CN102508935A - On-chip network mapping method based on ant-colony chaos genetic algorithm - Google Patents

Abstract

Disclosed is an on-chip network mapping method based on the ant-colony chaos genetic algorithm. The standard ant-colony algorithm is basically used and the genetic algorithm is introduced in the on-chip network mapping method, parameters about each ant are coded by real numbers, codes of the ants are utilized as chromosome in the genetic algorithm, and algorithm parameters of coded ants are adjusted by the genetic algorithm in each iteration. During running of the algorithm, recycled results of each iteration of the algorithm are monitored, if the fact that the algorithm is trapped in a local optimum solution is monitored, mutation probability of the genetic algorithm is increased by a method of introducing a chaos model, and further, the parameters of the ant-colony algorithm are changed by means of the genetic algorithm. By the aid of the on-chip network mapping method, capability of the anti-colony chaos genetic algorithm for searching the solution space can be improved effectively, and trapping in the local optimum solution is avoided. In addition, the on-chip network mapping method has excellent practical values and wide application prospect for solution of massive on-chip network mapping.

Description

A kind of network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm

Technical field

The present invention relates to the network-on-chip mapping method, specifically a kind ofly can effectively improve the search efficiency of mapping algorithm, avoid algorithm to be absorbed in the mapping method of locally optimal solution, be a kind of network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm in solution space.

Background technology

Along with semiconductor process techniques is stepped into improving constantly of integrated level on nanometer stage and the sheet, the global wires time-delay rises to and is several times as much as the chip clock, and traditional architectures can't satisfy SOC(system on a chip) (System-on-a-Chip, SoC) communication requirement.Though the conventional single-core processor chips can improve performance through the method for further raising integrated level, yet its expense will be too huge, all kinds of relevant bottleneck problems also are difficult to all the more solve.Therefore the way that adopts a plurality of relative simple processor to replace single processor is suggested, and receives extensive concern.Further increase along with check figure on the sheet; Conventional architectures; Can't satisfy the communicating between multi-kernel demand like cross bar switch, stratification bus etc., for this reason to the network-on-chip of multi-core technology (Network-on-Chip, NoC) framework is suggested; And, become one of scheme that solves the tool potentiality of multinuclear SOC(system on a chip) communication issue already because it has incontrovertible advantage at aspects such as extensibility, reusabilities.

In the design process of NoC system, after basic communication frameworks such as system topological and routing policy were confirmed, the right matter of utmost importance of demand side was the network-on-chip mapping problems.The good mappings method can be so that entire system all makes moderate progress in the performance of each side such as energy consumption, time-delay, fault-tolerance, and therefore for many years, no matter great research enthusiasm is all revealed to this issue table in industry or educational circles.

Particularly; Mapping process is meant on the basis of the traffic between known NoC architecture and IP kernel; By someway each IP kernel being assigned among the NoC on each resource node; To realize the mutual corresponding process of application-specific and NoC architecture, the quality of mapping result then draws through the comparison object function.With two-dimensional mesh trellis NoC is example, and the NoC mapping process is as shown in Figure 1.

Then need at first provide two definition with mathematical way accurate description NoC mapping process.

Definition 1: (V is oriented acyclic weighted graph E) to given application characteristic figure G, and vertex v i ∈ V among the figure representes an IP kernel of carrying out particular task; Directed arc ei, j ∈ E, the correspondence between expression vertex v i and the vj, its weight wi, j representes the traffic between vi and the vj, bi, j then are communication bandwidth requirement between vi and the vj.

Definition 2: (R P) is digraph to given NoC architectural feature figure P.Summit ri ∈ R among the figure, a resource node among the expression NoC; Directed arc pi, j ∈ P representes that from the summit ri is to the path of rj; Manhatton distance between hi, j represent from ri to rj; Bi, j are path pi, the maximum communication bandwidth that j can provide.

Then the NoC mapping process is: given G and P, and the mapping function map () of searching G → P requires objective function to optimize as far as possible, and satisfies following constraint simultaneously:

$\∀v_i\∈V\⇒\mathrm{map}\left(v_i\right)\∈R---\left(1\right)$

$\∀v_i\≠v_j\⇔\mathrm{map}\left(v_i\right)\≠\mathrm{map}\left(v_j\right)---\left(2\right)$

size(G)≤size(P) (3)

$\∀b_{i,j}\≤B_{i,j}---\left(4\right)$

Wherein, formula (1), (2) are in order to guarantee the corresponding one by one of IP kernel and resource node, and formula (3), (4) then guarantee scale and bandwidth requirement.

