CN103428804B - Find mapping scheme and network code location method between network-on-chip task and node - Google Patents

Abstract

The invention discloses mapping scheme and network code location method between a kind of searching network-on-chip task and node, it is applied in the mapping on wireless NoC for multicast, when the distribution of search optimal node and routing scheme, not only consider that clean culture task total power consumption and response time are as the preferential index of assessed form, ensure that the throughput of wherein multicast task is maximum and Internet Transmission that is multicast task is least in power-consuming simultaneously.Thus make the performance of this preferred conceptual design network out out of employing reach best, and and other is to the Measures compare finding the priority scheme mapped between task and node under multiple target function, and there is complexity, realize simple feature.

Description

Method for searching mapping scheme between network-on-chip task and node and network coding position

Technical Field

The invention belongs to the field of integrated circuit design, particularly relates to a method for searching a mapping scheme between tasks and nodes and a network coding position of a wireless network on chip, and aims at the task allocation of the wireless network on chip and the searching process of the network coding path and the coding position.

Background

In the information market, with the continuous expansion of the microchip technology scale, large-scale system on chip becomes possible. As a brand new communication architecture for designing a multi-core chip, the network on chip can well solve the problem of complex inter-chip global communication in the design of a system on chip. However, due to the two-dimensional metal connection of the network on chip, the network on chip has the problems of high delay and high energy consumption when multi-hop transmission is performed, and the reliability of the network on chip system cannot be guaranteed when the processor is scaled up. The wireless network-on-Chip (wireless network-on-Chip) replaces the wired connection with the wireless connection, inserts a faster wireless transmission channel between the network-on-Chip nodes, increases the transmission range of the network-on-Chip nodes, and reduces the hop count of the information transmission path in the multi-core system, so as to achieve the purposes of reducing the system delay and energy consumption and increasing the system throughput rate, and further improve the overall performance of the system.

Network coding is an information exchange technology combining coding and routing, and the core idea is to allow information processing at an intermediate node in a network to achieve the theoretical upper bound of information flow by improving coding efficiency. The network coding can improve the throughput rate and bandwidth utilization rate of the network, the network robustness and the network safety, and under the condition that the communication network has broadcasting, the application of the network coding is a theoretical upper bound which can reach the information flow rate proposed by the Shannon theorem. And the wireless transmitting node of the wireless network-on-chip system is usually provided with an on-chip antenna, which can well meet the requirement that the nodes can broadcast in the multicast transmission network.

By looking up new and widely gathering literature, we find that the methods for finding the mapping scheme between tasks and nodes on the network on chip have been disclosed in the following categories:

the method is characterized in that a NoC mapping based on an ant colony optimization algorithm, computer engineering and application 2005, 41(18): 7-10' is adopted to realize network-on-chip mapping by adopting an effective ant colony optimization algorithm, and communication power consumption of a system is reduced as much as possible while a task is mapped to a network-on-chip node. The objective function of the method is power consumption, and the method is roughly as follows:

according to the number of ants, tasks are circularly distributed to each unallocated node according to the probability (the probability is the probability of distributing a node to a certain task calculated in a mode), all the nodes are distributed to the corresponding tasks once in a circulation, then a two-exchange method is adopted to carry out local search, the optimal solution with the minimum power consumption is found out, the pheromone of the ant corresponding to the optimal solution is updated by the power consumption value pheromone attenuation degree of the optimal solution and the current pheromone value, and after the process is iterated for multiple times, the distribution scheme which enables the minimum power consumption is finally selected.

The mapping scheme obtained by the method has smaller communication power consumption, but does not consider the delay performance of the system.

Document "popsicle, li, tommy et al. NoC mapping method oriented to energy consumption and latency. e. 2008, 36 (5): 937-. The calculation formula of the objective function cost of the method is as follows:

cost=λ×E+(1-λ)×VAR

wherein, λ is a proportionality coefficient, E is energy consumption, and VAR is a link load variance. The method is substantially as follows:

according to the number of ants, tasks are circularly distributed to each unallocated node according to the probability (the probability is the probability of distributing a node to a certain task calculated in a mode), all the nodes are distributed to the corresponding tasks once in a circulation, then a two-exchange method is adopted to carry out local search, a solution with minimum power consumption is found out, the pheromone of the ant corresponding to the optimal solution is updated by the attenuation degree of the pheromone of the power consumption value of the optimal solution and the value of the current pheromone, and after the process is iterated for multiple times, a distribution scheme for enabling the objective function cost to be minimum is finally selected.

This method does not consider that communication power consumption and execution time are completely different physical quantities, and the obtained mapping scheme does not necessarily have excellent performance.

The document Tanglei, Shashi Kumar, Atwo-steping genetic algorithm for mapping task mapping stoetworkworkworkworkandworkkouchhichicithicharchitecture, proceedings software Euromosymposiumon Digital System Design2003, Antalya, Turkey: IEEE,2003: 180-. The method is substantially as follows:

(1) firstly, assuming that the time delays of all edges in a task flow graph are equal and are a constant, namely an average; functional units (e.g., processors, memory, etc.) are allocated to each task such that the overall latency of the system is minimized

(2) And (4) using the edge delay of the real data flow graph, further optimizing, and distributing the functional units to the nodes of the NoC.

The method divides the whole optimization stage into two steps, thereby reducing the complexity of calculation and shortening the calculation time, but the energy consumption and the time delay of the obtained mapping scheme can not be simultaneously minimum.

The literature YiouChen, JianhaoHu, GengshengChen, xiangling, "energy and delay-aware mapping for real-time digital processing system on network chip functions, LasVegas, USA,2010, pp.375-378. It is proposed herein that:

(1) the data flow diagram defined autonomously obtains complete link information by using the weight and the connection relation of the data flow diagram, and can objectively reflect performance parameters such as delay, power consumption and the like of a digital signal system before mapping.

(2) Selecting a basic genetic algorithm as a prototype of a mapping algorithm, and enabling the algorithm to adapt to the NoC mapping problem through some improvements so as to obtain better algorithm convergence characteristics; meanwhile, a power consumption model, a delay model and a multi-target model are established aiming at the mapping target, and compared with a traditional mathematical model of power consumption and delay, the method can more accurately represent the performance change of the system before and after mapping.

The mapping method is simple in principle, fewer in steps and good in application value. However, the mapping method can only be applied to applications with unicast tasks, and is still deficient for applications including multicast tasks because the related performance after mapping of multicast tasks is not considered.

Disclosure of Invention

The present invention is directed to overcome the above-mentioned drawbacks of the prior art, and provides a method for searching a mapping scheme between tasks and nodes and a network coding position of a wireless network-on-chip, which provides a method for searching for a wireless network-on-chip with multiple tasks and multiple nodes to minimize the total power consumption and response time of unicast tasks.

A further object of the present invention is to provide a method for finding a mapping scheme between tasks and nodes and a network coding location of a wireless network on chip, which ensures maximum throughput of multicast tasks and minimum network transmission power consumption of multicast tasks.

In order to achieve the purpose of the invention, the invention adopts the technical scheme that:

a method for searching a mapping scheme and a network coding position between a network-on-chip task and a node comprises the following steps:

1) randomly generating a scheme set comprising K allocation schemesAnd for each allocation scheme X = (X)₁,x₂,...,x_N) Setting a probability p of a cross_taskAnd probability of mutation q_taskAnd the number of iterations T of the task map_taskAnd let variable t record the number of times of task mapping iteration_task=0 and the number of solutions Z =1 obtained by initialization, and sets the maximum number of solutions Z_max；

2) Find a set of solutionsIn each scheme, the power consumption and the response time of the wireless network-on-chip unicast task are recorded, and the maximum power consumption and the maximum response time are recorded as a worst power consumption value and a worst response time value;

3) calculate the scheme groupThe size of the adaptability value reflects the priority of the scheme, the higher the value is, the priority is according to the total work of the unicast taskThe consumption and the response time are low;

4) group of schemesIs randomly divided intoSelecting the most superior schemes in each scheme group according to the priority relation determined in the step 3 to form a priority scheme group

5) Group of priority schemesThe schemes in (1) are paired randomly, and the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step 1_iInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

6) With the mutation probability q set in step 1_iChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

7) The sequence of the fitness value calculated in step 3 from small to large, from the group of solutionsIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

8) Group of schemesAndare combined into a scheme group

9) Set of calculation schemesThe wireless network-on-chip unicasts the total power consumption and the response time under each scheme. During calculation, firstly, judging whether a multicast task source and a destination node of each scheme in a scheme group are mapped to different wireless network-on-chip nodes, if so, directly assigning a worst power consumption value and a worst corresponding time value to the power consumption and response time value of the scheme, and if not, calculating the scheme groupThe wireless network-on-chip unicasts the total power consumption and the response time under each scheme.

