping techniques research " in the method mentioned, utilize calculating formula P=PN+PPEGo out power consumption P of wireless network-on-chip, wherein, PNnullIt it is the power consumption fixed value that on only relevant with carrier number on electromagnetic waveguide transmission line wireless network-on-chip, RF-I link consumes,Can be according to M.Frank Chang，Article " CMP network-on-chip overlaid with multi-band RF-interconnect " that Jason Cong et al. delivered in 2008 or M.Frank Chang，The method that Jason Cong et al. mentions in the article " Power Reduction of CMP Communication Networks via RF-Interconnects " delivered for 2008 calculates PN, PPEIt it is the power consumption sum of all nodes on wireless network-on-chip.
3-2 response time is the time interval being input to output from Flow chart task, and its calculating formula is D=TP+TD+TW, wherein, TPFor being input to the process time of all tasks on outgoing route, TDFor the data transfer time sum being input on outgoing route, TWFor transmitting procedure is queued up and the time waited due to the congested packet caused.
3-3 is to scheme group JtIn every kind of scheme, it is judged that whether it meets following four condition: a task can only be mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time；If being unsatisfactory for, then its power consumption is set to scheme group JtMiddle power consumption maximum, response time is set to scheme group JtResponse time greatest length；
4, use the non-dominated ranking method of Deb, exclusive method, banker method, challenge tournament method, recurrence method, quick sort, (within 2007, Science Press publishes the quick sort improved, " multi-objective Evolutionary Algorithm and application thereof " that author is Zheng Jinhua is introduced) or one of the method such as the fitness value calculation method introduced in calendar year 2001 publishes an article SPEA2:Improving the strength Pareto evolutionary algorithm of Zitzler calculate the fitness value of each scheme in scheme group, the priority of this fitness value size reflection scheme, it is worth the least priority the highest, so-called priority level is little for excellent by power consumption and response time；
5, by scheme group JtIn K scheme be randomly divided intoIndividual scheme group, wherein the scheme number in L Shi Mei scheme group, select one priority scheme group Q of prepreerence scheme composition in every scheme group according to the dominance relation determined in the 3rd stept；
6, by priority scheme group QtIn scheme press random pair, described pairing refers to be grouped according to two schemes one group, and with the Probability p of intersection set in step 2iThe numbering of the node in any one same position in two schemes in every pair of scheme is interchangeable, more all be combined by every pair of scheme after exchanging is obtained scheme group Qt′；
7, with the mutation probability q set in step 2iChange scheme group QtIn ', the numbering of the node on any one position in each scheme, obtains scheme group Gt；
8, the fitness value calculated by the 4th step order from small to large, from scheme group JtMiddle selectionIndividual scheme, with scheme group GtScheme combine, constitute and comprise new scheme group G of K schemet′。
9, by scheme group JtAnd Gt' it is merged into scheme group Rt。
10, numerical procedure group RtIn network-on-chip power consumption and response time under every kind of scheme；
null11、Use NPGA、SPEA2、PESA、PAES、MGAMOO、(within 2007, Science Press publishes MOMGA，One of " multi-objective Evolutionary Algorithm and application thereof " that author is Zheng Jinhua is introduced) or the method such as the boundary set generation method introduced in the article A Fast Elitist Non-dominated Sorting Genetic Algorithm for Multi-objective Optimization:NSGA-II being published on magazine IEEE Transactions on Evolutionary Computation volume six for 2002 of Deb，By scheme group RtIn 2K scheme be divided into multiple group, each group represents a boundary set, and to little group #, the priority of the scheme numbering the group bigger than numbering of the scheme in less group is high；So-called priority level is little for excellent by power consumption and response time.
12, new scheme group F not having scheme is initializedt, then from the group that the 11st step has been divided, by the numbering of group order from big to small, successively the scheme in group is joined FtIn, until scheme group FtIn scheme number more than K, then by the scheme of group that is eventually adding from FtMiddle taking-up；
13, determining in the 11st step the order of priority of each scheme in each group, concrete grammar is as follows:
13-1 makes the distance value of each scheme in group be 0, and described distance value represents that the program is the tightst with contacting of other scheme；
The size of the power consumption of each scheme that 13-2 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that power consumption is less comes before the scheme that power consumption is bigger；
The response time size of each scheme that 13-3 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that response time is less comes before the scheme that response time is bigger；
13-4, for scheme ordering in step 13-2, calculates the power consumption distance of each scheme, the power consumption of each scheme distance for come it below the 1st, the 2nd ..., i-th (value of i meets relational expression: i≤log22K) power consumption of individual scheme respectively correspondingly deducts the 1st come before it, the 2nd ..., the difference addition obtained by the power consumption of i-th scheme, then divided by the value of i gained.