Summary of the invention

The problem that the present invention will solve is: in order effectively to improve NoC entire system performance, need to strengthen the search capability of mapping algorithm.

Technical scheme of the present invention is: a kind of network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm; Through ant crowd Chaos Genetic Algorithm each IP kernel is assigned among the network-on-chip NoC on each resource node; The mapping of realization network-on-chip, said ant crowd Chaos Genetic Algorithm is: be the basis with the standard ant group algorithm, to the parameter employing real coding of every ant; And be encoded to the chromosome in the genetic algorithm with this; Each is taken turns and uses genetic algorithm that the ant group algorithm parameter of encoding is adjusted in the iteration at ant group algorithm, promptly through selection in the genetic algorithm and interlace operation chromosome is adjusted, and upgrades the parameter of ant group algorithm; Simultaneously; Each result who takes turns iteration is monitored, be absorbed in locally optimal solution, then introduce the mutation probability that chaotic model strengthens genetic algorithm if monitor ant group algorithm; And then revise the ant group algorithm parameter through genetic algorithm again; Optimum solution until ant group algorithm satisfies real system chip design needs, and promptly the power consumption of NoC is minimum with time-delay, and iteration finishes; Accomplish the network-on-chip mapping according to optimum solution, the said algorithm that monitors is absorbed in locally optimal solution and is meant that algorithm epicycle optimum solution and last round of optimum solution equate.

Concrete steps are:

1) initiation parameter and the completion initialization procedure to SOC(system on a chip) is set: the maximum cycle that ant group algorithm is set; The heuristic factor of information; Expect the heuristic factor; And the ant number is set, then every ant is positioned over reference position separately, according to the initial solution under the initiation parameter generation standard ant group algorithm of front;

2) structure iterative solution: in the process of construction solution, suppose that k ant is assigned to IP kernel Pi on the resource node Rj with probability at the t time circulation time:

$p_{i,j}^k\left(t\right)=\left\{\begin{array}{c}\frac{{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}}{\underset{j\&NotElement;{\mathrm{tabu}}_k}{\mathrm{\&Sigma;}}{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}},j\&NotElement;{\mathrm{tabu}}_{\mathrm{<figure-callout\; id="0"\; label="k"\; filenames="HDA0000093231960000031.png"\; state="\{\{state\}\}">k</figure-callout>}}\\ 0,j\&Element;{\mathrm{tabu}}_k\end{array}\right.---\left(5\right)$

Set tabu _k(k=1,2 ..., M), M refers to the sum of IP kernel, tabu _kBe used to write down the IP kernel that ant k had distributed, τ _{I, j}(t) be illustrated in circulation time the t time, IP kernel Pi is assigned to the pheromones intensity on the resource node Rj, η _{I, j}(t) be meant that IP kernel Pi is assigned to the heuristic information on the resource node Rj, α in the formula, β are respectively the heuristic factor of information and the heuristic factor of expectation, α, and β is provided with initial value by step 1) in the first time during iteration;

3) pheromones is upgraded: represent the lasting degree (0＜ρ＜1) of pheromones, Δ τ with parameter ρ _{I, j}Be the pheromones increment:

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k---\left(9\right)$

represent in this circulation ant k dispense path (quantity of information that stays on the Pi → Rj), computing formula is:

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k=\left\{\begin{array}{c}\frac{1}{\cos t\left(k\right)},\mathrm{map}\left(k\right)\mathrm{includes}(P_i\&RightArrow;R_j)\\ 0,\mathrm{else}\end{array}\right.---\left(10\right)$

In the formula (10); Cost (k) for ant k according to step 2) cost of the allocative decision that obtains, optimum solution is to have separating of minimum cost, the definition of said cost function is according to different target requirements and different; Make optimum solution maximum to the contribution of quantity of information; When to the maximized optimization problem of objective function, then the 1/cost (k) in the formula (10) is changed to cost (k), and this moment, optimum solution was to have separating of maximum cost; After all ants are accomplished once circulation, according to following formula to each dispense path (quantity of information on the Pi → Rj) is upgraded:

τ _i，j(t+1)＝ρ×τ _i，j(t)+Δτ _i，j (11)

4) judge optimum solution: being solved to of iterative solution during ant group algorithm once circulates:

41) from IP kernel set P, press probability

Select a unappropriated IP kernel Pi to be assigned on the Rj, and add this nuclear to tabu _kIn;

42) repeat the N step, all be assigned on the corresponding resource tabu up to all IP kernels _kFull;

After one time ant crowd iterative loop is accomplished, in all ants, select optimum solution, if optimum solution satisfies condition, iteration finishes, if do not satisfy, then carries out step 5) and gets into loop iteration next time;

5) upgrade the ant group algorithm parameter with genetic algorithm, and chaos algorithm avoids being absorbed in locally optimal solution: when separating of obtaining of ant group algorithm can not meet the demands, use genetic algorithm to upgrade the parameter of ant group algorithm; Use real number α; β, Q carries out gene code to each ant, and the chromosome of each ant is promptly used (α; β, Q) expression:

α＝x×α _f+(1-x)×α _m (12)

β＝x×β _f+(1-x)×β _m (13)

Q＝x×Q _f+(1-x)×Q _m (14)

Realize the survival of the fittest through the wheel disc probability, select two populations, their parameter (α; β Q) is hybridized in proportion, and then makes a variation according to the parameter of certain probability to them; The variation probability is here adjusted by chaotic model, and the result is preserved, repeat M/2 time after; The parameter of all ants all is updated, and gets into the iterative process of next round ant group algorithm then, promptly gets back to step 1);

The initial value of said hybridization scale-up factor x is made as 0.5, when detecting algorithm and be absorbed in local optimum, adopts chaotic model that the x value is adjusted:

${\mathrm{cx}}_n^{m+1}=4{\mathrm{cx}}_n^m(1-x_n^m)---\left(15\right)$

In the formula; N the Chaos Variable that m iteration of

expression obtains,

and

upgrades according to the hybridization scale-up factor x of formula (15) to genetic algorithm.

Parameter coding to every ant comprises: to the heuristic factor of information, expect that the heuristic factor and pheromones intensity encodes.

The present invention is the basis with the standard ant group algorithm; Introduce genetic algorithm simultaneously; Parameter to every ant adopts real coding, and is encoded to the chromosome in the genetic algorithm with this, takes turns at each and uses genetic algorithm that the ant group algorithm parameter of encoding is adjusted in the iteration.In the algorithm operational process; The present invention is also to each is taken turns the round-robin result and monitors in the algorithm; Be absorbed in locally optimal solution if monitor algorithm, then strengthen the mutation probability of genetic algorithm, and then revise the ant group algorithm parameter through genetic algorithm again through the method for introducing chaotic model.The present invention can effectively improve the search capability of algorithm to solution space, avoids it to stagnate in locally optimal solution, has found the solution good practical value and application prospects for extensive network-on-chip mapping problems.

Description of drawings

Fig. 1 is a NoC mapping process synoptic diagram.

Fig. 2 is an ant crowd Chaos Genetic Algorithm schematic flow sheet of the present invention.

Fig. 3 is a MPEG4 demoder task decomposing schematic representation.

Fig. 4 is a MPEG4 demoder different mappings synoptic diagram as a result.

Fig. 5 is that the present invention is to MPEG4 demoder mapping result synoptic diagram.

Fig. 6 mapping result that is the present invention and standard ant group algorithm under task image at random is synoptic diagram relatively.

Embodiment

Based on the ant group algorithm of heredity mainly is in order to solve the dependence of traditional ant group algorithm to initiation parameter, to hybridize through the parameter of selecting suitable population, thereby and constantly parameter is upgraded with certain probability variation.Be absorbed in local optimum for fear of algorithm, also introduce chaotic model, strengthen mutation probability.Main flow process of the present invention is as shown in Figure 2.

Initiation parameter is set and accomplishes initialization procedure system.Be specially maximum cycle is set, the heuristic factor of information is expected the heuristic factor, and the ant number is set, and then every ant is positioned over reference position separately, and then forms initial the separating of algorithm.

Next be the construction process that a new round is separated.In the process of construction solution, suppose that k ant is assigned to IP kernel i on the resource node j with probability

at the t time circulation time.