10) Group of schemesAll 2K schemes are divided into a plurality of groups, each group represents a boundary set, and the groups are numbered, wherein the schemes in the groups with smaller numbers have higher priority than the schemes in the groups with larger numbers; the priority is preferably low in power consumption and response time.

11) Initializing a new recipe set without recipesThen adding the schemes in the subgroups into the subgroups divided from the 10 th step in sequence according to the descending order of the numbers of the subgroupsIn (1) up to scheme groupThe number of schemes in the scheme exceeds K, and then all the schemes of the group added finally are taken out.

12) 10) the specific method steps for determining the priority of the schemes in each group are as follows:

12-1 makes the distance value of each scheme in the subgroup 0, and the distance value indicates whether the scheme is closely related to other schemes.

12-2, arranging the schemes in a reverse order according to the power consumption of each scheme calculated in the step 9), wherein the scheme with lower power consumption is arranged in front of the scheme with higher power consumption,

12-3 calculating power consumption distances of the schemes, each power consumption distance of a scheme being a value obtained by subtracting a power consumption of a scheme ranked before from a power consumption of a scheme ranked after it,

12-4, arranging the schemes in a reverse order according to the response time sizes of the schemes calculated in the step 9), wherein the scheme with the smaller response time is arranged in front of the scheme with the larger response time,

12-5 calculating response time distances for the respective plans, the response time distance for each plan being a value obtained by subtracting the response time of the plan that is ranked before from the response time of the plan that is ranked after it,

12-6 add the response time distance and the power consumption distance of each solution to obtain a solution distance value,

12-7, the least power consuming scheme and the least response time scheme are ranked first, and then the remaining schemes are ranked according to distance, and are ranked behind the least power consuming scheme and the least response time scheme, and the scheme with the larger distance value is ranked before the scheme with the smaller distance value, so that the priority of the scheme ranked in the front is higher than that of the scheme ranked in the rear.

13) Adding the selected scheme into the scheme group according to the priority level in the group taken out from the step 12)Up to the scheme groupThe number of the schemes in the scheme is K.

14) Group of schemesIs randomly divided intoA scheme group, wherein L is the number of the schemes in each scheme group, and the scheme with the highest priority in each scheme group is selected according to the priority relation determined in the steps 10) and 12) to form a priority scheme group

15) Group of priority schemesThe schemes in (1) are randomly paired again, and the probability p of the cross set in the step 1) is used_taskInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

16) With the probability of variation q set in step 1)_taskChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

17) The set of solutions determined according to step 10)Priority order of the middle schemes, from scheme group in order of priority from high to lowIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

18. If t is_task<T_taskThen t is_task=t_task+1, then return to step 2), continue iteration; otherwise go to step 19);

19) calculate the scheme groupThe scheme J with the maximum adaptability value, namely simultaneously minimizing the power consumption and the response time_minAs a task mapping scheme between tasks and nodes of the wireless network on chip.

Wherein, the method for calculating the power consumption and the response time task of the unicast task of the wireless network on chip under each mapping scheme in the step 2) is as follows:

the response time in the task mapping is the longest path delay from input to output and comprises the processing delay of all tasks in the path and the transmission time of data exchange between the tasks.

According to an embodiment of the present invention, the method further comprises the steps of:

i. obtaining scheme J according to task mapping_minDetermining the mapping node position of the multicast task source task and the multicast task destination task on the wireless on-chip networkAndfrom X, according to the relation between multicast tasks in the task flow graph_sourceEliminating nodes of a wireless network on chip mapped by isolated multicast task source nodes, and leaving a multicast task source task with an association relation and a node set mapped by a target task thereof by using vectorsAndand (4) showing. The isolation means that the multicast source node does not have the same sink node with other multicast source nodes or the transmission path thereof does not have any conflict with other multicast transmissions, and the multicast source node is called as isolated. The association relationship refers to the situation that several multicast sources have the same sink and transmission paths cross each other, and these multicast sources and sinks are said to have the association relationship.

ii. calculating X_source' where each source node goes to X_des' all paths of each destination node in the system are recorded by using a matrix total _ path, wherein a = j × m, and path _ ID = [ b ]₁,b₂,...b_a]Recording the total number of paths from each source to each destination node;

with vector Scheme = (y)₁,y₂,...y_a) To indicate a scheme for coding path selection, y_kThe number of a path in the current coding path selection scheme is shown, and the path is the first path on the total _ pathRandomly generating a scheme group containing P distribution schemesAnd Scheme = (y) for each coding path₁,y₂,...y_a) Setting a probability p of a cross_codingAnd probability of mutation q_codingAnd the number of code mapping iterations T_codingAnd let a variable t record the number of iterations_coding=0；

Finding a set of protocolsThe throughput rate of the multicast task and the network transmission power consumption under each scheme;

v. calculating a scheme group by adopting a non-dominated sorting method of Deb (published by scientific publishing Co. 2007, introduced in 'Multi-object evolutionary Algorithm and application thereof' authored by Zhengjinhua)The size of the adaptability value reflects the priority of the scheme, the higher the value is, the higher the priority is, the priority is superior according to the high transmission throughput rate of the multicast task and the low network power consumption;

grouping protocolsIs randomly divided intoA scheme group, wherein N is the number of the schemes in each scheme group, and the most superior schemes in each scheme group are selected according to the priority relationship determined in the step v to form a priority scheme group

Grouping priority schemesThe schemes in (1) are paired randomly, and the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step i_codingInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

Using the mutation probability q set in step i_codingChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

From the scheme group, in descending order of the fitness value calculated in the step vIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

x. group of schemesAndare combined into a scheme group

Calculating a set of recipes according to the method of step ivThe multicast task throughput rate and the network power consumption under each scheme;

scheme groups were generated using the boundary set generation method introduced in the article AFastElitiston-dominated SortingGenetic Algorithm for Multi-objective optimization, NSGA-II, published by Deb in 2002 in the journal IEEETransactionsonEvolutionarComputationSevThe 2P schemes in the group are divided into a plurality of groups, each group represents a boundary set, and the groups are numbered, and the schemes in the groups with smaller numbers have higher priority than the schemes in the groups with larger numbers; the priority is preferably high in terms of high throughput and low network power consumption;

initializing a new recipe set without recipesThen, from the small groups divided by the xii step, adding the schemes in the small groups into the small groups in sequence according to the sequence of the numbers of the small groups from large to smallIn (1) up to scheme groupThe number of the schemes in the group exceeds P, and then all the schemes of the group added finally are taken out.

Determining the priority of each scheme in each group in the step xii by the following specific method:

xiv-1 makes the distance value of each scheme in the group 0, and the distance value indicates whether the scheme is closely related to other schemes;

xiv-2 carries out reverse order arrangement on the schemes according to the network power consumption of each scheme calculated in the step xi, wherein the scheme with lower power consumption is arranged in front of the scheme with higher power consumption;

xiv-3 calculating the network power consumption distance of each scheme, wherein the power consumption distance of each scheme is the value obtained by subtracting the power consumption of the scheme ranked in front of the scheme from the power consumption of the scheme ranked behind the scheme;

xiv-4, arranging the schemes according to the throughput rates of the schemes calculated in the step xi, wherein the scheme with the higher throughput rate is arranged in front of the scheme with the lower throughput rate;

xiv-5 calculating the throughput rate distance of each scheme, wherein the throughput rate distance of each scheme is the value obtained by subtracting the throughput rate of the scheme ranked in front of the scheme from the throughput rate of the scheme ranked in back of the scheme;

xiv-6, adding the throughput rate distance and the power consumption distance of each scheme to obtain a distance value of the scheme;

xiv-7 ranks the scheme with the maximum throughput rate and the scheme with the minimum network power consumption at the top, then sorts the rest schemes according to the distance, and ranks the schemes behind the scheme with the maximum throughput rate and the scheme with the minimum network power consumption, wherein the scheme with the larger distance value ranks at the top of the scheme with the smaller distance value, so that the priority of the scheme ranked at the top is higher than that of the scheme ranked at the bottom;

xv. adding the selected priority scheme into the scheme group from the group obtained in the step xiiiUp to the scheme groupThe number of the schemes is P;

grouping protocols xviIs randomly divided intoSelecting the most superior schemes in each scheme group according to the priority relationship determined by the xii step and the xiv step to form a priority scheme group V_t'；