13-5 is for scheme ordering in step 13-3, calculate the response time distance of each scheme, the response time distance of each scheme is come it below the 1st, 2nd ..., the power consumption of i-th scheme respectively correspondingly deducts the 1st come before it, 2nd, ..., the difference addition obtained by the power consumption of i-th scheme, then divided by the value of i gained；
Response time distance and the power consumption of each scheme are obtained the distance value of scheme apart from addition by 13-6；
Scheme is ranked up by 13-7 according to distance size, and the scheme that distance value is bigger comes before the scheme that distance value is less, and the priority ratio so coming scheme above comes the priority height of scheme below；
14, in the group taken out the 12nd step after have passed through the 13rd step sequence, according to priority height selection scheme adds scheme group Ft, until scheme group FtIn scheme number be K；
15, by scheme group FtIn K scheme be randomly divided intoIndividual scheme group, wherein the scheme number in L Shi Mei scheme group, combine, according to the 10th step and the 12nd step, the dominance relation determined and select one priority scheme group F of prepreerence scheme composition in every scheme groupt′；
16, by priority scheme group FtScheme in ' random pair again, the Probability p of the intersection to set in step 2iThe numbering of the node in any one same position in two schemes in every pair of scheme is interchangeable, more all be combined by every pair of scheme after exchanging is obtained scheme group Ht；
17, the probability q of the variation to set in step 2iChange scheme group HtIn the numbering of the node on any one position in each scheme, obtain scheme group Ht′；
18, combine, according to the 11st step and the 13rd step, scheme group F determinedtThe order of priority of middle scheme, according to priority order from high to low, from scheme group FtMiddle selectionIndividual scheme, with scheme group Ht' scheme combine, constitute and comprise new scheme group J of K schemet+1；
If 19 t < T, then t=t+1, then return to step 3, otherwise enter step 20；
20, scheme group J is checkedtIn each scheme whether meet following four condition: a task can only be mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time；If being unsatisfactory for, then the fitness value of the program is set to maximum；Otherwise, scheme group J is calculated according to the method described in step 4tIn the fitness value of each scheme, select the scheme that the scheme that fitness value is minimum, i.e. power consumption and response time are the most minimum, as the optimum task of g topological structure to be selected to internodal mapping scheme Jtg。
If 21 g < G, then g=g+1, enter step 2；Otherwise, step 22 is entered；
22, topological structure T to be selected is obtained1,T2,…TgCorresponding G kind optimal case is respectivelyUse the non-dominated ranking method of Deb, exclusive method, banker method, challenge tournament method, recurrence method, quick sort, (within 2007, Science Press publishes the quick sort of improvement, one of " multi-objective Evolutionary Algorithm and application thereof " that author is Zheng Jinhua is introduced) or the method such as the fitness value calculation method introduced in calendar year 2001 publishes an article SPEA2:Improving the strength Pareto evolutionary algorithm of Zitzler, calculateIn the fitness value of each scheme, select the scheme that fitness is minimum, as the task of wireless network-on-chip to internodal mapping scheme, using topological g corresponding for the program as the topological structure of wireless network-on-chip.
Below with by the duty mapping of the task image shown in Fig. 1 to the node of the network-on-chip shown in Fig. 2, and as a example by selecting in the topological structure shown in Fig. 4, Fig. 5 and Fig. 6 so that power consumption and the most minimum topological structure of response time, illustrate.Having 6 tasks in task image shown in Fig. 1, the network-on-chip shown in Fig. 2 has 16 nodes.The present embodiment is contemplated to find in each topological structure that a kind of wireless network-on-chip 6 shown in Fig. 1 task being assigned to shown in Fig. 3 generates, and in all topological structure, the power consumption of wireless network-on-chip and response time minimize the mapping scheme of value simultaneously after looking for mapping.For the ease of describing, we use vector X=(x1,x2,x3,x4,x5,x6) represent a kind of scheme that 6 tasks of network-on-chip map, the x in the nth position of vector X to 16 nodesn(0≤xn≤ M-1) represent n-th duty mapping node serial number to certain node, i.e. xnIllustrate and the task of numbered n is allocated to xthnNode.