$p_{i,j}^k\left(t\right)=\left\{\begin{array}{c}\frac{{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}}{\underset{j\&NotElement;{\mathrm{tabu}}_k}{\mathrm{\&Sigma;}}{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}},j\&NotElement;{\mathrm{tabu}}_{\mathrm{<figure-callout\; id="0"\; label="k"\; filenames="HDA0000093231960000031.png"\; state="\{\{state\}\}">k</figure-callout>}}\\ 0,j\&Element;{\mathrm{tabu}}_k\end{array}\right.---\left(5\right)$

Tabu _k(k=1,2 ..., M) being used to write down the IP kernel that ant k had distributed, M refers to the sum of IP kernel.τ _{I, j}(t) be illustrated in circulation time the t time, IP kernel Pi is assigned to the pheromones intensity on the resource node Rj, the renewal of pheromones will be described in detail below.η _{I, j}(t) be meant that IP kernel Pi is assigned to the heuristic information on the resource node Rj.α in the formula, β are respectively the heuristic factor of information and the heuristic factor of expectation.α, β is provided with initial value by initialization system in the first time during iteration, after iteration of the every execution of ant group algorithm, is upgraded by genetic algorithm later on.

A good mapping is always hoped to be assigned to most important IP kernel on the strongest resource node of communication capacity.Therefore, the definition heuristic information is as follows:

η _i，j(t)＝VIP(i)/Comm(j) (6)

η _{I, j}(t) embody the resonable degree of IP kernel Pi being distributed to resource node Rj.The communication capacity of Comm (j) expression Rj in NoC, VIP (i) represents the significance level of Pi in application characteristic figure.

$\mathrm{Comm}\left(j\right)=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{h}_{i,j}---\left(7\right)$

$\mathrm{VIP}\left(i\right)=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_{i,j}+\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_{j,i}---\left(8\right)$

W in the formula _{I, j}And w _{J, i}Be the traffic, N is the node number, h _{I, j}The jumping figure (hop) of expression from node ri to node rj process.

Solution procedure is following:

1) from IP kernel set P, presses probability

Select a unappropriated IP kernel Pi to be assigned on the resource node Rj, and add this nuclear to tabu _kIn;

2) repeat the N step, all be assigned on the corresponding resource tabu up to all nuclear _kFull.

After one time ant crowd iterative loop is accomplished, in all ants, select optimum solution, get into loop iteration next time then.

Next be the pheromones renewal process.Along with program run, the pheromones that stayed in the past dies away, and in the process that every ant is selected, has new pheromones to add simultaneously again.Represent the lasting degree (0＜ρ＜1) of pheromones, Δ τ with parameter ρ _{I, j}Be the pheromones increment.

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k---\left(9\right)$

represent in this circulation ant k dispense path (quantity of information that stays on the Pj → Ri), computing formula is:

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k=\left\{\begin{array}{c}\frac{1}{\cos t\left(k\right)},\mathrm{map}\left(k\right)\mathrm{includes}(P_i\&RightArrow;R_j)\\ 0,\mathrm{else}\end{array}\right.---\left(10\right)$

In the formula, cost (k) is the cost of ant k according to the allocative decision of top step completion.Optimum solution has minimum cost, so maximum to the contribution of quantity of information.As for the definition of cost function, according to different target requirements, its definition also can be different.Certainly, if to the maximized optimization problem of objective function, then the 1/cost (k) in the formula (10) should be changed to cost (k), because this moment, optimum solution was to have separating of maximum cost, so should do this adjustment so that its contribution to quantity of information is maximum.After all ants were accomplished once circulation, (quantity of information on the Pi → Rj) was upgraded to each dispense path according to following formula.

τ _i，j(t+1)＝ρ×τ _i，j(t)+Δτ _i，j (11)

Be the step of upgrading ant group algorithm with genetic algorithm then; It also is one of step of the present invention's most critical of being different from traditional ant group algorithm; Parameter to every ant adopts real coding; And be encoded to the chromosome in the genetic algorithm with this; Each is taken turns and uses genetic algorithm that the ant group algorithm parameter of encoding is adjusted in the iteration at ant group algorithm, promptly through selection in the genetic algorithm and interlace operation chromosome is adjusted, and for parameter too sensitive issue is set to solve ant group algorithm itself.When separating of obtaining of ant group algorithm can not meet the demands, use genetic algorithm to upgrade the parameter of ant group algorithm, can guarantee the carrying out of ant group algorithm.For fear of the complicated coding and decoding process of genetic algorithm, use real number α here, β, Q carries out gene code to each ant.The chromosome of each ant is promptly used (α, β, Q) expression so.