Will prioritize scheme group V_t' the scheme is then randomly paired, with the probability p of crossover set in step iii_codingInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

xviii. using the probability of variation q set in step iii_codingChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

The set of protocols determined according to step xiiPriority order of the middle schemes, from scheme group in order of priority from high to lowIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing P schemes

xx. if t_coding<T_codingThen t is_coding=t_coding+1 and then returning to step iv, otherwise step xxi;

xxi. calculating a scheme group by adopting a non-dominated sorting method of Deb (introduced in 'Multi-objective evolutionary Algorithm and application thereof' published by scientific Press 2007, whose author is Zhengjinhua)The scheme with the maximum adaptability value, namely the scheme with the minimum network transmission power consumption and the maximum throughput rate, is used as a mapping scheme for the coding mapping of the wireless on-chip network;

xxii, judging whether the throughput rate of the coding mapping scheme obtained at the moment is the minimum value specified in the step iv and whether the power consumption is the maximum value specified in the step iv, if the throughput rate is not the minimum value and the power consumption is not the maximum value, recording the mapping scheme, and meanwhile, Z = Z +1, then entering the step xxiii, otherwise, returning to the step 2);

xxiii if Z is less than or equal to Z_maxReturning to the step 2), otherwise, entering the step xxiv;

xxiv. calculation of Z by Deb's non-dominated ranking method (published by scientific Press 2007, introduced in "Multi-object evolutionary Algorithm and its applications" by Zhengjinhua)_maxAnd selecting the scheme with the maximum adaptability value, namely the scheme with the maximum multicast task throughput rate and the minimum multicast task transmission power consumption as the final mapping selection scheme, and recording the number of the coding point and the path selected by coding mapping.

In the above step iv, the solution setThe method for the throughput rate of the multicast task and the network transmission power consumption under each scheme comprises the following steps:

the network transmission power consumption refers to power consumption caused by data transmission of a wireless routing node in a network, and comprises arbitration power consumption, coding power consumption and wireless channel transmission power consumption of the node; the relevant calculation parameters are referred to the values of the parameters mentioned in the article "design of Multi-channel WirelessNoCtoImprovenanone-ChipCommunicationCapacity", published by DanZHao, YiWang, JianLi and TakamaroKikkawa.

Calculating the throughput rate and the network transmission power consumption under each scheme, and firstly judging whether the path combination of the scheme meets the condition that the multicast task transmission can reach the maximum multicast rate; if the condition that the maximum multicast rate can be reached is met, obtaining the coding point of the path combination, and calculating the throughput rate and the network transmission power consumption of the mapping scheme; if the condition of the maximum multicast rate cannot be reached, the throughput rate of the multicast task of the mapping scheme is a minimum value, and the minimum value can be 0. The network power consumption of multicast task transmission of the mapping scheme is a maximum value.

The transmission throughput rate (measured by the amount of multicast information transmitted in the network per unit time) of the multicast task is calculated according to the following formula:

$R=\underset{s\&Element;S}{\mathrm{\&Sigma;}}\frac{f_s\&CenterDot;n_s}{T_M}$

where S is a set of multicast source nodes, S is one of the source nodes, f_sRepresenting source nodes s to s in a multicast graphTraffic of the individual destination nodes, n_sRepresents the number of destination nodes corresponding to the source node s, T_MIs the transmission time of the multicast task.

In the step ii, the calculation method of total _ path is related to a routing algorithm by using a Mesh topology structure adopted by the wireless NoC, and specifically includes:

ii-1X_source' with j source nodes, X_des' there are m destination nodes; b₀= 0; let the notation x =1, y =1; for a = j × m dimension array path _ ID = [ b =₁,b₂,...b_a]Assignment, specifically:

(1) let p =1, q =1;

(2) selection of X_source' the p-th source node (let its node number n)_s) And X_des' q-th destination node (let its node number n)_d) And the coordinates of the nodes on the Mesh topology are (x) according to the node numbers of the nodes_s,y_s) And (x)_d,y_d). Wherein, $x_s=n_s-1-\sqrt{M}\&CenterDot;\mathrm{floor}\left(\frac{n_s-1}{\sqrt{M}}\right),$ $y_s=\mathrm{floor}\left(\frac{n_s-1}{\sqrt{M}}\right),$ $x_d=n_d-1-\sqrt{M}\&CenterDot;\mathrm{floor}\left(\frac{n_d-1}{\sqrt{M}}\right),$ $y_d=\mathrm{floor}\left(\frac{n_d-1}{\sqrt{M}}\right),$ floor (·) is a floor rounding function;

(3) computing a slave node n_sTo node n_dIs that the number of shortest routing paths is $\mathrm{Max}\_\mathrm{path}\_\mathrm{num}=\frac{\left(\right|x_s-x_d|+|y_s-y_d\left|\right)!}{|x_s-x_d|!|y_s-y_d|!},$ The paths are numbered sequentially, the path number value is denoted by e, and each path is sequentially denoted byLet c = (q-1) · j + p, b_c= Max _ path _ num; then p = p + 1;

(4) if p > j, q = q +1, entering the step (5), otherwise, returning to the step (2);

(5) if q is greater than m, finishing assignment and entering step ii-2; otherwise, returning to the step (2) by letting p =1;

ii-2 initializationLine ofAll zero array total _ path of a column;

ii-3 selection of X_source' the X-th source node and X_des' let k = (y-1) · j + x;

ii-4 from Total _ PathLine start to lineLine, recording the total b from current source to current destination node_kNumbering each node on each path; initializing g =1, g' =1, e =1; the detailed steps are as follows:

I. selecting the e-th path, wherein r is the number of nodes on the path, the source node is the 1 st node on the path, the node connected with the source node on the path is the 2 nd node on the path, the rest nodes are analogized, and the destination node is the r-th node on the path;

II, assigning the node number at the g' th position in the current path to an element at the g th row and column position of the h row of the array total _ path; then g '= g' +1, g = g + 1;

if g' is less than or equal to r, returning to the step II, otherwise, entering the step IV;

iv, the total _ path line number h = h +1, the path number e = e +1, and go to step V;

v, if e is less than or equal to b_kEntering step I; otherwise, entering step ii-5;

ii-5 at this time, from X_source' the X-th source node to X_des' all shortest routing paths of the y-th destination node have been recorded; index x = x +1, go to step ii-6;

ii-6 if x > j, then ii-7 is entered. Otherwise, returning to ii-3;

ii-7 notation y = y +1, if y < = m, let x =1, return ii-3; otherwise, it indicates that the assignment to total _ path has been completed.

Compared with the prior art, the invention has the technical effects that:

aiming at the mapping of the multicast application on the wireless NoC, when searching the optimal node distribution and routing scheme, the invention not only considers the total power consumption and the response time of the unicast task as the priority indexes of the assessment scheme, but also ensures the maximum throughput rate of the multicast task and the minimum network transmission power consumption of the multicast task. Therefore, the performance of the network designed by the optimized scheme is the best, and compared with other methods for searching the priority scheme of mapping between tasks and nodes under a plurality of objective functions, the method has the characteristics of complexity and simplicity in realization, and the method of the invention can be obtained by simulation and has high convergence speed.

Drawings

Fig. 1 is a task diagram consisting of 7 tasks, which is enumerated by an embodiment of the present invention.

Fig. 2 illustrates an on-chip network consisting of 16 nodes according to an embodiment of the present invention.

In the figure, 1 is a unicast task node, 2 is traffic between tasks, 3 is a multicast source task node, 4 is a multicast destination task node, 5 is a processing node of a wireless NoC, and 6 is a routing node of the wireless NoC.

Detailed Description

The present invention will be described in further detail with reference to test examples and specific embodiments. It should be understood that the scope of the above-described subject matter is not limited to the following examples, and any techniques implemented based on the disclosure of the present invention are within the scope of the present invention.

The method of the invention is mainly oriented to the task flow diagram with multicast tasks. The task flow graph has two task nodes of unicast and multicast, data transmission between the unicast task nodes adopts a unicast transmission mode, and a routing algorithm of the unicast task nodes is determined by a topological structure adopted by the wireless NoC. The data transmission between the multicast task nodes adopts a broadcast transmission mode and a full path routing algorithm. The full-path routing algorithm means that when a routing path can select any forward link from a source node to a destination node (the forward link is the transmission direction from the source node to the destination node in the routing algorithm specified by the topology) and the mapping of the task flow graph on the wireless on-chip network is considered, the mapping performance of the unicast task and the mapping performance of the multicast task need to be considered. The invention provides a mapping method for searching the mapping between tasks and nodes when the total power consumption and the response time of a unicast task are simultaneously minimized and the throughput rate of a multicast task is maximum and the network transmission power consumption of the multicast task is minimum and selecting a corresponding multicast node routing path to construct a network code for a multi-task multi-node wireless network-on-chip. The mapping scheme obtained by the method has high performance, high convergence speed and low complexity.