The step finding the power consumption scheme the most minimum with response time is as follows:
1, generating topological structure T1 to be selected of mapping scheme between 3 wireless network-on-chip tasks and node, T2, T3 correspond to respectively: 2D-Mesh topology, 2D-DeBruijn topology, and 3d-Mesh topology, respectively correspondingly such as Fig. 4, shown in Fig. 5, Fig. 6.Initializing g=1, topological Numbers G=3 the most to be selected, now, in global layer, wireless routing node generates Mesh topology as required；
2, to the 1st topology to be selected, i.e. Mesh topological structure, stochastic generation one comprises scheme group J of 20 allocative decisionstnull，All schemes are successively by 1,2,...,20 numberings，Each allocative decision vector representation in scheme group is: (11,1,15,9,0,15),(4,1,0,8，1,6),(7,5,15,8,7,8),(10,11,5,0,6,8),(0,6,7,6,15,9),(14,11,11,0,8，5),(15,5,14,0,15,1),(3,0,8,7,1,9),(2,2,1,8,2,8),(3,13,8,15,5,8),(8,8,9,12,10,2),(14,0,3,2,15,5),(5,2,4,9,0,8),(7,2,15,11,1,11),(7,1,15,2,1,11),(9,12,3,9,10,11),(11,3,12,14,14,6),(11,12,4,7,1,9),(4,8,12,12,9,4),(1,15,6,9,3,9).Setting the probability of intersection of each scheme as 0.4, the probability of variation is 0.01, and total iterations is 100, and records the variable t=0 of the number of times of iteration；
3, scheme group J is obtainedtIn the power consumption of network-on-chip and response time under every kind of scheme, concrete grammar is as follows:
3-1, the computing formula of power consumption (P) use: P=PN+PPE, PNThe power consumption fixed value consumed for RF-I link on the wireless network-on-chip that carrier number on electromagnetic waveguide transmission line is relevant,PEP is the power consumption sum of all nodes on wireless network-on-chip.On network, the power consumption of all process nodes while i.e. determining task image, can determine the power consumption of task by estimating to obtain to the operand of each task equal to the process time of all tasks in task image, the process time of each task.In this example, for various different schemes, PPEValue perseverance be 1.18." the power consumption calculation method proposed in Power Reduction of CMP Communication Networks via RF-Interconnects mono-literary composition calculates power consumption P that on the wireless network-on-chip of various scheme, RF-I link consumes to the article using M.Frank Chang, Jason Cong et al. to deliver in 2008N, it is respectively as follows: 18.05,12.57,20.36,15.13,14.47,15.31,24.81,13.81,15.51,16.03,10.92,18.23,12.59,14.48,16.53,13.51,14.73,14.48,9.72,17.73.
3-2, the calculating formula of response time (D) use: D=P+TR+TW, P is the process time sum being input to all nodes on outgoing route, TR is the time delay sum being input to all links on outgoing route, and TW is to be input to the queuing of packet and waiting time sum on outgoing route.In this example, the response time of each scheme is respectively as follows: 0.16,0.1601,0.1601,0.1601,0.1601,0.1601,0.1602,0.1601,0.1601,0.1601,0.1601,0.1601,0.16,0.16,0.1601,0.1601,0.1601,0.1601,0.16,0.1601.
3-3, judge whether these 20 kinds of schemes meet following four condition: a task can only be mapped on a node；The processing speed of data should be greater than data and reaches rate；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time.Through judging, serial number 1,2, the scheme of 3,5,6,7,9,10,11,14,15,16,17,19,20 is unsatisfactory for condition, and the power consumption of these schemes is set to maximum in scheme group, and i.e. 24.81, response time is set to maximum in scheme group, i.e. 0.1601.
4, the non-dominated ranking method (Science Press published in 2007, and author is " multi-objective Evolutionary Algorithm and application thereof " middle introduction of Zheng Jinhua) using Deb calculates scheme group JtIn the fitness value of each scheme be:
0.7895,1.6316,0.1579,1,1.1579,0.1579,0,1.3684,0.4737,0.4737,1.6316,0.4737,1.8947,1.6316,0.4737,1.6316,1,1.2632,2,0.7895, the priority of this value size reflection scheme, it is worth the least priority the highest.
5, by scheme group JtIn 20 schemes be randomly divided into 10 scheme groups, every scheme group there are 2 schemes, the prepreerence scheme of every scheme group, a prepreerence scheme one new scheme group Q of composition then will selected is selected from all scheme groups according to the dominance relation determined in the 3rd stept, new scheme group QtIn scheme be: (4,1,0,8,1,6), (10,11,5,0,6,8), (0,6,7,6,15,9), (3,0,8,7,1,9), (2,2,1,8,2,8), (8,8,9,12,10,2), (5,2,4,9,0,8), (9,12,3,9,10,11), (11,12,4,7,1,9), (4,8,12,12,9,3).
6, by scheme group QtIn scheme press random pair, described pairing refers to be grouped according to two schemes one group, and with the probability 0.4 of intersection set in step 1, the numbering of the node on random site between every pair of scheme is interchangeable, more all be combined by every pair of scheme after exchanging is obtained scheme group Qt', scheme group QtScheme in ' is: (10,1,0,8,1,6), (4,11,5,0,6,8), (0,6,7,6,15,9), (3,0,8,7,1,9), (2,2,1,8,2,8), (8,8,9,12,10,2), (5,2,4,9,0,8), (9,12,3,9,10,11), (11,12,4,7,1,9), (4,8,12,12,9,3).