α＝x×α _f+(1-x)×α _m (12)

β＝x×β _f+(1-x)×β _m (13)

Q＝x×Q _f+(1-x)×Q _m (14)

Particularly, realize selecting the superior and eliminating the inferior through wheel disc probability (Roulette).Select two suitable populations, (α, β Q) are hybridized (Crossover) according to certain ratio with their parameter.And then make a variation (Mutate) according to the parameter of certain probability to them, the result is preserved.After repeating M/2 time, the parameter of all ants all is updated, and gets into the iterative process of next round ant group algorithm then.

In formula (12)-(14), the initial value of hybridization scale-up factor x all is set to 0.5.Certainly,, when detecting algorithm and be absorbed in local optimum, will adopt chaotic model that the x value is adjusted according to following step along with the operation of program, particularly:

When this optimum solution equates with the optimum solution of last time, mean to be absorbed in local optimum, stagnate for fear of algorithm, utilize chaos that initial value is changed extremely responsive characteristic, it is as follows that coefficient is at this moment carried out the chaos operation.

${\mathrm{cx}}_n^{m+1}=4{\mathrm{cx}}_n^m(1-x_n^m)---\left(15\right)$

In the formula; N the Chaos Variable that m iteration of expression obtains,

and

upgrades according to the hybridization scale-up factor x of formula (15) to genetic algorithm.

Through specific embodiment enforcement of the present invention is described below.

Embodiment 1

Be checking the present invention, with algorithm application on the mapping problems of MPEG4 demoder.The MPEG4 demoder can be decomposed into 12 tasks, and is as shown in Figure 3, gives 12 IP kernels them then and carries out respectively, and how the mapping problems of this moment promptly gets on these 12 IP kernels NoC that to be placed into a scale be 3x4.

Before formal application the present invention addressed this problem, we needed the mathematic(al) representation of first objective definition function cost.

At first, system energy consumption is defined as:

$E\left(C\right)=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_{i,j}\&times;h_{i,j}---\left(16\right)$

W in the formula _{I, j}Be the traffic, N is the node number, h _{I, j}The jumping figure (hop) of expression from node ri to node rj process.For the simplified design model, This document assumes that h _{I, j}Be between two nodes manhatton distance (| x _Ri-x _Rj|+| y _Ri-y _Rj|).So the target of optimizing energy cost is to minimize each internodal weighted manhattan apart from sum, exactly that communication task is heavier several IP kernels are assigned to next-door neighbour's node, realize the communication neighbourization.

Next consider the time-delay influence, go out time-delay suc as formula shown in (17) from source node i to destination node j:

T _i，j＝(T _b+T _w)×h _i，j+T _b(B-1) (17)

In the formula, Tb for no when congested frame data pass through a switch and a needed time of link, packet header is in the switching node place average latency when congested in order to exist for Tw, B is the frame number that comprises in the packet.Wherein Tb and B are constant coefficient.Can find out that by (17) formula Ti, j depend on parameter Tw and h _{I, j}Under bandwidth constraint, the transmission delay Tw that reduces data can realize through alleviating data congestion, and the key of alleviate congestion just is the balance link load.The balance link load minimizes link load variance exactly.So optimal delay with link load variance (VAR (L)) as index:

$\mathrm{VAR}\left(L\right)=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{[\mathrm{Load}\left(l_i\right)-\mathrm{Load}\left(l_{\mathrm{avg}}\right)]}^2/M---\left(18\right)$

M is the link sum in the formula, and Load (li) is the charge capacity of link li, and Load (lavg) is the average link charge capacity.Obviously, communication delay optimization evenly distributes communication task exactly.