Fig. 1 shows a task diagram with 7 tasks, and fig. 2 shows a network on chip with 16 nodes, which is used for explaining the present invention specifically based on this embodiment, and is not intended to limit the present invention. The present embodiment is to find a mapping scheme for distributing the 7 tasks shown in fig. 1 to the 16 nodes of the wireless network on chip shown in fig. 2 and designing a routing path, so that the total power consumption and the response time of the network on chip after mapping reach minimum values at the same time, and the maximum throughput of the multicast tasks and the minimum power consumption of the multicast task network are ensured. For convenience of description, we use vector X = (X)₁,x₂,x₃,x₄,x₅,x₆,x₇) One solution to represent the mapping of 7 tasks of a network on chip to 16 nodes, X at the nth position of the vector X_n(1≤x_n≦ 16) represents the number that the nth task maps to the node, i.e., x_nIndicates that the task with the number n is allocated to the x-th task_nThe node of (2).

The steps of finding the scheme which simultaneously minimizes the total power consumption and the response time and simultaneously ensures the maximum throughput rate of the multicast tasks and the minimum network power consumption of the multicast tasks are as follows:

the steps of task mapping are as follows:

1. randomly generating a scheme group comprising 20 allocation schemesEach allocation scheme in the scheme group is represented by a vector as: (11151071632),(810814101112),(12121014847),(166255412),(9811105615),(54113841),(1316916493),(48121511614),(4987377),(841035107),(10241341016),(101112101213),(2147314716),(2132613144),(5471413152),(121551012137),(7114101459),(14111015131310),(15169104316),(71711698). Setting the probability of intersection of each scheme as p_task=0.4, the probability of mutation is q_task=0.1, number of task mapping iterations is T_task=100, and let a variable t record the number of iterations_taskAnd =0. The number of schemes obtained by initialization is Z =1, and the maximum number of solutions is set to be Z_max=2。

2. Determining a set of recipesPower consumption and response time of wireless network on chip under each mapping scheme

2-1 calculation parameters for Power consumption reference is made to simulation parameters set forth in the paper "design Multi-channel WirelessNoCcto ImproveOn-ChipCommunication Capacity" published by DanZHao, YiWang, Jian Li and TakamaroKikkawa. A32 nm UWB connection is selected, the data rate is 20Gbps, and the power consumption per bit is 2.7 pJ. In this example, the total power consumption of each scheme is: 0.2444,0.2445,0.2445,0.2446,0.2445,0.2445,0.2445,0.2445,0.2445,0.2445,0.2445,0.2445,0.2444,0.2444,0.2446,0.2445,0.2445,0.2444,0.2445,0.2445.

The 2-2 response time is the longest path delay from input to output, and comprises the sum of the task processing time of all processing nodes on the path from the wireless input node to the output node and the link transmission time between the nodes. In this example, the response time of each scheme is: 0.1058,0.1067,0.1067,0.1072,0.1067,0.1066,0.1066,0.1064,0.1072,0.1062,0.1069,0.1067,0.1064,0.1064,0.1072,0.1069,0.1064,0.1061,0.1067,0.1069.

3. The scheme group is calculated by adopting a non-dominated sorting method of Deb (introduced in multi-objective evolutionary algorithm and application thereof published by scientific publishing agency in 2007 and written by Zhengjinhua)The fitness values of the schemes in (1) are respectively as follows: 2.0000,1.2632,1.0526,0.0526,1.1579,0.9474,0.6842,1.4211,0.2105,1.5789,0.6842,0.6842,1.7368,1.7368,0.0526,0.6842,1.4211,1.8947,0.3684,0.3684.

4. Group schemesRandomly dividing 20 schemes into 10 scheme groups, selecting the most superior scheme from each scheme group according to the priority determined in step 3, and combining the most superior scheme from all scheme groups into a new scheme groupNew set of solutionsThe scheme in (1) is respectively as follows: (11151071632),(12121014847),(9811105615),(48121511614),(841035107),(10241341016),(2147314716),(121551012137),(14111015131310),(15169104316).

5. Group schemesThe schemes in (1) are paired randomly, and the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step 1_task=0.4 interchanges the node numbers at random positions between each pair of schemes, and then all the interchanged pairs of schemes are combined to obtain a scheme groupScheme groupThe scheme in (1) is respectively as follows: (11151071632),(12121014847),(9811105615),(48121511614),(8410341016),(1024135107),(2147314716),(121551012137),(14111015131310),(15169104316)

6. Using the mutation probability q set in step 1_task=0.1 change scheme groupThe serial numbers of the nodes at random positions in each scheme in the scheme group are obtainedScheme groupThe scheme in (1) is respectively as follows: (11151071632),(12121014847),(9811105615),(48121511614),(8410341016),(1024135107),(2143314716),(121551012137),(13111015131310),(15169104316)

7. From scheme group to scheme group in the order of smaller fitness to larger fitnessSelect 10 schemes from the group of schemesAre combined together to form a new scheme group containing 20 schemesIs (11151071632), (12121014847), (9811105615), (48121511614), (8410341016), (1024135107), (2143314716), (121551012137), (13111015131310), (15169104316), (12121014847), (9811105615), (810814101112), (48121511614), (7114101459), (841)035107),(2147314716),(2132613144),(14111015131310),(11151071632)。

8. Group schemesAndare combined into a scheme group containing 40 schemesNamely (11151071632), (810814101112), (12121014847), (166255412), (9811105615), (54113841), (1316916493), (48121511614), (4987377), (841035107), (10241341016), (101112101213), (2147314716), (2132613144), (5471413152), (121551012137), (7114101459), (14111015131310), (15169104316), (71711698), (11151071632), (12121014847), (9811105615), (48121511614), (8410341016), (1024135107), (2143314716), (121551012137), (13111015131310), (15169104316), (12121014847), (9811105615), (810814101112), (48121511614), (7114101459), (841035107), (2147314716), (2132613144), (14111015131310), (11151071632).

9. Set of calculation schemesThe power consumption and the response time of the wireless network-on-chip under each scheme are judged, then whether the multicast task source and the destination node of each scheme in the scheme group are mapped to different wireless network-on-chip nodes is judged, if yes, the scheme group is usedUpdating the power consumption and response time value of the scheme by the worst power consumption value (0.2446) and the worst response time value (0.1072), wherein the total power consumption corresponding to each scheme is respectively as follows:

0.2444，0.2445，0.2445，0.2446，0.2445，0.2445，0.2445，0.2445，0.2445，0.2445，0.2445，0.2445，0.2444，0.2444，0.2446，0.2445，0.2445，0.2444，0.2445，0.2445，0.2444，0.2446，0.2446，0.2446，0.2446，0.2446，0.2446，0.2446，0.2446，0.2446，0.2445，0.2445，0.2445，0.2445，0.2445，0.2445，0.2444，0.2444，0.2444，0.2444，。

the response time of each mapping scheme is respectively as follows:

0.1058，0.1067，0.1067，0.1072，0.1067，0.1066，0.1066，0.1064，0.1072，0.1062，0.1069，0.1067，0.1064，0.1064，0.1072，0.1069，0.1064，0.1061，0.1067，0.1069，0.1058，0.1072，0.1072，0.1072，0.1072，0.1072，0.1072，0.1072，0.1072，0.1072，0.1067，0.1067，0.1067，0.1064，0.1064，0.1062，0.1064，0.1064，0.1061，0.1058。

10. the scheme groups were generated using the boundary set generation method introduced in the article AFastElitiston-dominated SortingGenetic Algorithm for Multi-objective optimization, NSGA-II, published by Deb in 2002 in the journal IEEETransactionsonEvolutionarycomputing volume sixThe 40 schemes in (1) are divided into 13 groups, each group represents a boundary set, and the schemes in the small group have higher priority than the schemes in the large group for group numbering; the priority is preferably set to reduce the total power consumption and the response time.