7, scheme group Q is changed with the mutation probability 0.01 set in step 1tIn ', the numbering of the node on random site in each scheme, obtains scheme group Gt, scheme group GtIn scheme be: (10,1,0,8,1,6), (4,11,5,0,6,8), (0,6,7,6,15,9), (3,0,8,7,1,9), (2,2,1,8,2,8), (8,8,9,12,10,2), (5,2,4,9,0,8), (9,12,3,9,10,11), (11,12,4,7,1,9), (4,8,12,12,9,3).
8, by fitness order from small to large, from scheme group Jt10 schemes of middle selection, with scheme group GtScheme combine, constitute and comprise new scheme group G of 20 schemest′。
9, by scheme group JtAnd Gt' it is merged into scheme group R comprising 40 schemestnull，That is: (11，1，15，9，0，15),(4，1，0，8，1，6),(7，5，15，8，7，8),(10，11，5，0，6，8),(0，6，7，6，15，9),(14，11，11，0，8，5),(15，5，14，0，15，1),(3，0，8，7，1，9),(2，2，1，8，2，8),(3，1，3，8，15，5，8),(8，8，9，12，10，2),(14，0，3，2，15，5),(5，2，4，9，0，8),(7，2，15，11，1，11),(7，1，15，2，1，11),(9，12，3，9，10，11),(11，3，12，14，14，6),(11，12，4，7，1，9),(4，8，12，12，9，4),(1，15，6，9，3，9),(4，1，0，8，1，6),(10，11，5，0，6，8),(0，6，7，6，15，9),(3，0，8，7，1，9),(2，2，1，8，2，8),(8，8，9，12，10，2),(5，2，4，9，0，8),(9，12，3，9，10，11),(11，12，4，7，1，9),(4，8，12，12，9，3),(11，3，12，14，14，6),(0，6，7，6，15，9),(11，12，4，7，1，9),(3，0，8，7，1，9),(4，1，0，8，1，6),(8，8，9，12，10，2),(7，2，15，11，1，11),(9，12，3，9，10，11),(5，2，4，9，0，8),(4，8，12，12，9，4).
10, numerical procedure group RtIn network-on-chip power consumption and response time under every kind of scheme, the power consumption of each scheme is respectively as follows:
18.05,12.57,20.36,15.13,14.47,15.31,24.81,13.81,15.51,16.03,10.92,18.23,12.59,14.48,16.53,13.51,14.73,14.48,9.72,17.73,12.57,15.13,14.47,13.81,15.51,10.92,12.59,13.51,14.48,13.48,14.73,14.47,14.48,13.81,12.57,10.92,14.48,13.51,12.59,9.72；The response time of each scheme is respectively as follows: 0.16,0.1601,0.1601,0.1601,0.1601,0.1601,0.1602,0.1601,0. 1601,0.1601,0.1601,0.1601,0.16,0.16,0.1601,0.1601,0.1601,0.1601,0.16,
0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.16,0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.1601,0.16,0.1601,0.16,0.16。
11, use one of boundary set generation method that Deb introduces in the article A Fast Elitist Non-dominated Sorting Genetic Algorithm for Multi-objective Optimization:NSGA-II being published on magazine IEEE Transactions on Evolutionary Computation volume six for 2002 by scheme group RtIn 40 schemes be divided into 13 groups, each group represents a boundary set, and to little group #, the priority of the scheme numbering the group bigger than numbering of the scheme in less group is high；So-called priority level is little for excellent by power consumption and response time.nullScheme in the group 1 obtained as stated above has scheme 19，Scheme 40，Scheme inside group 2 has scheme 13,Scheme 27,Scheme 39,Scheme in group 3 has scheme 2,Scheme 11,Scheme 21,Scheme 26,Scheme 35,Scheme 36,Scheme inside group 4 has scheme 16,Scheme 28,Scheme 38,Scheme in group 5 has scheme 8,Scheme 24,Scheme 34,Scheme in group 6 has scheme 14,Scheme 37,Scheme inside group 7 has scheme 18,Scheme 29,Scheme 33,Scheme in group 8 has scheme 5,Scheme 23,Scheme 32,Scheme inside group 9 has scheme 4,Scheme 17,Scheme 22,Scheme 30,Scheme 31,Scheme in group 10 has scheme 1,Scheme 20,Scheme inside group 11 has scheme 9,Scheme 10,Scheme 12,Scheme 15,Scheme 25,Scheme in group 12 has scheme 3,Scheme 6，Scheme in group 13 has scheme 7.
12, new scheme group F not having scheme is initializedt, then from the group that the 10th step has been divided, by the numbering of group order from small to large, from group 1 to group 13, successively the scheme in group is joined FtIn, until scheme group FtIn scheme number more than 20, now by the scheme of group (i.e. group 7) that is eventually adding from FtMiddle taking-up.
13, determining the order of priority of each scheme in each group obtained in the 11st step, concrete grammar is as follows:
13-1 makes the distance value of each scheme in group be 0, and described distance value represents that the program is the tightst with contacting of other scheme.