Can find out that from (16), (18) formula we hope optimization system energy consumption E (C) on the one hand, hope balance link supported V AR (L) on the other hand.In order to reach the purpose of combined optimization, our objective definition function definition is following:

cost＝λ×E(C)+(1-λ)VAR(L) (19)

In the formula, λ is a scale-up factor, is used for regulating communication energy consumption and the proportion of time-delay at cost function, and span is (0,1).When λ=1, optimize the communication energy consumption; Communication delay is optimized in λ=0 o'clock.Need need adjust the λ value through concrete in actual use, such as when the optimization of communication energy consumption is more important, the λ value (0.5,1]; When communication delay is more important, and the λ value [0,0.5), the concrete value of λ is adjusted according to actual needs.

Final test findings is as shown in Figure 4.As can beappreciated from fig. 4, when λ=0.5, the optimization of communication energy consumption obviously is weaker than the communication energy optimization of λ=0 o'clock, and the optimization of link load is weaker than the link load optimization of λ=1 o'clock.The optimization of λ=0.5 is to unite consideration communication energy consumption and link load balance.That is to say and under the situation that satisfies the requirement of communication energy consumption, to optimize load balance through the λ value is adjusted.Or satisfying under the situation of bandwidth requirement, optimize the communication energy consumption.Fig. 5 is the mapping result of λ=0.5 o'clock.Place at random and compare, on the communication energy consumption, reduced by 31%, optimized 56% on the link load.

Embodiment 2

For demonstrating fully advantage of the present invention, generated a series of task image at random, and stressed different optimization aim (λ=1,0.5,0) NoC has been carried out mapping optimization, it compares mapping result with the standard ant group algorithm then.In the present embodiment, (19) formula among the embodiment 1 is still adopted in the definition of objective function.Net result is as shown in Figure 6.Fig. 6 has shown λ=1,0.5, the comparing result of the communication energy consumption of 0 o'clock each mapping scheme and link load variance and standard ant group algorithm.Can find out that from figure the present invention obviously is superior to traditional ant group algorithm.λ=1 o'clock mapping scheme reduces by 11% than reference scheme cost, and λ=0.5 o'clock mapping scheme reduces by 4% than reference scheme, and λ=0 o'clock mapping scheme reduces by 1% than reference scheme.

The present invention can effectively improve the search capability of mapping algorithm to solution space, avoids it to be absorbed in locally optimal solution, to improving the network-on-a-chip overall performance, reduces communication overhead, reduces communication delay, and positive excellent application value is arranged.

Claims (3)

1. network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm; It is characterized in that each IP kernel being assigned among the network-on-chip NoC on each resource node through ant crowd Chaos Genetic Algorithm; The mapping of realization network-on-chip, said ant crowd Chaos Genetic Algorithm is: be the basis with the standard ant group algorithm, to the parameter employing real coding of every ant; And be encoded to the chromosome in the genetic algorithm with this; Each is taken turns and uses genetic algorithm that the ant group algorithm parameter of encoding is adjusted in the iteration at ant group algorithm, promptly through selection in the genetic algorithm and interlace operation chromosome is adjusted, and upgrades the parameter of ant group algorithm; Simultaneously; Each result who takes turns iteration is monitored, be absorbed in locally optimal solution, then introduce the mutation probability that chaotic model strengthens genetic algorithm if monitor ant group algorithm; And then revise the ant group algorithm parameter through genetic algorithm again; Optimum solution until ant group algorithm satisfies real system chip design needs, and promptly the power consumption of NoC is minimum with time-delay, and iteration finishes; Accomplish the network-on-chip mapping according to optimum solution, the said algorithm that monitors is absorbed in locally optimal solution and is meant that algorithm epicycle optimum solution and last round of optimum solution equate.

2. a kind of network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm according to claim 1 is characterized in that concrete steps are:

1) initiation parameter and the completion initialization procedure to SOC(system on a chip) is set: the maximum cycle that ant group algorithm is set; The heuristic factor of information; Expect the heuristic factor; And the ant number is set, then every ant is positioned over reference position separately, according to the initial solution under the initiation parameter generation standard ant group algorithm of front;

2) structure iterative solution: in the process of construction solution, suppose that k ant is assigned to IP kernel Pi on the resource node Rj with probability

at the t time circulation time:

$p_{i,j}^k\left(t\right)=\left\{\begin{array}{c}\frac{{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}}{\underset{j\&NotElement;{\mathrm{tabu}}_k}{\mathrm{\&Sigma;}}{\left[{\mathrm{\&tau;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&alpha;}}\&times;{\left[{\mathrm{\&eta;}}_{i,j}\left(t\right)\right]}^{\mathrm{\&beta;}}},j\&NotElement;{\mathrm{tabu}}_k\\ 0,j\&Element;{\mathrm{tabu}}_k\end{array}\right.---\left(5\right)$

Set tabu _k(k=1,2 ..., M), M refers to the sum of IP kernel, tabu _kBe used to write down the IP kernel that ant k had distributed, τ _{I, j}(t) be illustrated in circulation time the t time, IP kernel Pi is assigned to the pheromones intensity on the resource node Rj, η _{I, j}(t) be meant that IP kernel Pi is assigned to the heuristic information on the resource node Rj, α in the formula, β are respectively the heuristic factor of information and the heuristic factor of expectation, α, and β is provided with initial value by step 1) in the first time during iteration;

3) pheromones is upgraded: represent the lasting degree (0＜ρ＜1) of pheromones, Δ τ with parameter ρ _{I, j}Be the pheromones increment:

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k---\left(9\right)$

represent in this circulation ant k dispense path (quantity of information that stays on the Pi → Rj), computing formula is:

$\mathrm{\&Delta;}{\mathrm{\&tau;}}_{i,j}^k=\left\{\begin{array}{c}\frac{1}{\cos t\left(k\right)},\mathrm{map}\left(k\right)\mathrm{includes}(P_i\&RightArrow;R_j)\\ 0,\mathrm{else}\end{array}\right.---\left(10\right)$

In the formula (10); Cost (k) for ant k according to step 2) cost of the allocative decision that obtains, optimum solution is to have separating of minimum cost, the definition of said cost function is according to different target requirements and different; Make optimum solution maximum to the contribution of quantity of information; When to the maximized optimization problem of objective function, then the 1/cost (k) in the formula (10) is changed to cost (k), and this moment, optimum solution was to have separating of maximum cost; After all ants are accomplished once circulation, according to following formula to each dispense path (quantity of information on the Pi → Rj) is upgraded:

τ _i，j(t+1)＝ρ×τ _i，j(t)+Δτ _i，j (11)

4) judge optimum solution: being solved to of iterative solution during ant group algorithm once circulates:

41) from IP kernel set P, press probability

Select a unappropriated IP kernel Pi to be assigned on the Rj, and add this nuclear to tabu _kIn;

42) repeat the N step, all be assigned on the corresponding resource tabu up to all IP kernels _kFull;

After one time ant crowd iterative loop is accomplished, in all ants, select optimum solution, if optimum solution satisfies condition, iteration finishes, if do not satisfy, then carries out step 5) and gets into loop iteration next time;

5) upgrade the ant group algorithm parameter with genetic algorithm, and chaos algorithm avoids being absorbed in locally optimal solution: when separating of obtaining of ant group algorithm can not meet the demands, use genetic algorithm to upgrade the parameter of ant group algorithm; Use real number α; β, Q carries out gene code to each ant, and the chromosome of each ant is promptly used (α; β, Q) expression:

α＝x×α _f+(1-x)×α _m (12)

β＝x×β _f+(1-x)×β _m (13)

Q＝x×Q _f+(1-x)×Q _m (14)

Realize the survival of the fittest through the wheel disc probability, select two populations, their parameter (α; β Q) is hybridized in proportion, and then makes a variation according to the parameter of certain probability to them; The variation probability is here adjusted by chaotic model, and the result is preserved, repeat M/2 time after; The parameter of all ants all is updated, and gets into the iterative process of next round ant group algorithm then, promptly gets back to step 1);

The initial value of said hybridization scale-up factor x is made as 0.5, when detecting algorithm and be absorbed in local optimum, adopts chaotic model that the x value is adjusted:

${\mathrm{cx}}_n^{m+1}=4{\mathrm{cx}}_n^m(1-x_n^m)---\left(15\right)$

In the formula; N the Chaos Variable that m iteration of

expression obtains,

and

upgrades according to the hybridization scale-up factor x of formula (15) to genetic algorithm.

3. a kind of network-on-chip mapping method based on ant crowd Chaos Genetic Algorithm according to claim 1 and 2 is characterized in that the parameter coding of every ant is comprised: to the heuristic factor of information, expect that the heuristic factor and pheromones intensity encodes.

Images