The schemes in panel 1 obtained by the above method are scheme 1, scheme 21, scheme 40; the schemes in the group 2 include scheme 18, scheme 39; among the protocols in panel 3 are protocol 13, protocol 14, protocol 37, protocol 38; the schemes in the group 4 include scheme 10, scheme 36; among the protocols in panel 5 are scheme 8, scheme 17, scheme 34, scheme 35; the protocols in panel 6 are scheme 2, scheme 33; the schemes in the group 7 include scheme 5, scheme 32; scheme 3 and scheme 31 are arranged in the group 8; the protocol in panel 9 is scheme 6; the protocols within panel 10 are scheme 7, scheme 11, scheme 16; the schemes within the group 11 are scheme 19, scheme 20; scheme 9 is arranged in the small group 12; the protocols in panel 13 are scheme 4, scheme 15, scheme 22, scheme 23, scheme 24, scheme 25, scheme 26, scheme 27, scheme 28, scheme 29, scheme 30.

11. Initializing a new recipe group without recipesThen, from the subgroups divided in the 10 th step, the schemes in the subgroups are added to the subgroups from subgroup 1 to subgroup 13 in the order of the numbers of the subgroups from small to largeIn (1) up to scheme groupThe number of schemes in the group exceeds 20, and the schemes in the group added last (namely the group 8) are all taken out.

12. And (3) determining the priority of each scheme in each group obtained in the step 10, wherein the specific method comprises the following steps:

12-1 makes the distance value of each scheme in the subgroup 0, and the distance value indicates whether the scheme is closely related to other schemes.

12-2, arranging the schemes in a reverse order according to the power consumption of each scheme calculated in the step 9, wherein the scheme with lower power consumption is arranged in front of the scheme with higher power consumption,

12-3 calculating power consumption distances of the schemes, each power consumption distance of a scheme being a value obtained by subtracting a power consumption of a scheme ranked before from a power consumption of a scheme ranked after it,

12-4, according to the response time of every scheme calculated in step 9, making reverse order arrangement of the schemes, the scheme whose response time is smaller is arranged in front of the scheme whose response time is larger,

12-5 calculating response time distances for the respective plans, the response time distance for each plan being a value obtained by subtracting the response time of the plan that is ranked before from the response time of the plan that is ranked after it,

12-6 add the response time distance and the power consumption distance of each solution to obtain a solution distance value,

12-7, the least power consuming scheme and the least response time scheme are ranked first, and then the remaining schemes are ranked according to distance, and are ranked behind the least power consuming scheme and the least response time scheme, and the scheme with the larger distance value is ranked before the scheme with the smaller distance value, so that the priority of the scheme ranked in the front is higher than that of the scheme ranked in the rear.

13. Adding schemes into the scheme group according to the selection of the priorities in the group sorted from the 12 th stepUp to the scheme groupThe number of the schemes in the scheme group is 20, and after the scheme group is completedThe scheme in (1) is as follows: (11151071632),(11151071632),(11151071632),(14111015131310),(14111015131310),(2147314716),(2132613144),(2147314716),(2132613144),(841035107),(841035107),(48121511614),(7114101459),(48121511614),(7114101459),(810814101112),(810814101112),(9811105615),(9811105615),(12121014847).

14. Group schemesRandomly dividing 20 schemes into 10 scheme groups, wherein each scheme group comprises 2 schemes, selecting the scheme with the highest priority in each scheme group according to the priority determined in the 10 th and the 12 th steps, and then selecting the scheme with the highest priority from all the scheme groupsA preferred scheme is formed into a new scheme groupScheme groupThe scheme in (1) comprises (11151071632), (11151071632), (14111015131310), (2132613144), (2132613144), (841035107), (7114101459), (7114101459), (810814101112), (9811105615)

15. Group schemesThe schemes in (1) are then randomly paired with the probability p of crossover set in step 1_task=0.4 interchanges the node numbers at random positions between each pair of schemes, and then all pairs are combined to obtain a scheme groupNamely: (11151071632),(11151071632),(14112613144),(2131015131310),(2132613144),(841035107),(7114101459),(7114101459),(810814101112),(9811105615)

16. With the probability of variation q set in step 1_task=0.1 change scheme groupThe serial numbers of the nodes at random positions in each scheme in the scheme group are obtainedNamely: (11151071632),(11151071632),(14112612144),(2121015131310),(2132613144),(841034107),(7114101459),(7114101459),(810814101112),(9911105715).

17. From scheme group to scheme group in the order of smaller fitness to larger fitnessSelect 10 schemes from the group of schemesAre combined together to form a new scheme group containing 20 schemesThe solutions are (11151071632), (11151071632), (14112612144), (2121015131310), (2132613144), (841034107), (7114101459), (7114101459), (810814101112), (9911105715), (48121511614), (7114101459), (810814101112), (810814101112), (9811105615), (9811105615), (12121014847), (11151071632), (11151071632), (11151071632).

18. If the number of iterations t_task<100, then t_task=t_task+1, then returning to the step 2 to perform a new round of iterative operation; otherwise, entering step 19;

19. the scheme group is calculated by adopting a non-dominated sorting method of Deb (introduced in multi-objective evolutionary algorithm and application thereof published by scientific publishing agency in 2007 and written by Zhengjinhua)The scheme with the minimum adaptability value, namely the scheme J with the minimum total power consumption and response time simultaneously_min(11, 15, 10,7, 16,3, 2) as a task-to-node task mapping scheme for the network-on-chip.

The steps of encoding the map are as follows:

i. obtaining scheme J according to task mapping_minDetermining the mapping node position X of the multicast task source task and the multicast task destination task on the wireless on-chip network_sourceIs (15, 10, 7), destination task maps position X_desIs (16, 3, 2). Based on the relation between multicast tasks in FIG. 1Except for the isolated multicast source node 7 and the corresponding destination node 2, a multicast task source and destination node set with an association relationship is left, and the multicast task source and destination node set comprises the following steps: x_source' = (10, 15) and X_des'=（3,16）。

ii. calculating X_source' where each source node goes to X_des' all paths of each destination node in the method are calculated as follows:

ii-1X_source' having 2 Source nodes, X_des' there are 2 destination nodes. b₀And =0. Let the notation x =1, y = 1. For a 4-dimensional array path _ ID = [ b =₁,b₂,...b₄]And assigning according to the method as follows:

(1) let p =1, q =1;

(2) selection of X_source' 1 st Source node (node number 10) and X_des' the 1 st destination node (node number 3) is found to have coordinates (1, 2) and (2, 0), respectively, from the node number.

(3) Calculating the number of the shortest routing paths from the source node to the destination node as Max _ path _ num =3, numbering the paths in sequence, and indicating the path number value as e, wherein each path is sequentially indicated asLet c =1, b₁And = 1. Then p = p + 1;

(4) if p > j, q = q +1, entering the step (5), otherwise, returning to the step (2);

(5) if q is greater than m, finishing assignment and entering step ii-2; otherwise let p =1 and go back to step (2).

ii-2 initialize the all zero array total _ path of 8 rows and 8 columns;

ii-3 selection of X_source' the 1 st Source node (node 10) and X_des' 1 st destination node (node 3), let k = 1.

ii-4 records the node numbers of the total 3 paths from the h =1 th line to the h' =3 th line of the total _ path from the current source to the current destination node. Initialization g =1, g' =1, e = 1. The detailed steps are as follows:

I. and selecting a 1 st path, wherein the number of nodes on the path is 4, the source node (node 10) is the 1 st node on the path, the node 11 is the 2 nd node on the path, the node 7 is the third node on the path, and the destination node (node 3) is the 4 th node on the path.

Assigning the number (10) of the 1 st node in the current path to the element at the 1 st row and 1 st column position of the array total _ path. Then g '= g' +1, g = g + 1;

if g' is less than or equal to 4, returning to the step II, otherwise, entering the step IV;

iv, the total _ path line number h = h +1, the path number e = e +1, and go to step V;

v, if e is less than or equal to 3, entering the step I; otherwise, entering step ii-5;

ii-5 at this time, from X_source' 1 st Source node (node 10) to X_des' all shortest route paths of the 1 st destination node (node 3) have been recorded as [10,11,7,3,0,0,0]. The index x = x + 1. Entering step ii-6;

ii-6 if x >2, then ii-7 is entered. Otherwise returning ii-3, ii-7 marking y = y +1, if y < =2, letting x =1, returning ii-3; otherwise, finishing the assignment of the total _ path to obtain a matrix total _ path as follows:

$\mathrm{total}\_\mathrm{path}=\left[\begin{array}{cccccccc}10& 11& 7& 3& 0& 0& 0& 0\\ 10& 6& 7& 3& 0& 0& 0& 0\\ 10& 6& 2& 3& 0& 0& 0& 0\\ 15& 11& 7& 3& 0& 0& 0& 0\\ 10& 11& 12& 16& 0& 0& 0& 0\\ 10& 11& 15& 16& 0& 0& 0& 0\\ 10& 14& 15& 16& 0& 0& 0& 0\\ 15& 16& 0& 0& 0& 0& 0& 0\end{array}\right]$

path _ ID is [3131 ].