The size of the power consumption of each scheme that 13-2 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that power consumption is less comes before the scheme that power consumption is bigger.
The response time size of each scheme that 13-3 calculates according to the 10th step carries out inverted order arrangement to scheme, and the scheme that response time is less comes before the scheme that response time is bigger.
13-4 is for scheme ordering in 13-2, calculate the power consumption distance of each scheme, the power consumption distance of each scheme is come it below the 1st, 2nd ..., the power consumption of the 5th scheme respectively correspondingly deducts the 1st come before it, 2nd, ..., the difference addition obtained by the power consumption of the 5th scheme, then divided by the value of 5 gained.
13-5 is for scheme ordering in 13-3, calculate the response time distance of each scheme, the response time distance of each scheme is come it below the 1st, 2nd ..., the power consumption of the 5th scheme respectively correspondingly deducts the 1st come before it, 2nd, ..., the difference addition obtained by the power consumption of the 5th scheme, then divided by the value of 5 gained.
Response time distance and the power consumption of each scheme are obtained the distance value of scheme apart from addition by 13-6.
Scheme is ranked up by 13-7 according to distance size, and the scheme that distance value is bigger comes before the scheme that distance value is less, and the priority ratio so coming scheme above comes the priority height of scheme below.
14, the group after the 13rd step sequence, according to priority height selection scheme adds scheme group Ft, until scheme group FtIn scheme number be 20, after completing, scheme group FtnullIn scheme be: (4，8，12，12，9，4),(4，8，12，12，9，4),(5，2，4，9，0，8),(5，2，4，9，0，8),(5，2，4，9，0，8),(4，1，0，8，1，6),(8，8，9，12，10， 2),(4，1，0，8，1，6),(8，8，9，12，10，2),(4，1，0，8，1，6),(8，8，9，12，10，2),(9，12，3，9，10，11),(9，12，3，9，10，11),(9，12，3，9，10，11),(3，0，8，7，1，9),(3，0，8，7，1，9),(3，0，8，7，1，9),(7，2，15，11，1，11),(7，2，15，11，1，11),(11，12，4，7，1，9).
15, by scheme group FtIn 20 schemes be randomly divided into 10 scheme groups, every scheme group there are 2 schemes, combine, according to the 10th and 12 steps, the dominance relation determined and select the prepreerence scheme of every scheme group, a prepreerence scheme one new scheme group F of composition then will selected from all scheme groupst', scheme group FtScheme in ' includes (4,8,12,12,9,4), (5,2,4,9,0,8), (5,2,4,9,0,8), (8,8,9,12,10,2), (8,8,9,12,10,2), (8,8,9,12,10,2), (9,12,3,9,10,11), (3,0,8,7,1,9), (3,0,8,7,1,9), (7,2,15,11,1,11).
16, by scheme group FtScheme in ' random pair again, the numbering of the node on random site between every pair of scheme is interchangeable by the probability 0.4 of the intersection to set in step 1, more all being combined is obtained scheme group Ht, it may be assumed that (4,8,12,12,9,4), (5,2,4,9,0,8), (8,2,4,9,0,8), (5,8,9,12,10,2), (8,8,9,12,10,2), (8,8,9,12,10,2), (9,12,3,9,10,11), (3,0,8,7,1,9), (3,0,8,7,1,9), (7,2,15,11,1,11).
17, the probability 0.01 of the variation to set in step 1 changes scheme group HtIn the numbering of the node on random site in each scheme, obtain scheme group Ht', it may be assumed that (4,8,12,12,9,4), (5,2,4,9,0,8), (8,2,4,9,0,8), (5,8,9,12,10,2), (8,8,9,12,10,2), (8,8,9,12,10,2), (9,12,3,9,10,11), (3,0,8,7,1,9), (3,0,8,7,1,9), (7,2,15,11,1,11).
18, by fitness order from small to large, from scheme group Ft10 schemes of middle selection, with scheme group Ht' scheme combine, constitute and comprise new scheme group J of 20 schemest+1 。
If 19 iterations t are less than 100, then t=t+1, then return to step 2, carry out new round interative computation, otherwise enter step 20；
Whether each scheme in scheme group that 20, checks again for meets following four condition, and a task can only be mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is less than or equal to provided maximum bandwidth；The iterative boundary of feedback control loop is less than or equal to iteration time.The scheme being unsatisfactory for condition is set to maximum, the scheme of condition will be met, calculate scheme group J according to the method described in step 4100In the fitness value of each scheme, the scheme minimum by fitness value, i.e. scheme (5,2,9,9,5,4) as the optimum task of g in wireless network-on-chip topological structure to be selected to internodal mapping scheme.