Randomly generating a set of 20 allocation schemesRespectively as follows: (1321),(1131),(3131),(3121),(1111),(2111),(3121),(2131),(3111),(1111),(2121),(2121),(2121),(2111),(2131),(3111),(2131),(2121),(2131),(3111). For each braidCode path scheme setting cross probability p_coding=0.5 and mutation probability q_coding=0.1, and number of code mapping iterations T_coding=100, and let a variable t record the number of iterations_coding=0；

Set of judgment schemesIf the path combination under each scheme meets the condition that the multicast task transmission can reach the maximum multicast rate, the throughput rate is set to be 0, the power consumption is set to be 10, and the scheme group is obtainedThe throughput rate of the multicast task under the other various schemes is as follows: 0,0,0,0,0,0,0,0,26.7,0,0,0,0,0,0,26.7,0,0,0, 26.7.

The power consumption is: 10,10,10,10,10,10,10,10,0.0001,10,10,10,10,10,10,0.0001,10,10,10,0.0001.

v. calculating scheme group by adopting non-dominated sorting method of DebThe fitness value of each scheme in the method is as follows: 0.8421,0.8421,0.8421,0.8421,0.8421,0.8421,0.8421,0.8421,1.8947,0.8421,0.8421,0.8421,0.8421,0.8421,0.8421,1.8947,0.8421,0.8421,0.8421,1.8947.

Grouping protocolsOf the 20 schemes was randomly divided into 10 scheme subgroups, with 2 schemes in each scheme subgroup. Selecting the most superior schemes in each scheme group according to the priority relation determined in the step v to form a priority scheme groupComprises the following steps: (1321),(3131),(1111),(3121),(3111),(2121),(2121),(3111),(2131),(3111).

Grouping priority schemesThe schemes in (1) are paired randomly, wherein the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step iii_coding=0.5 interchanges the numbers of nodes at any same position on two schemes in each pair of schemes, and then combines all the interchanged pairs of schemes to obtain a scheme groupComprises the following steps: (1131),(3321),(1111),(3121),(3121),(2111),(2111),(3121),(2131),(3111).

Using the mutation probability q set in step i_coding=0.1 change scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme groupComprises the following steps: (1131),(3121),(1121),(3121),(3121),(2111),(2111),(3121),(3131),(3111).

From the scheme group, in descending order of the fitness value calculated in the step vSelect 10 schemes from the group of schemesAre combined together to form a new scheme group containing 20 schemesComprises the following steps: (1131),(3121),(1121),(3121),(3121),(2111),(2111),(3121),(3131),(3111),(2121),(2121),(2111),(2131),(2131),(2121),(2131),(3111),(3111),(3111).

x. group of schemesAndare combined into a scheme groupComprises the following steps: (1321),(1131),(3131),(3121),(1111),(2111),(3121),(2131),(3111),(1111),(2121),(2121),(2121),(2111),(2131),(3111),(2131),(2121),(2131),(3111),(1131),(3121),(1121),(3121),(3121),(2111),(2111),(3121),(3131),(2121),(2121),(2111),(2131),(2131),(2121),(2131),(0000),(0000),(0000),(0000).

xi. judgment scheme groupIf the path combination under each scheme meets the condition that the multicast task transmission can reach the maximum multicast rate, the throughput rate is set to be 0, the power consumption is set to be 10, and the scheme group is obtainedThe throughput rates of the multicast tasks under the other various schemes are 0,0,0,0,0,0,0,26.7,0,0,0,0,0,0, 0,0,0,0,0.

The network power consumption is respectively: 10,10,10,10,10,10, 10,0.0001,10,10,10,10,10,10, 10,10,10,10, 10;

scheme groups were generated using the boundary set generation method introduced in the article AFastElitiston-dominated SortingGenetic Algorithm for Multi-objective optimization, NSGA-II, published by Deb in 2002 in the journal IEEETransactionsonEvolutionarComputationSevThe 40 schemes in the group are divided into 2 subgroups, each subgroup represents a boundary set, and the subgroups are numbered, and the schemes in the subgroups with smaller numbers have higher priority than the schemes in the subgroups with larger numbers; the priority is preferably high in terms of high throughput and low network power consumption. The schemes in panel 1 obtained as described above are scheme 9, scheme 16, scheme 20; the remaining schemes are all the schemes in group 2.

Xiii initializes a new recipe set without recipesThen, from the group divided in the xii step, adding the schemes in the groups into the groups in sequence according to the sequence of the numbers of the groups from 1 to 2In (1) up to scheme groupThe number of schemes in the group exceeds 20, and then all the schemes in the group (group 2) added at last are taken out.

Determining the priority of each scheme in each group in the step xii by the following specific method:

xiv-1 gives each solution in the subgroup a distance value of 0, said distance value indicating whether the solution is closely related to other solutions.

xiv-2 carries out reverse order arrangement on the schemes according to the network power consumption of each scheme calculated in the step xi, the scheme with lower power consumption is arranged in front of the scheme with higher power consumption,

xiv-3 calculates the network power consumption distance of each solution, the power consumption distance of each solution being the value obtained by subtracting the power consumption of the solution arranged in front of it from the power consumption of the solution arranged behind it,

xiv-4 arranges the schemes according to the throughput rate of each scheme calculated in the step xi, the scheme with higher throughput rate is arranged in front of the scheme with lower throughput rate,

xiv-5 calculates the throughput distance of each scheme, the throughput distance of each scheme being the value obtained by subtracting the throughput of the scheme preceding it from the throughput of the scheme following it,

xiv-6 adds the throughput distance and the power consumption distance of each scheme to obtain the distance value of the scheme,

xiv-7 ranks the scheme with the highest throughput rate and the scheme with the lowest network power consumption at the top, and then sorts the remaining schemes according to the distance, and ranks the schemes with the highest throughput rate and the scheme with the lowest network power consumption at the back, wherein the scheme with the larger distance value ranks at the top of the scheme with the smaller distance value, so that the priority of the scheme ranked at the top is higher than that of the scheme ranked at the back.

xv. adding the scheme group according to the selection scheme of the priority level from the small group taken in the xiii stepUp toScheme groupThe number of schemes in the scheme is 20.Comprises the following steps: (3111),(3111),(3111),(1321),(2131),(1131),(3131),(3121),(1111),(2111),(3121),(2131),(1111),(2121),(2121),(2121),(2111),(2131),(2131),(2121).

Grouping protocols xviThe 20 schemes in the scheme group are randomly divided into 10 scheme groups, each scheme group comprises 2 schemes, and the most superior scheme in each scheme group is selected according to the priority relation determined in the step xii to form a priority scheme group V_t'。V_t' is (3111), (3111), (2131), (3131), (1111), (3121), (1111), (2121), (2111), (2131).

Will prioritize scheme group V_t' the scheme is then randomly paired, with the probability p of crossover set in step iii_coding=0.5 interchanges the numbers of nodes at any same position on two schemes in each pair of schemes, and then combines all the interchanged pairs of schemes to obtain a scheme groupComprises the following steps: (3111),(3111),(2131),(3131),(1111),(3121),(1121),(2111),(2131),(2111).

xviii. using the probability of variation q set in step iii_coding=0.1 change scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme groupComprises the following steps: (3111),(3111),(2131),(3131),(1111),(3121),(1121),(2111),(2131),(2111).