If 21 g < G, then g=g+1, enter step 2, repeat above-mentioned in steps, otherwise enter step 22；
22, the topological structure 2D-Mesh to be selected topology obtained, 2D-Debruijn topology, 3 kinds of optimal cases corresponding to 3D-Mesh topology are respectively as follows: 5, 2, 9, 9, 5, 4), (6, 11, 11, 7, 9, 7), (0, 5, 0, 0, 4, 0), the fitness value calculating 3 kinds of optimal cases according to the method described in step 4 is respectively 0.1655, 0.2635, 0.17, select the scheme that fitness value is minimum, i.e. scheme (5, 2, 9, 9, 5, 4), as the task of wireless network-on-chip to internodal mapping scheme, by topology corresponding for the program, i.e. 2D-Mesh topology is as the topological structure of wireless network-on-chip.One inventive point of the present invention selects optimum topology exactly.To different task images, its optimum topology is probably different.In the present embodiment, by the duty mapping of the task image shown in Fig. 1 to the node of the network-on-chip shown in Fig. 2, and select in the topological structure shown in Fig. 4, Fig. 5 and Fig. 6 so that power consumption and the most minimum topological structure of response time.
Above in conjunction with accompanying drawing, the detailed description of the invention of the present invention is described in detail, but the present invention is not restricted to above-mentioned embodiment, in the case of without departing from the spirit and scope of claims hereof, those skilled in the art may be made that various amendment or remodeling.
Claims (1)
1. a method for designing for mapping scheme and topological structure, its feature between task and node in network-on-chip
It is, with vector X=(x1, x2... xn..., xN) represent what N number of task of network-on-chip mapped to M node
A kind of scheme, the x in the nth position of vector XnRepresent n-th duty mapping joint to certain node
Point numbering, i.e. xnIllustrate and the task of numbered n is allocated to xthnNode, wherein, xnValue is
More than or equal to 0, and simultaneously less than or equal to M-1；Comprise the steps:
1, topological structure T to be selected of mapping scheme between wireless network-on-chip task and node is generated1,T2,...,TG,
The mapping scheme of optimum is selected in different topological structure to be selected；Concretely comprise the following steps step 2 and arrive step 20, initially
Change g=1；
2, to g topology Tg to be selected, stochastic generation one comprises scheme group J of K allocative decisiont, and
For each allocative decision X=(x1, x2... xn..., xN) set a Probability p intersectediProbability q with variationi,
And total iterations T, and make the variable t=0 of the number of times of record iteration；
3, scheme group J is obtainedtIn network-on-chip power consumption and response time under every kind of scheme:
Utilize calculating formula P=PN+PPEObtain power consumption P of wireless network-on-chip, wherein, PNBe only with electromagnetic wave
Lead the power consumption fixed value that on the wireless network-on-chip that carrier number on transmission line is relevant, RF-I link consumes, PPEIt it is nothing
The power consumption sum of all nodes on line network-on-chip；
Response time is the time interval being input to output from Flow chart task, and its calculating formula is
D=TP+TD+TW, wherein, TPFor being input to the process time of all tasks on outgoing route, TDFor input
Data transfer time sum on outgoing route, TWFor transmitting procedure is arranged due to the congested packet caused
Team and the time of wait；
To scheme group JtIn every kind of scheme, it is judged that whether it meets following four condition: a task can only be reflected
It is mapped on a node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme is little
In or provided maximum bandwidth is provided；The iterative boundary of feedback control loop is less than or equal to iteration time；If no
Meet, then its power consumption is set to scheme group JtMiddle power consumption maximum, response time is set to scheme group JtResponse time
Greatest length；
4, scheme group J is calculatedtIn the fitness value of each scheme, this fitness value size reflection scheme
Priority, it is the highest to be worth the least priority, and so-called priority level is little for excellent by power consumption and response time；
5, by scheme group JtIn K scheme be randomly divided intoIndividual scheme group, wherein in L Shi Mei scheme group
Scheme number, select in every scheme group prepreerence scheme composition according to the dominance relation determined in the 3rd step
One priority scheme group Qt；
6, by priority scheme group QtIn scheme press random pair, described pairing refers to according to two schemes one
Group is grouped, and the Probability p of the intersection to set in step 2iTwo schemes in every pair of scheme are taken up an official post
The numbering of the node in a same position of anticipating is interchangeable, the most again by all by every pair of scheme after exchanging
It is combined and obtains scheme group Qt′；
7, with the mutation probability q set in step 2iChange scheme group QtIn any one position in each scheme in '
The numbering of the node put, obtains scheme group Gt；
8, the fitness value calculated by the 4th step order from small to large, from scheme group JtMiddle selection
Individual scheme, with scheme group GtScheme combine, constitute and comprise new scheme group G of K schemet′；
9, by scheme group JtAnd Gt' it is merged into scheme group Rt；
10, numerical procedure group RtIn network-on-chip power consumption and response time under every kind of scheme；
11, by scheme group RtIn 2K scheme be divided into multiple group, each group represents a boundary set,
To little group #, the priority of the scheme numbering the group bigger than numbering of the scheme in less group is high；
So-called priority level is little for excellent by power consumption and response time；
12, new scheme group F not having scheme is initializedt, then from the group that the 11st step has been divided,
By the numbering of group order from big to small, successively the scheme in group is joined FtIn, until scheme group Ft
In scheme number more than K, then by the scheme of group that is eventually adding from FtIn all take out；
13, determining in the 11st step the order of priority of each scheme in each group, concrete grammar is as follows:
13-1 makes the distance value of each scheme in group be 0, and described distance value represents the program and other scheme
Contact the tightst；
The size of the power consumption of each scheme that 13-2 calculates according to the 10th step carries out inverted order arrangement, merit to scheme
Consume less scheme to come before the scheme that power consumption is bigger；
13-3 carries out inverted order arrangement according to the response time size of each scheme that the 10th step calculates to scheme,
The scheme that response time is less comes before the scheme that response time is bigger；
13-4, for scheme ordering in step 13-2, calculates the power consumption distance of each scheme, each scheme
Power consumption distance is come it below the 1st, the 2nd ..., the power consumption of i-th scheme respectively correspondingly deducts
Come the 1st before it, the 2nd ..., the difference addition obtained by the power consumption of i-th scheme, then removes
With the value of i gained, wherein the value of i meets relational expression: i≤log22K；
13-5, for scheme ordering in step 13-3, calculates the response time distance of each scheme, Mei Gefang
The response time distance of case is come it below the 1st, the 2nd ..., the power consumption of i-th scheme is respectively
Deduct the 1st come before it accordingly, the 2nd ..., the poor phase obtained by the power consumption of i-th scheme
Add, then divided by the value of i gained；
Response time distance and the power consumption of each scheme are obtained the distance value of scheme apart from addition by 13-6；
Scheme is ranked up by 13-7 according to distance size, and it is less that the scheme that distance value is bigger comes distance value
Before scheme, the priority ratio so coming scheme above comes the priority height of scheme below；
14, in the group taken out the 12nd step after have passed through the 13rd step sequence according to priority just
Selection scheme adds scheme group Ft, until scheme group FtIn scheme number be K；
15, by scheme group FtIn K scheme be randomly divided intoIndividual scheme group, wherein L Shi Mei scheme group
In scheme number, combine, according to the 10th step and the 12nd step, the dominance relation determined and select in every scheme group
Prepreerence scheme one priority scheme group F of compositiont′；
16, by priority scheme group FtScheme in ' random pair again, the Probability p of the intersection to set in step 2i
The numbering of the node in any one same position in two schemes in every pair of scheme is interchangeable, then
Again all be combined by every pair of scheme after exchanging is obtained scheme group Ht；
17, the probability q of the variation to set in step 2iChange scheme group HtIn in each scheme
The numbering of the node on any one position, obtains scheme group Ht′；
18, combine, according to the 11st step and the 13rd step, scheme group F determinedtThe order of priority of middle scheme, according to
Priority order from high to low, from scheme group FtMiddle selectionIndividual scheme, with scheme group Ht' scheme
Combine, constitute new scheme group J comprising K schemet+1；
If 19 t < T, then t=t+1, then return to step 3 repeat above-mentioned in steps, otherwise enter
Step 20；
20, scheme group J is checkedtIn each scheme whether meet following four condition: a task can only be mapped to
On one node；The processing speed of data reaches rate more than data；The required bandwidth of mapping scheme less than or
Thered is provided maximum bandwidth is provided；The iterative boundary of feedback control loop is less than or equal to iteration time；If being unsatisfactory for,
Then the fitness value of the program is set to maximum；Otherwise, scheme is calculated according to the method described in step 4
Group JtIn the fitness value of each scheme, when selecting the scheme that fitness value is minimum, i.e. power consumption and response
Between simultaneously minimum scheme, as the optimum task of g topological structure to be selected to internodal mapping
Scheme Jtg, enter step 21；
If 21 g < G, then g=g+1, enter step 2, repeat above-mentioned in steps；Otherwise,
Enter step 22；
22, topological structure T to be selected is obtained1,T2,...TGCorresponding G kind optimal case is respectively
CalculateIn the fitness value of each scheme, select the scheme that fitness is minimum, as nothing
The task of line network-on-chip is to internodal mapping scheme, using topological g corresponding for the program as wireless online