The set of protocols determined according to step xiiPriority order of the middle schemes, from scheme group in order of priority from high to lowSelect 10 schemes from the group of schemesAre combined together to form a new scheme group containing 20 schemesIs (3111), (3111), (2131), (3131), (1111), (3121), (1121), (2111), (2131), (2111), (2121), (2121), (2121), (2111), (2131), (2131), (2121), (3111), (3111), (3111).

xx. if t_coding<T_codingThen t is_coding=t_coding+1 and then returning to step iv, otherwise step xxi;

calculating a set of protocols according to the method described in step ivThe scheme with the minimum adaptability value is used as the scheme of the wireless network-on-chip coding mappingA mapping scheme. The scheme is (3111).

xxii. judge whether the throughput of the coding mapping scheme obtained at this time is not 0 and the power consumption is 10, since the throughput of the scheme is 26.7 and the power consumption is 0.0001, record the mapping scheme and Z = Z +1, and then proceed to step xxiii. Otherwise, return to step 2 of the task mapping part.

xxiii. if Z ≦ 2, return to step 2 of the task mapping section, otherwise, go to step xxiv.

xxiv. get 2 mapping schemes. Scheme 1 multicast task throughput rate is 26.7 and power consumption is 0.0001. Scheme 2 multicast task throughput is 20 and power consumption is 0.0006. Adopting a non-dominated sorting method of Deb (published by scientific publishing agency in 2007, introduced in 'multi-objective evolutionary algorithm and application thereof' authored in Zhengjinhua) to obtain the fitness of 1 and 0 respectively based on the multicast task throughput rate and the multicast task transmission power consumption in 2 mapping schemes, and selecting the scheme with the maximum fitness, namely the scheme with the maximum multicast task throughput rate and the minimum multicast task transmission power consumption, as the final mapping selection scheme. The resulting protocol is scheme 1: the task mapping scheme is (11151071632) and the coding mapping scheme is (3111), i.e. the selected path is $\left[\begin{array}{cccc}10& 6& 2& 3\\ 15& 11& 7& 3\\ 10& 11& 12& 16\\ 15& 16& 0& 0\end{array}\right]$

The corresponding encoding point is the routing node 11, i.e. the corresponding processing node PE 11.

While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the above embodiments, and various modifications or alterations can be made by those skilled in the art without departing from the spirit and scope of the claims of the present application.

Claims (5)

1. A method for searching a mapping scheme and a network coding position between a network-on-chip task and a node is characterized by comprising the following steps:

1) randomly generating a scheme set comprising K allocation schemesAnd for each allocation scheme X ═ X₁,x₂,...,x_N) Setting a probability p of a cross_taskAnd probability of mutation q_taskAnd task mapping iterationsNumber T_taskAnd let variable t record the number of times of task mapping iteration_taskThe maximum number of solutions Z is set to 0 and the number of solutions Z obtained by initialization is 1_max；

2) Find set of protocols J_tIn each scheme, the power consumption and the response time of the wireless network-on-chip unicast task are recorded, and the maximum power consumption and the maximum response time are recorded as a worst power consumption value and a worst response time value;

3) calculate scheme group J_tThe size of the adaptability value reflects the priority of the scheme, the higher the value is, the higher the priority is, the priority is superior according to the total power consumption and the response time of the unicast task;

4) scheme group J_tIs randomly divided intoSelecting the most superior schemes in each scheme group according to the priority relation determined in the step 3 to form a priority scheme group

5) Group of priority schemesThe schemes in (1) are paired randomly, and the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step 1_iInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

6) With the mutation probability q set in step 1_iChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

7) The sequence of the fitness values calculated in step 3 from small to large is from scheme group J_tIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

8) Group of schemesAndare combined into a scheme group

9) Set of calculation schemesIn the calculation, firstly, whether a multicast task source and a destination node of each scheme in a scheme group are mapped to different wireless network-on-chip nodes is judged, if yes, the worst power consumption value and the worst corresponding time value are directly assigned to the power consumption and response time value of the scheme, and if not, the scheme group is calculatedThe wireless network-on-chip unicasts the total power consumption and the response time under each scheme;

10) scheme group R_tAll 2K schemes are divided into a plurality of groups, each group represents a boundary set, and the groups are numbered, wherein the schemes in the groups with smaller numbers have higher priority than the schemes in the groups with larger numbers; the priority is preferably in terms of small power consumption and response time;

11) initializing a new recipe set without recipesThen adding the schemes in the subgroups into the subgroups divided from the 10 th step in sequence according to the descending order of the numbers of the subgroupsIn (1) up to scheme groupThe number of the schemes in the scheme exceeds K, and then all the schemes of the group added finally are taken out;

12) 10) the specific method steps for determining the priority of the schemes in each group are as follows:

12-1, enabling the distance value of each scheme in the group to be 0, wherein the distance value represents whether the scheme is closely related to other schemes or not;

12-2, arranging the schemes in a reverse order according to the power consumption of each scheme calculated in the step 9), wherein the scheme with lower power consumption is arranged in front of the scheme with higher power consumption;

12-3 calculating power consumption distances of the schemes, the power consumption distance of each scheme being a value obtained by subtracting the power consumption of the scheme ranked in front of the scheme from the power consumption of the scheme ranked behind the scheme;

12-4, arranging the schemes in a reverse order according to the response time of each scheme calculated in the step 9), wherein the scheme with shorter response time is arranged in front of the scheme with longer response time;

12-5 calculating response time distances for the respective plans, the response time distance for each plan being a value obtained by subtracting the response time of the plan that is ranked before from the response time of the plan that is ranked after the response time of the plan that is ranked before the response time of the plan that;

12-6, adding the response time distance and the power consumption distance of each scheme to obtain a distance value of the scheme;

12-7, arranging the scheme with the minimum power consumption and the scheme with the minimum response time at the top, and then sequencing the rest schemes according to the distance, wherein the scheme with the minimum power consumption and the scheme with the minimum response time are arranged at the back, the scheme with the larger distance value is arranged at the front of the scheme with the smaller distance value, and thus the priority of the scheme arranged at the front is higher than that of the scheme arranged at the back;

13) adding the selected scheme into the scheme group according to the priority level in the group taken out from the step 12)Up to the scheme groupThe number of the schemes is K;

14) group of schemesIs randomly divided intoA scheme group, wherein L is the number of the schemes in each scheme group, and the scheme with the highest priority in each scheme group is selected according to the priority relation determined in the steps 10) and 12) to form a priority scheme group

15) Group of priority schemesIn (1)Randomly pairing the schemes again, and setting the probability p of the cross in the step 1)_taskInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

16) With the probability of variation q set in step 1)_taskChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

17) The set of solutions determined according to step 10)Priority order of the middle schemes, from scheme group in order of priority from high to lowIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

18. If t is_task＜T_taskThen t is_task＝t_task+1, then return to step 2), continue iteration; otherwise go to step 19);

19) calculate the scheme groupThe scheme J with the maximum adaptability value, namely simultaneously minimizing the power consumption and the response time_minThe method is used as a task mapping scheme from the task of the wireless network on chip to the nodes;

wherein the ratio of K, L,are all positive integers.

2. The method for finding mapping schemes and network coding positions between network-on-chip tasks and nodes according to claim 1, wherein the method for calculating the power consumption and response time of the wireless network-on-chip unicast task under each scheme in step 2) is as follows: the power consumption of the wireless network-on-chip unicast task in the task mapping is the sum of the power consumption of all processing nodes and the power consumption of data transmission among the nodes; the response time in the task map is the longest path delay from input to output, and includes the processing delay of all tasks in the path and the transmission time of data exchange between the tasks.

3. The method for finding mapping schemes and network coding positions between network-on-chip tasks and nodes according to claim 1 or 2, further comprising the steps of:

i. obtaining scheme J according to task mapping_minDetermining the mapping node position of the multicast task source task and the multicast task destination task on the wireless on-chip networkAndfrom X, according to the relation between multicast tasks in the task flow graph_sourceElimination of isolated multicast task source node mappingsThe nodes of the network on the line are left with multicast task source tasks with the cross-linking relation and node sets mapped by the target tasks thereof respectively by using vectorsAndrepresents;

ii. calculating X_source' where each source node goes to X_des' all paths of each destination node in the network are recorded by using a matrix total _ path, wherein a is j × m, and path _ ID is [ b ]₁,b₂,...b_a]Recording the total number of paths from each source to each destination node;

using vector Scheme ═ y₁,y₂,...y_a) To represent one scheme of coding path selection; y is_kThe number of a path in the current coding path selection scheme is shown, and the path is the first path on the total _ pathA strip path; randomly generating a scheme group containing P distribution schemesAnd for each coding path Scheme (y)₁,y₂,...y_a) Setting a probability p of a cross_codingAnd probability of mutation q_codingAnd the number of code mapping iterations T_codingAnd let a variable t record the number of iterations_coding＝0；

Finding a set of protocolsThe throughput rate of the multicast task and the network transmission power consumption under each scheme;

v. calculate the set of plansThe size of the adaptability value reflects the priority of the scheme, the higher the value is, the higher the priority is, the priority is superior according to the high transmission throughput rate of the multicast task and the low network power consumption;

grouping protocolsIs randomly divided intoA scheme group, wherein N is the number of the schemes in each scheme group, and the most superior schemes in each scheme group are selected according to the priority relationship determined in the step v to form a priority scheme group

Grouping priority schemesThe schemes in (1) are paired randomly, and the pairing refers to grouping according to a group of two schemes and setting the probability p of intersection in step i_codingInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

Using the mutation probability q set in step i_codingChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

From the scheme group, in descending order of the fitness value calculated in the step vIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing K schemes

x. group of schemesAndare combined into a scheme group

Calculating a set of recipes according to the method of step ivThe multicast task throughput rate and the network power consumption under each scheme;

will scheme groupThe 2P schemes in the group are divided into a plurality of groups, each group represents a boundary set, and the groups are numbered, and the schemes in the groups with smaller numbers have higher priority than the schemes in the groups with larger numbers; the priority is preferably high in terms of high throughput and low network power consumption;

initializing a new noneScheme group of schemesThen, from the small groups divided by the xii step, adding the schemes in the small groups into the small groups in sequence according to the sequence of the numbers of the small groups from large to smallIn (1) up to scheme groupThe number of the schemes in the group exceeds P, and then all the schemes added in the group are taken out;

determining the priority of each scheme in each group in the step xii by the following specific method:

xiv-1 makes the distance value of each scheme in the group 0, and the distance value indicates whether the scheme is closely related to other schemes;

xiv-2 carries out reverse order arrangement on the schemes according to the network power consumption of each scheme calculated in the step xi, wherein the scheme with lower power consumption is arranged in front of the scheme with higher power consumption;

xiv-3 calculating the network power consumption distance of each scheme, wherein the power consumption distance of each scheme is the value obtained by subtracting the power consumption of the scheme ranked in front of the scheme from the power consumption of the scheme ranked behind the scheme;

xiv-4, arranging the schemes according to the throughput rates of the schemes calculated in the step xi, wherein the scheme with the higher throughput rate is arranged in front of the scheme with the lower throughput rate;

xiv-5 calculating the throughput rate distance of each scheme, wherein the throughput rate distance of each scheme is the value obtained by subtracting the throughput rate of the scheme ranked in front of the scheme from the throughput rate of the scheme ranked in back of the scheme;

xiv-6, adding the throughput rate distance and the power consumption distance of each scheme to obtain a distance value of the scheme;

xiv-7 ranks the scheme with the maximum throughput rate and the scheme with the minimum network power consumption at the top, then sorts the rest schemes according to the distance, and ranks the schemes behind the scheme with the maximum throughput rate and the scheme with the minimum network power consumption, wherein the scheme with the larger distance value ranks at the top of the scheme with the smaller distance value, so that the priority of the scheme ranked at the top is higher than that of the scheme ranked at the bottom;

xv. adding the selected priority scheme into the scheme group from the group obtained in the step xiiiUp to the scheme groupThe number of the schemes is P;

grouping protocols xviIs randomly divided intoSelecting the most superior schemes in each scheme group according to the priority relationship determined by the xii step and the xiv step to form a priority scheme group V_t'；

Will prioritize scheme group V_t' the scheme is then randomly paired, with the probability p of crossover set in step iii_codingInterchanging the serial numbers of the nodes at any same position on two schemes in each pair of schemes, and then combining all interchanged pairs of schemes to obtain a scheme group

xviii. using the probability of variation q set in step iii_codingChanging scheme groupThe serial number of the node at any position in each scheme in the scheme obtains a scheme group

The set of protocols determined according to step xiiPriority order of the middle schemes, from scheme group in order of priority from high to lowIn selectionIndividual schemes, and groups of schemesAre combined together to form a new scheme group containing P schemes

xx. if t_coding＜T_codingThen t is_coding＝t_coding+1 and then returning to step iv, otherwise step xxi;

calculate scheme groupThe scheme with the maximum adaptability value, namely the scheme with the minimum network transmission power consumption and the maximum throughput rate, is used as a mapping scheme for the coding mapping of the wireless on-chip network;

xxii, judging whether the throughput rate of the coding mapping scheme obtained at the moment is the minimum value specified in the step iv and whether the power consumption is the maximum value specified in the step iv, if the throughput rate is not the minimum value and the power consumption is not the maximum value, recording the mapping scheme, and meanwhile, Z is Z +1, and then entering the step xxiii; otherwise, returning to the step 2);

xxiii if Z is less than or equal to Z_maxThen, thenReturning to the step 2), otherwise, entering a step xxiv;

xxiv. calculating Z_maxSelecting the scheme with the maximum adaptability value, namely the scheme with the maximum multicast task throughput rate and the minimum multicast task transmission power consumption, as the final mapping selection scheme, and recording the path selected by coding mapping and the number of the coding point;

wherein the ratio of P, N,are all positive integers.

4. The method for finding mapping scheme and network coding position between network-on-chip task and node as claimed in claim 3, wherein in the step iv, the solution setThe method for the throughput rate of the multicast task and the network transmission power consumption under each scheme comprises the following steps:

calculating the throughput rate of the multicast task and the network transmission power consumption under each scheme, and firstly judging whether the path combination of the scheme meets the condition that the multicast task transmission can reach the maximum multicast rate; if the condition that the maximum multicast rate can be reached is met, obtaining the coding point of the path combination, and calculating the throughput rate of the multicast task and the network transmission power consumption of the mapping scheme; if the condition of the maximum multicast rate cannot be reached, the throughput rate of the multicast task of the mapping scheme is a minimum value, and the network power consumption of the multicast task transmission of the mapping scheme is a maximum value; the network transmission power consumption refers to power consumption caused by data transmission of a wireless routing node in a network, and comprises arbitration power consumption, coding power consumption and wireless channel transmission power consumption of the node;

the multicast task transmission throughput rate is calculated according to the following formula:

where S is a set of multicast source nodes, S is one of the source nodes, f_sRepresenting the traffic from a source node s to a single destination node in a multicast graph, n_sRepresents the number of destination nodes corresponding to the source node s, T_MIs the transmission time of the multicast task.

5. The method for finding a mapping scheme between network-on-chip tasks and nodes and a network coding position according to claim 4, wherein in the step ii, the total _ path calculation method is related to a routing algorithm by a Mesh topology structure adopted by the wireless NoC, and the specific steps are as follows:

ii-1X_source' with j source nodes, X_des' there are m destination nodes; b₀0; let the reference x be 1 and y be 1; for a dimension array of a ═ j × m, path _ ID ═ b₁,b₂,...b_a]Assignment, specifically:

(1) let p be 1 and q be 1;

(2) selection of X_source' the p-th source node in the node list is n_sAnd X_des'the q-th destination node of' is numbered n_dAnd the coordinates of the nodes on the Mesh topology are (x) according to the node numbers of the nodes_s,y_s) And (x)_d,y_d) (ii) a Wherein, floor (·) is a floor rounding function;

(3) computing a slave node n_sTo node n_dIs that the number of shortest routing paths isThe paths are numbered sequentially, the path number value is denoted by e, and each path is sequentially denoted byLet c be (q-1) · j + p, b_cMax _ path _ num; then p ═ p + 1;

(4) if p > j, q is q +1, and the process proceeds to step (5); otherwise, returning to the step (2);

(5) if q is greater than m, finishing assignment and entering step ii-2; otherwise, making p equal to 1, and returning to the step (2);

ii-2 initializationLine ofAll zero array total _ path of a column;

ii-3 selection of X_source' the X-th source node and X_desThe ith destination node of' let k ═ y-1 · j + x;

ii-4 from Total _ PathLine start to lineLine, recording the total b from current source to current destination node_kNumbering each node on each path; initializing g-1, g' -1, e-1; the detailed steps are as follows:

I. selecting the e-th path, wherein r is the number of nodes on the path, the source node is the 1 st node on the path, the node connected with the source node on the path is the 2 nd node on the path, the rest nodes are analogized, and the destination node is the r-th node on the path;

II, assigning the node number at the g' th position in the current path to an element at the g th row and column position of the h row of the array total _ path; then g' +1, g + 1;

if g' is less than or equal to r, returning to the step II, otherwise, entering the step IV;

iv, entering step V, if the total _ path number h is h +1, and the path number e is e + 1;

v, if e is less than or equal to b_kEntering step I; otherwise, entering step ii-5; (ii) a

ii-5 at this time, from X_source' the X-th source node to X_des' all shortest routing paths of the y-th destination node have been recorded; the index x ═ x +1, enter step ii-6;

ii-6 if x > j, entering ii-7, otherwise returning to ii-3;

ii-7 with the index y ═ y +1, if y < ═ m, let x ═ 1, return ii-3; otherwise, it indicates that the assignment to total _ path has been completed.