The topological structure of network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410027528.8A CN103761212B (en) | 2014-01-21 | 2014-01-21 | The method for designing of mapping scheme and topological structure between task and node in network-on-chip |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410027528.8A CN103761212B (en) | 2014-01-21 | 2014-01-21 | The method for designing of mapping scheme and topological structure between task and node in network-on-chip |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103761212A CN103761212A (en) | 2014-04-30 |
| CN103761212B true CN103761212B (en) | 2016-08-31 |
Family
ID=50528452
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410027528.8A Expired - Fee Related CN103761212B (en) | 2014-01-21 | 2014-01-21 | The method for designing of mapping scheme and topological structure between task and node in network-on-chip |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103761212B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104270308A (en) * | 2014-10-15 | 2015-01-07 | 重庆大学 | On-radio-frequency-piece network application mapping method facing unbalanced communication feature |
| CN105187313B (en) * | 2015-09-25 | 2018-05-01 | 东北大学 | A kind of Survey on network-on-chip topology and its adaptive routing method |
| CN107483541A (en) * | 2017-07-17 | 2017-12-15 | 广东工业大学 | A kind of online task immigration method based on rolling time horizon |
| CN107391247B (en) * | 2017-07-21 | 2020-06-26 | 同济大学 | Breadth-first greedy mapping method for network-on-chip application |
| CN108416164B (en) * | 2018-03-23 | 2021-09-28 | 南京大学 | Three-dimensional network-on-chip temperature reconstruction system based on limited number of temperature sensors |
| CN114116599A (en) * | 2021-12-17 | 2022-03-01 | 广东工业大学 | Three-dimensional network-on-chip topological structure design method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101799838A (en) * | 2010-01-22 | 2010-08-11 | 电子科技大学 | Method for seeking mapping scheme between tasks and nodes of network on chip |
| CN103428804A (en) * | 2013-07-31 | 2013-12-04 | 电子科技大学 | Method for searching mapping scheme between tasks and nodes of network-on-chip (NoC) and network code position |
-
2014
- 2014-01-21 CN CN201410027528.8A patent/CN103761212B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101799838A (en) * | 2010-01-22 | 2010-08-11 | 电子科技大学 | Method for seeking mapping scheme between tasks and nodes of network on chip |
| CN103428804A (en) * | 2013-07-31 | 2013-12-04 | 电子科技大学 | Method for searching mapping scheme between tasks and nodes of network-on-chip (NoC) and network code position |
Non-Patent Citations (1)
| Title |
|---|
| 三维片上网络拓扑研究;陈亦欧 等;《电信科学》;20090430;第25卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103761212A (en) | 2014-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103761212B (en) | The method for designing of mapping scheme and topological structure between task and node in network-on-chip | |
| CN101799838B (en) | Method for seeking mapping scheme between tasks and nodes of network on chip | |
| Trik et al. | Providing an Adaptive Routing along with a Hybrid Selection Strategy to Increase Efficiency in NoC‐Based Neuromorphic Systems | |
| CN104145281A (en) | Neural network computing apparatus and system, and method therefor | |
| CN105634974B (en) | Route determining methods and device in software defined network | |
| CN106934454B (en) | Test-schedule method in network on three-dimensional chip based on Petri network | |
| Xu et al. | Unified multi‐objective mapping for network‐on‐chip using genetic‐based hyper‐heuristic algorithms | |
| Wang et al. | Latency-aware mapping for 3D NoC using rank-based multi-objective genetic algorithm | |
| Wang et al. | An effective optimization algorithm for application mapping in network-on-chip designs | |
| CN102325089A (en) | Fat tree type network-on-chip mapping method based on differential evolution and predatory search strategy | |
| Vinay Kumar et al. | Multi-culture diversity based self adaptive particle swarm optimization for optimal floorplanning | |
| Selvitopi et al. | A recursive hypergraph bipartitioning framework for reducing bandwidth and latency costs simultaneously | |
| Yang et al. | Trust-based scheduling strategy for cloud workflow applications | |
| Xu et al. | An efficient energy and thermal-aware mapping for regular network-on-chip | |
| Le et al. | Performance-driven assignment and mapping for reliable networks-on-chips | |
| Tutsch et al. | Comparison of network-on-chip topologies for multicore systems considering multicast and local traffic | |
| Gabis et al. | Bi-objective cost function for adaptive routing in network-on-chip | |
| Morgan et al. | Multi-objective optimization of NoC standard architectures using Genetic Algorithms | |
| Ababei | Efficient congestion-oriented custom network-on-chip topology synthesis | |
| Kundu et al. | A comparative performance evaluation of network-on-chip architectures under self-similar traffic | |
| CN109582986A (en) | A kind of method of network on three-dimensional chip test port selection optimization under power consumption limit | |
| Cao et al. | Fast turn restriction algorithm to build deadlock-free modular chiplet integration systems | |
| Kumar et al. | A support vector regression-based approach to predict the performance of 2d & 3d on-chip communication architectures | |
| Rezaei-Ravari et al. | Dynamic clustering-based routing scheme for 2D-mesh networks-on-chip | |
| Cao et al. | LBDR: A load-balanced deadlock-free routing strategy for chiplet systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160831 Termination date: 20190121 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |